Hugging Face's logo

EmbeddedLLM
/
Inkling-Small-MXFP4

Image-Text-to-Text
Transformers
Safetensors
inkling_mm_model
amd-quark
mxfp4
rocm
vllm
inkling
conversational
8-bit precision
quark
Model card Files
xet
 Community
pstan commited on
Commit
a13dccd
·
verified ·
1 Parent(s): cd6295d

Add files using upload-large-folder tool

Browse files
Files changed (43) hide show
  1. .gitattributes +1 -0
  2. chat_template.jinja +129 -0
  3. config.json +700 -0
  4. model-00001-of-00032.safetensors +3 -0
  5. model-00002-of-00032.safetensors +3 -0
  6. model-00003-of-00032.safetensors +3 -0
  7. model-00004-of-00032.safetensors +3 -0
  8. model-00005-of-00032.safetensors +3 -0
  9. model-00006-of-00032.safetensors +3 -0
  10. model-00007-of-00032.safetensors +3 -0
  11. model-00008-of-00032.safetensors +3 -0
  12. model-00009-of-00032.safetensors +3 -0
  13. model-00010-of-00032.safetensors +3 -0
  14. model-00011-of-00032.safetensors +3 -0
  15. model-00012-of-00032.safetensors +3 -0
  16. model-00013-of-00032.safetensors +3 -0
  17. model-00014-of-00032.safetensors +3 -0
  18. model-00015-of-00032.safetensors +3 -0
  19. model-00016-of-00032.safetensors +3 -0
  20. model-00017-of-00032.safetensors +3 -0
  21. model-00018-of-00032.safetensors +3 -0
  22. model-00019-of-00032.safetensors +3 -0
  23. model-00020-of-00032.safetensors +3 -0
  24. model-00021-of-00032.safetensors +3 -0
  25. model-00022-of-00032.safetensors +3 -0
  26. model-00023-of-00032.safetensors +3 -0
  27. model-00024-of-00032.safetensors +3 -0
  28. model-00025-of-00032.safetensors +3 -0
  29. model-00026-of-00032.safetensors +3 -0
  30. model-00027-of-00032.safetensors +3 -0
  31. model-00028-of-00032.safetensors +3 -0
  32. model-00029-of-00032.safetensors +3 -0
  33. model-00030-of-00032.safetensors +3 -0
  34. model-00031-of-00032.safetensors +3 -0
  35. model-00032-of-00032.safetensors +3 -0
  36. model.safetensors.index.json +1133 -0
  37. mtp.safetensors +3 -0
  38. processor_config.json +46 -0
  39. quantize_quark.py +1025 -0
  40. special_tokens_map.json +22 -0
  41. tiktoken/tokenizer.model +3 -0
  42. tokenizer.json +3 -0
  43. tokenizer_config.json +508 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
chat_template.jinja ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set effort_map = {"none": 0.0, "minimal": 0.1, "low": 0.2, "medium": 0.7, "high": 0.9, "max": 0.99} -%}
2
+ {%- set role_token = {"user": "<|message_user|>", "assistant": "<|message_model|>", "system": "<|message_system|>", "tool": "<|message_tool|>"} -%}
3
+
4
+ {%- macro emit_thinking_effort() -%}
5
+ {%- set eff = reasoning_effort if reasoning_effort is defined and reasoning_effort is not none else 0.9 -%}
6
+ {%- if eff is string -%}
7
+ {%- set key = eff | trim -%}
8
+ {%- if key not in effort_map -%}
9
+ {{- raise_exception("Unknown reasoning_effort: " ~ eff) -}}
10
+ {%- endif -%}
11
+ {%- set num = effort_map[key] -%}
12
+ {%- else -%}
13
+ {%- set num = eff | float -%}
14
+ {%- endif -%}
15
+ {%- if num < 0.0 or num > 0.99 -%}
16
+ {{- raise_exception("reasoning_effort must be in [0.0, 0.99]") -}}
17
+ {%- endif -%}
18
+ {{- "<|message_system|><|content_text|>Thinking effort level: " -}}
19
+ {%- if num == 0.0 -%}0{%- else -%}{{ num }}{%- endif -%}
20
+ {{- "<|end_message|>" -}}
21
+ {%- endmacro -%}
22
+
23
+ {%- if tools -%}
24
+ {%- set tool_state = namespace(specs=[]) -%}
25
+ {%- for tool in tools -%}
26
+ {%- set fn = tool.function if tool.function is defined else tool -%}
27
+ {%- set spec = {
28
+ "description": (fn.description if fn.description is defined and fn.description else ""),
29
+ "name": fn.name,
30
+ "parameters": (fn.parameters if fn.parameters is defined and fn.parameters else {}),
31
+ "type": (tool.type if tool.type is defined and tool.type else "function"),
32
+ } -%}
33
+ {%- set tool_state.specs = tool_state.specs + [spec] -%}
34
+ {%- endfor -%}
35
+ {{- "<|message_system|>tool_declare<|content_xml|>" -}}
36
+ {{- tool_state.specs | tojson(sort_keys=true, separators=(",", ":")) -}}
37
+ {{- "<|end_message|>" -}}
38
+ {%- endif -%}
39
+
40
+ {%- set state = namespace(effort_emitted=false) -%}
41
+ {%- for message in messages -%}
42
+ {%- if message.role not in role_token -%}
43
+ {{- raise_exception("Unknown message role: " ~ message.role) -}}
44
+ {%- endif -%}
45
+ {%- if not state.effort_emitted and message.role != "system" -%}
46
+ {{- emit_thinking_effort() -}}
47
+ {%- set state.effort_emitted = true -%}
48
+ {%- endif -%}
49
+
50
+ {%- set rtok = role_token[message.role] -%}
51
+
52
+ {%- if message.role == "tool" -%}
53
+ {%- set tool_name_state = namespace(name="") -%}
54
+ {%- if message.name is defined and message.name -%}
55
+ {%- set tool_name_state.name = message.name -%}
56
+ {%- elif message.tool_call_id is defined and message.tool_call_id -%}
57
+ {%- for prev in messages -%}
58
+ {%- if prev.role == "assistant" and prev.tool_calls -%}
59
+ {%- for tc in prev.tool_calls -%}
60
+ {%- if tc.id is defined and tc.id == message.tool_call_id and tc.function.name is defined -%}
61
+ {%- set tool_name_state.name = tc.function.name -%}
62
+ {%- endif -%}
63
+ {%- endfor -%}
64
+ {%- endif -%}
65
+ {%- endfor -%}
66
+ {%- endif -%}
67
+ {{- rtok -}}
68
+ {%- if tool_name_state.name -%}{{- tool_name_state.name -}}{%- endif -%}
69
+ {{- "<|content_text|>" -}}
70
+ {%- if message.content is string -%}{{- message.content -}}{%- endif -%}
71
+ {{- "<|end_message|>" -}}
72
+
73
+ {%- else -%}
74
+ {%- if message.role == "assistant" and message.reasoning_content is defined and message.reasoning_content -%}
75
+ {{- "<|message_model|><|content_thinking|>" ~ message.reasoning_content ~ "<|end_message|>" -}}
76
+ {%- endif -%}
77
+
78
+ {%- if message.content is string -%}
79
+ {{- rtok ~ "<|content_text|>" ~ message.content ~ "<|end_message|>" -}}
80
+ {%- elif message.content -%}
81
+ {%- for part in message.content -%}
82
+ {%- if part is string -%}
83
+ {{- rtok ~ "<|content_text|>" ~ part ~ "<|end_message|>" -}}
84
+ {%- elif part.type is not defined or part.type in ("text", "input_text") -%}
85
+ {%- set text_part = (part.text if part.text is defined and part.text is string else "") -%}
86
+ {{- rtok ~ "<|content_text|>" ~ text_part ~ "<|end_message|>" -}}
87
+ {%- elif part.type in ("image", "input_image", "image_url") -%}
88
+ {{- rtok ~ "<|content_image|><|unused_200054|><|end_message|>" -}}
89
+ {%- elif part.type in ("audio", "input_audio", "audio_url") -%}
90
+ {{- rtok ~ "<|content_audio_input|><|unused_200053|><|audio_end|><|end_message|>" -}}
91
+ {%- else -%}
92
+ {{- raise_exception("Unsupported content part type: " ~ part.type) -}}
93
+ {%- endif -%}
94
+ {%- endfor -%}
95
+ {%- endif -%}
96
+
97
+ {%- if message.role == "assistant" and message.tool_calls -%}
98
+ {%- for tc in message.tool_calls -%}
99
+ {%- set fn = tc.function -%}
100
+ {%- if fn.name is not defined or fn.name is not string -%}
101
+ {{- raise_exception("tool call function name must be a string") -}}
102
+ {%- endif -%}
103
+ {%- set args = fn.arguments if fn.arguments is defined and fn.arguments else {} -%}
104
+ {%- if args is string -%}
105
+ {{- raise_exception("tool call arguments must be a parsed object, not a JSON string; canonicalize upstream") -}}
106
+ {%- endif -%}
107
+ {%- if args is not mapping -%}
108
+ {{- raise_exception("tool call arguments must be an object") -}}
109
+ {%- endif -%}
110
+ {{- "<|message_model|>" ~ fn.name ~ "<|content_invoke_tool_json|>" -}}
111
+ {{- '{"name":' ~ (fn.name | tojson(sort_keys=true, separators=(",", ":"))) ~ ',"args":' -}}
112
+ {{- (args | tojson(sort_keys=true, separators=(",", ":"))) -}}
113
+ {{- "}<|end_message|>" -}}
114
+ {%- endfor -%}
115
+ {%- endif -%}
116
+
117
+ {%- if message.role == "assistant" -%}
118
+ {{- "<|content_model_end_sampling|>" -}}
119
+ {%- endif -%}
120
+ {%- endif -%}
121
+ {%- endfor -%}
122
+
123
+ {%- if not state.effort_emitted -%}
124
+ {{- emit_thinking_effort() -}}
125
+ {%- endif -%}
126
+
127
+ {%- if add_generation_prompt -%}
128
+ {{- "<|message_model|>" -}}
129
+ {%- endif -%}
config.json ADDED
@@ -0,0 +1,700 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "InklingForConditionalGeneration"
4
+ ],
5
+ "model_type": "inkling_mm_model",
6
+ "eos_token_id": 200006,
7
+ "text_config": {
8
+ "model_max_length": 1048576,
9
+ "torch_dtype": "bfloat16",
10
+ "hidden_size": 4096,
11
+ "num_hidden_layers": 42,
12
+ "vocab_size": 201024,
13
+ "num_attention_heads": 32,
14
+ "num_key_value_heads": 8,
15
+ "head_dim": 128,
16
+ "d_rel": 16,
17
+ "rel_extent": 1024,
18
+ "q_bias": false,
19
+ "o_bias": false,
20
+ "log_scaling_n_floor": 128000,
21
+ "log_scaling_alpha": 0.1,
22
+ "rms_norm_eps": 1e-06,
23
+ "use_embed_norm": true,
24
+ "local_layer_ids": [
25
+ 0,
26
+ 1,
27
+ 2,
28
+ 3,
29
+ 4,
30
+ 6,
31
+ 7,
32
+ 8,
33
+ 9,
34
+ 10,
35
+ 12,
36
+ 13,
37
+ 14,
38
+ 15,
39
+ 16,
40
+ 18,
41
+ 19,
42
+ 20,
43
+ 21,
44
+ 22,
45
+ 24,
46
+ 25,
47
+ 26,
48
+ 27,
49
+ 28,
50
+ 30,
51
+ 31,
52
+ 32,
53
+ 33,
54
+ 34,
55
+ 36,
56
+ 37,
57
+ 38,
58
+ 39,
59
+ 40
60
+ ],
61
+ "dense_mlp_idx": 2,
62
+ "use_sconv": true,
63
+ "sconv_kernel_size": 4,
64
+ "unpadded_vocab_size": 200058,
65
+ "logits_mup_width_multiplier": 16.0,
66
+ "final_logit_softcapping": null,
67
+ "swa_head_dim": 128,
68
+ "swa_num_attention_heads": 32,
69
+ "swa_num_key_value_heads": 8,
70
+ "sliding_window_size": 512,
71
+ "n_routed_experts": 256,
72
+ "num_experts_per_tok": 6,
73
+ "n_shared_experts": 2,
74
+ "shared_expert_sink": true,
75
+ "dense_intermediate_size": 16384,
76
+ "intermediate_size": 2048,
77
+ "route_scale": 8.0,
78
+ "use_gate_bias": true,
79
+ "gate_activation": "sigmoid",
80
+ "norm_after_topk": true,
81
+ "use_global_scale": true
82
+ },
83
+ "audio_config": {
84
+ "decoder_dmodel": 4096,
85
+ "n_mel_bins": 80,
86
+ "mel_vocab_size": 16,
87
+ "bias": false,
88
+ "dmel_min_value": -7.0,
89
+ "dmel_max_value": 2.0,
90
+ "use_audio_norm": true,
91
+ "audio_mode": "dmel"
92
+ },
93
+ "vision_config": {
94
+ "vision_encoder_type": "hmlp",
95
+ "decoder_dmodel": 4096,
96
+ "patch_size": 40,
97
+ "temporal_patch_size": 2,
98
+ "n_channels": 3,
99
+ "n_layers": 4,
100
+ "use_vision_norm": true
101
+ },
102
+ "mtp_config": {
103
+ "num_nextn_predict_layers": 8,
104
+ "chain_hidden_post_norm": false,
105
+ "local_layer_ids": [
106
+ 0,
107
+ 2,
108
+ 4,
109
+ 5,
110
+ 6,
111
+ 7
112
+ ]
113
+ },
114
+ "quantization_config": {
115
+ "global_quant_config": {
116
+ "input_tensors": {
117
+ "dtype": "fp4",
118
+ "is_dynamic": true,
119
+ "qscheme": "per_group",
120
+ "ch_axis": -1,
121
+ "group_size": 32,
122
+ "block_size": null,
123
+ "symmetric": null,
124
+ "round_method": "half_even",
125
+ "scale_type": "float",
126
+ "scale_format": "e8m0",
127
+ "scale_calculation_mode": "even",
128
+ "mx_element_dtype": null,
129
+ "observer_cls": "PerBlockMXObserver",
130
+ "is_scale_quant": false,
131
+ "enable_buffer_reuse": false,
132
+ "max_input_numel": 4194304
133
+ },
134
+ "output_tensors": null,
135
+ "weight": {
136
+ "dtype": "fp4",
137
+ "is_dynamic": false,
138
+ "qscheme": "per_group",
139
+ "ch_axis": -1,
140
+ "group_size": 32,
141
+ "block_size": null,
142
+ "symmetric": null,
143
+ "round_method": "half_even",
144
+ "scale_type": "float",
145
+ "scale_format": "e8m0",
146
+ "scale_calculation_mode": "even",
147
+ "mx_element_dtype": null,
148
+ "observer_cls": "PerBlockMXObserver",
149
+ "is_scale_quant": false,
150
+ "enable_buffer_reuse": false,
151
+ "max_input_numel": 4194304
152
+ },
153
+ "bias": null,
154
+ "target_device": null
155
+ },
156
+ "exclude": [
157
+ "model.audio.encoder",
158
+ "model.llm.layers.0.attn.k_sconv",
159
+ "model.llm.layers.0.attn.v_sconv",
160
+ "model.llm.layers.0.attn.wk_dv",
161
+ "model.llm.layers.0.attn.wo_ud",
162
+ "model.llm.layers.0.attn.wq_du",
163
+ "model.llm.layers.0.attn.wr_du",
164
+ "model.llm.layers.0.attn.wv_dv",
165
+ "model.llm.layers.0.attn_sconv",
166
+ "model.llm.layers.0.mlp.w13_dn",
167
+ "model.llm.layers.0.mlp.w2_md",
168
+ "model.llm.layers.0.mlp_sconv",
169
+ "model.llm.layers.1.attn.k_sconv",
170
+ "model.llm.layers.1.attn.v_sconv",
171
+ "model.llm.layers.1.attn.wk_dv",
172
+ "model.llm.layers.1.attn.wo_ud",
173
+ "model.llm.layers.1.attn.wq_du",
174
+ "model.llm.layers.1.attn.wr_du",
175
+ "model.llm.layers.1.attn.wv_dv",
176
+ "model.llm.layers.1.attn_sconv",
177
+ "model.llm.layers.1.mlp.w13_dn",
178
+ "model.llm.layers.1.mlp.w2_md",
179
+ "model.llm.layers.1.mlp_sconv",
180
+ "model.llm.layers.10.attn.k_sconv",
181
+ "model.llm.layers.10.attn.v_sconv",
182
+ "model.llm.layers.10.attn.wk_dv",
183
+ "model.llm.layers.10.attn.wo_ud",
184
+ "model.llm.layers.10.attn.wq_du",
185
+ "model.llm.layers.10.attn.wr_du",
186
+ "model.llm.layers.10.attn.wv_dv",
187
+ "model.llm.layers.10.attn_sconv",
188
+ "model.llm.layers.10.mlp.gate",
189
+ "model.llm.layers.10.mlp_sconv",
190
+ "model.llm.layers.11.attn.k_sconv",
191
+ "model.llm.layers.11.attn.v_sconv",
192
+ "model.llm.layers.11.attn.wk_dv",
193
+ "model.llm.layers.11.attn.wo_ud",
194
+ "model.llm.layers.11.attn.wq_du",
195
+ "model.llm.layers.11.attn.wr_du",
196
+ "model.llm.layers.11.attn.wv_dv",
197
+ "model.llm.layers.11.attn_sconv",
198
+ "model.llm.layers.11.mlp.gate",
199
+ "model.llm.layers.11.mlp_sconv",
200
+ "model.llm.layers.12.attn.k_sconv",
201
+ "model.llm.layers.12.attn.v_sconv",
202
+ "model.llm.layers.12.attn.wk_dv",
203
+ "model.llm.layers.12.attn.wo_ud",
204
+ "model.llm.layers.12.attn.wq_du",
205
+ "model.llm.layers.12.attn.wr_du",
206
+ "model.llm.layers.12.attn.wv_dv",
207
+ "model.llm.layers.12.attn_sconv",
208
+ "model.llm.layers.12.mlp.gate",
209
+ "model.llm.layers.12.mlp_sconv",
210
+ "model.llm.layers.13.attn.k_sconv",
211
+ "model.llm.layers.13.attn.v_sconv",
212
+ "model.llm.layers.13.attn.wk_dv",
213
+ "model.llm.layers.13.attn.wo_ud",
214
+ "model.llm.layers.13.attn.wq_du",
215
+ "model.llm.layers.13.attn.wr_du",
216
+ "model.llm.layers.13.attn.wv_dv",
217
+ "model.llm.layers.13.attn_sconv",
218
+ "model.llm.layers.13.mlp.gate",
219
+ "model.llm.layers.13.mlp_sconv",
220
+ "model.llm.layers.14.attn.k_sconv",
221
+ "model.llm.layers.14.attn.v_sconv",
222
+ "model.llm.layers.14.attn.wk_dv",
223
+ "model.llm.layers.14.attn.wo_ud",
224
+ "model.llm.layers.14.attn.wq_du",
225
+ "model.llm.layers.14.attn.wr_du",
226
+ "model.llm.layers.14.attn.wv_dv",
227
+ "model.llm.layers.14.attn_sconv",
228
+ "model.llm.layers.14.mlp.gate",
229
+ "model.llm.layers.14.mlp_sconv",
230
+ "model.llm.layers.15.attn.k_sconv",
231
+ "model.llm.layers.15.attn.v_sconv",
232
+ "model.llm.layers.15.attn.wk_dv",
233
+ "model.llm.layers.15.attn.wo_ud",
234
+ "model.llm.layers.15.attn.wq_du",
235
+ "model.llm.layers.15.attn.wr_du",
236
+ "model.llm.layers.15.attn.wv_dv",
237
+ "model.llm.layers.15.attn_sconv",
238
+ "model.llm.layers.15.mlp.gate",
239
+ "model.llm.layers.15.mlp_sconv",
240
+ "model.llm.layers.16.attn.k_sconv",
241
+ "model.llm.layers.16.attn.v_sconv",
242
+ "model.llm.layers.16.attn.wk_dv",
243
+ "model.llm.layers.16.attn.wo_ud",
244
+ "model.llm.layers.16.attn.wq_du",
245
+ "model.llm.layers.16.attn.wr_du",
246
+ "model.llm.layers.16.attn.wv_dv",
247
+ "model.llm.layers.16.attn_sconv",
248
+ "model.llm.layers.16.mlp.gate",
249
+ "model.llm.layers.16.mlp_sconv",
250
+ "model.llm.layers.17.attn.k_sconv",
251
+ "model.llm.layers.17.attn.v_sconv",
252
+ "model.llm.layers.17.attn.wk_dv",
253
+ "model.llm.layers.17.attn.wo_ud",
254
+ "model.llm.layers.17.attn.wq_du",
255
+ "model.llm.layers.17.attn.wr_du",
256
+ "model.llm.layers.17.attn.wv_dv",
257
+ "model.llm.layers.17.attn_sconv",
258
+ "model.llm.layers.17.mlp.gate",
259
+ "model.llm.layers.17.mlp_sconv",
260
+ "model.llm.layers.18.attn.k_sconv",
261
+ "model.llm.layers.18.attn.v_sconv",
262
+ "model.llm.layers.18.attn.wk_dv",
263
+ "model.llm.layers.18.attn.wo_ud",
264
+ "model.llm.layers.18.attn.wq_du",
265
+ "model.llm.layers.18.attn.wr_du",
266
+ "model.llm.layers.18.attn.wv_dv",
267
+ "model.llm.layers.18.attn_sconv",
268
+ "model.llm.layers.18.mlp.gate",
269
+ "model.llm.layers.18.mlp_sconv",
270
+ "model.llm.layers.19.attn.k_sconv",
271
+ "model.llm.layers.19.attn.v_sconv",
272
+ "model.llm.layers.19.attn.wk_dv",
273
+ "model.llm.layers.19.attn.wo_ud",
274
+ "model.llm.layers.19.attn.wq_du",
275
+ "model.llm.layers.19.attn.wr_du",
276
+ "model.llm.layers.19.attn.wv_dv",
277
+ "model.llm.layers.19.attn_sconv",
278
+ "model.llm.layers.19.mlp.gate",
279
+ "model.llm.layers.19.mlp_sconv",
280
+ "model.llm.layers.2.attn.k_sconv",
281
+ "model.llm.layers.2.attn.v_sconv",
282
+ "model.llm.layers.2.attn.wk_dv",
283
+ "model.llm.layers.2.attn.wo_ud",
284
+ "model.llm.layers.2.attn.wq_du",
285
+ "model.llm.layers.2.attn.wr_du",
286
+ "model.llm.layers.2.attn.wv_dv",
287
+ "model.llm.layers.2.attn_sconv",
288
+ "model.llm.layers.2.mlp.experts.w13_weight",
289
+ "model.llm.layers.2.mlp.experts.w2_weight",
290
+ "model.llm.layers.2.mlp.gate",
291
+ "model.llm.layers.2.mlp_sconv",
292
+ "model.llm.layers.20.attn.k_sconv",
293
+ "model.llm.layers.20.attn.v_sconv",
294
+ "model.llm.layers.20.attn.wk_dv",
295
+ "model.llm.layers.20.attn.wo_ud",
296
+ "model.llm.layers.20.attn.wq_du",
297
+ "model.llm.layers.20.attn.wr_du",
298
+ "model.llm.layers.20.attn.wv_dv",
299
+ "model.llm.layers.20.attn_sconv",
300
+ "model.llm.layers.20.mlp.gate",
301
+ "model.llm.layers.20.mlp_sconv",
302
+ "model.llm.layers.21.attn.k_sconv",
303
+ "model.llm.layers.21.attn.v_sconv",
304
+ "model.llm.layers.21.attn.wk_dv",
305
+ "model.llm.layers.21.attn.wo_ud",
306
+ "model.llm.layers.21.attn.wq_du",
307
+ "model.llm.layers.21.attn.wr_du",
308
+ "model.llm.layers.21.attn.wv_dv",
309
+ "model.llm.layers.21.attn_sconv",
310
+ "model.llm.layers.21.mlp.gate",
311
+ "model.llm.layers.21.mlp_sconv",
312
+ "model.llm.layers.22.attn.k_sconv",
313
+ "model.llm.layers.22.attn.v_sconv",
314
+ "model.llm.layers.22.attn.wk_dv",
315
+ "model.llm.layers.22.attn.wo_ud",
316
+ "model.llm.layers.22.attn.wq_du",
317
+ "model.llm.layers.22.attn.wr_du",
318
+ "model.llm.layers.22.attn.wv_dv",
319
+ "model.llm.layers.22.attn_sconv",
320
+ "model.llm.layers.22.mlp.gate",
321
+ "model.llm.layers.22.mlp_sconv",
322
+ "model.llm.layers.23.attn.k_sconv",
323
+ "model.llm.layers.23.attn.v_sconv",
324
+ "model.llm.layers.23.attn.wk_dv",
325
+ "model.llm.layers.23.attn.wo_ud",
326
+ "model.llm.layers.23.attn.wq_du",
327
+ "model.llm.layers.23.attn.wr_du",
328
+ "model.llm.layers.23.attn.wv_dv",
329
+ "model.llm.layers.23.attn_sconv",
330
+ "model.llm.layers.23.mlp.gate",
331
+ "model.llm.layers.23.mlp_sconv",
332
+ "model.llm.layers.24.attn.k_sconv",
333
+ "model.llm.layers.24.attn.v_sconv",
334
+ "model.llm.layers.24.attn.wk_dv",
335
+ "model.llm.layers.24.attn.wo_ud",
336
+ "model.llm.layers.24.attn.wq_du",
337
+ "model.llm.layers.24.attn.wr_du",
338
+ "model.llm.layers.24.attn.wv_dv",
339
+ "model.llm.layers.24.attn_sconv",
340
+ "model.llm.layers.24.mlp.gate",
341
+ "model.llm.layers.24.mlp_sconv",
342
+ "model.llm.layers.25.attn.k_sconv",
343
+ "model.llm.layers.25.attn.v_sconv",
344
+ "model.llm.layers.25.attn.wk_dv",
345
+ "model.llm.layers.25.attn.wo_ud",
346
+ "model.llm.layers.25.attn.wq_du",
347
+ "model.llm.layers.25.attn.wr_du",
348
+ "model.llm.layers.25.attn.wv_dv",
349
+ "model.llm.layers.25.attn_sconv",
350
+ "model.llm.layers.25.mlp.gate",
351
+ "model.llm.layers.25.mlp_sconv",
352
+ "model.llm.layers.26.attn.k_sconv",
353
+ "model.llm.layers.26.attn.v_sconv",
354
+ "model.llm.layers.26.attn.wk_dv",
355
+ "model.llm.layers.26.attn.wo_ud",
356
+ "model.llm.layers.26.attn.wq_du",
357
+ "model.llm.layers.26.attn.wr_du",
358
+ "model.llm.layers.26.attn.wv_dv",
359
+ "model.llm.layers.26.attn_sconv",
360
+ "model.llm.layers.26.mlp.gate",
361
+ "model.llm.layers.26.mlp_sconv",
362
+ "model.llm.layers.27.attn.k_sconv",
363
+ "model.llm.layers.27.attn.v_sconv",
364
+ "model.llm.layers.27.attn.wk_dv",
365
+ "model.llm.layers.27.attn.wo_ud",
366
+ "model.llm.layers.27.attn.wq_du",
367
+ "model.llm.layers.27.attn.wr_du",
368
+ "model.llm.layers.27.attn.wv_dv",
369
+ "model.llm.layers.27.attn_sconv",
370
+ "model.llm.layers.27.mlp.gate",
371
+ "model.llm.layers.27.mlp_sconv",
372
+ "model.llm.layers.28.attn.k_sconv",
373
+ "model.llm.layers.28.attn.v_sconv",
374
+ "model.llm.layers.28.attn.wk_dv",
375
+ "model.llm.layers.28.attn.wo_ud",
376
+ "model.llm.layers.28.attn.wq_du",
377
+ "model.llm.layers.28.attn.wr_du",
378
+ "model.llm.layers.28.attn.wv_dv",
379
+ "model.llm.layers.28.attn_sconv",
380
+ "model.llm.layers.28.mlp.gate",
381
+ "model.llm.layers.28.mlp_sconv",
382
+ "model.llm.layers.29.attn.k_sconv",
383
+ "model.llm.layers.29.attn.v_sconv",
384
+ "model.llm.layers.29.attn.wk_dv",
385
+ "model.llm.layers.29.attn.wo_ud",
386
+ "model.llm.layers.29.attn.wq_du",
387
+ "model.llm.layers.29.attn.wr_du",
388
+ "model.llm.layers.29.attn.wv_dv",
389
+ "model.llm.layers.29.attn_sconv",
390
+ "model.llm.layers.29.mlp.gate",
391
+ "model.llm.layers.29.mlp_sconv",
392
+ "model.llm.layers.3.attn.k_sconv",
393
+ "model.llm.layers.3.attn.v_sconv",
394
+ "model.llm.layers.3.attn.wk_dv",
395
+ "model.llm.layers.3.attn.wo_ud",
396
+ "model.llm.layers.3.attn.wq_du",
397
+ "model.llm.layers.3.attn.wr_du",
398
+ "model.llm.layers.3.attn.wv_dv",
399
+ "model.llm.layers.3.attn_sconv",
400
+ "model.llm.layers.3.mlp.gate",
401
+ "model.llm.layers.3.mlp_sconv",
402
+ "model.llm.layers.30.attn.k_sconv",
403
+ "model.llm.layers.30.attn.v_sconv",
404
+ "model.llm.layers.30.attn.wk_dv",
405
+ "model.llm.layers.30.attn.wo_ud",
406
+ "model.llm.layers.30.attn.wq_du",
407
+ "model.llm.layers.30.attn.wr_du",
408
+ "model.llm.layers.30.attn.wv_dv",
409
+ "model.llm.layers.30.attn_sconv",
410
+ "model.llm.layers.30.mlp.gate",
411
+ "model.llm.layers.30.mlp_sconv",
412
+ "model.llm.layers.31.attn.k_sconv",
413
+ "model.llm.layers.31.attn.v_sconv",
414
+ "model.llm.layers.31.attn.wk_dv",
415
+ "model.llm.layers.31.attn.wo_ud",
416
+ "model.llm.layers.31.attn.wq_du",
417
+ "model.llm.layers.31.attn.wr_du",
418
+ "model.llm.layers.31.attn.wv_dv",
419
+ "model.llm.layers.31.attn_sconv",
420
+ "model.llm.layers.31.mlp.gate",
421
+ "model.llm.layers.31.mlp_sconv",
422
+ "model.llm.layers.32.attn.k_sconv",
423
+ "model.llm.layers.32.attn.v_sconv",
424
+ "model.llm.layers.32.attn.wk_dv",
425
+ "model.llm.layers.32.attn.wo_ud",
426
+ "model.llm.layers.32.attn.wq_du",
427
+ "model.llm.layers.32.attn.wr_du",
428
+ "model.llm.layers.32.attn.wv_dv",
429
+ "model.llm.layers.32.attn_sconv",
430
+ "model.llm.layers.32.mlp.gate",
431
+ "model.llm.layers.32.mlp_sconv",
432
+ "model.llm.layers.33.attn.k_sconv",
433
+ "model.llm.layers.33.attn.v_sconv",
434
+ "model.llm.layers.33.attn.wk_dv",
435
+ "model.llm.layers.33.attn.wo_ud",
436
+ "model.llm.layers.33.attn.wq_du",
437
+ "model.llm.layers.33.attn.wr_du",
438
+ "model.llm.layers.33.attn.wv_dv",
439
+ "model.llm.layers.33.attn_sconv",
440
+ "model.llm.layers.33.mlp.gate",
441
+ "model.llm.layers.33.mlp_sconv",
442
+ "model.llm.layers.34.attn.k_sconv",
443
+ "model.llm.layers.34.attn.v_sconv",
444
+ "model.llm.layers.34.attn.wk_dv",
445
+ "model.llm.layers.34.attn.wo_ud",
446
+ "model.llm.layers.34.attn.wq_du",
447
+ "model.llm.layers.34.attn.wr_du",
448
+ "model.llm.layers.34.attn.wv_dv",
449
+ "model.llm.layers.34.attn_sconv",
450
+ "model.llm.layers.34.mlp.gate",
451
+ "model.llm.layers.34.mlp_sconv",
452
+ "model.llm.layers.35.attn.k_sconv",
453
+ "model.llm.layers.35.attn.v_sconv",
454
+ "model.llm.layers.35.attn.wk_dv",
455
+ "model.llm.layers.35.attn.wo_ud",
456
+ "model.llm.layers.35.attn.wq_du",
457
+ "model.llm.layers.35.attn.wr_du",
458
+ "model.llm.layers.35.attn.wv_dv",
459
+ "model.llm.layers.35.attn_sconv",
460
+ "model.llm.layers.35.mlp.gate",
461
+ "model.llm.layers.35.mlp_sconv",
462
+ "model.llm.layers.36.attn.k_sconv",
463
+ "model.llm.layers.36.attn.v_sconv",
464
+ "model.llm.layers.36.attn.wk_dv",
465
+ "model.llm.layers.36.attn.wo_ud",
466
+ "model.llm.layers.36.attn.wq_du",
467
+ "model.llm.layers.36.attn.wr_du",
468
+ "model.llm.layers.36.attn.wv_dv",
469
+ "model.llm.layers.36.attn_sconv",
470
+ "model.llm.layers.36.mlp.gate",
471
+ "model.llm.layers.36.mlp_sconv",
472
+ "model.llm.layers.37.attn.k_sconv",
473
+ "model.llm.layers.37.attn.v_sconv",
474
+ "model.llm.layers.37.attn.wk_dv",
475
+ "model.llm.layers.37.attn.wo_ud",
476
+ "model.llm.layers.37.attn.wq_du",
477
+ "model.llm.layers.37.attn.wr_du",
478
+ "model.llm.layers.37.attn.wv_dv",
479
+ "model.llm.layers.37.attn_sconv",
480
+ "model.llm.layers.37.mlp.gate",
481
+ "model.llm.layers.37.mlp_sconv",
482
+ "model.llm.layers.38.attn.k_sconv",
483
+ "model.llm.layers.38.attn.v_sconv",
484
+ "model.llm.layers.38.attn.wk_dv",
485
+ "model.llm.layers.38.attn.wo_ud",
486
+ "model.llm.layers.38.attn.wq_du",
487
+ "model.llm.layers.38.attn.wr_du",
488
+ "model.llm.layers.38.attn.wv_dv",
489
+ "model.llm.layers.38.attn_sconv",
490
+ "model.llm.layers.38.mlp.gate",
491
+ "model.llm.layers.38.mlp_sconv",
492
+ "model.llm.layers.39.attn.k_sconv",
493
+ "model.llm.layers.39.attn.v_sconv",
494
+ "model.llm.layers.39.attn.wk_dv",
495
+ "model.llm.layers.39.attn.wo_ud",
496
+ "model.llm.layers.39.attn.wq_du",
497
+ "model.llm.layers.39.attn.wr_du",
498
+ "model.llm.layers.39.attn.wv_dv",
499
+ "model.llm.layers.39.attn_sconv",
500
+ "model.llm.layers.39.mlp.gate",
501
+ "model.llm.layers.39.mlp_sconv",
502
+ "model.llm.layers.4.attn.k_sconv",
503
+ "model.llm.layers.4.attn.v_sconv",
504
+ "model.llm.layers.4.attn.wk_dv",
505
+ "model.llm.layers.4.attn.wo_ud",
506
+ "model.llm.layers.4.attn.wq_du",
507
+ "model.llm.layers.4.attn.wr_du",
508
+ "model.llm.layers.4.attn.wv_dv",
509
+ "model.llm.layers.4.attn_sconv",
510
+ "model.llm.layers.4.mlp.gate",
511
+ "model.llm.layers.4.mlp_sconv",
512
+ "model.llm.layers.40.attn.k_sconv",
513
+ "model.llm.layers.40.attn.v_sconv",
514
+ "model.llm.layers.40.attn.wk_dv",
515
+ "model.llm.layers.40.attn.wo_ud",
516
+ "model.llm.layers.40.attn.wq_du",
517
+ "model.llm.layers.40.attn.wr_du",
518
+ "model.llm.layers.40.attn.wv_dv",
519
+ "model.llm.layers.40.attn_sconv",
520
+ "model.llm.layers.40.mlp.gate",
521
+ "model.llm.layers.40.mlp_sconv",
522
+ "model.llm.layers.41.attn.k_sconv",
523
+ "model.llm.layers.41.attn.v_sconv",
524
+ "model.llm.layers.41.attn.wk_dv",
525
+ "model.llm.layers.41.attn.wo_ud",
526
+ "model.llm.layers.41.attn.wq_du",
527
+ "model.llm.layers.41.attn.wr_du",
528
+ "model.llm.layers.41.attn.wv_dv",
529
+ "model.llm.layers.41.attn_sconv",
530
+ "model.llm.layers.41.mlp.gate",
531
+ "model.llm.layers.41.mlp_sconv",
532
+ "model.llm.layers.5.attn.k_sconv",
533
+ "model.llm.layers.5.attn.v_sconv",
534
+ "model.llm.layers.5.attn.wk_dv",
535
+ "model.llm.layers.5.attn.wo_ud",
536
+ "model.llm.layers.5.attn.wq_du",
537
+ "model.llm.layers.5.attn.wr_du",
538
+ "model.llm.layers.5.attn.wv_dv",
539
+ "model.llm.layers.5.attn_sconv",
540
+ "model.llm.layers.5.mlp.gate",
541
+ "model.llm.layers.5.mlp_sconv",
542
+ "model.llm.layers.6.attn.k_sconv",
543
+ "model.llm.layers.6.attn.v_sconv",
544
+ "model.llm.layers.6.attn.wk_dv",
545
+ "model.llm.layers.6.attn.wo_ud",
546
+ "model.llm.layers.6.attn.wq_du",
547
+ "model.llm.layers.6.attn.wr_du",
548
+ "model.llm.layers.6.attn.wv_dv",
549
+ "model.llm.layers.6.attn_sconv",
550
+ "model.llm.layers.6.mlp.gate",
551
+ "model.llm.layers.6.mlp_sconv",
552
+ "model.llm.layers.7.attn.k_sconv",
553
+ "model.llm.layers.7.attn.v_sconv",
554
+ "model.llm.layers.7.attn.wk_dv",
555
+ "model.llm.layers.7.attn.wo_ud",
556
+ "model.llm.layers.7.attn.wq_du",
557
+ "model.llm.layers.7.attn.wr_du",
558
+ "model.llm.layers.7.attn.wv_dv",
559
+ "model.llm.layers.7.attn_sconv",
560
+ "model.llm.layers.7.mlp.gate",
561
+ "model.llm.layers.7.mlp_sconv",
562
+ "model.llm.layers.8.attn.k_sconv",
563
+ "model.llm.layers.8.attn.v_sconv",
564
+ "model.llm.layers.8.attn.wk_dv",
565
+ "model.llm.layers.8.attn.wo_ud",
566
+ "model.llm.layers.8.attn.wq_du",
567
+ "model.llm.layers.8.attn.wr_du",
568
+ "model.llm.layers.8.attn.wv_dv",
569
+ "model.llm.layers.8.attn_sconv",
570
+ "model.llm.layers.8.mlp.gate",
571
+ "model.llm.layers.8.mlp_sconv",
572
+ "model.llm.layers.9.attn.k_sconv",
573
+ "model.llm.layers.9.attn.v_sconv",
574
+ "model.llm.layers.9.attn.wk_dv",
575
+ "model.llm.layers.9.attn.wo_ud",
576
+ "model.llm.layers.9.attn.wq_du",
577
+ "model.llm.layers.9.attn.wr_du",
578
+ "model.llm.layers.9.attn.wv_dv",
579
+ "model.llm.layers.9.attn_sconv",
580
+ "model.llm.layers.9.mlp.gate",
581
+ "model.llm.layers.9.mlp_sconv",
582
+ "model.mtp.layers.0.input_proj",
583
+ "model.mtp.layers.0.transformer_block.attn.k_sconv",
584
+ "model.mtp.layers.0.transformer_block.attn.v_sconv",
585
+ "model.mtp.layers.0.transformer_block.attn.wk_dv",
586
+ "model.mtp.layers.0.transformer_block.attn.wo_ud",
587
+ "model.mtp.layers.0.transformer_block.attn.wq_du",
588
+ "model.mtp.layers.0.transformer_block.attn.wr_du",
589
+ "model.mtp.layers.0.transformer_block.attn.wv_dv",
590
+ "model.mtp.layers.0.transformer_block.attn_sconv",
591
+ "model.mtp.layers.0.transformer_block.mlp.w13_dn",
592
+ "model.mtp.layers.0.transformer_block.mlp.w2_md",
593
+ "model.mtp.layers.0.transformer_block.mlp_sconv",
594
+ "model.mtp.layers.1.input_proj",
595
+ "model.mtp.layers.1.transformer_block.attn.k_sconv",
596
+ "model.mtp.layers.1.transformer_block.attn.v_sconv",
597
+ "model.mtp.layers.1.transformer_block.attn.wk_dv",
598
+ "model.mtp.layers.1.transformer_block.attn.wo_ud",
599
+ "model.mtp.layers.1.transformer_block.attn.wq_du",
600
+ "model.mtp.layers.1.transformer_block.attn.wr_du",
601
+ "model.mtp.layers.1.transformer_block.attn.wv_dv",
602
+ "model.mtp.layers.1.transformer_block.attn_sconv",
603
+ "model.mtp.layers.1.transformer_block.mlp.w13_dn",
604
+ "model.mtp.layers.1.transformer_block.mlp.w2_md",
605
+ "model.mtp.layers.1.transformer_block.mlp_sconv",
606
+ "model.mtp.layers.2.input_proj",
607
+ "model.mtp.layers.2.transformer_block.attn.k_sconv",
608
+ "model.mtp.layers.2.transformer_block.attn.v_sconv",
609
+ "model.mtp.layers.2.transformer_block.attn.wk_dv",
610
+ "model.mtp.layers.2.transformer_block.attn.wo_ud",
611
+ "model.mtp.layers.2.transformer_block.attn.wq_du",
612
+ "model.mtp.layers.2.transformer_block.attn.wr_du",
613
+ "model.mtp.layers.2.transformer_block.attn.wv_dv",
614
+ "model.mtp.layers.2.transformer_block.attn_sconv",
615
+ "model.mtp.layers.2.transformer_block.mlp.w13_dn",
616
+ "model.mtp.layers.2.transformer_block.mlp.w2_md",
617
+ "model.mtp.layers.2.transformer_block.mlp_sconv",
618
+ "model.mtp.layers.3.input_proj",
619
+ "model.mtp.layers.3.transformer_block.attn.k_sconv",
620
+ "model.mtp.layers.3.transformer_block.attn.v_sconv",
621
+ "model.mtp.layers.3.transformer_block.attn.wk_dv",
622
+ "model.mtp.layers.3.transformer_block.attn.wo_ud",
623
+ "model.mtp.layers.3.transformer_block.attn.wq_du",
624
+ "model.mtp.layers.3.transformer_block.attn.wr_du",
625
+ "model.mtp.layers.3.transformer_block.attn.wv_dv",
626
+ "model.mtp.layers.3.transformer_block.attn_sconv",
627
+ "model.mtp.layers.3.transformer_block.mlp.w13_dn",
628
+ "model.mtp.layers.3.transformer_block.mlp.w2_md",
629
+ "model.mtp.layers.3.transformer_block.mlp_sconv",
630
+ "model.mtp.layers.4.input_proj",
631
+ "model.mtp.layers.4.transformer_block.attn.k_sconv",
632
+ "model.mtp.layers.4.transformer_block.attn.v_sconv",
633
+ "model.mtp.layers.4.transformer_block.attn.wk_dv",
634
+ "model.mtp.layers.4.transformer_block.attn.wo_ud",
635
+ "model.mtp.layers.4.transformer_block.attn.wq_du",
636
+ "model.mtp.layers.4.transformer_block.attn.wr_du",
637
+ "model.mtp.layers.4.transformer_block.attn.wv_dv",
638
+ "model.mtp.layers.4.transformer_block.attn_sconv",
639
+ "model.mtp.layers.4.transformer_block.mlp.w13_dn",
640
+ "model.mtp.layers.4.transformer_block.mlp.w2_md",
641
+ "model.mtp.layers.4.transformer_block.mlp_sconv",
642
+ "model.mtp.layers.5.input_proj",
643
+ "model.mtp.layers.5.transformer_block.attn.k_sconv",
644
+ "model.mtp.layers.5.transformer_block.attn.v_sconv",
645
+ "model.mtp.layers.5.transformer_block.attn.wk_dv",
646
+ "model.mtp.layers.5.transformer_block.attn.wo_ud",
647
+ "model.mtp.layers.5.transformer_block.attn.wq_du",
648
+ "model.mtp.layers.5.transformer_block.attn.wr_du",
649
+ "model.mtp.layers.5.transformer_block.attn.wv_dv",
650
+ "model.mtp.layers.5.transformer_block.attn_sconv",
651
+ "model.mtp.layers.5.transformer_block.mlp.w13_dn",
652
+ "model.mtp.layers.5.transformer_block.mlp.w2_md",
653
+ "model.mtp.layers.5.transformer_block.mlp_sconv",
654
+ "model.mtp.layers.6.input_proj",
655
+ "model.mtp.layers.6.transformer_block.attn.k_sconv",
656
+ "model.mtp.layers.6.transformer_block.attn.v_sconv",
657
+ "model.mtp.layers.6.transformer_block.attn.wk_dv",
658
+ "model.mtp.layers.6.transformer_block.attn.wo_ud",
659
+ "model.mtp.layers.6.transformer_block.attn.wq_du",
660
+ "model.mtp.layers.6.transformer_block.attn.wr_du",
661
+ "model.mtp.layers.6.transformer_block.attn.wv_dv",
662
+ "model.mtp.layers.6.transformer_block.attn_sconv",
663
+ "model.mtp.layers.6.transformer_block.mlp.w13_dn",
664
+ "model.mtp.layers.6.transformer_block.mlp.w2_md",
665
+ "model.mtp.layers.6.transformer_block.mlp_sconv",
666
+ "model.mtp.layers.7.input_proj",
667
+ "model.mtp.layers.7.transformer_block.attn.k_sconv",
668
+ "model.mtp.layers.7.transformer_block.attn.v_sconv",
669
+ "model.mtp.layers.7.transformer_block.attn.wk_dv",
670
+ "model.mtp.layers.7.transformer_block.attn.wo_ud",
671
+ "model.mtp.layers.7.transformer_block.attn.wq_du",
672
+ "model.mtp.layers.7.transformer_block.attn.wr_du",
673
+ "model.mtp.layers.7.transformer_block.attn.wv_dv",
674
+ "model.mtp.layers.7.transformer_block.attn_sconv",
675
+ "model.mtp.layers.7.transformer_block.mlp.w13_dn",
676
+ "model.mtp.layers.7.transformer_block.mlp.w2_md",
677
+ "model.mtp.layers.7.transformer_block.mlp_sconv",
678
+ "model.visual.layers.linear_0",
679
+ "model.visual.layers.linear_1",
680
+ "model.visual.layers.linear_2",
681
+ "model.visual.layers.linear_3"
682
+ ],
683
+ "algo_config": null,
684
+ "softmax_quant_spec": null,
685
+ "quant_method": "quark",
686
+ "layer_type_quant_config": {},
687
+ "layer_quant_config": {},
688
+ "kv_cache_quant_config": {},
689
+ "kv_cache_post_rope": false,
690
+ "quant_mode": "eager_mode",
691
+ "version": "0.12.post1+rocm72.torch2.11",
692
+ "export": {
693
+ "kv_cache_group": [],
694
+ "min_kv_scale": 0.0,
695
+ "pack_method": "reorder",
696
+ "weight_format": "real_quantized",
697
+ "weight_merge_groups": null
698
+ }
699
+ }
700
+ }
model-00001-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74dce232aeffc11bbe86652934799fd8d76d1a6e322ac68b00c9f75f8d92f7ea
3
+ size 11008035764
model-00002-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b11e28903ed4f8d88ffaecf33b0cab2e0a557ee624cb2b86af160829a61a8773
3
+ size 3525521144
model-00003-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b02a1bfe46752435056ec19b6428c993a88db09c5d67865fd317f838d9f67f08
3
+ size 4731326792
model-00004-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a7d1247cd4541d16a572098ccb1dd819ad39e56619f4afb53cb69e9dd0792a9
3
+ size 4596957916
model-00005-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35e43ebed10a3ee20b748145aa44b4d17fecedacbdcfe39fdd7cffb22df09f8c
3
+ size 3477138308
model-00006-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:788596c975af4ef72b8dca1a95971cab5f812cd9523b6ce74f103e4ccc47909c
3
+ size 3961855308
model-00007-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be83838b2fa7e1d8f71c3d1c040fcc10339cf14dfe0cba110d86e51960bc2c63
3
+ size 4972656110
model-00008-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7cc8c8597977a0835cb4a0dfade6571fe1a53db372aabd9dd0b15dff3703e756
3
+ size 3632409908
model-00009-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0d0b17b9f1e4f0a857abcb0c8d4ea26739a876fbb9c592528194ddd1611c98a3
3
+ size 5189658596
model-00010-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:233192697a93d729f86d5abd7b0030e22b2bd810dc5692bc8fa4095e09a1eac3
3
+ size 3676558996
model-00011-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7c0e4e84bee9fef32601d8e116b012f4962bd3165ca541eb1057f1e1c796f69d
3
+ size 4947387220
model-00012-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c80f40750191bacafb1864b679891970b4b955341fcbcbc50f0ac54d49cfe9d
3
+ size 5081635392
model-00013-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9f60019efd7631dc80768cf674be872eb11f4bf92bd49f07bd2fcdee8107000
3
+ size 5022781112
model-00014-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b86df7155272e488b45287337494264548f06227ca452858aafb183107a60425
3
+ size 4915946056
model-00015-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7b1d1f9d7f0852f02c9a0611ebb35f7f51d4bb2e59e415e048bd87279e3fac69
3
+ size 3615534862
model-00016-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7817b4c2dbdc0c35bda53899fd1ddf44b49af98e4845d817306fbdfb9c0c4fff
3
+ size 4632691844
model-00017-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8dd59d79ea1630992a11be1c3c81efe7051c8de5464d3807c9a7335ad3f4e1a
3
+ size 8067962804
model-00018-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f045411ea8ee9c1641f62bd9209ad6c6938b5aec7ce32c3f38f42676294c5b1
3
+ size 3851986820
model-00019-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a73f6407c242a9908b935b603bb6825c3f043d8496915ad4ad727995f07de842
3
+ size 5136558292
model-00020-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:604b73b20e18de686813c2bffb07deb2b70a5ea0333a5751307a94d340bf9356
3
+ size 4714473644
model-00021-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:10ce7399ecabc95e57b4c1482487851c04d1883b4083ab4c70e8932ae9b09430
3
+ size 5020791260
model-00022-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eff8e779e2f25da369b3a31de51f31e20de06a7ee6dfca179482d51c7565811f
3
+ size 5281036592
model-00023-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d094871ddc61a41189123381e2efe271420b1dc11e2a1dafac4dcf3034676f7f
3
+ size 4844582264
model-00024-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3754e6346c4268ab0467bf11e732eec35d5130192f395b14b6a69acac9155e7
3
+ size 4723077136
model-00025-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f6cade48d16fc3958331627adedbdaa5c6fdd05bda3c034363877d3d682b6cb6
3
+ size 4852912484
model-00026-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bd4ae77a6faf3ca28a023c47534d8e545a4367fccccd7b687173efb70bd29844
3
+ size 4972678356
model-00027-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0494a86de23a1e24467d94d8599371692d4e1f75e730e463a69c0a67eff3243d
3
+ size 4701975132
model-00028-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1e70270cfb894e116713cdab0eb10df6bcf34ab5e72aebdcdfae67d51de72168
3
+ size 4815146480
model-00029-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aca2957cb0724f01d92e8a6c09bac0f618fbf70e06705cab6d75d6b251b5929b
3
+ size 4850864560
model-00030-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f84ab58bd7e193f4fca0c67eecfb816951b4dbac70bd7447ddd4e182605b9ff4
3
+ size 6248246224
model-00031-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dcf9a3dfa5e45190f6cdb2911aa6c3ecdf84cf3b649c84469100d8da5ea69e0f
3
+ size 4571810020
model-00032-of-00032.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:583d63ff0cd27b299e1afd2ff6e1568fff501f85af4ccfbc43d8a33ec4fde98b
3
+ size 4775343280
model.safetensors.index.json ADDED
@@ -0,0 +1,1133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 162881386036
4
+ },
5
+ "weight_map": {
6
+ "model.llm.layers.22.mlp.gate.global_scale": "model-00008-of-00032.safetensors",
7
+ "model.llm.layers.0.attn.q_norm.weight": "model-00008-of-00032.safetensors",
8
+ "model.llm.layers.12.mlp_norm.weight": "model-00008-of-00032.safetensors",
9
+ "model.llm.layers.13.mlp_norm.weight": "model-00008-of-00032.safetensors",
10
+ "model.llm.layers.14.mlp.shared_experts.shared_w13_weight": "model-00008-of-00032.safetensors",
11
+ "model.llm.layers.15.mlp.gate.weight": "model-00008-of-00032.safetensors",
12
+ "model.llm.layers.16.mlp.shared_experts.shared_w2_weight": "model-00008-of-00032.safetensors",
13
+ "model.llm.layers.19.attn.q_norm.weight": "model-00008-of-00032.safetensors",
14
+ "model.llm.layers.23.attn.wk_dv.weight": "model-00008-of-00032.safetensors",
15
+ "model.llm.layers.25.attn.k_sconv.weight": "model-00008-of-00032.safetensors",
16
+ "model.llm.layers.26.attn.rel_logits_proj.proj": "model-00008-of-00032.safetensors",
17
+ "model.llm.layers.27.mlp.experts.w2_weight": "model-00008-of-00032.safetensors",
18
+ "model.llm.layers.27.mlp.experts.w2_weight_scale": "model-00008-of-00032.safetensors",
19
+ "model.llm.layers.30.attn.wr_du.weight": "model-00008-of-00032.safetensors",
20
+ "model.llm.layers.34.attn.wv_dv.weight": "model-00008-of-00032.safetensors",
21
+ "model.llm.layers.36.attn.wr_du.weight": "model-00008-of-00032.safetensors",
22
+ "model.llm.layers.37.mlp_sconv.weight": "model-00008-of-00032.safetensors",
23
+ "model.llm.layers.38.mlp.experts.w13_weight": "model-00008-of-00032.safetensors",
24
+ "model.llm.layers.38.mlp.experts.w13_weight_scale": "model-00008-of-00032.safetensors",
25
+ "model.llm.layers.39.mlp.shared_experts.shared_w13_weight": "model-00008-of-00032.safetensors",
26
+ "model.llm.layers.4.mlp.gate.weight": "model-00008-of-00032.safetensors",
27
+ "model.llm.layers.6.attn.wr_du.weight": "model-00008-of-00032.safetensors",
28
+ "model.llm.layers.6.mlp_sconv.weight": "model-00008-of-00032.safetensors",
29
+ "model.llm.layers.8.attn.wr_du.weight": "model-00008-of-00032.safetensors",
30
+ "model.llm.layers.9.attn.wr_du.weight": "model-00008-of-00032.safetensors",
31
+ "model.llm.layers.22.mlp.gate.bias": "model-00002-of-00032.safetensors",
32
+ "model.llm.layers.26.mlp.gate.global_scale": "model-00002-of-00032.safetensors",
33
+ "model.llm.layers.29.mlp.gate.global_scale": "model-00002-of-00032.safetensors",
34
+ "model.llm.layers.15.attn_norm.weight": "model-00002-of-00032.safetensors",
35
+ "model.llm.layers.16.attn.v_sconv.weight": "model-00002-of-00032.safetensors",
36
+ "model.llm.layers.16.attn_norm.weight": "model-00002-of-00032.safetensors",
37
+ "model.llm.layers.17.attn_sconv.weight": "model-00002-of-00032.safetensors",
38
+ "model.llm.layers.22.attn.q_norm.weight": "model-00002-of-00032.safetensors",
39
+ "model.llm.layers.22.attn_sconv.weight": "model-00002-of-00032.safetensors",
40
+ "model.llm.layers.22.mlp.experts.w2_weight": "model-00002-of-00032.safetensors",
41
+ "model.llm.layers.22.mlp.experts.w2_weight_scale": "model-00002-of-00032.safetensors",
42
+ "model.llm.layers.26.attn.wq_du.weight": "model-00002-of-00032.safetensors",
43
+ "model.llm.layers.27.mlp.experts.w13_weight": "model-00002-of-00032.safetensors",
44
+ "model.llm.layers.27.mlp.experts.w13_weight_scale": "model-00002-of-00032.safetensors",
45
+ "model.llm.layers.27.mlp_sconv.weight": "model-00002-of-00032.safetensors",
46
+ "model.llm.layers.30.attn.v_sconv.weight": "model-00002-of-00032.safetensors",
47
+ "model.llm.layers.30.mlp.shared_experts.shared_w2_weight": "model-00002-of-00032.safetensors",
48
+ "model.llm.layers.31.attn.k_sconv.weight": "model-00002-of-00032.safetensors",
49
+ "model.llm.layers.32.mlp.gate.weight": "model-00002-of-00032.safetensors",
50
+ "model.llm.layers.35.attn.k_sconv.weight": "model-00002-of-00032.safetensors",
51
+ "model.llm.layers.37.attn.k_sconv.weight": "model-00002-of-00032.safetensors",
52
+ "model.llm.layers.37.attn_sconv.weight": "model-00002-of-00032.safetensors",
53
+ "model.llm.layers.5.attn.wo_ud.weight": "model-00002-of-00032.safetensors",
54
+ "model.llm.layers.21.mlp.gate.bias": "model-00007-of-00032.safetensors",
55
+ "model.llm.layers.28.mlp.gate.global_scale": "model-00007-of-00032.safetensors",
56
+ "model.llm.layers.0.attn.v_sconv.weight": "model-00007-of-00032.safetensors",
57
+ "model.llm.layers.0.mlp.global_scale": "model-00007-of-00032.safetensors",
58
+ "model.llm.layers.11.attn.wk_dv.weight": "model-00007-of-00032.safetensors",
59
+ "model.llm.layers.11.mlp.experts.w2_weight": "model-00007-of-00032.safetensors",
60
+ "model.llm.layers.11.mlp.experts.w2_weight_scale": "model-00007-of-00032.safetensors",
61
+ "model.llm.layers.11.mlp_norm.weight": "model-00007-of-00032.safetensors",
62
+ "model.llm.layers.12.attn.wq_du.weight": "model-00007-of-00032.safetensors",
63
+ "model.llm.layers.13.attn.v_sconv.weight": "model-00007-of-00032.safetensors",
64
+ "model.llm.layers.14.attn.wo_ud.weight": "model-00007-of-00032.safetensors",
65
+ "model.llm.layers.15.attn.wk_dv.weight": "model-00007-of-00032.safetensors",
66
+ "model.llm.layers.15.mlp_sconv.weight": "model-00007-of-00032.safetensors",
67
+ "model.llm.layers.17.attn.rel_logits_proj.proj": "model-00007-of-00032.safetensors",
68
+ "model.llm.layers.17.mlp.shared_experts.shared_w13_weight": "model-00007-of-00032.safetensors",
69
+ "model.llm.layers.19.mlp.gate.weight": "model-00007-of-00032.safetensors",
70
+ "model.llm.layers.2.attn.wo_ud.weight": "model-00007-of-00032.safetensors",
71
+ "model.llm.layers.2.mlp_sconv.weight": "model-00007-of-00032.safetensors",
72
+ "model.llm.layers.24.mlp.gate.weight": "model-00007-of-00032.safetensors",
73
+ "model.llm.layers.25.mlp.experts.w13_weight": "model-00007-of-00032.safetensors",
74
+ "model.llm.layers.25.mlp.experts.w13_weight_scale": "model-00007-of-00032.safetensors",
75
+ "model.llm.layers.28.attn.wo_ud.weight": "model-00007-of-00032.safetensors",
76
+ "model.llm.layers.3.attn.wk_dv.weight": "model-00007-of-00032.safetensors",
77
+ "model.llm.layers.3.mlp_norm.weight": "model-00007-of-00032.safetensors",
78
+ "model.llm.layers.30.attn_norm.weight": "model-00007-of-00032.safetensors",
79
+ "model.llm.layers.30.mlp.shared_experts.shared_w13_weight": "model-00007-of-00032.safetensors",
80
+ "model.llm.layers.32.attn.rel_logits_proj.proj": "model-00007-of-00032.safetensors",
81
+ "model.llm.layers.32.mlp_sconv.weight": "model-00007-of-00032.safetensors",
82
+ "model.llm.layers.33.attn.wr_du.weight": "model-00007-of-00032.safetensors",
83
+ "model.llm.layers.34.attn.wr_du.weight": "model-00007-of-00032.safetensors",
84
+ "model.llm.layers.35.attn_sconv.weight": "model-00007-of-00032.safetensors",
85
+ "model.llm.layers.35.mlp.gate.weight": "model-00007-of-00032.safetensors",
86
+ "model.llm.layers.36.mlp.shared_experts.shared_w13_weight": "model-00007-of-00032.safetensors",
87
+ "model.llm.layers.38.attn.k_norm.weight": "model-00007-of-00032.safetensors",
88
+ "model.llm.layers.4.mlp_sconv.weight": "model-00007-of-00032.safetensors",
89
+ "model.llm.layers.6.attn.wo_ud.weight": "model-00007-of-00032.safetensors",
90
+ "model.llm.layers.8.mlp.experts.w2_weight": "model-00007-of-00032.safetensors",
91
+ "model.llm.layers.8.mlp.experts.w2_weight_scale": "model-00007-of-00032.safetensors",
92
+ "model.llm.layers.2.mlp.gate.bias": "model-00001-of-00032.safetensors",
93
+ "model.llm.layers.7.mlp.gate.global_scale": "model-00001-of-00032.safetensors",
94
+ "model.llm.layers.11.attn_sconv.weight": "model-00001-of-00032.safetensors",
95
+ "model.llm.layers.13.attn.k_sconv.weight": "model-00001-of-00032.safetensors",
96
+ "model.llm.layers.18.mlp_norm.weight": "model-00001-of-00032.safetensors",
97
+ "model.llm.layers.2.attn.wq_du.weight": "model-00001-of-00032.safetensors",
98
+ "model.llm.layers.2.mlp.experts.w13_weight": "model-00001-of-00032.safetensors",
99
+ "model.llm.layers.2.mlp.shared_experts.shared_w13_weight": "model-00001-of-00032.safetensors",
100
+ "model.llm.layers.21.attn_norm.weight": "model-00001-of-00032.safetensors",
101
+ "model.llm.layers.22.mlp.shared_experts.shared_w2_weight": "model-00001-of-00032.safetensors",
102
+ "model.llm.layers.26.attn.v_sconv.weight": "model-00001-of-00032.safetensors",
103
+ "model.llm.layers.28.mlp.experts.w13_weight": "model-00001-of-00032.safetensors",
104
+ "model.llm.layers.28.mlp.experts.w13_weight_scale": "model-00001-of-00032.safetensors",
105
+ "model.llm.layers.32.attn.q_norm.weight": "model-00001-of-00032.safetensors",
106
+ "model.llm.layers.34.mlp.gate.weight": "model-00001-of-00032.safetensors",
107
+ "model.llm.layers.9.mlp.gate.global_scale": "model-00006-of-00032.safetensors",
108
+ "model.llm.layers.1.attn.k_sconv.weight": "model-00006-of-00032.safetensors",
109
+ "model.llm.layers.1.attn.wr_du.weight": "model-00006-of-00032.safetensors",
110
+ "model.llm.layers.11.attn.wo_ud.weight": "model-00006-of-00032.safetensors",
111
+ "model.llm.layers.13.attn.k_norm.weight": "model-00006-of-00032.safetensors",
112
+ "model.llm.layers.16.mlp.gate.weight": "model-00006-of-00032.safetensors",
113
+ "model.llm.layers.16.mlp_norm.weight": "model-00006-of-00032.safetensors",
114
+ "model.llm.layers.18.attn.wr_du.weight": "model-00006-of-00032.safetensors",
115
+ "model.llm.layers.18.mlp_sconv.weight": "model-00006-of-00032.safetensors",
116
+ "model.llm.layers.19.attn_sconv.weight": "model-00006-of-00032.safetensors",
117
+ "model.llm.layers.2.attn.rel_logits_proj.proj": "model-00006-of-00032.safetensors",
118
+ "model.llm.layers.20.attn.k_sconv.weight": "model-00006-of-00032.safetensors",
119
+ "model.llm.layers.21.attn.q_norm.weight": "model-00006-of-00032.safetensors",
120
+ "model.llm.layers.21.mlp.shared_experts.shared_w2_weight": "model-00006-of-00032.safetensors",
121
+ "model.llm.layers.23.attn_norm.weight": "model-00006-of-00032.safetensors",
122
+ "model.llm.layers.23.mlp.experts.w2_weight": "model-00006-of-00032.safetensors",
123
+ "model.llm.layers.23.mlp.experts.w2_weight_scale": "model-00006-of-00032.safetensors",
124
+ "model.llm.layers.23.mlp.shared_experts.shared_w13_weight": "model-00006-of-00032.safetensors",
125
+ "model.llm.layers.24.attn.q_norm.weight": "model-00006-of-00032.safetensors",
126
+ "model.llm.layers.26.attn.wo_ud.weight": "model-00006-of-00032.safetensors",
127
+ "model.llm.layers.26.attn.wv_dv.weight": "model-00006-of-00032.safetensors",
128
+ "model.llm.layers.27.attn.k_sconv.weight": "model-00006-of-00032.safetensors",
129
+ "model.llm.layers.27.attn.v_sconv.weight": "model-00006-of-00032.safetensors",
130
+ "model.llm.layers.27.mlp_norm.weight": "model-00006-of-00032.safetensors",
131
+ "model.llm.layers.28.mlp_sconv.weight": "model-00006-of-00032.safetensors",
132
+ "model.llm.layers.29.attn_norm.weight": "model-00006-of-00032.safetensors",
133
+ "model.llm.layers.3.attn.v_sconv.weight": "model-00006-of-00032.safetensors",
134
+ "model.llm.layers.30.mlp.experts.w13_weight": "model-00006-of-00032.safetensors",
135
+ "model.llm.layers.30.mlp.experts.w13_weight_scale": "model-00006-of-00032.safetensors",
136
+ "model.llm.layers.31.attn.wq_du.weight": "model-00006-of-00032.safetensors",
137
+ "model.llm.layers.33.mlp_sconv.weight": "model-00006-of-00032.safetensors",
138
+ "model.llm.layers.34.mlp.shared_experts.shared_w13_weight": "model-00006-of-00032.safetensors",
139
+ "model.llm.layers.35.attn.rel_logits_proj.proj": "model-00006-of-00032.safetensors",
140
+ "model.llm.layers.36.attn.wq_du.weight": "model-00006-of-00032.safetensors",
141
+ "model.llm.layers.37.attn.wq_du.weight": "model-00006-of-00032.safetensors",
142
+ "model.llm.layers.38.mlp.shared_experts.shared_w13_weight": "model-00006-of-00032.safetensors",
143
+ "model.llm.layers.39.attn.k_norm.weight": "model-00006-of-00032.safetensors",
144
+ "model.llm.layers.39.attn.wq_du.weight": "model-00006-of-00032.safetensors",
145
+ "model.llm.layers.4.attn.q_norm.weight": "model-00006-of-00032.safetensors",
146
+ "model.llm.layers.4.mlp.shared_experts.shared_w2_weight": "model-00006-of-00032.safetensors",
147
+ "model.llm.layers.40.attn.k_sconv.weight": "model-00006-of-00032.safetensors",
148
+ "model.llm.layers.40.attn.wq_du.weight": "model-00006-of-00032.safetensors",
149
+ "model.llm.layers.40.mlp.gate.weight": "model-00006-of-00032.safetensors",
150
+ "model.llm.layers.5.attn.wr_du.weight": "model-00006-of-00032.safetensors",
151
+ "model.llm.layers.6.attn.k_norm.weight": "model-00006-of-00032.safetensors",
152
+ "model.llm.layers.6.mlp_norm.weight": "model-00006-of-00032.safetensors",
153
+ "model.llm.layers.7.attn.k_norm.weight": "model-00006-of-00032.safetensors",
154
+ "model.llm.layers.8.attn.wv_dv.weight": "model-00006-of-00032.safetensors",
155
+ "model.llm.layers.9.attn.v_sconv.weight": "model-00006-of-00032.safetensors",
156
+ "model.llm.layers.9.mlp.gate.weight": "model-00006-of-00032.safetensors",
157
+ "model.llm.layers.40.mlp.gate.global_scale": "model-00005-of-00032.safetensors",
158
+ "model.llm.layers.10.attn.k_sconv.weight": "model-00005-of-00032.safetensors",
159
+ "model.llm.layers.11.attn.rel_logits_proj.proj": "model-00005-of-00032.safetensors",
160
+ "model.llm.layers.13.mlp.shared_experts.shared_w2_weight": "model-00005-of-00032.safetensors",
161
+ "model.llm.layers.16.attn.q_norm.weight": "model-00005-of-00032.safetensors",
162
+ "model.llm.layers.17.attn.v_sconv.weight": "model-00005-of-00032.safetensors",
163
+ "model.llm.layers.17.attn.wk_dv.weight": "model-00005-of-00032.safetensors",
164
+ "model.llm.layers.17.attn.wv_dv.weight": "model-00005-of-00032.safetensors",
165
+ "model.llm.layers.22.attn.wr_du.weight": "model-00005-of-00032.safetensors",
166
+ "model.llm.layers.3.attn.q_norm.weight": "model-00005-of-00032.safetensors",
167
+ "model.llm.layers.33.attn_norm.weight": "model-00005-of-00032.safetensors",
168
+ "model.llm.layers.36.attn.q_norm.weight": "model-00005-of-00032.safetensors",
169
+ "model.llm.layers.39.mlp.experts.w13_weight": "model-00005-of-00032.safetensors",
170
+ "model.llm.layers.39.mlp.experts.w13_weight_scale": "model-00005-of-00032.safetensors",
171
+ "model.llm.layers.4.mlp.experts.w2_weight": "model-00005-of-00032.safetensors",
172
+ "model.llm.layers.4.mlp.experts.w2_weight_scale": "model-00005-of-00032.safetensors",
173
+ "model.llm.layers.41.attn.k_norm.weight": "model-00005-of-00032.safetensors",
174
+ "model.llm.layers.2.mlp.gate.global_scale": "model-00003-of-00032.safetensors",
175
+ "model.llm.layers.33.mlp.gate.bias": "model-00003-of-00032.safetensors",
176
+ "model.llm.layers.36.mlp.gate.global_scale": "model-00003-of-00032.safetensors",
177
+ "model.llm.layers.0.attn_sconv.weight": "model-00003-of-00032.safetensors",
178
+ "model.llm.layers.15.attn.k_sconv.weight": "model-00003-of-00032.safetensors",
179
+ "model.llm.layers.15.mlp.experts.w13_weight": "model-00003-of-00032.safetensors",
180
+ "model.llm.layers.15.mlp.experts.w13_weight_scale": "model-00003-of-00032.safetensors",
181
+ "model.llm.layers.16.attn.wv_dv.weight": "model-00003-of-00032.safetensors",
182
+ "model.llm.layers.17.attn.wq_du.weight": "model-00003-of-00032.safetensors",
183
+ "model.llm.layers.19.attn.wq_du.weight": "model-00003-of-00032.safetensors",
184
+ "model.llm.layers.20.attn.k_norm.weight": "model-00003-of-00032.safetensors",
185
+ "model.llm.layers.22.mlp_sconv.weight": "model-00003-of-00032.safetensors",
186
+ "model.llm.layers.24.mlp.experts.w13_weight": "model-00003-of-00032.safetensors",
187
+ "model.llm.layers.24.mlp.experts.w13_weight_scale": "model-00003-of-00032.safetensors",
188
+ "model.llm.layers.25.attn_norm.weight": "model-00003-of-00032.safetensors",
189
+ "model.llm.layers.27.attn.k_norm.weight": "model-00003-of-00032.safetensors",
190
+ "model.llm.layers.27.attn.wv_dv.weight": "model-00003-of-00032.safetensors",
191
+ "model.llm.layers.29.attn.wo_ud.weight": "model-00003-of-00032.safetensors",
192
+ "model.llm.layers.33.attn.v_sconv.weight": "model-00003-of-00032.safetensors",
193
+ "model.llm.layers.33.mlp_norm.weight": "model-00003-of-00032.safetensors",
194
+ "model.llm.layers.37.attn_norm.weight": "model-00003-of-00032.safetensors",
195
+ "model.llm.layers.38.attn.wv_dv.weight": "model-00003-of-00032.safetensors",
196
+ "model.llm.layers.39.attn.wk_dv.weight": "model-00003-of-00032.safetensors",
197
+ "model.llm.layers.4.attn.wo_ud.weight": "model-00003-of-00032.safetensors",
198
+ "model.llm.layers.4.attn_sconv.weight": "model-00003-of-00032.safetensors",
199
+ "model.llm.layers.5.attn_norm.weight": "model-00003-of-00032.safetensors",
200
+ "model.llm.layers.13.mlp.gate.global_scale": "model-00004-of-00032.safetensors",
201
+ "model.llm.layers.11.mlp.shared_experts.shared_w2_weight": "model-00004-of-00032.safetensors",
202
+ "model.llm.layers.6.mlp.experts.w13_weight": "model-00004-of-00032.safetensors",
203
+ "model.llm.layers.6.mlp.experts.w13_weight_scale": "model-00004-of-00032.safetensors",
204
+ "model.llm.layers.8.mlp.experts.w13_weight": "model-00004-of-00032.safetensors",
205
+ "model.llm.layers.8.mlp.experts.w13_weight_scale": "model-00004-of-00032.safetensors",
206
+ "model.llm.layers.16.mlp.gate.bias": "model-00011-of-00032.safetensors",
207
+ "model.llm.layers.20.mlp.gate.bias": "model-00011-of-00032.safetensors",
208
+ "model.llm.layers.20.mlp.gate.global_scale": "model-00011-of-00032.safetensors",
209
+ "model.llm.layers.27.mlp.gate.bias": "model-00011-of-00032.safetensors",
210
+ "model.llm.layers.0.attn_norm.weight": "model-00011-of-00032.safetensors",
211
+ "model.llm.layers.1.attn.k_norm.weight": "model-00011-of-00032.safetensors",
212
+ "model.llm.layers.12.attn.k_sconv.weight": "model-00011-of-00032.safetensors",
213
+ "model.llm.layers.13.attn.wq_du.weight": "model-00011-of-00032.safetensors",
214
+ "model.llm.layers.14.attn.rel_logits_proj.proj": "model-00011-of-00032.safetensors",
215
+ "model.llm.layers.14.mlp.shared_experts.shared_w2_weight": "model-00011-of-00032.safetensors",
216
+ "model.llm.layers.15.attn.wq_du.weight": "model-00011-of-00032.safetensors",
217
+ "model.llm.layers.16.attn.wq_du.weight": "model-00011-of-00032.safetensors",
218
+ "model.llm.layers.18.attn.wk_dv.weight": "model-00011-of-00032.safetensors",
219
+ "model.llm.layers.2.mlp_norm.weight": "model-00011-of-00032.safetensors",
220
+ "model.llm.layers.20.attn.wo_ud.weight": "model-00011-of-00032.safetensors",
221
+ "model.llm.layers.20.mlp_sconv.weight": "model-00011-of-00032.safetensors",
222
+ "model.llm.layers.21.attn_sconv.weight": "model-00011-of-00032.safetensors",
223
+ "model.llm.layers.22.attn.wo_ud.weight": "model-00011-of-00032.safetensors",
224
+ "model.llm.layers.25.mlp.shared_experts.shared_w13_weight": "model-00011-of-00032.safetensors",
225
+ "model.llm.layers.26.mlp.experts.w13_weight": "model-00011-of-00032.safetensors",
226
+ "model.llm.layers.26.mlp.experts.w13_weight_scale": "model-00011-of-00032.safetensors",
227
+ "model.llm.layers.27.attn.wr_du.weight": "model-00011-of-00032.safetensors",
228
+ "model.llm.layers.29.mlp.gate.weight": "model-00011-of-00032.safetensors",
229
+ "model.llm.layers.3.mlp.shared_experts.shared_w2_weight": "model-00011-of-00032.safetensors",
230
+ "model.llm.layers.32.attn.k_sconv.weight": "model-00011-of-00032.safetensors",
231
+ "model.llm.layers.35.attn.q_norm.weight": "model-00011-of-00032.safetensors",
232
+ "model.llm.layers.35.mlp_sconv.weight": "model-00011-of-00032.safetensors",
233
+ "model.llm.layers.38.attn.q_norm.weight": "model-00011-of-00032.safetensors",
234
+ "model.llm.layers.39.mlp.experts.w2_weight": "model-00011-of-00032.safetensors",
235
+ "model.llm.layers.39.mlp.experts.w2_weight_scale": "model-00011-of-00032.safetensors",
236
+ "model.llm.layers.4.mlp.shared_experts.shared_w13_weight": "model-00011-of-00032.safetensors",
237
+ "model.llm.layers.40.attn.k_norm.weight": "model-00011-of-00032.safetensors",
238
+ "model.llm.layers.41.attn.v_sconv.weight": "model-00011-of-00032.safetensors",
239
+ "model.llm.layers.6.mlp.experts.w2_weight": "model-00011-of-00032.safetensors",
240
+ "model.llm.layers.6.mlp.experts.w2_weight_scale": "model-00011-of-00032.safetensors",
241
+ "model.llm.layers.7.attn.v_sconv.weight": "model-00011-of-00032.safetensors",
242
+ "model.llm.layers.8.attn.k_norm.weight": "model-00011-of-00032.safetensors",
243
+ "model.llm.layers.8.attn.rel_logits_proj.proj": "model-00011-of-00032.safetensors",
244
+ "model.visual.layers.norm_0.weight": "model-00011-of-00032.safetensors",
245
+ "model.llm.layers.13.mlp.gate.bias": "model-00009-of-00032.safetensors",
246
+ "model.llm.layers.16.mlp.gate.global_scale": "model-00009-of-00032.safetensors",
247
+ "model.llm.layers.19.mlp.gate.bias": "model-00009-of-00032.safetensors",
248
+ "model.llm.layers.21.mlp.gate.global_scale": "model-00009-of-00032.safetensors",
249
+ "model.llm.layers.32.mlp.gate.global_scale": "model-00009-of-00032.safetensors",
250
+ "model.llm.layers.4.mlp.gate.bias": "model-00009-of-00032.safetensors",
251
+ "model.llm.layers.0.attn.k_norm.weight": "model-00009-of-00032.safetensors",
252
+ "model.llm.layers.0.attn.wq_du.weight": "model-00009-of-00032.safetensors",
253
+ "model.llm.layers.0.mlp.w13_dn.weight": "model-00009-of-00032.safetensors",
254
+ "model.llm.layers.1.attn_sconv.weight": "model-00009-of-00032.safetensors",
255
+ "model.llm.layers.11.attn.k_sconv.weight": "model-00009-of-00032.safetensors",
256
+ "model.llm.layers.13.attn.rel_logits_proj.proj": "model-00009-of-00032.safetensors",
257
+ "model.llm.layers.14.mlp.gate.weight": "model-00009-of-00032.safetensors",
258
+ "model.llm.layers.15.attn.rel_logits_proj.proj": "model-00009-of-00032.safetensors",
259
+ "model.llm.layers.15.mlp.shared_experts.shared_w2_weight": "model-00009-of-00032.safetensors",
260
+ "model.llm.layers.16.attn.rel_logits_proj.proj": "model-00009-of-00032.safetensors",
261
+ "model.llm.layers.17.attn.k_norm.weight": "model-00009-of-00032.safetensors",
262
+ "model.llm.layers.18.attn_sconv.weight": "model-00009-of-00032.safetensors",
263
+ "model.llm.layers.20.attn.wv_dv.weight": "model-00009-of-00032.safetensors",
264
+ "model.llm.layers.22.mlp_norm.weight": "model-00009-of-00032.safetensors",
265
+ "model.llm.layers.23.attn.wo_ud.weight": "model-00009-of-00032.safetensors",
266
+ "model.llm.layers.23.attn.wq_du.weight": "model-00009-of-00032.safetensors",
267
+ "model.llm.layers.23.attn.wv_dv.weight": "model-00009-of-00032.safetensors",
268
+ "model.llm.layers.24.attn.wq_du.weight": "model-00009-of-00032.safetensors",
269
+ "model.llm.layers.26.attn.q_norm.weight": "model-00009-of-00032.safetensors",
270
+ "model.llm.layers.26.mlp.gate.weight": "model-00009-of-00032.safetensors",
271
+ "model.llm.layers.27.attn.wq_du.weight": "model-00009-of-00032.safetensors",
272
+ "model.llm.layers.28.attn.k_sconv.weight": "model-00009-of-00032.safetensors",
273
+ "model.llm.layers.28.mlp.experts.w2_weight": "model-00009-of-00032.safetensors",
274
+ "model.llm.layers.28.mlp.experts.w2_weight_scale": "model-00009-of-00032.safetensors",
275
+ "model.llm.layers.29.mlp_sconv.weight": "model-00009-of-00032.safetensors",
276
+ "model.llm.layers.32.attn_norm.weight": "model-00009-of-00032.safetensors",
277
+ "model.llm.layers.34.attn.v_sconv.weight": "model-00009-of-00032.safetensors",
278
+ "model.llm.layers.34.mlp.shared_experts.shared_w2_weight": "model-00009-of-00032.safetensors",
279
+ "model.llm.layers.35.attn.wk_dv.weight": "model-00009-of-00032.safetensors",
280
+ "model.llm.layers.36.attn.wo_ud.weight": "model-00009-of-00032.safetensors",
281
+ "model.llm.layers.37.attn.v_sconv.weight": "model-00009-of-00032.safetensors",
282
+ "model.llm.layers.37.mlp.gate.weight": "model-00009-of-00032.safetensors",
283
+ "model.llm.layers.38.attn.v_sconv.weight": "model-00009-of-00032.safetensors",
284
+ "model.llm.layers.39.attn.q_norm.weight": "model-00009-of-00032.safetensors",
285
+ "model.llm.layers.4.mlp.experts.w13_weight": "model-00009-of-00032.safetensors",
286
+ "model.llm.layers.4.mlp.experts.w13_weight_scale": "model-00009-of-00032.safetensors",
287
+ "model.llm.layers.41.attn.rel_logits_proj.proj": "model-00009-of-00032.safetensors",
288
+ "model.llm.layers.41.mlp.experts.w2_weight": "model-00009-of-00032.safetensors",
289
+ "model.llm.layers.41.mlp.experts.w2_weight_scale": "model-00009-of-00032.safetensors",
290
+ "model.llm.layers.41.mlp.gate.weight": "model-00009-of-00032.safetensors",
291
+ "model.llm.layers.5.attn.k_sconv.weight": "model-00009-of-00032.safetensors",
292
+ "model.llm.layers.5.mlp.gate.weight": "model-00009-of-00032.safetensors",
293
+ "model.llm.layers.5.mlp_norm.weight": "model-00009-of-00032.safetensors",
294
+ "model.llm.layers.7.attn.rel_logits_proj.proj": "model-00009-of-00032.safetensors",
295
+ "model.llm.layers.7.attn.wr_du.weight": "model-00009-of-00032.safetensors",
296
+ "model.llm.layers.8.attn.k_sconv.weight": "model-00009-of-00032.safetensors",
297
+ "model.llm.layers.9.attn.rel_logits_proj.proj": "model-00009-of-00032.safetensors",
298
+ "model.visual.layers.linear_2.weight": "model-00009-of-00032.safetensors",
299
+ "model.llm.layers.11.mlp.gate.global_scale": "model-00015-of-00032.safetensors",
300
+ "model.llm.layers.37.mlp.gate.bias": "model-00015-of-00032.safetensors",
301
+ "model.llm.layers.1.mlp.global_scale": "model-00015-of-00032.safetensors",
302
+ "model.llm.layers.12.attn.q_norm.weight": "model-00015-of-00032.safetensors",
303
+ "model.llm.layers.12.mlp.experts.w2_weight": "model-00015-of-00032.safetensors",
304
+ "model.llm.layers.12.mlp.experts.w2_weight_scale": "model-00015-of-00032.safetensors",
305
+ "model.llm.layers.12.mlp.shared_experts.shared_w13_weight": "model-00015-of-00032.safetensors",
306
+ "model.llm.layers.19.mlp.experts.w13_weight": "model-00015-of-00032.safetensors",
307
+ "model.llm.layers.19.mlp.experts.w13_weight_scale": "model-00015-of-00032.safetensors",
308
+ "model.llm.layers.22.attn_norm.weight": "model-00015-of-00032.safetensors",
309
+ "model.llm.layers.29.attn.q_norm.weight": "model-00015-of-00032.safetensors",
310
+ "model.llm.layers.29.mlp.shared_experts.shared_w2_weight": "model-00015-of-00032.safetensors",
311
+ "model.llm.layers.31.attn.wk_dv.weight": "model-00015-of-00032.safetensors",
312
+ "model.llm.layers.31.attn.wv_dv.weight": "model-00015-of-00032.safetensors",
313
+ "model.llm.layers.32.attn.wk_dv.weight": "model-00015-of-00032.safetensors",
314
+ "model.llm.layers.32.mlp.shared_experts.shared_w13_weight": "model-00015-of-00032.safetensors",
315
+ "model.llm.layers.34.attn.q_norm.weight": "model-00015-of-00032.safetensors",
316
+ "model.llm.layers.7.attn_sconv.weight": "model-00015-of-00032.safetensors",
317
+ "model.llm.layers.3.mlp.gate.global_scale": "model-00016-of-00032.safetensors",
318
+ "model.llm.layers.10.mlp.experts.w13_weight": "model-00016-of-00032.safetensors",
319
+ "model.llm.layers.10.mlp.experts.w13_weight_scale": "model-00016-of-00032.safetensors",
320
+ "model.llm.layers.15.attn.wo_ud.weight": "model-00016-of-00032.safetensors",
321
+ "model.llm.layers.23.mlp.experts.w13_weight": "model-00016-of-00032.safetensors",
322
+ "model.llm.layers.23.mlp.experts.w13_weight_scale": "model-00016-of-00032.safetensors",
323
+ "model.llm.layers.30.mlp.gate.weight": "model-00016-of-00032.safetensors",
324
+ "model.llm.layers.31.attn.wo_ud.weight": "model-00016-of-00032.safetensors",
325
+ "model.llm.layers.36.attn_sconv.weight": "model-00016-of-00032.safetensors",
326
+ "model.llm.layers.5.attn_sconv.weight": "model-00016-of-00032.safetensors",
327
+ "model.llm.layers.24.mlp.gate.global_scale": "model-00010-of-00032.safetensors",
328
+ "model.llm.layers.1.attn.wk_dv.weight": "model-00010-of-00032.safetensors",
329
+ "model.llm.layers.1.mlp.w2_md.weight": "model-00010-of-00032.safetensors",
330
+ "model.llm.layers.10.attn.rel_logits_proj.proj": "model-00010-of-00032.safetensors",
331
+ "model.llm.layers.10.attn.wk_dv.weight": "model-00010-of-00032.safetensors",
332
+ "model.llm.layers.17.attn.wr_du.weight": "model-00010-of-00032.safetensors",
333
+ "model.llm.layers.22.attn.wk_dv.weight": "model-00010-of-00032.safetensors",
334
+ "model.llm.layers.24.attn_norm.weight": "model-00010-of-00032.safetensors",
335
+ "model.llm.layers.24.attn_sconv.weight": "model-00010-of-00032.safetensors",
336
+ "model.llm.layers.26.mlp_sconv.weight": "model-00010-of-00032.safetensors",
337
+ "model.llm.layers.33.mlp.experts.w13_weight": "model-00010-of-00032.safetensors",
338
+ "model.llm.layers.33.mlp.experts.w13_weight_scale": "model-00010-of-00032.safetensors",
339
+ "model.llm.layers.34.mlp_sconv.weight": "model-00010-of-00032.safetensors",
340
+ "model.llm.layers.35.attn_norm.weight": "model-00010-of-00032.safetensors",
341
+ "model.llm.layers.35.mlp.experts.w2_weight": "model-00010-of-00032.safetensors",
342
+ "model.llm.layers.35.mlp.experts.w2_weight_scale": "model-00010-of-00032.safetensors",
343
+ "model.llm.layers.36.mlp_norm.weight": "model-00010-of-00032.safetensors",
344
+ "model.llm.layers.38.mlp.gate.weight": "model-00010-of-00032.safetensors",
345
+ "model.llm.layers.39.mlp_sconv.weight": "model-00010-of-00032.safetensors",
346
+ "model.llm.layers.40.mlp_norm.weight": "model-00010-of-00032.safetensors",
347
+ "model.llm.layers.40.mlp_sconv.weight": "model-00010-of-00032.safetensors",
348
+ "model.llm.layers.41.attn.wr_du.weight": "model-00010-of-00032.safetensors",
349
+ "model.llm.layers.41.attn.wv_dv.weight": "model-00010-of-00032.safetensors",
350
+ "model.llm.layers.41.mlp.shared_experts.shared_w13_weight": "model-00010-of-00032.safetensors",
351
+ "model.llm.layers.8.attn.wk_dv.weight": "model-00010-of-00032.safetensors",
352
+ "model.visual.layers.linear_0.weight": "model-00010-of-00032.safetensors",
353
+ "model.llm.layers.11.mlp.gate.bias": "model-00014-of-00032.safetensors",
354
+ "model.llm.layers.12.mlp.gate.global_scale": "model-00014-of-00032.safetensors",
355
+ "model.llm.layers.14.mlp.gate.global_scale": "model-00014-of-00032.safetensors",
356
+ "model.llm.layers.3.mlp.gate.bias": "model-00014-of-00032.safetensors",
357
+ "model.llm.layers.35.mlp.gate.bias": "model-00014-of-00032.safetensors",
358
+ "model.llm.layers.40.mlp.gate.bias": "model-00014-of-00032.safetensors",
359
+ "model.llm.layers.10.attn_norm.weight": "model-00014-of-00032.safetensors",
360
+ "model.llm.layers.11.attn.wv_dv.weight": "model-00014-of-00032.safetensors",
361
+ "model.llm.layers.12.attn_norm.weight": "model-00014-of-00032.safetensors",
362
+ "model.llm.layers.12.mlp.gate.weight": "model-00014-of-00032.safetensors",
363
+ "model.llm.layers.13.mlp.shared_experts.shared_w13_weight": "model-00014-of-00032.safetensors",
364
+ "model.llm.layers.14.mlp_sconv.weight": "model-00014-of-00032.safetensors",
365
+ "model.llm.layers.16.attn.k_sconv.weight": "model-00014-of-00032.safetensors",
366
+ "model.llm.layers.16.mlp.shared_experts.shared_w13_weight": "model-00014-of-00032.safetensors",
367
+ "model.llm.layers.17.mlp.experts.w13_weight": "model-00014-of-00032.safetensors",
368
+ "model.llm.layers.17.mlp.experts.w13_weight_scale": "model-00014-of-00032.safetensors",
369
+ "model.llm.layers.17.mlp.gate.weight": "model-00014-of-00032.safetensors",
370
+ "model.llm.layers.19.attn.wk_dv.weight": "model-00014-of-00032.safetensors",
371
+ "model.llm.layers.19.attn.wr_du.weight": "model-00014-of-00032.safetensors",
372
+ "model.llm.layers.2.attn.k_sconv.weight": "model-00014-of-00032.safetensors",
373
+ "model.llm.layers.2.attn_norm.weight": "model-00014-of-00032.safetensors",
374
+ "model.llm.layers.23.attn.k_sconv.weight": "model-00014-of-00032.safetensors",
375
+ "model.llm.layers.28.attn.wv_dv.weight": "model-00014-of-00032.safetensors",
376
+ "model.llm.layers.31.attn_norm.weight": "model-00014-of-00032.safetensors",
377
+ "model.llm.layers.32.mlp.experts.w13_weight": "model-00014-of-00032.safetensors",
378
+ "model.llm.layers.32.mlp.experts.w13_weight_scale": "model-00014-of-00032.safetensors",
379
+ "model.llm.layers.32.mlp.shared_experts.shared_w2_weight": "model-00014-of-00032.safetensors",
380
+ "model.llm.layers.33.attn.wk_dv.weight": "model-00014-of-00032.safetensors",
381
+ "model.llm.layers.33.attn.wo_ud.weight": "model-00014-of-00032.safetensors",
382
+ "model.llm.layers.34.attn.rel_logits_proj.proj": "model-00014-of-00032.safetensors",
383
+ "model.llm.layers.35.mlp.shared_experts.shared_w2_weight": "model-00014-of-00032.safetensors",
384
+ "model.llm.layers.37.attn.rel_logits_proj.proj": "model-00014-of-00032.safetensors",
385
+ "model.llm.layers.39.mlp_norm.weight": "model-00014-of-00032.safetensors",
386
+ "model.llm.layers.40.attn.wv_dv.weight": "model-00014-of-00032.safetensors",
387
+ "model.llm.layers.40.attn_sconv.weight": "model-00014-of-00032.safetensors",
388
+ "model.llm.layers.5.attn.v_sconv.weight": "model-00014-of-00032.safetensors",
389
+ "model.llm.layers.6.mlp.shared_experts.shared_w13_weight": "model-00014-of-00032.safetensors",
390
+ "model.llm.layers.8.mlp_norm.weight": "model-00014-of-00032.safetensors",
391
+ "model.llm.layers.9.attn.q_norm.weight": "model-00014-of-00032.safetensors",
392
+ "model.llm.layers.30.mlp.gate.bias": "model-00012-of-00032.safetensors",
393
+ "model.llm.layers.9.mlp.gate.bias": "model-00012-of-00032.safetensors",
394
+ "model.llm.layers.0.attn.wo_ud.weight": "model-00012-of-00032.safetensors",
395
+ "model.llm.layers.1.attn_norm.weight": "model-00012-of-00032.safetensors",
396
+ "model.llm.layers.10.attn.wq_du.weight": "model-00012-of-00032.safetensors",
397
+ "model.llm.layers.10.mlp.gate.weight": "model-00012-of-00032.safetensors",
398
+ "model.llm.layers.12.attn.rel_logits_proj.proj": "model-00012-of-00032.safetensors",
399
+ "model.llm.layers.13.attn_sconv.weight": "model-00012-of-00032.safetensors",
400
+ "model.llm.layers.17.mlp.experts.w2_weight": "model-00012-of-00032.safetensors",
401
+ "model.llm.layers.17.mlp.experts.w2_weight_scale": "model-00012-of-00032.safetensors",
402
+ "model.llm.layers.17.mlp_sconv.weight": "model-00012-of-00032.safetensors",
403
+ "model.llm.layers.18.attn.k_sconv.weight": "model-00012-of-00032.safetensors",
404
+ "model.llm.layers.19.attn.k_sconv.weight": "model-00012-of-00032.safetensors",
405
+ "model.llm.layers.19.attn.wv_dv.weight": "model-00012-of-00032.safetensors",
406
+ "model.llm.layers.20.mlp.shared_experts.shared_w13_weight": "model-00012-of-00032.safetensors",
407
+ "model.llm.layers.21.attn.wo_ud.weight": "model-00012-of-00032.safetensors",
408
+ "model.llm.layers.24.mlp.shared_experts.shared_w13_weight": "model-00012-of-00032.safetensors",
409
+ "model.llm.layers.25.attn.rel_logits_proj.proj": "model-00012-of-00032.safetensors",
410
+ "model.llm.layers.27.attn_sconv.weight": "model-00012-of-00032.safetensors",
411
+ "model.llm.layers.28.attn.k_norm.weight": "model-00012-of-00032.safetensors",
412
+ "model.llm.layers.28.attn.wr_du.weight": "model-00012-of-00032.safetensors",
413
+ "model.llm.layers.3.mlp.shared_experts.shared_w13_weight": "model-00012-of-00032.safetensors",
414
+ "model.llm.layers.30.attn.wo_ud.weight": "model-00012-of-00032.safetensors",
415
+ "model.llm.layers.30.mlp_norm.weight": "model-00012-of-00032.safetensors",
416
+ "model.llm.layers.31.attn.v_sconv.weight": "model-00012-of-00032.safetensors",
417
+ "model.llm.layers.31.mlp_sconv.weight": "model-00012-of-00032.safetensors",
418
+ "model.llm.layers.32.mlp.experts.w2_weight": "model-00012-of-00032.safetensors",
419
+ "model.llm.layers.32.mlp.experts.w2_weight_scale": "model-00012-of-00032.safetensors",
420
+ "model.llm.layers.35.attn.wq_du.weight": "model-00012-of-00032.safetensors",
421
+ "model.llm.layers.35.mlp.experts.w13_weight": "model-00012-of-00032.safetensors",
422
+ "model.llm.layers.35.mlp.experts.w13_weight_scale": "model-00012-of-00032.safetensors",
423
+ "model.llm.layers.35.mlp_norm.weight": "model-00012-of-00032.safetensors",
424
+ "model.llm.layers.37.mlp.shared_experts.shared_w13_weight": "model-00012-of-00032.safetensors",
425
+ "model.llm.layers.40.attn.wo_ud.weight": "model-00012-of-00032.safetensors",
426
+ "model.llm.layers.7.attn.q_norm.weight": "model-00012-of-00032.safetensors",
427
+ "model.llm.layers.8.attn.wq_du.weight": "model-00012-of-00032.safetensors",
428
+ "model.llm.layers.37.mlp.gate.global_scale": "model-00013-of-00032.safetensors",
429
+ "model.llm.layers.6.mlp.gate.global_scale": "model-00013-of-00032.safetensors",
430
+ "model.llm.layers.0.mlp.w2_md.weight": "model-00013-of-00032.safetensors",
431
+ "model.llm.layers.1.attn.rel_logits_proj.proj": "model-00013-of-00032.safetensors",
432
+ "model.llm.layers.10.attn.k_norm.weight": "model-00013-of-00032.safetensors",
433
+ "model.llm.layers.12.mlp.experts.w13_weight": "model-00013-of-00032.safetensors",
434
+ "model.llm.layers.12.mlp.experts.w13_weight_scale": "model-00013-of-00032.safetensors",
435
+ "model.llm.layers.14.attn.v_sconv.weight": "model-00013-of-00032.safetensors",
436
+ "model.llm.layers.18.attn.wv_dv.weight": "model-00013-of-00032.safetensors",
437
+ "model.llm.layers.20.attn.wq_du.weight": "model-00013-of-00032.safetensors",
438
+ "model.llm.layers.20.mlp.shared_experts.shared_w2_weight": "model-00013-of-00032.safetensors",
439
+ "model.llm.layers.24.attn.wr_du.weight": "model-00013-of-00032.safetensors",
440
+ "model.llm.layers.25.attn.wk_dv.weight": "model-00013-of-00032.safetensors",
441
+ "model.llm.layers.26.mlp.shared_experts.shared_w13_weight": "model-00013-of-00032.safetensors",
442
+ "model.llm.layers.29.mlp.experts.w2_weight": "model-00013-of-00032.safetensors",
443
+ "model.llm.layers.29.mlp.experts.w2_weight_scale": "model-00013-of-00032.safetensors",
444
+ "model.llm.layers.3.attn.wq_du.weight": "model-00013-of-00032.safetensors",
445
+ "model.llm.layers.33.attn.wq_du.weight": "model-00013-of-00032.safetensors",
446
+ "model.llm.layers.33.mlp.gate.weight": "model-00013-of-00032.safetensors",
447
+ "model.llm.layers.38.attn_norm.weight": "model-00013-of-00032.safetensors",
448
+ "model.llm.layers.39.attn.wo_ud.weight": "model-00013-of-00032.safetensors",
449
+ "model.llm.layers.4.attn.v_sconv.weight": "model-00013-of-00032.safetensors",
450
+ "model.llm.layers.40.mlp.experts.w2_weight": "model-00013-of-00032.safetensors",
451
+ "model.llm.layers.40.mlp.experts.w2_weight_scale": "model-00013-of-00032.safetensors",
452
+ "model.llm.layers.40.mlp.shared_experts.shared_w13_weight": "model-00013-of-00032.safetensors",
453
+ "model.llm.layers.6.attn.q_norm.weight": "model-00013-of-00032.safetensors",
454
+ "model.llm.layers.6.attn_sconv.weight": "model-00013-of-00032.safetensors",
455
+ "model.llm.layers.9.attn_norm.weight": "model-00013-of-00032.safetensors",
456
+ "model.llm.layers.12.mlp.gate.bias": "model-00018-of-00032.safetensors",
457
+ "model.llm.layers.18.mlp.gate.bias": "model-00018-of-00032.safetensors",
458
+ "model.llm.layers.41.mlp.gate.global_scale": "model-00018-of-00032.safetensors",
459
+ "model.llm.layers.6.mlp.gate.bias": "model-00018-of-00032.safetensors",
460
+ "model.llm.layers.0.mlp_sconv.weight": "model-00018-of-00032.safetensors",
461
+ "model.llm.layers.14.attn.wv_dv.weight": "model-00018-of-00032.safetensors",
462
+ "model.llm.layers.16.attn.wk_dv.weight": "model-00018-of-00032.safetensors",
463
+ "model.llm.layers.17.attn.q_norm.weight": "model-00018-of-00032.safetensors",
464
+ "model.llm.layers.18.attn.wo_ud.weight": "model-00018-of-00032.safetensors",
465
+ "model.llm.layers.19.attn.k_norm.weight": "model-00018-of-00032.safetensors",
466
+ "model.llm.layers.21.mlp.experts.w2_weight": "model-00018-of-00032.safetensors",
467
+ "model.llm.layers.21.mlp.experts.w2_weight_scale": "model-00018-of-00032.safetensors",
468
+ "model.llm.layers.22.attn.wv_dv.weight": "model-00018-of-00032.safetensors",
469
+ "model.llm.layers.22.mlp.shared_experts.shared_w13_weight": "model-00018-of-00032.safetensors",
470
+ "model.llm.layers.23.attn_sconv.weight": "model-00018-of-00032.safetensors",
471
+ "model.llm.layers.24.attn.v_sconv.weight": "model-00018-of-00032.safetensors",
472
+ "model.llm.layers.24.mlp.shared_experts.shared_w2_weight": "model-00018-of-00032.safetensors",
473
+ "model.llm.layers.26.attn.k_norm.weight": "model-00018-of-00032.safetensors",
474
+ "model.llm.layers.27.mlp.shared_experts.shared_w13_weight": "model-00018-of-00032.safetensors",
475
+ "model.llm.layers.28.attn.q_norm.weight": "model-00018-of-00032.safetensors",
476
+ "model.llm.layers.28.mlp.gate.weight": "model-00018-of-00032.safetensors",
477
+ "model.llm.layers.29.attn.v_sconv.weight": "model-00018-of-00032.safetensors",
478
+ "model.llm.layers.3.mlp.experts.w2_weight": "model-00018-of-00032.safetensors",
479
+ "model.llm.layers.3.mlp.experts.w2_weight_scale": "model-00018-of-00032.safetensors",
480
+ "model.llm.layers.30.attn_sconv.weight": "model-00018-of-00032.safetensors",
481
+ "model.llm.layers.31.mlp.gate.weight": "model-00018-of-00032.safetensors",
482
+ "model.llm.layers.39.attn.v_sconv.weight": "model-00018-of-00032.safetensors",
483
+ "model.llm.layers.39.attn_sconv.weight": "model-00018-of-00032.safetensors",
484
+ "model.llm.layers.40.attn.q_norm.weight": "model-00018-of-00032.safetensors",
485
+ "model.llm.layers.40.attn.v_sconv.weight": "model-00018-of-00032.safetensors",
486
+ "model.llm.layers.40.attn.wk_dv.weight": "model-00018-of-00032.safetensors",
487
+ "model.llm.layers.41.mlp_norm.weight": "model-00018-of-00032.safetensors",
488
+ "model.llm.layers.5.attn.wv_dv.weight": "model-00018-of-00032.safetensors",
489
+ "model.llm.layers.7.attn.wo_ud.weight": "model-00018-of-00032.safetensors",
490
+ "model.llm.layers.8.mlp.gate.weight": "model-00018-of-00032.safetensors",
491
+ "model.llm.layers.9.attn.k_norm.weight": "model-00018-of-00032.safetensors",
492
+ "model.llm.layers.9.mlp.experts.w2_weight": "model-00018-of-00032.safetensors",
493
+ "model.llm.layers.9.mlp.experts.w2_weight_scale": "model-00018-of-00032.safetensors",
494
+ "model.llm.layers.9.mlp.shared_experts.shared_w13_weight": "model-00018-of-00032.safetensors",
495
+ "model.llm.layers.9.mlp_norm.weight": "model-00018-of-00032.safetensors",
496
+ "model.visual.layers.linear_1.weight": "model-00018-of-00032.safetensors",
497
+ "model.visual.layers.linear_3.weight": "model-00018-of-00032.safetensors",
498
+ "model.visual.layers.norm_1.weight": "model-00018-of-00032.safetensors",
499
+ "model.llm.layers.10.mlp.gate.bias": "model-00017-of-00032.safetensors",
500
+ "model.llm.layers.8.mlp.gate.global_scale": "model-00017-of-00032.safetensors",
501
+ "model.llm.layers.0.mlp_norm.weight": "model-00017-of-00032.safetensors",
502
+ "model.llm.layers.1.attn.q_norm.weight": "model-00017-of-00032.safetensors",
503
+ "model.llm.layers.1.mlp_sconv.weight": "model-00017-of-00032.safetensors",
504
+ "model.llm.layers.12.attn.v_sconv.weight": "model-00017-of-00032.safetensors",
505
+ "model.llm.layers.12.attn.wk_dv.weight": "model-00017-of-00032.safetensors",
506
+ "model.llm.layers.13.attn.wo_ud.weight": "model-00017-of-00032.safetensors",
507
+ "model.llm.layers.14.mlp.experts.w13_weight": "model-00017-of-00032.safetensors",
508
+ "model.llm.layers.14.mlp.experts.w13_weight_scale": "model-00017-of-00032.safetensors",
509
+ "model.llm.layers.16.attn.wo_ud.weight": "model-00017-of-00032.safetensors",
510
+ "model.llm.layers.19.mlp_sconv.weight": "model-00017-of-00032.safetensors",
511
+ "model.llm.layers.2.mlp.experts.w2_weight": "model-00017-of-00032.safetensors",
512
+ "model.llm.layers.20.attn.q_norm.weight": "model-00017-of-00032.safetensors",
513
+ "model.llm.layers.20.attn.v_sconv.weight": "model-00017-of-00032.safetensors",
514
+ "model.llm.layers.21.mlp.gate.weight": "model-00017-of-00032.safetensors",
515
+ "model.llm.layers.23.attn.k_norm.weight": "model-00017-of-00032.safetensors",
516
+ "model.llm.layers.23.mlp_sconv.weight": "model-00017-of-00032.safetensors",
517
+ "model.llm.layers.25.mlp.shared_experts.shared_w2_weight": "model-00017-of-00032.safetensors",
518
+ "model.llm.layers.28.attn.wk_dv.weight": "model-00017-of-00032.safetensors",
519
+ "model.llm.layers.28.attn_sconv.weight": "model-00017-of-00032.safetensors",
520
+ "model.llm.layers.29.attn.wr_du.weight": "model-00017-of-00032.safetensors",
521
+ "model.llm.layers.3.attn.rel_logits_proj.proj": "model-00017-of-00032.safetensors",
522
+ "model.llm.layers.31.mlp.shared_experts.shared_w13_weight": "model-00017-of-00032.safetensors",
523
+ "model.llm.layers.34.attn.wo_ud.weight": "model-00017-of-00032.safetensors",
524
+ "model.llm.layers.36.attn_norm.weight": "model-00017-of-00032.safetensors",
525
+ "model.llm.layers.38.attn.wo_ud.weight": "model-00017-of-00032.safetensors",
526
+ "model.llm.layers.38.attn.wr_du.weight": "model-00017-of-00032.safetensors",
527
+ "model.llm.layers.38.mlp.experts.w2_weight": "model-00017-of-00032.safetensors",
528
+ "model.llm.layers.38.mlp.experts.w2_weight_scale": "model-00017-of-00032.safetensors",
529
+ "model.llm.layers.38.mlp.shared_experts.shared_w2_weight": "model-00017-of-00032.safetensors",
530
+ "model.llm.layers.4.attn.wr_du.weight": "model-00017-of-00032.safetensors",
531
+ "model.llm.layers.4.attn.wv_dv.weight": "model-00017-of-00032.safetensors",
532
+ "model.llm.layers.41.attn.k_sconv.weight": "model-00017-of-00032.safetensors",
533
+ "model.llm.layers.6.attn.v_sconv.weight": "model-00017-of-00032.safetensors",
534
+ "model.llm.layers.7.attn.wv_dv.weight": "model-00017-of-00032.safetensors",
535
+ "model.llm.layers.9.mlp.shared_experts.shared_w2_weight": "model-00017-of-00032.safetensors",
536
+ "model.llm.layers.38.mlp.gate.global_scale": "model-00019-of-00032.safetensors",
537
+ "model.audio.encoder.weight": "model-00019-of-00032.safetensors",
538
+ "model.llm.embed_norm.weight": "model-00019-of-00032.safetensors",
539
+ "model.llm.layers.13.attn.wr_du.weight": "model-00019-of-00032.safetensors",
540
+ "model.llm.layers.14.attn.k_norm.weight": "model-00019-of-00032.safetensors",
541
+ "model.llm.layers.16.mlp.experts.w13_weight": "model-00019-of-00032.safetensors",
542
+ "model.llm.layers.16.mlp.experts.w13_weight_scale": "model-00019-of-00032.safetensors",
543
+ "model.llm.layers.16.mlp_sconv.weight": "model-00019-of-00032.safetensors",
544
+ "model.llm.layers.20.mlp_norm.weight": "model-00019-of-00032.safetensors",
545
+ "model.llm.layers.24.attn.k_sconv.weight": "model-00019-of-00032.safetensors",
546
+ "model.llm.layers.29.attn_sconv.weight": "model-00019-of-00032.safetensors",
547
+ "model.llm.layers.3.mlp.gate.weight": "model-00019-of-00032.safetensors",
548
+ "model.llm.layers.30.mlp.experts.w2_weight": "model-00019-of-00032.safetensors",
549
+ "model.llm.layers.30.mlp.experts.w2_weight_scale": "model-00019-of-00032.safetensors",
550
+ "model.llm.layers.34.attn.wk_dv.weight": "model-00019-of-00032.safetensors",
551
+ "model.llm.layers.37.attn.wk_dv.weight": "model-00019-of-00032.safetensors",
552
+ "model.llm.layers.37.attn.wo_ud.weight": "model-00019-of-00032.safetensors",
553
+ "model.llm.unembed.weight": "model-00019-of-00032.safetensors",
554
+ "model.llm.layers.15.mlp.gate.bias": "model-00022-of-00032.safetensors",
555
+ "model.llm.layers.23.mlp.gate.global_scale": "model-00022-of-00032.safetensors",
556
+ "model.llm.layers.24.mlp.gate.bias": "model-00022-of-00032.safetensors",
557
+ "model.llm.layers.25.mlp.gate.global_scale": "model-00022-of-00032.safetensors",
558
+ "model.llm.layers.27.mlp.gate.global_scale": "model-00022-of-00032.safetensors",
559
+ "model.llm.layers.32.mlp.gate.bias": "model-00022-of-00032.safetensors",
560
+ "model.llm.layers.33.mlp.gate.global_scale": "model-00022-of-00032.safetensors",
561
+ "model.llm.layers.34.mlp.gate.global_scale": "model-00022-of-00032.safetensors",
562
+ "model.llm.layers.36.mlp.gate.bias": "model-00022-of-00032.safetensors",
563
+ "model.llm.layers.5.mlp.gate.global_scale": "model-00022-of-00032.safetensors",
564
+ "model.audio.final_norm.weight": "model-00022-of-00032.safetensors",
565
+ "model.llm.layers.10.attn.q_norm.weight": "model-00022-of-00032.safetensors",
566
+ "model.llm.layers.11.mlp.gate.weight": "model-00022-of-00032.safetensors",
567
+ "model.llm.layers.11.mlp.shared_experts.shared_w13_weight": "model-00022-of-00032.safetensors",
568
+ "model.llm.layers.13.attn.q_norm.weight": "model-00022-of-00032.safetensors",
569
+ "model.llm.layers.13.attn.wk_dv.weight": "model-00022-of-00032.safetensors",
570
+ "model.llm.layers.14.attn.q_norm.weight": "model-00022-of-00032.safetensors",
571
+ "model.llm.layers.15.mlp.shared_experts.shared_w13_weight": "model-00022-of-00032.safetensors",
572
+ "model.llm.layers.17.mlp.shared_experts.shared_w2_weight": "model-00022-of-00032.safetensors",
573
+ "model.llm.layers.18.attn.q_norm.weight": "model-00022-of-00032.safetensors",
574
+ "model.llm.layers.18.attn.v_sconv.weight": "model-00022-of-00032.safetensors",
575
+ "model.llm.layers.18.attn.wq_du.weight": "model-00022-of-00032.safetensors",
576
+ "model.llm.layers.18.mlp.shared_experts.shared_w2_weight": "model-00022-of-00032.safetensors",
577
+ "model.llm.layers.19.attn.wo_ud.weight": "model-00022-of-00032.safetensors",
578
+ "model.llm.layers.19.mlp.experts.w2_weight": "model-00022-of-00032.safetensors",
579
+ "model.llm.layers.19.mlp.experts.w2_weight_scale": "model-00022-of-00032.safetensors",
580
+ "model.llm.layers.2.attn.wv_dv.weight": "model-00022-of-00032.safetensors",
581
+ "model.llm.layers.2.mlp.gate.weight": "model-00022-of-00032.safetensors",
582
+ "model.llm.layers.20.attn.rel_logits_proj.proj": "model-00022-of-00032.safetensors",
583
+ "model.llm.layers.20.attn.wk_dv.weight": "model-00022-of-00032.safetensors",
584
+ "model.llm.layers.21.attn.k_norm.weight": "model-00022-of-00032.safetensors",
585
+ "model.llm.layers.21.attn.wk_dv.weight": "model-00022-of-00032.safetensors",
586
+ "model.llm.layers.21.mlp_norm.weight": "model-00022-of-00032.safetensors",
587
+ "model.llm.layers.22.attn.k_norm.weight": "model-00022-of-00032.safetensors",
588
+ "model.llm.layers.22.attn.wq_du.weight": "model-00022-of-00032.safetensors",
589
+ "model.llm.layers.23.attn.wr_du.weight": "model-00022-of-00032.safetensors",
590
+ "model.llm.layers.25.attn.wo_ud.weight": "model-00022-of-00032.safetensors",
591
+ "model.llm.layers.25.mlp_norm.weight": "model-00022-of-00032.safetensors",
592
+ "model.llm.layers.25.mlp_sconv.weight": "model-00022-of-00032.safetensors",
593
+ "model.llm.layers.26.attn_norm.weight": "model-00022-of-00032.safetensors",
594
+ "model.llm.layers.27.attn.wk_dv.weight": "model-00022-of-00032.safetensors",
595
+ "model.llm.layers.27.attn_norm.weight": "model-00022-of-00032.safetensors",
596
+ "model.llm.layers.27.mlp.shared_experts.shared_w2_weight": "model-00022-of-00032.safetensors",
597
+ "model.llm.layers.29.attn.k_norm.weight": "model-00022-of-00032.safetensors",
598
+ "model.llm.layers.29.attn.wv_dv.weight": "model-00022-of-00032.safetensors",
599
+ "model.llm.layers.3.attn.k_sconv.weight": "model-00022-of-00032.safetensors",
600
+ "model.llm.layers.3.attn.wo_ud.weight": "model-00022-of-00032.safetensors",
601
+ "model.llm.layers.3.attn.wr_du.weight": "model-00022-of-00032.safetensors",
602
+ "model.llm.layers.31.attn.rel_logits_proj.proj": "model-00022-of-00032.safetensors",
603
+ "model.llm.layers.32.attn.wo_ud.weight": "model-00022-of-00032.safetensors",
604
+ "model.llm.layers.32.attn.wq_du.weight": "model-00022-of-00032.safetensors",
605
+ "model.llm.layers.33.attn.k_norm.weight": "model-00022-of-00032.safetensors",
606
+ "model.llm.layers.33.attn.rel_logits_proj.proj": "model-00022-of-00032.safetensors",
607
+ "model.llm.layers.33.attn_sconv.weight": "model-00022-of-00032.safetensors",
608
+ "model.llm.layers.34.attn.k_sconv.weight": "model-00022-of-00032.safetensors",
609
+ "model.llm.layers.34.attn_norm.weight": "model-00022-of-00032.safetensors",
610
+ "model.llm.layers.34.mlp.experts.w13_weight": "model-00022-of-00032.safetensors",
611
+ "model.llm.layers.34.mlp.experts.w13_weight_scale": "model-00022-of-00032.safetensors",
612
+ "model.llm.layers.35.attn.v_sconv.weight": "model-00022-of-00032.safetensors",
613
+ "model.llm.layers.36.attn.k_sconv.weight": "model-00022-of-00032.safetensors",
614
+ "model.llm.layers.36.mlp_sconv.weight": "model-00022-of-00032.safetensors",
615
+ "model.llm.layers.37.attn.wr_du.weight": "model-00022-of-00032.safetensors",
616
+ "model.llm.layers.37.mlp.shared_experts.shared_w2_weight": "model-00022-of-00032.safetensors",
617
+ "model.llm.layers.38.attn.rel_logits_proj.proj": "model-00022-of-00032.safetensors",
618
+ "model.llm.layers.38.mlp_sconv.weight": "model-00022-of-00032.safetensors",
619
+ "model.llm.layers.4.attn.wk_dv.weight": "model-00022-of-00032.safetensors",
620
+ "model.llm.layers.4.attn.wq_du.weight": "model-00022-of-00032.safetensors",
621
+ "model.llm.layers.40.attn.wr_du.weight": "model-00022-of-00032.safetensors",
622
+ "model.llm.layers.41.attn_norm.weight": "model-00022-of-00032.safetensors",
623
+ "model.llm.layers.5.mlp.shared_experts.shared_w13_weight": "model-00022-of-00032.safetensors",
624
+ "model.llm.layers.5.mlp.shared_experts.shared_w2_weight": "model-00022-of-00032.safetensors",
625
+ "model.llm.layers.7.mlp.experts.w2_weight": "model-00022-of-00032.safetensors",
626
+ "model.llm.layers.7.mlp.experts.w2_weight_scale": "model-00022-of-00032.safetensors",
627
+ "model.llm.layers.8.attn_sconv.weight": "model-00022-of-00032.safetensors",
628
+ "model.llm.layers.9.mlp_sconv.weight": "model-00022-of-00032.safetensors",
629
+ "model.llm.layers.15.mlp.gate.global_scale": "model-00020-of-00032.safetensors",
630
+ "model.llm.layers.11.attn.wq_du.weight": "model-00020-of-00032.safetensors",
631
+ "model.llm.layers.13.mlp_sconv.weight": "model-00020-of-00032.safetensors",
632
+ "model.llm.layers.17.attn.k_sconv.weight": "model-00020-of-00032.safetensors",
633
+ "model.llm.layers.22.attn.v_sconv.weight": "model-00020-of-00032.safetensors",
634
+ "model.llm.layers.29.mlp.experts.w13_weight": "model-00020-of-00032.safetensors",
635
+ "model.llm.layers.29.mlp.experts.w13_weight_scale": "model-00020-of-00032.safetensors",
636
+ "model.llm.layers.33.mlp.shared_experts.shared_w2_weight": "model-00020-of-00032.safetensors",
637
+ "model.llm.layers.36.mlp.experts.w13_weight": "model-00020-of-00032.safetensors",
638
+ "model.llm.layers.36.mlp.experts.w13_weight_scale": "model-00020-of-00032.safetensors",
639
+ "model.llm.layers.37.attn.wv_dv.weight": "model-00020-of-00032.safetensors",
640
+ "model.llm.layers.4.mlp_norm.weight": "model-00020-of-00032.safetensors",
641
+ "model.llm.layers.41.attn.wk_dv.weight": "model-00020-of-00032.safetensors",
642
+ "model.llm.layers.5.attn.k_norm.weight": "model-00020-of-00032.safetensors",
643
+ "model.llm.layers.7.mlp.shared_experts.shared_w13_weight": "model-00020-of-00032.safetensors",
644
+ "model.llm.layers.7.mlp_norm.weight": "model-00020-of-00032.safetensors",
645
+ "model.llm.layers.8.attn_norm.weight": "model-00020-of-00032.safetensors",
646
+ "model.llm.layers.11.attn.v_sconv.weight": "model-00024-of-00032.safetensors",
647
+ "model.llm.layers.15.attn.v_sconv.weight": "model-00024-of-00032.safetensors",
648
+ "model.llm.layers.2.attn.wr_du.weight": "model-00024-of-00032.safetensors",
649
+ "model.llm.layers.2.attn_sconv.weight": "model-00024-of-00032.safetensors",
650
+ "model.llm.layers.20.attn_sconv.weight": "model-00024-of-00032.safetensors",
651
+ "model.llm.layers.21.mlp.experts.w13_weight": "model-00024-of-00032.safetensors",
652
+ "model.llm.layers.21.mlp.experts.w13_weight_scale": "model-00024-of-00032.safetensors",
653
+ "model.llm.layers.21.mlp.shared_experts.shared_w13_weight": "model-00024-of-00032.safetensors",
654
+ "model.llm.layers.22.attn.rel_logits_proj.proj": "model-00024-of-00032.safetensors",
655
+ "model.llm.layers.22.mlp.experts.w13_weight": "model-00024-of-00032.safetensors",
656
+ "model.llm.layers.22.mlp.experts.w13_weight_scale": "model-00024-of-00032.safetensors",
657
+ "model.llm.layers.25.attn.k_norm.weight": "model-00024-of-00032.safetensors",
658
+ "model.llm.layers.29.attn.rel_logits_proj.proj": "model-00024-of-00032.safetensors",
659
+ "model.llm.layers.29.mlp.shared_experts.shared_w13_weight": "model-00024-of-00032.safetensors",
660
+ "model.llm.layers.3.mlp_sconv.weight": "model-00024-of-00032.safetensors",
661
+ "model.llm.layers.30.attn.wv_dv.weight": "model-00024-of-00032.safetensors",
662
+ "model.llm.layers.34.attn.k_norm.weight": "model-00024-of-00032.safetensors",
663
+ "model.llm.layers.36.attn.k_norm.weight": "model-00024-of-00032.safetensors",
664
+ "model.llm.layers.37.attn.k_norm.weight": "model-00024-of-00032.safetensors",
665
+ "model.llm.layers.38.attn_sconv.weight": "model-00024-of-00032.safetensors",
666
+ "model.llm.layers.38.mlp_norm.weight": "model-00024-of-00032.safetensors",
667
+ "model.llm.layers.39.attn.k_sconv.weight": "model-00024-of-00032.safetensors",
668
+ "model.llm.layers.39.attn.wv_dv.weight": "model-00024-of-00032.safetensors",
669
+ "model.llm.layers.39.mlp.gate.weight": "model-00024-of-00032.safetensors",
670
+ "model.llm.layers.7.mlp.gate.weight": "model-00024-of-00032.safetensors",
671
+ "model.llm.layers.8.mlp_sconv.weight": "model-00024-of-00032.safetensors",
672
+ "model.visual.final_norm.weight": "model-00024-of-00032.safetensors",
673
+ "model.llm.layers.10.mlp.gate.global_scale": "model-00023-of-00032.safetensors",
674
+ "model.llm.layers.39.mlp.gate.global_scale": "model-00023-of-00032.safetensors",
675
+ "model.llm.layers.41.mlp.gate.bias": "model-00023-of-00032.safetensors",
676
+ "model.llm.layers.10.attn.wv_dv.weight": "model-00023-of-00032.safetensors",
677
+ "model.llm.layers.10.mlp.shared_experts.shared_w13_weight": "model-00023-of-00032.safetensors",
678
+ "model.llm.layers.11.attn.wr_du.weight": "model-00023-of-00032.safetensors",
679
+ "model.llm.layers.12.attn.wv_dv.weight": "model-00023-of-00032.safetensors",
680
+ "model.llm.layers.20.mlp.experts.w13_weight": "model-00023-of-00032.safetensors",
681
+ "model.llm.layers.20.mlp.experts.w13_weight_scale": "model-00023-of-00032.safetensors",
682
+ "model.llm.layers.21.attn.wr_du.weight": "model-00023-of-00032.safetensors",
683
+ "model.llm.layers.22.mlp.gate.weight": "model-00023-of-00032.safetensors",
684
+ "model.llm.layers.23.attn.v_sconv.weight": "model-00023-of-00032.safetensors",
685
+ "model.llm.layers.23.mlp.shared_experts.shared_w2_weight": "model-00023-of-00032.safetensors",
686
+ "model.llm.layers.25.attn.q_norm.weight": "model-00023-of-00032.safetensors",
687
+ "model.llm.layers.26.mlp.shared_experts.shared_w2_weight": "model-00023-of-00032.safetensors",
688
+ "model.llm.layers.26.mlp_norm.weight": "model-00023-of-00032.safetensors",
689
+ "model.llm.layers.3.attn.wv_dv.weight": "model-00023-of-00032.safetensors",
690
+ "model.llm.layers.32.attn.k_norm.weight": "model-00023-of-00032.safetensors",
691
+ "model.llm.layers.32.attn_sconv.weight": "model-00023-of-00032.safetensors",
692
+ "model.llm.layers.33.attn.k_sconv.weight": "model-00023-of-00032.safetensors",
693
+ "model.llm.layers.37.attn.q_norm.weight": "model-00023-of-00032.safetensors",
694
+ "model.llm.layers.41.attn.q_norm.weight": "model-00023-of-00032.safetensors",
695
+ "model.llm.layers.5.mlp_sconv.weight": "model-00023-of-00032.safetensors",
696
+ "model.llm.layers.6.mlp.gate.weight": "model-00023-of-00032.safetensors",
697
+ "model.llm.layers.7.attn.wk_dv.weight": "model-00023-of-00032.safetensors",
698
+ "model.llm.layers.7.mlp.shared_experts.shared_w2_weight": "model-00023-of-00032.safetensors",
699
+ "model.llm.layers.7.mlp_sconv.weight": "model-00023-of-00032.safetensors",
700
+ "model.llm.layers.8.mlp.shared_experts.shared_w13_weight": "model-00023-of-00032.safetensors",
701
+ "model.llm.layers.9.mlp.experts.w13_weight": "model-00023-of-00032.safetensors",
702
+ "model.llm.layers.9.mlp.experts.w13_weight_scale": "model-00023-of-00032.safetensors",
703
+ "model.llm.layers.19.mlp.gate.global_scale": "model-00021-of-00032.safetensors",
704
+ "model.llm.layers.25.mlp.gate.bias": "model-00021-of-00032.safetensors",
705
+ "model.llm.layers.26.mlp.gate.bias": "model-00021-of-00032.safetensors",
706
+ "model.llm.layers.31.mlp.gate.bias": "model-00021-of-00032.safetensors",
707
+ "model.llm.layers.31.mlp.gate.global_scale": "model-00021-of-00032.safetensors",
708
+ "model.llm.layers.34.mlp.gate.bias": "model-00021-of-00032.safetensors",
709
+ "model.llm.layers.35.mlp.gate.global_scale": "model-00021-of-00032.safetensors",
710
+ "model.llm.layers.7.mlp.gate.bias": "model-00021-of-00032.safetensors",
711
+ "model.llm.layers.8.mlp.gate.bias": "model-00021-of-00032.safetensors",
712
+ "model.llm.layers.10.attn.wo_ud.weight": "model-00021-of-00032.safetensors",
713
+ "model.llm.layers.10.mlp.shared_experts.shared_w2_weight": "model-00021-of-00032.safetensors",
714
+ "model.llm.layers.11.attn.k_norm.weight": "model-00021-of-00032.safetensors",
715
+ "model.llm.layers.11.attn_norm.weight": "model-00021-of-00032.safetensors",
716
+ "model.llm.layers.14.attn.wr_du.weight": "model-00021-of-00032.safetensors",
717
+ "model.llm.layers.14.mlp_norm.weight": "model-00021-of-00032.safetensors",
718
+ "model.llm.layers.15.attn.k_norm.weight": "model-00021-of-00032.safetensors",
719
+ "model.llm.layers.15.attn.wv_dv.weight": "model-00021-of-00032.safetensors",
720
+ "model.llm.layers.16.attn_sconv.weight": "model-00021-of-00032.safetensors",
721
+ "model.llm.layers.18.mlp.experts.w13_weight": "model-00021-of-00032.safetensors",
722
+ "model.llm.layers.18.mlp.experts.w13_weight_scale": "model-00021-of-00032.safetensors",
723
+ "model.llm.layers.19.attn.rel_logits_proj.proj": "model-00021-of-00032.safetensors",
724
+ "model.llm.layers.2.attn.wk_dv.weight": "model-00021-of-00032.safetensors",
725
+ "model.llm.layers.20.attn_norm.weight": "model-00021-of-00032.safetensors",
726
+ "model.llm.layers.23.attn.rel_logits_proj.proj": "model-00021-of-00032.safetensors",
727
+ "model.llm.layers.24.attn.wk_dv.weight": "model-00021-of-00032.safetensors",
728
+ "model.llm.layers.24.mlp.experts.w2_weight": "model-00021-of-00032.safetensors",
729
+ "model.llm.layers.24.mlp.experts.w2_weight_scale": "model-00021-of-00032.safetensors",
730
+ "model.llm.layers.24.mlp_sconv.weight": "model-00021-of-00032.safetensors",
731
+ "model.llm.layers.25.attn.wv_dv.weight": "model-00021-of-00032.safetensors",
732
+ "model.llm.layers.26.attn.wr_du.weight": "model-00021-of-00032.safetensors",
733
+ "model.llm.layers.28.attn_norm.weight": "model-00021-of-00032.safetensors",
734
+ "model.llm.layers.28.mlp.shared_experts.shared_w2_weight": "model-00021-of-00032.safetensors",
735
+ "model.llm.layers.29.attn.k_sconv.weight": "model-00021-of-00032.safetensors",
736
+ "model.llm.layers.29.attn.wk_dv.weight": "model-00021-of-00032.safetensors",
737
+ "model.llm.layers.29.mlp_norm.weight": "model-00021-of-00032.safetensors",
738
+ "model.llm.layers.32.attn.wr_du.weight": "model-00021-of-00032.safetensors",
739
+ "model.llm.layers.32.attn.wv_dv.weight": "model-00021-of-00032.safetensors",
740
+ "model.llm.layers.33.mlp.shared_experts.shared_w13_weight": "model-00021-of-00032.safetensors",
741
+ "model.llm.layers.35.attn.k_norm.weight": "model-00021-of-00032.safetensors",
742
+ "model.llm.layers.35.attn.wo_ud.weight": "model-00021-of-00032.safetensors",
743
+ "model.llm.layers.35.mlp.shared_experts.shared_w13_weight": "model-00021-of-00032.safetensors",
744
+ "model.llm.layers.38.attn.k_sconv.weight": "model-00021-of-00032.safetensors",
745
+ "model.llm.layers.40.attn.rel_logits_proj.proj": "model-00021-of-00032.safetensors",
746
+ "model.llm.layers.41.mlp.shared_experts.shared_w2_weight": "model-00021-of-00032.safetensors",
747
+ "model.llm.layers.5.attn.wk_dv.weight": "model-00021-of-00032.safetensors",
748
+ "model.llm.layers.5.mlp.experts.w2_weight": "model-00021-of-00032.safetensors",
749
+ "model.llm.layers.5.mlp.experts.w2_weight_scale": "model-00021-of-00032.safetensors",
750
+ "model.llm.layers.6.attn.wk_dv.weight": "model-00021-of-00032.safetensors",
751
+ "model.llm.layers.6.attn.wq_du.weight": "model-00021-of-00032.safetensors",
752
+ "model.llm.layers.6.attn.wv_dv.weight": "model-00021-of-00032.safetensors",
753
+ "model.llm.layers.9.attn.k_sconv.weight": "model-00021-of-00032.safetensors",
754
+ "model.llm.layers.9.attn.wq_du.weight": "model-00021-of-00032.safetensors",
755
+ "model.llm.layers.30.mlp.gate.global_scale": "model-00025-of-00032.safetensors",
756
+ "model.llm.layers.0.attn.wk_dv.weight": "model-00025-of-00032.safetensors",
757
+ "model.llm.layers.10.attn.wr_du.weight": "model-00025-of-00032.safetensors",
758
+ "model.llm.layers.13.attn.wv_dv.weight": "model-00025-of-00032.safetensors",
759
+ "model.llm.layers.15.attn.wr_du.weight": "model-00025-of-00032.safetensors",
760
+ "model.llm.layers.15.mlp_norm.weight": "model-00025-of-00032.safetensors",
761
+ "model.llm.layers.16.mlp.experts.w2_weight": "model-00025-of-00032.safetensors",
762
+ "model.llm.layers.16.mlp.experts.w2_weight_scale": "model-00025-of-00032.safetensors",
763
+ "model.llm.layers.17.attn.wo_ud.weight": "model-00025-of-00032.safetensors",
764
+ "model.llm.layers.19.mlp.shared_experts.shared_w13_weight": "model-00025-of-00032.safetensors",
765
+ "model.llm.layers.19.mlp.shared_experts.shared_w2_weight": "model-00025-of-00032.safetensors",
766
+ "model.llm.layers.2.attn.k_norm.weight": "model-00025-of-00032.safetensors",
767
+ "model.llm.layers.2.attn.q_norm.weight": "model-00025-of-00032.safetensors",
768
+ "model.llm.layers.2.attn.v_sconv.weight": "model-00025-of-00032.safetensors",
769
+ "model.llm.layers.2.mlp.shared_experts.shared_w2_weight": "model-00025-of-00032.safetensors",
770
+ "model.llm.layers.20.attn.wr_du.weight": "model-00025-of-00032.safetensors",
771
+ "model.llm.layers.23.mlp_norm.weight": "model-00025-of-00032.safetensors",
772
+ "model.llm.layers.24.mlp_norm.weight": "model-00025-of-00032.safetensors",
773
+ "model.llm.layers.31.attn.q_norm.weight": "model-00025-of-00032.safetensors",
774
+ "model.llm.layers.31.mlp.experts.w2_weight": "model-00025-of-00032.safetensors",
775
+ "model.llm.layers.31.mlp.experts.w2_weight_scale": "model-00025-of-00032.safetensors",
776
+ "model.llm.layers.32.attn.v_sconv.weight": "model-00025-of-00032.safetensors",
777
+ "model.llm.layers.36.attn.wv_dv.weight": "model-00025-of-00032.safetensors",
778
+ "model.llm.layers.37.mlp.experts.w13_weight": "model-00025-of-00032.safetensors",
779
+ "model.llm.layers.37.mlp.experts.w13_weight_scale": "model-00025-of-00032.safetensors",
780
+ "model.llm.layers.38.attn.wk_dv.weight": "model-00025-of-00032.safetensors",
781
+ "model.llm.layers.41.attn.wo_ud.weight": "model-00025-of-00032.safetensors",
782
+ "model.llm.layers.5.attn.rel_logits_proj.proj": "model-00025-of-00032.safetensors",
783
+ "model.llm.layers.6.attn.rel_logits_proj.proj": "model-00025-of-00032.safetensors",
784
+ "model.llm.layers.9.attn.wo_ud.weight": "model-00025-of-00032.safetensors",
785
+ "model.llm.layers.9.attn.wv_dv.weight": "model-00025-of-00032.safetensors",
786
+ "model.llm.norm.weight": "model-00025-of-00032.safetensors",
787
+ "model.mtp.layers.0.embed_norm.weight": "mtp.safetensors",
788
+ "model.mtp.layers.0.hidden_norm.weight": "mtp.safetensors",
789
+ "model.mtp.layers.0.input_proj.weight": "mtp.safetensors",
790
+ "model.mtp.layers.0.transformer_block.attn.k_norm.weight": "mtp.safetensors",
791
+ "model.mtp.layers.0.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
792
+ "model.mtp.layers.0.transformer_block.attn.q_norm.weight": "mtp.safetensors",
793
+ "model.mtp.layers.0.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
794
+ "model.mtp.layers.0.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
795
+ "model.mtp.layers.0.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
796
+ "model.mtp.layers.0.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
797
+ "model.mtp.layers.0.transformer_block.attn.wq_du.weight": "mtp.safetensors",
798
+ "model.mtp.layers.0.transformer_block.attn.wr_du.weight": "mtp.safetensors",
799
+ "model.mtp.layers.0.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
800
+ "model.mtp.layers.0.transformer_block.attn_norm.weight": "mtp.safetensors",
801
+ "model.mtp.layers.0.transformer_block.attn_sconv.weight": "mtp.safetensors",
802
+ "model.mtp.layers.0.transformer_block.mlp.global_scale": "mtp.safetensors",
803
+ "model.mtp.layers.0.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
804
+ "model.mtp.layers.0.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
805
+ "model.mtp.layers.0.transformer_block.mlp_norm.weight": "mtp.safetensors",
806
+ "model.mtp.layers.0.transformer_block.mlp_sconv.weight": "mtp.safetensors",
807
+ "model.mtp.layers.1.embed_norm.weight": "mtp.safetensors",
808
+ "model.mtp.layers.1.hidden_norm.weight": "mtp.safetensors",
809
+ "model.mtp.layers.1.input_proj.weight": "mtp.safetensors",
810
+ "model.mtp.layers.1.transformer_block.attn.k_norm.weight": "mtp.safetensors",
811
+ "model.mtp.layers.1.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
812
+ "model.mtp.layers.1.transformer_block.attn.q_norm.weight": "mtp.safetensors",
813
+ "model.mtp.layers.1.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
814
+ "model.mtp.layers.1.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
815
+ "model.mtp.layers.1.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
816
+ "model.mtp.layers.1.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
817
+ "model.mtp.layers.1.transformer_block.attn.wq_du.weight": "mtp.safetensors",
818
+ "model.mtp.layers.1.transformer_block.attn.wr_du.weight": "mtp.safetensors",
819
+ "model.mtp.layers.1.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
820
+ "model.mtp.layers.1.transformer_block.attn_norm.weight": "mtp.safetensors",
821
+ "model.mtp.layers.1.transformer_block.attn_sconv.weight": "mtp.safetensors",
822
+ "model.mtp.layers.1.transformer_block.mlp.global_scale": "mtp.safetensors",
823
+ "model.mtp.layers.1.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
824
+ "model.mtp.layers.1.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
825
+ "model.mtp.layers.1.transformer_block.mlp_norm.weight": "mtp.safetensors",
826
+ "model.mtp.layers.1.transformer_block.mlp_sconv.weight": "mtp.safetensors",
827
+ "model.mtp.layers.2.embed_norm.weight": "mtp.safetensors",
828
+ "model.mtp.layers.2.hidden_norm.weight": "mtp.safetensors",
829
+ "model.mtp.layers.2.input_proj.weight": "mtp.safetensors",
830
+ "model.mtp.layers.2.transformer_block.attn.k_norm.weight": "mtp.safetensors",
831
+ "model.mtp.layers.2.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
832
+ "model.mtp.layers.2.transformer_block.attn.q_norm.weight": "mtp.safetensors",
833
+ "model.mtp.layers.2.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
834
+ "model.mtp.layers.2.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
835
+ "model.mtp.layers.2.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
836
+ "model.mtp.layers.2.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
837
+ "model.mtp.layers.2.transformer_block.attn.wq_du.weight": "mtp.safetensors",
838
+ "model.mtp.layers.2.transformer_block.attn.wr_du.weight": "mtp.safetensors",
839
+ "model.mtp.layers.2.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
840
+ "model.mtp.layers.2.transformer_block.attn_norm.weight": "mtp.safetensors",
841
+ "model.mtp.layers.2.transformer_block.attn_sconv.weight": "mtp.safetensors",
842
+ "model.mtp.layers.2.transformer_block.mlp.global_scale": "mtp.safetensors",
843
+ "model.mtp.layers.2.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
844
+ "model.mtp.layers.2.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
845
+ "model.mtp.layers.2.transformer_block.mlp_norm.weight": "mtp.safetensors",
846
+ "model.mtp.layers.2.transformer_block.mlp_sconv.weight": "mtp.safetensors",
847
+ "model.mtp.layers.3.embed_norm.weight": "mtp.safetensors",
848
+ "model.mtp.layers.3.hidden_norm.weight": "mtp.safetensors",
849
+ "model.mtp.layers.3.input_proj.weight": "mtp.safetensors",
850
+ "model.mtp.layers.3.transformer_block.attn.k_norm.weight": "mtp.safetensors",
851
+ "model.mtp.layers.3.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
852
+ "model.mtp.layers.3.transformer_block.attn.q_norm.weight": "mtp.safetensors",
853
+ "model.mtp.layers.3.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
854
+ "model.mtp.layers.3.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
855
+ "model.mtp.layers.3.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
856
+ "model.mtp.layers.3.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
857
+ "model.mtp.layers.3.transformer_block.attn.wq_du.weight": "mtp.safetensors",
858
+ "model.mtp.layers.3.transformer_block.attn.wr_du.weight": "mtp.safetensors",
859
+ "model.mtp.layers.3.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
860
+ "model.mtp.layers.3.transformer_block.attn_norm.weight": "mtp.safetensors",
861
+ "model.mtp.layers.3.transformer_block.attn_sconv.weight": "mtp.safetensors",
862
+ "model.mtp.layers.3.transformer_block.mlp.global_scale": "mtp.safetensors",
863
+ "model.mtp.layers.3.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
864
+ "model.mtp.layers.3.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
865
+ "model.mtp.layers.3.transformer_block.mlp_norm.weight": "mtp.safetensors",
866
+ "model.mtp.layers.3.transformer_block.mlp_sconv.weight": "mtp.safetensors",
867
+ "model.mtp.layers.4.embed_norm.weight": "mtp.safetensors",
868
+ "model.mtp.layers.4.hidden_norm.weight": "mtp.safetensors",
869
+ "model.mtp.layers.4.input_proj.weight": "mtp.safetensors",
870
+ "model.mtp.layers.4.transformer_block.attn.k_norm.weight": "mtp.safetensors",
871
+ "model.mtp.layers.4.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
872
+ "model.mtp.layers.4.transformer_block.attn.q_norm.weight": "mtp.safetensors",
873
+ "model.mtp.layers.4.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
874
+ "model.mtp.layers.4.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
875
+ "model.mtp.layers.4.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
876
+ "model.mtp.layers.4.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
877
+ "model.mtp.layers.4.transformer_block.attn.wq_du.weight": "mtp.safetensors",
878
+ "model.mtp.layers.4.transformer_block.attn.wr_du.weight": "mtp.safetensors",
879
+ "model.mtp.layers.4.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
880
+ "model.mtp.layers.4.transformer_block.attn_norm.weight": "mtp.safetensors",
881
+ "model.mtp.layers.4.transformer_block.attn_sconv.weight": "mtp.safetensors",
882
+ "model.mtp.layers.4.transformer_block.mlp.global_scale": "mtp.safetensors",
883
+ "model.mtp.layers.4.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
884
+ "model.mtp.layers.4.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
885
+ "model.mtp.layers.4.transformer_block.mlp_norm.weight": "mtp.safetensors",
886
+ "model.mtp.layers.4.transformer_block.mlp_sconv.weight": "mtp.safetensors",
887
+ "model.mtp.layers.5.embed_norm.weight": "mtp.safetensors",
888
+ "model.mtp.layers.5.hidden_norm.weight": "mtp.safetensors",
889
+ "model.mtp.layers.5.input_proj.weight": "mtp.safetensors",
890
+ "model.mtp.layers.5.transformer_block.attn.k_norm.weight": "mtp.safetensors",
891
+ "model.mtp.layers.5.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
892
+ "model.mtp.layers.5.transformer_block.attn.q_norm.weight": "mtp.safetensors",
893
+ "model.mtp.layers.5.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
894
+ "model.mtp.layers.5.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
895
+ "model.mtp.layers.5.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
896
+ "model.mtp.layers.5.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
897
+ "model.mtp.layers.5.transformer_block.attn.wq_du.weight": "mtp.safetensors",
898
+ "model.mtp.layers.5.transformer_block.attn.wr_du.weight": "mtp.safetensors",
899
+ "model.mtp.layers.5.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
900
+ "model.mtp.layers.5.transformer_block.attn_norm.weight": "mtp.safetensors",
901
+ "model.mtp.layers.5.transformer_block.attn_sconv.weight": "mtp.safetensors",
902
+ "model.mtp.layers.5.transformer_block.mlp.global_scale": "mtp.safetensors",
903
+ "model.mtp.layers.5.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
904
+ "model.mtp.layers.5.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
905
+ "model.mtp.layers.5.transformer_block.mlp_norm.weight": "mtp.safetensors",
906
+ "model.mtp.layers.5.transformer_block.mlp_sconv.weight": "mtp.safetensors",
907
+ "model.mtp.layers.6.embed_norm.weight": "mtp.safetensors",
908
+ "model.mtp.layers.6.hidden_norm.weight": "mtp.safetensors",
909
+ "model.mtp.layers.6.input_proj.weight": "mtp.safetensors",
910
+ "model.mtp.layers.6.transformer_block.attn.k_norm.weight": "mtp.safetensors",
911
+ "model.mtp.layers.6.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
912
+ "model.mtp.layers.6.transformer_block.attn.q_norm.weight": "mtp.safetensors",
913
+ "model.mtp.layers.6.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
914
+ "model.mtp.layers.6.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
915
+ "model.mtp.layers.6.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
916
+ "model.mtp.layers.6.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
917
+ "model.mtp.layers.6.transformer_block.attn.wq_du.weight": "mtp.safetensors",
918
+ "model.mtp.layers.6.transformer_block.attn.wr_du.weight": "mtp.safetensors",
919
+ "model.mtp.layers.6.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
920
+ "model.mtp.layers.6.transformer_block.attn_norm.weight": "mtp.safetensors",
921
+ "model.mtp.layers.6.transformer_block.attn_sconv.weight": "mtp.safetensors",
922
+ "model.mtp.layers.6.transformer_block.mlp.global_scale": "mtp.safetensors",
923
+ "model.mtp.layers.6.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
924
+ "model.mtp.layers.6.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
925
+ "model.mtp.layers.6.transformer_block.mlp_norm.weight": "mtp.safetensors",
926
+ "model.mtp.layers.6.transformer_block.mlp_sconv.weight": "mtp.safetensors",
927
+ "model.mtp.layers.7.embed_norm.weight": "mtp.safetensors",
928
+ "model.mtp.layers.7.hidden_norm.weight": "mtp.safetensors",
929
+ "model.mtp.layers.7.input_proj.weight": "mtp.safetensors",
930
+ "model.mtp.layers.7.transformer_block.attn.k_norm.weight": "mtp.safetensors",
931
+ "model.mtp.layers.7.transformer_block.attn.k_sconv.weight": "mtp.safetensors",
932
+ "model.mtp.layers.7.transformer_block.attn.q_norm.weight": "mtp.safetensors",
933
+ "model.mtp.layers.7.transformer_block.attn.rel_logits_proj.proj": "mtp.safetensors",
934
+ "model.mtp.layers.7.transformer_block.attn.v_sconv.weight": "mtp.safetensors",
935
+ "model.mtp.layers.7.transformer_block.attn.wk_dv.weight": "mtp.safetensors",
936
+ "model.mtp.layers.7.transformer_block.attn.wo_ud.weight": "mtp.safetensors",
937
+ "model.mtp.layers.7.transformer_block.attn.wq_du.weight": "mtp.safetensors",
938
+ "model.mtp.layers.7.transformer_block.attn.wr_du.weight": "mtp.safetensors",
939
+ "model.mtp.layers.7.transformer_block.attn.wv_dv.weight": "mtp.safetensors",
940
+ "model.mtp.layers.7.transformer_block.attn_norm.weight": "mtp.safetensors",
941
+ "model.mtp.layers.7.transformer_block.attn_sconv.weight": "mtp.safetensors",
942
+ "model.mtp.layers.7.transformer_block.mlp.global_scale": "mtp.safetensors",
943
+ "model.mtp.layers.7.transformer_block.mlp.w13_dn.weight": "mtp.safetensors",
944
+ "model.mtp.layers.7.transformer_block.mlp.w2_md.weight": "mtp.safetensors",
945
+ "model.mtp.layers.7.transformer_block.mlp_norm.weight": "mtp.safetensors",
946
+ "model.mtp.layers.7.transformer_block.mlp_sconv.weight": "mtp.safetensors",
947
+ "model.llm.layers.23.mlp.gate.bias": "model-00026-of-00032.safetensors",
948
+ "model.llm.layers.4.mlp.gate.global_scale": "model-00026-of-00032.safetensors",
949
+ "model.llm.layers.5.mlp.gate.bias": "model-00026-of-00032.safetensors",
950
+ "model.llm.layers.0.attn.wr_du.weight": "model-00026-of-00032.safetensors",
951
+ "model.llm.layers.0.attn.wv_dv.weight": "model-00026-of-00032.safetensors",
952
+ "model.llm.layers.1.attn.wo_ud.weight": "model-00026-of-00032.safetensors",
953
+ "model.llm.layers.1.attn.wv_dv.weight": "model-00026-of-00032.safetensors",
954
+ "model.llm.layers.1.mlp_norm.weight": "model-00026-of-00032.safetensors",
955
+ "model.llm.layers.10.mlp_sconv.weight": "model-00026-of-00032.safetensors",
956
+ "model.llm.layers.12.attn_sconv.weight": "model-00026-of-00032.safetensors",
957
+ "model.llm.layers.12.mlp.shared_experts.shared_w2_weight": "model-00026-of-00032.safetensors",
958
+ "model.llm.layers.14.attn_norm.weight": "model-00026-of-00032.safetensors",
959
+ "model.llm.layers.14.attn_sconv.weight": "model-00026-of-00032.safetensors",
960
+ "model.llm.layers.14.mlp.experts.w2_weight": "model-00026-of-00032.safetensors",
961
+ "model.llm.layers.14.mlp.experts.w2_weight_scale": "model-00026-of-00032.safetensors",
962
+ "model.llm.layers.16.attn.wr_du.weight": "model-00026-of-00032.safetensors",
963
+ "model.llm.layers.17.attn_norm.weight": "model-00026-of-00032.safetensors",
964
+ "model.llm.layers.18.attn.k_norm.weight": "model-00026-of-00032.safetensors",
965
+ "model.llm.layers.18.attn_norm.weight": "model-00026-of-00032.safetensors",
966
+ "model.llm.layers.19.attn.v_sconv.weight": "model-00026-of-00032.safetensors",
967
+ "model.llm.layers.19.attn_norm.weight": "model-00026-of-00032.safetensors",
968
+ "model.llm.layers.19.mlp_norm.weight": "model-00026-of-00032.safetensors",
969
+ "model.llm.layers.21.attn.wq_du.weight": "model-00026-of-00032.safetensors",
970
+ "model.llm.layers.21.attn.wv_dv.weight": "model-00026-of-00032.safetensors",
971
+ "model.llm.layers.23.attn.q_norm.weight": "model-00026-of-00032.safetensors",
972
+ "model.llm.layers.23.mlp.gate.weight": "model-00026-of-00032.safetensors",
973
+ "model.llm.layers.24.attn.rel_logits_proj.proj": "model-00026-of-00032.safetensors",
974
+ "model.llm.layers.24.attn.wv_dv.weight": "model-00026-of-00032.safetensors",
975
+ "model.llm.layers.25.attn.wq_du.weight": "model-00026-of-00032.safetensors",
976
+ "model.llm.layers.26.attn.wk_dv.weight": "model-00026-of-00032.safetensors",
977
+ "model.llm.layers.27.attn.wo_ud.weight": "model-00026-of-00032.safetensors",
978
+ "model.llm.layers.28.attn.v_sconv.weight": "model-00026-of-00032.safetensors",
979
+ "model.llm.layers.28.mlp.shared_experts.shared_w13_weight": "model-00026-of-00032.safetensors",
980
+ "model.llm.layers.29.attn.wq_du.weight": "model-00026-of-00032.safetensors",
981
+ "model.llm.layers.3.attn.k_norm.weight": "model-00026-of-00032.safetensors",
982
+ "model.llm.layers.30.attn.k_norm.weight": "model-00026-of-00032.safetensors",
983
+ "model.llm.layers.30.attn.q_norm.weight": "model-00026-of-00032.safetensors",
984
+ "model.llm.layers.30.mlp_sconv.weight": "model-00026-of-00032.safetensors",
985
+ "model.llm.layers.31.attn.k_norm.weight": "model-00026-of-00032.safetensors",
986
+ "model.llm.layers.31.mlp_norm.weight": "model-00026-of-00032.safetensors",
987
+ "model.llm.layers.33.attn.wv_dv.weight": "model-00026-of-00032.safetensors",
988
+ "model.llm.layers.33.mlp.experts.w2_weight": "model-00026-of-00032.safetensors",
989
+ "model.llm.layers.33.mlp.experts.w2_weight_scale": "model-00026-of-00032.safetensors",
990
+ "model.llm.layers.36.attn.wk_dv.weight": "model-00026-of-00032.safetensors",
991
+ "model.llm.layers.39.attn.rel_logits_proj.proj": "model-00026-of-00032.safetensors",
992
+ "model.llm.layers.39.attn.wr_du.weight": "model-00026-of-00032.safetensors",
993
+ "model.llm.layers.39.mlp.shared_experts.shared_w2_weight": "model-00026-of-00032.safetensors",
994
+ "model.llm.layers.4.attn.rel_logits_proj.proj": "model-00026-of-00032.safetensors",
995
+ "model.llm.layers.41.attn_sconv.weight": "model-00026-of-00032.safetensors",
996
+ "model.llm.layers.5.attn.q_norm.weight": "model-00026-of-00032.safetensors",
997
+ "model.llm.layers.5.mlp.experts.w13_weight": "model-00026-of-00032.safetensors",
998
+ "model.llm.layers.5.mlp.experts.w13_weight_scale": "model-00026-of-00032.safetensors",
999
+ "model.llm.layers.7.attn.k_sconv.weight": "model-00026-of-00032.safetensors",
1000
+ "model.llm.layers.8.mlp.shared_experts.shared_w2_weight": "model-00026-of-00032.safetensors",
1001
+ "model.visual.layers.norm_2.weight": "model-00026-of-00032.safetensors",
1002
+ "model.llm.layers.14.mlp.gate.bias": "model-00027-of-00032.safetensors",
1003
+ "model.llm.layers.17.mlp.gate.bias": "model-00027-of-00032.safetensors",
1004
+ "model.llm.layers.18.mlp.gate.global_scale": "model-00027-of-00032.safetensors",
1005
+ "model.llm.layers.1.attn.wq_du.weight": "model-00027-of-00032.safetensors",
1006
+ "model.llm.layers.10.mlp.experts.w2_weight": "model-00027-of-00032.safetensors",
1007
+ "model.llm.layers.10.mlp.experts.w2_weight_scale": "model-00027-of-00032.safetensors",
1008
+ "model.llm.layers.11.mlp_sconv.weight": "model-00027-of-00032.safetensors",
1009
+ "model.llm.layers.13.attn_norm.weight": "model-00027-of-00032.safetensors",
1010
+ "model.llm.layers.13.mlp.experts.w13_weight": "model-00027-of-00032.safetensors",
1011
+ "model.llm.layers.13.mlp.experts.w13_weight_scale": "model-00027-of-00032.safetensors",
1012
+ "model.llm.layers.13.mlp.experts.w2_weight": "model-00027-of-00032.safetensors",
1013
+ "model.llm.layers.13.mlp.experts.w2_weight_scale": "model-00027-of-00032.safetensors",
1014
+ "model.llm.layers.15.attn_sconv.weight": "model-00027-of-00032.safetensors",
1015
+ "model.llm.layers.18.mlp.shared_experts.shared_w13_weight": "model-00027-of-00032.safetensors",
1016
+ "model.llm.layers.20.mlp.gate.weight": "model-00027-of-00032.safetensors",
1017
+ "model.llm.layers.21.attn.k_sconv.weight": "model-00027-of-00032.safetensors",
1018
+ "model.llm.layers.21.mlp_sconv.weight": "model-00027-of-00032.safetensors",
1019
+ "model.llm.layers.24.attn.wo_ud.weight": "model-00027-of-00032.safetensors",
1020
+ "model.llm.layers.25.mlp.gate.weight": "model-00027-of-00032.safetensors",
1021
+ "model.llm.layers.30.attn.k_sconv.weight": "model-00027-of-00032.safetensors",
1022
+ "model.llm.layers.33.attn.q_norm.weight": "model-00027-of-00032.safetensors",
1023
+ "model.llm.layers.28.mlp.gate.bias": "model-00032-of-00032.safetensors",
1024
+ "model.llm.layers.29.mlp.gate.bias": "model-00032-of-00032.safetensors",
1025
+ "model.llm.layers.11.attn.q_norm.weight": "model-00032-of-00032.safetensors",
1026
+ "model.llm.layers.13.mlp.gate.weight": "model-00032-of-00032.safetensors",
1027
+ "model.llm.layers.15.attn.q_norm.weight": "model-00032-of-00032.safetensors",
1028
+ "model.llm.layers.17.mlp_norm.weight": "model-00032-of-00032.safetensors",
1029
+ "model.llm.layers.18.mlp.experts.w2_weight": "model-00032-of-00032.safetensors",
1030
+ "model.llm.layers.18.mlp.experts.w2_weight_scale": "model-00032-of-00032.safetensors",
1031
+ "model.llm.layers.25.attn_sconv.weight": "model-00032-of-00032.safetensors",
1032
+ "model.llm.layers.3.mlp.experts.w13_weight": "model-00032-of-00032.safetensors",
1033
+ "model.llm.layers.3.mlp.experts.w13_weight_scale": "model-00032-of-00032.safetensors",
1034
+ "model.llm.layers.30.attn.rel_logits_proj.proj": "model-00032-of-00032.safetensors",
1035
+ "model.llm.layers.30.attn.wk_dv.weight": "model-00032-of-00032.safetensors",
1036
+ "model.llm.layers.31.mlp.shared_experts.shared_w2_weight": "model-00032-of-00032.safetensors",
1037
+ "model.llm.layers.34.attn.wq_du.weight": "model-00032-of-00032.safetensors",
1038
+ "model.llm.layers.37.mlp.experts.w2_weight": "model-00032-of-00032.safetensors",
1039
+ "model.llm.layers.37.mlp.experts.w2_weight_scale": "model-00032-of-00032.safetensors",
1040
+ "model.llm.layers.40.mlp.shared_experts.shared_w2_weight": "model-00032-of-00032.safetensors",
1041
+ "model.llm.layers.41.attn.wq_du.weight": "model-00032-of-00032.safetensors",
1042
+ "model.llm.layers.41.mlp_sconv.weight": "model-00032-of-00032.safetensors",
1043
+ "model.llm.layers.5.attn.wq_du.weight": "model-00032-of-00032.safetensors",
1044
+ "model.llm.layers.7.attn.wq_du.weight": "model-00032-of-00032.safetensors",
1045
+ "model.llm.layers.7.attn_norm.weight": "model-00032-of-00032.safetensors",
1046
+ "model.llm.layers.8.attn.v_sconv.weight": "model-00032-of-00032.safetensors",
1047
+ "model.llm.layers.39.mlp.gate.bias": "model-00029-of-00032.safetensors",
1048
+ "model.llm.layers.1.mlp.w13_dn.weight": "model-00029-of-00032.safetensors",
1049
+ "model.llm.layers.10.attn_sconv.weight": "model-00029-of-00032.safetensors",
1050
+ "model.llm.layers.12.attn.k_norm.weight": "model-00029-of-00032.safetensors",
1051
+ "model.llm.layers.12.attn.wr_du.weight": "model-00029-of-00032.safetensors",
1052
+ "model.llm.layers.15.mlp.experts.w2_weight": "model-00029-of-00032.safetensors",
1053
+ "model.llm.layers.15.mlp.experts.w2_weight_scale": "model-00029-of-00032.safetensors",
1054
+ "model.llm.layers.21.attn.v_sconv.weight": "model-00029-of-00032.safetensors",
1055
+ "model.llm.layers.24.attn.k_norm.weight": "model-00029-of-00032.safetensors",
1056
+ "model.llm.layers.25.attn.wr_du.weight": "model-00029-of-00032.safetensors",
1057
+ "model.llm.layers.26.mlp.experts.w2_weight": "model-00029-of-00032.safetensors",
1058
+ "model.llm.layers.26.mlp.experts.w2_weight_scale": "model-00029-of-00032.safetensors",
1059
+ "model.llm.layers.27.attn.rel_logits_proj.proj": "model-00029-of-00032.safetensors",
1060
+ "model.llm.layers.28.mlp_norm.weight": "model-00029-of-00032.safetensors",
1061
+ "model.llm.layers.31.attn_sconv.weight": "model-00029-of-00032.safetensors",
1062
+ "model.llm.layers.35.attn.wv_dv.weight": "model-00029-of-00032.safetensors",
1063
+ "model.llm.layers.36.mlp.gate.weight": "model-00029-of-00032.safetensors",
1064
+ "model.llm.layers.37.mlp_norm.weight": "model-00029-of-00032.safetensors",
1065
+ "model.llm.layers.4.attn.k_norm.weight": "model-00029-of-00032.safetensors",
1066
+ "model.llm.layers.40.attn_norm.weight": "model-00029-of-00032.safetensors",
1067
+ "model.llm.layers.6.attn.k_sconv.weight": "model-00029-of-00032.safetensors",
1068
+ "model.llm.layers.6.attn_norm.weight": "model-00029-of-00032.safetensors",
1069
+ "model.llm.layers.7.mlp.experts.w13_weight": "model-00029-of-00032.safetensors",
1070
+ "model.llm.layers.7.mlp.experts.w13_weight_scale": "model-00029-of-00032.safetensors",
1071
+ "model.llm.layers.0.attn.rel_logits_proj.proj": "model-00028-of-00032.safetensors",
1072
+ "model.llm.layers.12.attn.wo_ud.weight": "model-00028-of-00032.safetensors",
1073
+ "model.llm.layers.14.attn.wq_du.weight": "model-00028-of-00032.safetensors",
1074
+ "model.llm.layers.16.attn.k_norm.weight": "model-00028-of-00032.safetensors",
1075
+ "model.llm.layers.25.attn.v_sconv.weight": "model-00028-of-00032.safetensors",
1076
+ "model.llm.layers.25.mlp.experts.w2_weight": "model-00028-of-00032.safetensors",
1077
+ "model.llm.layers.25.mlp.experts.w2_weight_scale": "model-00028-of-00032.safetensors",
1078
+ "model.llm.layers.27.attn.q_norm.weight": "model-00028-of-00032.safetensors",
1079
+ "model.llm.layers.28.attn.rel_logits_proj.proj": "model-00028-of-00032.safetensors",
1080
+ "model.llm.layers.28.attn.wq_du.weight": "model-00028-of-00032.safetensors",
1081
+ "model.llm.layers.3.attn_sconv.weight": "model-00028-of-00032.safetensors",
1082
+ "model.llm.layers.31.attn.wr_du.weight": "model-00028-of-00032.safetensors",
1083
+ "model.llm.layers.34.mlp_norm.weight": "model-00028-of-00032.safetensors",
1084
+ "model.llm.layers.35.attn.wr_du.weight": "model-00028-of-00032.safetensors",
1085
+ "model.llm.layers.36.mlp.experts.w2_weight": "model-00028-of-00032.safetensors",
1086
+ "model.llm.layers.36.mlp.experts.w2_weight_scale": "model-00028-of-00032.safetensors",
1087
+ "model.llm.layers.36.mlp.shared_experts.shared_w2_weight": "model-00028-of-00032.safetensors",
1088
+ "model.llm.layers.38.attn.wq_du.weight": "model-00028-of-00032.safetensors",
1089
+ "model.llm.layers.41.mlp.experts.w13_weight": "model-00028-of-00032.safetensors",
1090
+ "model.llm.layers.41.mlp.experts.w13_weight_scale": "model-00028-of-00032.safetensors",
1091
+ "model.llm.layers.6.mlp.shared_experts.shared_w2_weight": "model-00028-of-00032.safetensors",
1092
+ "model.llm.layers.8.attn.q_norm.weight": "model-00028-of-00032.safetensors",
1093
+ "model.llm.layers.8.attn.wo_ud.weight": "model-00028-of-00032.safetensors",
1094
+ "model.llm.layers.9.attn.wk_dv.weight": "model-00028-of-00032.safetensors",
1095
+ "model.llm.layers.17.mlp.gate.global_scale": "model-00031-of-00032.safetensors",
1096
+ "model.llm.layers.38.mlp.gate.bias": "model-00031-of-00032.safetensors",
1097
+ "model.llm.layers.10.mlp_norm.weight": "model-00031-of-00032.safetensors",
1098
+ "model.llm.layers.14.attn.wk_dv.weight": "model-00031-of-00032.safetensors",
1099
+ "model.llm.layers.20.mlp.experts.w2_weight": "model-00031-of-00032.safetensors",
1100
+ "model.llm.layers.20.mlp.experts.w2_weight_scale": "model-00031-of-00032.safetensors",
1101
+ "model.llm.layers.34.mlp.experts.w2_weight": "model-00031-of-00032.safetensors",
1102
+ "model.llm.layers.34.mlp.experts.w2_weight_scale": "model-00031-of-00032.safetensors",
1103
+ "model.llm.layers.4.attn.k_sconv.weight": "model-00031-of-00032.safetensors",
1104
+ "model.llm.layers.40.mlp.experts.w13_weight": "model-00031-of-00032.safetensors",
1105
+ "model.llm.layers.40.mlp.experts.w13_weight_scale": "model-00031-of-00032.safetensors",
1106
+ "model.llm.embed.weight": "model-00030-of-00032.safetensors",
1107
+ "model.llm.layers.0.attn.k_sconv.weight": "model-00030-of-00032.safetensors",
1108
+ "model.llm.layers.1.attn.v_sconv.weight": "model-00030-of-00032.safetensors",
1109
+ "model.llm.layers.10.attn.v_sconv.weight": "model-00030-of-00032.safetensors",
1110
+ "model.llm.layers.11.mlp.experts.w13_weight": "model-00030-of-00032.safetensors",
1111
+ "model.llm.layers.11.mlp.experts.w13_weight_scale": "model-00030-of-00032.safetensors",
1112
+ "model.llm.layers.12.mlp_sconv.weight": "model-00030-of-00032.safetensors",
1113
+ "model.llm.layers.14.attn.k_sconv.weight": "model-00030-of-00032.safetensors",
1114
+ "model.llm.layers.18.attn.rel_logits_proj.proj": "model-00030-of-00032.safetensors",
1115
+ "model.llm.layers.18.mlp.gate.weight": "model-00030-of-00032.safetensors",
1116
+ "model.llm.layers.21.attn.rel_logits_proj.proj": "model-00030-of-00032.safetensors",
1117
+ "model.llm.layers.22.attn.k_sconv.weight": "model-00030-of-00032.safetensors",
1118
+ "model.llm.layers.26.attn.k_sconv.weight": "model-00030-of-00032.safetensors",
1119
+ "model.llm.layers.26.attn_sconv.weight": "model-00030-of-00032.safetensors",
1120
+ "model.llm.layers.27.mlp.gate.weight": "model-00030-of-00032.safetensors",
1121
+ "model.llm.layers.3.attn_norm.weight": "model-00030-of-00032.safetensors",
1122
+ "model.llm.layers.30.attn.wq_du.weight": "model-00030-of-00032.safetensors",
1123
+ "model.llm.layers.31.mlp.experts.w13_weight": "model-00030-of-00032.safetensors",
1124
+ "model.llm.layers.31.mlp.experts.w13_weight_scale": "model-00030-of-00032.safetensors",
1125
+ "model.llm.layers.32.mlp_norm.weight": "model-00030-of-00032.safetensors",
1126
+ "model.llm.layers.34.attn_sconv.weight": "model-00030-of-00032.safetensors",
1127
+ "model.llm.layers.36.attn.rel_logits_proj.proj": "model-00030-of-00032.safetensors",
1128
+ "model.llm.layers.36.attn.v_sconv.weight": "model-00030-of-00032.safetensors",
1129
+ "model.llm.layers.39.attn_norm.weight": "model-00030-of-00032.safetensors",
1130
+ "model.llm.layers.4.attn_norm.weight": "model-00030-of-00032.safetensors",
1131
+ "model.llm.layers.9.attn_sconv.weight": "model-00030-of-00032.safetensors"
1132
+ }
1133
+ }
mtp.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aac3b906b63c529ef7b0c9fc724f7860b5d0de58af21d39bd2013580dac7d7a5
3
+ size 4463845360
processor_config.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_token": "<|unused_200053|>",
3
+ "audio_bos_token": "<|content_audio_input|>",
4
+ "dmel_max_value": 2.0,
5
+ "dmel_min_value": -7.0,
6
+ "feature_extractor": {
7
+ "audio_token_duration_s": 0.05,
8
+ "feature_extractor_type": "InklingFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 800,
11
+ "n_fft": 1600,
12
+ "padding_side": "right",
13
+ "padding_value": 0.0,
14
+ "return_attention_mask": true,
15
+ "sampling_rate": 16000,
16
+ "window_size": 1600,
17
+ "window_size_multiplier": 2.0
18
+ },
19
+ "image_processor": {
20
+ "do_convert_rgb": true,
21
+ "do_normalize": true,
22
+ "do_rescale": true,
23
+ "do_resize": true,
24
+ "image_mean": [
25
+ 0.48145466,
26
+ 0.4578275,
27
+ 0.40821073
28
+ ],
29
+ "image_processor_type": "InklingImageProcessor",
30
+ "image_std": [
31
+ 0.26862954,
32
+ 0.26130258,
33
+ 0.27577711
34
+ ],
35
+ "resample": 3,
36
+ "rescale_factor": 0.00392156862745098,
37
+ "size": {
38
+ "height": 40,
39
+ "width": 40
40
+ }
41
+ },
42
+ "image_token": "<|unused_200054|>",
43
+ "image_bos_token": "<|content_image|>",
44
+ "num_dmel_bins": 16,
45
+ "processor_class": "InklingProcessor"
46
+ }
quantize_quark.py ADDED
@@ -0,0 +1,1025 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #
2
+ # Copyright (C) 2023, Advanced Micro Devices, Inc. All rights reserved.
3
+ # SPDX-License-Identifier: MIT
4
+ #
5
+ # Adopted from https://github.com/amd/Quark/blob/release/0.12/examples/torch/language_modeling/llm_ptq/quantize_quark.py
6
+
7
+ import argparse
8
+ import json
9
+ import os
10
+ import sys
11
+ import warnings
12
+ from pathlib import Path
13
+
14
+ import torch
15
+ from huggingface_hub import snapshot_download
16
+
17
+ from quark.common.profiler import GlobalProfiler, ProfileStep
18
+ from quark.common.utils.log import ScreenLogger
19
+ from quark.torch import (
20
+ LLMTemplate,
21
+ ModelQuantizer,
22
+ RuntimeOptions,
23
+ export_gguf,
24
+ export_onnx,
25
+ export_safetensors,
26
+ import_model_from_safetensors,
27
+ load_params,
28
+ save_params,
29
+ )
30
+ from quark.torch.export.api import _move_quantizer_to_dict
31
+ from quark.torch.quantization.config.config import load_quant_algo_config_from_file
32
+ from quark.torch.quantization import file2file_quantization
33
+ from quark.torch.utils import TPDeviceManager
34
+
35
+ # TODO: Using sys.path.append is bad practice.
36
+ sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), "..")))
37
+
38
+ from quark.contrib.llm_eval import eval_model
39
+ from quark.torch.utils.llm import (
40
+ check_compatibility_before_quantization,
41
+ get_calib_dataloader,
42
+ get_model,
43
+ get_tokenizer,
44
+ maybe_save_preprocessors,
45
+ preprocess_for_quantization,
46
+ )
47
+
48
+ logger = ScreenLogger(__name__)
49
+ quark_is_linear_weight_tensor = file2file_quantization._is_linear_weight_tensor
50
+ quark_quantize_and_save_safetensor_shard = file2file_quantization._quantize_and_save_safetensor_shard
51
+
52
+ # set CUDA_VISIBLE_DEVICES for profiling
53
+ if "CUDA_VISIBLE_DEVICES" not in os.environ:
54
+ os.environ["CUDA_VISIBLE_DEVICES"] = "0"
55
+
56
+ # The code below demonstrates how to register custom model templates and
57
+ # quantization schemes. If you need to add support for a new model architecture
58
+ # or define custom quantization configurations, uncomment and modify this section.
59
+ #
60
+ # To use:
61
+ # 1. Uncomment the code below
62
+ # 2. Modify the templates and/or schemes to match your model's architecture and/or quantization scheme
63
+ # 3. Run quantize_quark.py with your custom --quant_scheme name if new quantization schemes are registered
64
+ #
65
+
66
+ # from quark.torch.quantization.config.config import (
67
+ # Int8PerTensorSpec,
68
+ # QLayerConfig,
69
+ # )
70
+
71
+ # # --- Custom Model Templates ---
72
+ # # Define templates for model architectures not in the built-in list.
73
+ # # Model: internlm/internlm2-chat-7b
74
+ # internlm2_template = LLMTemplate(
75
+ # model_type="internlm2",
76
+ # kv_layers_name=["*wqkv"],
77
+ # q_layer_name="*wqkv",
78
+ # exclude_layers_name=["lm_head"],
79
+ # )
80
+ # LLMTemplate.register_template(internlm2_template)
81
+ # print(f"[INFO]: Registered template '{internlm2_template.model_type}'")
82
+
83
+ if "inkling_mm_model" not in LLMTemplate.list_available():
84
+ inkling_template = LLMTemplate(
85
+ model_type="inkling_mm_model",
86
+ kv_layers_name=None,
87
+ q_layer_name=None,
88
+ exclude_layers_name=[],
89
+ )
90
+ LLMTemplate.register_template(inkling_template)
91
+ print("[INFO]: Registered template 'inkling_mm_model'")
92
+
93
+
94
+ def _is_inkling_file2file_weight_tensor(tensor_name: str) -> bool:
95
+ if quark_is_linear_weight_tensor(tensor_name):
96
+ return True
97
+
98
+ parts = tensor_name.split(".")
99
+ return (
100
+ len(parts) == 7
101
+ and parts[0] == "model"
102
+ and parts[1] == "llm"
103
+ and parts[2] == "layers"
104
+ and parts[3].isdigit()
105
+ and parts[4] == "mlp"
106
+ and parts[5] == "experts"
107
+ and parts[6] in ("w13_weight", "w2_weight")
108
+ )
109
+
110
+
111
+ def _is_inkling_routed_expert_weight(tensor_name: str) -> bool:
112
+ parts = tensor_name.split(".")
113
+ return (
114
+ len(parts) == 7
115
+ and parts[0] == "model"
116
+ and parts[1] == "llm"
117
+ and parts[2] == "layers"
118
+ and parts[3].isdigit()
119
+ and int(parts[3]) >= 3
120
+ and parts[4] == "mlp"
121
+ and parts[5] == "experts"
122
+ and parts[6] in ("w13_weight", "w2_weight")
123
+ )
124
+
125
+
126
+ def _inkling_expert_chunk_size() -> int:
127
+ raw = os.environ.get("INKLING_QUARK_EXPERT_CHUNK_SIZE", "8")
128
+ try:
129
+ chunk_size = int(raw)
130
+ except ValueError as exc:
131
+ raise ValueError(f"INKLING_QUARK_EXPERT_CHUNK_SIZE must be an integer, got {raw!r}") from exc
132
+ if chunk_size < 1:
133
+ raise ValueError(f"INKLING_QUARK_EXPERT_CHUNK_SIZE must be >= 1, got {chunk_size}")
134
+ return chunk_size
135
+
136
+
137
+ def _fp4_nonzero_code_fraction(packed_weight: torch.Tensor) -> float:
138
+ flat = packed_weight.detach().reshape(-1)
139
+ if flat.numel() == 0:
140
+ return 0.0
141
+ max_sample = 1_000_000
142
+ if flat.numel() > max_sample:
143
+ stride = (flat.numel() + max_sample - 1) // max_sample
144
+ flat = flat[::stride][:max_sample]
145
+ low = flat & 0x0F
146
+ high = (flat >> 4) & 0x0F
147
+ nonzero = ((low != 0) & (low != 8)).sum() + ((high != 0) & (high != 8)).sum()
148
+ return float(nonzero.detach().cpu().item()) / float(2 * flat.numel())
149
+
150
+
151
+ def _quantize_weight_tensor(
152
+ tensor: torch.Tensor,
153
+ tensor_name: str,
154
+ layer_name: str,
155
+ weight_config,
156
+ ) -> tuple[torch.Tensor, torch.Tensor]:
157
+ quantized_tensors: dict[str, torch.Tensor] = {}
158
+ file2file_quantization._single_stage_quantize_weight(
159
+ tensor=tensor,
160
+ tensor_name=tensor_name,
161
+ layer_name=layer_name,
162
+ weight_config=weight_config,
163
+ quantized_tensors=quantized_tensors,
164
+ output_weight_map=None,
165
+ safetensor_filename="",
166
+ )
167
+ return quantized_tensors[tensor_name].contiguous(), quantized_tensors[tensor_name + "_scale"].contiguous()
168
+
169
+
170
+ def _quantize_inkling_routed_expert_tensor(
171
+ tensor_name: str,
172
+ tensor: torch.Tensor,
173
+ layer_config,
174
+ ) -> tuple[torch.Tensor, torch.Tensor]:
175
+ weight_config = layer_config.weight
176
+ assert isinstance(weight_config, file2file_quantization.QTensorConfig), (
177
+ f"weight config for {tensor_name} must be QTensorConfig"
178
+ )
179
+ if tensor.dim() != 3:
180
+ raise ValueError(f"{tensor_name}: expected stacked expert tensor with 3 dims, got {tuple(tensor.shape)}")
181
+
182
+ num_experts = tensor.shape[0]
183
+ rows = tensor.shape[1]
184
+ chunk_size = min(_inkling_expert_chunk_size(), num_experts)
185
+ print(
186
+ "[INKLING-F2F] chunked MXFP4 quantization "
187
+ f"tensor={tensor_name} shape={tuple(tensor.shape)} dtype={tensor.dtype} "
188
+ f"chunk_size={chunk_size}"
189
+ )
190
+
191
+ packed_out = None
192
+ scale_out = None
193
+ for expert_start in range(0, num_experts, chunk_size):
194
+ expert_end = min(expert_start + chunk_size, num_experts)
195
+ chunk = tensor[expert_start:expert_end].contiguous()
196
+ packed_chunk, scale_chunk = _quantize_weight_tensor(
197
+ chunk,
198
+ tensor_name,
199
+ ".".join(tensor_name.split(".")[:-1]),
200
+ weight_config,
201
+ )
202
+ scale_chunk = scale_chunk.reshape((expert_end - expert_start) * rows, -1).contiguous()
203
+
204
+ if packed_out is None:
205
+ packed_out = torch.empty(
206
+ (num_experts, *packed_chunk.shape[1:]),
207
+ dtype=packed_chunk.dtype,
208
+ device=packed_chunk.device,
209
+ )
210
+ scale_out = torch.empty(
211
+ (num_experts * rows, scale_chunk.shape[1]),
212
+ dtype=scale_chunk.dtype,
213
+ device=scale_chunk.device,
214
+ )
215
+ print(
216
+ "[INKLING-F2F] allocated output "
217
+ f"tensor={tensor_name} packed_shape={tuple(packed_out.shape)} "
218
+ f"scale_shape={tuple(scale_out.shape)}"
219
+ )
220
+
221
+ packed_out[expert_start:expert_end].copy_(packed_chunk)
222
+ scale_out[expert_start * rows : expert_end * rows].copy_(scale_chunk)
223
+ nonzero_frac = _fp4_nonzero_code_fraction(packed_chunk)
224
+ scale_min = int(scale_chunk.detach().min().cpu().item())
225
+ scale_max = int(scale_chunk.detach().max().cpu().item())
226
+ print(
227
+ "[INKLING-F2F] chunk done "
228
+ f"tensor={tensor_name} experts={expert_start}:{expert_end} "
229
+ f"fp4_nonzero_code_frac={nonzero_frac:.6f} scale_min={scale_min} scale_max={scale_max}"
230
+ )
231
+ if nonzero_frac < 0.1:
232
+ print(
233
+ "[INKLING-F2F][WARN] suspicious mostly-zero FP4 chunk "
234
+ f"tensor={tensor_name} experts={expert_start}:{expert_end} "
235
+ f"fp4_nonzero_code_frac={nonzero_frac:.6f}"
236
+ )
237
+ del chunk, packed_chunk, scale_chunk
238
+ file2file_quantization._empty_cache_if_cuda(tensor.device)
239
+
240
+ assert packed_out is not None and scale_out is not None
241
+ sentinels = [0, 28, 29, 64, 127, 255]
242
+ for expert_id in sentinels:
243
+ if expert_id >= num_experts:
244
+ continue
245
+ nonzero_frac = _fp4_nonzero_code_fraction(packed_out[expert_id : expert_id + 1])
246
+ print(
247
+ "[INKLING-F2F] sentinel "
248
+ f"tensor={tensor_name} expert={expert_id} fp4_nonzero_code_frac={nonzero_frac:.6f}"
249
+ )
250
+ if nonzero_frac < 0.1:
251
+ print(
252
+ "[INKLING-F2F][WARN] suspicious mostly-zero sentinel "
253
+ f"tensor={tensor_name} expert={expert_id} fp4_nonzero_code_frac={nonzero_frac:.6f}"
254
+ )
255
+ return packed_out, scale_out
256
+
257
+
258
+ def _inkling_quantize_and_save_safetensor_shard(
259
+ safetensor_path: str,
260
+ export_path: str,
261
+ quant_config,
262
+ device: str | torch.device,
263
+ *,
264
+ keep_excluded_layers_as_original_model_state: bool,
265
+ model_dtype: torch.dtype,
266
+ keep_original_model_state_tensor_names_set: set[str] | None = None,
267
+ weight_converters: list | None = None,
268
+ output_weight_map: dict[str, str] | None = None,
269
+ input_scale_dict: dict[str, torch.Tensor] | None = None,
270
+ hf_model_config: dict | None = None,
271
+ source_weight_map: dict[str, str] | None = None,
272
+ scale_inv_cache: dict[str, torch.Tensor] | None = None,
273
+ presharded_weights: dict[str, int] | None = None,
274
+ **kwargs,
275
+ ) -> None:
276
+ if kwargs:
277
+ print(f"[INKLING-F2F] ignoring Quark shard kwargs: {sorted(kwargs)}")
278
+ safetensor_filename = os.path.basename(safetensor_path)
279
+ logger.info(f"Loading {safetensor_filename}...")
280
+ tensors = file2file_quantization._load_safetensor_with_recover(
281
+ safetensor_path=safetensor_path,
282
+ quant_config=quant_config,
283
+ device=device,
284
+ keep_excluded_layers_as_original_model_state=keep_excluded_layers_as_original_model_state,
285
+ hf_model_config=hf_model_config,
286
+ weight_map=source_weight_map,
287
+ scale_inv_cache=scale_inv_cache,
288
+ keep_original_model_state_tensor_names_set=keep_original_model_state_tensor_names_set,
289
+ model_dtype=model_dtype,
290
+ presharded_weights=presharded_weights,
291
+ )
292
+
293
+ if weight_converters:
294
+ tensors = file2file_quantization._apply_weight_converters(tensors, weight_converters)
295
+
296
+ quantized_tensors: dict[str, torch.Tensor] = {}
297
+
298
+ for tensor_name, tensor in tensors.items():
299
+ if tensor_name.endswith((".weight_packed", ".weight_scale", ".weight_shape")):
300
+ continue
301
+
302
+ if output_weight_map is not None:
303
+ output_weight_map[tensor_name] = safetensor_filename
304
+
305
+ layer_name = ".".join(tensor_name.split(".")[:-1])
306
+ layer_config = file2file_quantization._get_layer_quant_config_by_tensor_name(
307
+ tensor_name=tensor_name,
308
+ quant_config=quant_config,
309
+ tensor_loaded=tensor,
310
+ )
311
+
312
+ if layer_config is not None and _is_inkling_routed_expert_weight(tensor_name):
313
+ packed_weight, scale = _quantize_inkling_routed_expert_tensor(tensor_name, tensor, layer_config)
314
+ quantized_tensors[tensor_name] = packed_weight
315
+ quantized_tensors[tensor_name + "_scale"] = scale
316
+ if output_weight_map is not None:
317
+ output_weight_map[tensor_name + "_scale"] = safetensor_filename
318
+ elif layer_config is not None:
319
+ weight_config = layer_config.weight
320
+ assert isinstance(weight_config, file2file_quantization.QTensorConfig), (
321
+ f"weight config for {layer_name} must be QTensorConfig"
322
+ )
323
+ packed_weight, scale = _quantize_weight_tensor(tensor, tensor_name, layer_name, weight_config)
324
+ quantized_tensors[tensor_name] = packed_weight
325
+ quantized_tensors[tensor_name + "_scale"] = scale
326
+ if output_weight_map is not None:
327
+ output_weight_map[tensor_name + "_scale"] = safetensor_filename
328
+ if input_scale_dict is not None:
329
+ if layer_name in input_scale_dict:
330
+ input_scale_key = layer_name + ".input_scale"
331
+ quantized_tensors[input_scale_key] = input_scale_dict[layer_name].contiguous()
332
+ if output_weight_map is not None:
333
+ output_weight_map[input_scale_key] = safetensor_filename
334
+ else:
335
+ logger.warning(f"Input scale not found for layer: {layer_name}")
336
+ else:
337
+ quantized_tensors[tensor_name] = tensor
338
+
339
+ del tensors
340
+ file2file_quantization._empty_cache_if_cuda(device)
341
+
342
+ output_path = os.path.join(export_path, safetensor_filename)
343
+ file2file_quantization.save_file(quantized_tensors, output_path)
344
+ output_size_mb = os.path.getsize(output_path) / (1024 * 1024)
345
+ logger.info(f"Saved {safetensor_filename} ({output_size_mb:.1f}MB)")
346
+
347
+
348
+ def _patch_inkling_file2file_weight_matcher() -> None:
349
+ file2file_quantization._is_linear_weight_tensor = _is_inkling_file2file_weight_tensor
350
+ file2file_quantization._quantize_and_save_safetensor_shard = _inkling_quantize_and_save_safetensor_shard
351
+
352
+
353
+ def _inkling_exclude_layers(hf_model_config: dict) -> list[str]:
354
+ text_config = hf_model_config.get("text_config") or hf_model_config
355
+ num_layers = int(text_config["num_hidden_layers"])
356
+ dense_mlp_idx = int(text_config.get("dense_mlp_idx", 2))
357
+
358
+ if num_layers not in (42, 66) or dense_mlp_idx != 2:
359
+ raise RuntimeError(
360
+ f"Unexpected model config: num_hidden_layers={num_layers}, dense_mlp_idx={dense_mlp_idx}. "
361
+ )
362
+
363
+ return [
364
+ "model.audio*",
365
+ "model.visual*",
366
+ "model.mtp*",
367
+ "model.llm.embed*",
368
+ "model.llm.unembed",
369
+ "model.llm.norm",
370
+ "model.llm.embed_norm",
371
+ "model.llm.layers.0.*",
372
+ "model.llm.layers.1.*",
373
+ "model.llm.layers.2.*",
374
+ "model.llm.layers.*.attn*",
375
+ "model.llm.layers.*.*sconv",
376
+ "model.llm.layers.*.mlp.gate",
377
+ "model.llm.layers.*.mlp.shared_experts*",
378
+ "model.llm.layers.*.*norm",
379
+ ]
380
+
381
+ # # --- Custom Quantization Schemes ---
382
+ # # Define custom quantization schemes using Quark's public QuantizationSpec classes.
383
+ # # These schemes can then be used via --quant_scheme <scheme_name>.
384
+ # # INT8 weight-only quantization
385
+ # int8_wo_scheme = QLayerConfig(weight=Int8PerTensorSpec().to_quantization_spec())
386
+ # LLMTemplate.register_scheme("int8_wo", config=int8_wo_scheme)
387
+ # print(f"[INFO]: Registered quantization scheme 'int8_wo'")
388
+
389
+
390
+ def _get_hf_model_config(model_dir: str) -> dict:
391
+ """Read config.json from the model directory without loading the model."""
392
+ config_path = os.path.join(model_dir, "config.json")
393
+ with open(config_path) as f:
394
+ return json.load(f)
395
+
396
+
397
+ def _build_quant_config(args: argparse.Namespace, model_config_type: str):
398
+ """Build quant_config from args and model_config_type (shared by normal and file-to-file paths)."""
399
+ if model_config_type not in LLMTemplate.list_available():
400
+ error_msg = (
401
+ f"\n[ERROR]: Model type '{model_config_type}' is not supported.\n\n"
402
+ f"Available templates: {LLMTemplate.list_available()}\n\n"
403
+ f"To add support for this model, uncomment and modify the 'Custom Model Templates'\n"
404
+ f"section at the top of this file to register a template for '{model_config_type}'.\n"
405
+ )
406
+ raise ValueError(error_msg)
407
+ template = LLMTemplate.get(model_config_type)
408
+
409
+ # Load algorithm configs from files if provided
410
+ algo_configs = {}
411
+ if args.quant_algo_config_file is not None:
412
+ for algo_name, algo_config_file in args.quant_algo_config_file:
413
+ algo_configs[algo_name] = load_quant_algo_config_from_file(algo_config_file)
414
+ print(f"[INFO]: Loaded algorithm configuration for {algo_name} from {algo_config_file}.")
415
+
416
+ # Build layer_config if --layer_quant_scheme is provided
417
+ layer_config = {}
418
+ if args.layer_quant_scheme is not None:
419
+ for layer_info in args.layer_quant_scheme:
420
+ layer_name = layer_info[0]
421
+ layer_scheme = layer_info[1]
422
+ layer_config[layer_name] = layer_scheme
423
+
424
+ quant_config = template.get_config(
425
+ scheme=args.quant_scheme,
426
+ algorithm=args.quant_algo,
427
+ kv_cache_scheme=args.kv_cache_dtype,
428
+ min_kv_scale=args.min_kv_scale,
429
+ layer_config=layer_config,
430
+ attention_scheme=args.attention_dtype,
431
+ exclude_layers=args.exclude_layers,
432
+ algo_configs=algo_configs if algo_configs else None,
433
+ )
434
+ quant_config.keep_prequantized_layers = not args.no_keep_prequantized_layers
435
+ return quant_config
436
+
437
+
438
+ def main(args: argparse.Namespace) -> None:
439
+ if args.revision is not None and os.path.isdir(args.model_dir):
440
+ raise ValueError(
441
+ f"The argument --revision {args.revision} is not supported using a local directory: {args.model_dir}"
442
+ )
443
+ elif not os.path.isdir(args.model_dir):
444
+ args.model_dir = snapshot_download(args.model_dir, revision=args.revision)
445
+
446
+ # Initialize global profiler
447
+ profiler = GlobalProfiler(output_path=os.path.join(args.output_dir, "quark_profile.yaml"))
448
+
449
+ # File-to-file quantization mode: bypass model loading, calibration and quantization,
450
+ # directly quantize safetensors files shard-by-shard and export.
451
+ if args.file2file_quantization:
452
+ print("\n[INFO]: File-to-file quantization mode enabled.")
453
+ hf_model_config = _get_hf_model_config(args.model_dir)
454
+ architectures = hf_model_config.get("architectures", [])
455
+ model_config_type = hf_model_config.get("model_type", architectures[0] if architectures else None)
456
+ if model_config_type == "inkling_mm_model":
457
+ _patch_inkling_file2file_weight_matcher()
458
+ args.exclude_layers = _inkling_exclude_layers(hf_model_config)
459
+ print(
460
+ f"[INFO]: Using hardcoded Inkling exclude_layers "
461
+ f"({len(args.exclude_layers)} patterns)."
462
+ )
463
+ num_layers = int((hf_model_config.get("text_config") or hf_model_config)["num_hidden_layers"])
464
+ print(f"[INFO]: Inkling file-to-file matcher will quantize routed experts in layers 3-{num_layers - 1}.")
465
+ quant_config = _build_quant_config(args, model_config_type)
466
+
467
+ print("\n[INFO]: Quantizing safetensors shards directly (file-to-file) ...")
468
+
469
+ weight_converters = LLMTemplate.get(model_config_type).f2f_weight_converters
470
+ if weight_converters:
471
+ logger.info(f"Applying {len(weight_converters)} weight converter(s) for model type '{model_config_type}'")
472
+
473
+ with profiler.scope(ProfileStep.FILE_TO_FILE_QUANTIZATION):
474
+ quantizer = ModelQuantizer(quant_config)
475
+ quantizer.direct_quantize_checkpoint(
476
+ pretrained_model_path=args.model_dir,
477
+ save_path=args.output_dir,
478
+ weight_converters=weight_converters,
479
+ keep_excluded_layers_as_original_model_state=args.keep_excluded_layers_as_original_model_state,
480
+ )
481
+
482
+ print(f"[INFO]: File-to-file quantization output saved to {args.output_dir}")
483
+ return
484
+
485
+ # 1. Define original model
486
+ model = None
487
+ # Load the pretrained model for quantization or for reload later (the old way).
488
+ if not args.model_reload or args.import_model_dir:
489
+ print("\n[INFO]: Loading model ...")
490
+
491
+ # We currently use CPU memory to load large models because GPU memory is typically smaller.
492
+ # The model will be dispatched to different GPUs based on the total number of GPUs specified by torchrun --nproc-per-node.
493
+ # TODO:
494
+ # The current method results in high CPU memory consumption due to multiple copies of the same model.
495
+ # We plan to address this in the future by implementing a more efficient way to dispatch the model to devices.
496
+ if args.use_tp:
497
+ device = "cpu"
498
+ else:
499
+ device = args.device
500
+
501
+ try:
502
+ with profiler.scope(ProfileStep.MODEL_LOADING):
503
+ model, _ = get_model(
504
+ args.model_dir,
505
+ args.data_type,
506
+ device,
507
+ args.multi_gpu,
508
+ args.multi_device,
509
+ args.model_attn_implementation,
510
+ trust_remote_code=args.trust_remote_code,
511
+ )
512
+ except torch.OutOfMemoryError as exception:
513
+ if torch.cuda.device_count() <= 1:
514
+ raise torch.OutOfMemoryError(
515
+ f"Out of memory error when loading the model {args.model_dir}. Only one device visible; this model does not fit on a single GPU."
516
+ ) from exception
517
+ elif not args.multi_gpu:
518
+ raise torch.OutOfMemoryError(
519
+ f"Out of memory error when loading the model {args.model_dir}. Consider using `--multi_gpu` as {torch.cuda.device_count()} devices are available."
520
+ ) from exception
521
+ else:
522
+ raise torch.OutOfMemoryError(
523
+ f"Out of memory error when loading the model {args.model_dir}. The model does not fit even with `--multi_gpu` across {torch.cuda.device_count()} devices. Consider using file-to-file quantization with `--file2file_quantization`, or make more GPU memory available."
524
+ ) from exception
525
+
526
+ # Check model compatibility with current Transformers version
527
+ print("\n[INFO]: Checking model compatibility ...")
528
+ check_compatibility_before_quantization(model, raise_on_error=False)
529
+
530
+ if args.use_tp:
531
+ TPDeviceManager.tp_mesh_init()
532
+
533
+ # 2. (Optional) Reload quantized model
534
+ if args.params_load:
535
+ print("\nRestore quantized model from json and safetensors file ...")
536
+ model = load_params(model, json_path=args.json_path, safetensors_path=args.safetensors_path)
537
+ args.skip_quantization = True
538
+ elif args.model_reload:
539
+ # Use import_model_dir if provided (separate quantized checkpoint), otherwise model_dir is the checkpoint itself.
540
+ reload_dir = args.import_model_dir or args.model_dir
541
+ print("\nRestore quantized model from hf_format safetensors file ...")
542
+ model = import_model_from_safetensors(
543
+ model=model,
544
+ model_dir=reload_dir,
545
+ multi_device=args.multi_device,
546
+ trust_remote_code=args.trust_remote_code,
547
+ attn_implementation=args.model_attn_implementation,
548
+ device="cpu" if args.use_tp else args.device,
549
+ multi_gpu=args.multi_gpu,
550
+ )
551
+ args.skip_quantization = True
552
+
553
+ architectures = getattr(model.config, "architectures", None) or []
554
+ model_type = (
555
+ model.config.model_type
556
+ if hasattr(model.config, "model_type")
557
+ else (architectures[0] if architectures else None)
558
+ )
559
+ tokenizer = get_tokenizer(
560
+ args.model_dir, max_seq_len=args.seq_len, model_type=model_type, trust_remote_code=args.trust_remote_code
561
+ )
562
+
563
+ # Detect multimodality from the model config's sub-modality keys instead of a
564
+ # hardcoded model_type whitelist — every HF VLM/ALM config exposes one of these
565
+ # (vision_config / audio_config / image_config / video_config).
566
+ multimodal = any(
567
+ getattr(model.config, k, None) is not None
568
+ for k in ("vision_config", "audio_config", "image_config", "video_config")
569
+ )
570
+
571
+ if args.use_tp:
572
+ if TPDeviceManager._tp_mesh is not None:
573
+ _move_quantizer_to_dict(model.model)
574
+
575
+ device = TPDeviceManager._device
576
+ tp_mesh = TPDeviceManager._tp_mesh
577
+
578
+ model.tensor_parallel(tp_mesh)
579
+ model.to(device)
580
+ else:
581
+ warnings.warn(
582
+ "Quark tensor parallelism is not initialized properly. Please check the torchrun settings.",
583
+ UserWarning,
584
+ stacklevel=2,
585
+ )
586
+ return
587
+
588
+ # 3. Define calibration dataloader(still need this step for weight only and dynamic quantization in Quark for current version.)
589
+ print("\n[INFO]: Loading dataset ...")
590
+
591
+ # When the model is small, accelerate will place it on the last device
592
+ main_device = model.device if args.multi_gpu or args.multi_device else args.device
593
+
594
+ with profiler.scope(ProfileStep.DATASET_LOADING):
595
+ calib_dataloader = get_calib_dataloader(
596
+ dataset_name=args.dataset,
597
+ tokenizer=tokenizer,
598
+ batch_size=args.batch_size,
599
+ num_calib_data=args.num_calib_data,
600
+ seqlen=args.seq_len,
601
+ device=main_device,
602
+ )
603
+
604
+ # 4. Quantization
605
+ if not args.skip_quantization:
606
+ preprocess_for_quantization(model)
607
+
608
+ architectures = getattr(model.config, "architectures", None) or []
609
+ model_config_type = (
610
+ model.config.model_type
611
+ if hasattr(model.config, "model_type")
612
+ else (architectures[0] if architectures else None)
613
+ )
614
+
615
+ quant_config = _build_quant_config(args, model_config_type)
616
+
617
+ if getattr(args, "kv_cache_post_rope", False):
618
+ if hasattr(quant_config, "kv_cache_post_rope"):
619
+ quant_config.kv_cache_post_rope = True
620
+ else:
621
+ warnings.warn(
622
+ "--kv_cache_post_rope specified but quant_config has no 'kv_cache_post_rope' field; flag ignored.",
623
+ RuntimeWarning,
624
+ stacklevel=2,
625
+ )
626
+
627
+ # In-place replacement of model modules with quantized versions
628
+ quantizer = ModelQuantizer(quant_config, args.multi_device)
629
+ model = quantizer.quantize_model(model, calib_dataloader)
630
+ args.exclude_layers = quantizer.config.exclude
631
+
632
+ # After quantization, freeze models - moving from soft weights that are quantized on the fly
633
+ # to e.g. `QuantLinear.weight` actually holding the fake quantized weights.
634
+ runtime_options = None
635
+ if args.enable_native_inference:
636
+ runtime_options = RuntimeOptions(
637
+ native_linear_mode=args.native_linear_mode,
638
+ )
639
+ model = quantizer.freeze(model, runtime_options=runtime_options)
640
+
641
+ if args.model_export is not None:
642
+ # Save pre-processors (tokenizer, image processor, etc.).
643
+ export_dir = Path(args.output_dir)
644
+ export_dir.mkdir(parents=True, exist_ok=True)
645
+ maybe_save_preprocessors(
646
+ args.model_dir,
647
+ export_dir,
648
+ trust_remote_code=args.trust_remote_code,
649
+ )
650
+
651
+ if args.custom_mode != "quark" and args.export_weight_format == "fake_quantized":
652
+ raise ValueError("Exporting with 'fake_quantized' only supports custom_mode=quark")
653
+
654
+ # Export option 1: hugging-face safetensors format
655
+ if "hf_format" in args.model_export:
656
+ print("\n[INFO]: Exporting hugging face format safetensors...")
657
+ with profiler.scope(ProfileStep.EXPORT_HF_SAFETENSORS), torch.no_grad():
658
+ export_safetensors(
659
+ model=model,
660
+ output_dir=args.output_dir,
661
+ custom_mode=args.custom_mode,
662
+ weight_format=args.export_weight_format,
663
+ pack_method=args.pack_method,
664
+ )
665
+
666
+ # Export option 2: onnx
667
+ if "onnx" in args.model_export:
668
+ print("\n[INFO]: Exporting onnx graph...")
669
+ with profiler.scope(ProfileStep.EXPORT_ONNX), torch.inference_mode():
670
+ batch_iter = iter(calib_dataloader)
671
+ input_args = next(batch_iter)
672
+ if "uint4" in args.quant_scheme or "int4" in args.quant_scheme:
673
+ uint4_int4_flag = True
674
+ else:
675
+ uint4_int4_flag = False
676
+
677
+ export_onnx(
678
+ model=model, output_dir=args.output_dir, input_args=input_args, uint4_int4_flag=uint4_int4_flag
679
+ )
680
+
681
+ # Export option 3: gguf
682
+ if "gguf" in args.model_export:
683
+ print("\n[INFO]: Exporting gguf model...")
684
+ with profiler.scope(ProfileStep.EXPORT_GGUF), torch.inference_mode():
685
+ export_gguf(model, output_dir=args.output_dir, model_type=model_type, tokenizer_path=args.model_dir)
686
+
687
+ if args.torch_compile:
688
+ print("\n[INFO]: Calling PyTorch 2 torch.compile...")
689
+ # Note: The model after torch.compile may not be able to export to other format
690
+ model = torch.compile(model)
691
+
692
+ if args.params_save:
693
+ save_params(model, model_type=model_type, export_dir=args.save_dir)
694
+
695
+ if not args.skip_evaluation:
696
+ print("\n[INFO]: Evaluating ...")
697
+
698
+ with profiler.scope(ProfileStep.MODEL_EVALUATION):
699
+ args.use_ppl_eval_model = True
700
+ eval_model(
701
+ args,
702
+ model,
703
+ main_device,
704
+ save_metrics_to_csv=args.save_metrics_to_csv,
705
+ output_dir=args.metrics_output_dir,
706
+ multimodal=multimodal,
707
+ )
708
+
709
+ if args.use_tp:
710
+ TPDeviceManager.tp_cleanup()
711
+
712
+
713
+ if __name__ == "__main__":
714
+ parser = argparse.ArgumentParser(description=__doc__)
715
+ # Argument for model
716
+ parser.add_argument(
717
+ "--model_dir",
718
+ help="Specify where the HuggingFace model is. This example support Llama, OPT models",
719
+ required=True,
720
+ )
721
+ parser.add_argument(
722
+ "--revision",
723
+ help="HuggingFace Hub revision (branch, tag, or commit) to download when --model_dir is a Hub model ID. "
724
+ "Triggers snapshot_download so all files come from the same revision.",
725
+ default=None,
726
+ )
727
+ parser.add_argument("--device", help="Device for running the quantizer", default="cuda", choices=["cuda", "cpu"])
728
+ parser.add_argument(
729
+ "--multi_gpu",
730
+ nargs="?",
731
+ const="auto",
732
+ default=None,
733
+ choices=["auto", "balanced"],
734
+ help="Enable multi-GPU mode. 'auto': default accelerate device map. "
735
+ "'balanced': use auto-adjusted device map for better GPU memory balance.",
736
+ )
737
+ parser.add_argument(
738
+ "--model_attn_implementation",
739
+ help="The attention implementation to use in the model",
740
+ default="eager",
741
+ choices=["eager", "sdpa", "flash_attention_2"],
742
+ )
743
+ parser.add_argument(
744
+ "--multi_device",
745
+ action="store_true",
746
+ help="we allow you to use this mode to run a model quantization that exceeds the size of your gpu memory if you use args.multi_gpu and still run into OOM "
747
+ "now it only supports thr common quantization without algorithms, please note that this can lead to very slow quantization.",
748
+ )
749
+
750
+ # Argument for calibration dataset
751
+ parser.add_argument(
752
+ "--dataset",
753
+ help="Dataset for calibration",
754
+ default="pileval",
755
+ choices=[
756
+ "pileval",
757
+ "wikitext",
758
+ "cnn_dailymail",
759
+ "pileval_for_awq_benchmark",
760
+ "wikitext_for_gptq_benchmark",
761
+ "HuggingFaceH4/ultrachat_200k",
762
+ "ScienceQA",
763
+ ],
764
+ )
765
+ parser.add_argument(
766
+ "--data_type", help="Datatype of the model", default="auto", choices=["auto", "float16", "bfloat16", "float32"]
767
+ )
768
+ parser.add_argument("--seq_len", type=int, help="Sequence length of data", default=512)
769
+ parser.add_argument("--batch_size", help="Batch size for calibration.", type=int, default=1)
770
+ parser.add_argument("--num_calib_data", help="Number of samples for calibration.", type=int, default=512)
771
+
772
+ # Argument for quantization
773
+ parser.add_argument("--skip_quantization", action="store_true")
774
+ parser.add_argument(
775
+ "--file2file_quantization",
776
+ action="store_true",
777
+ help="Enable file-to-file quantization mode. Quantizes safetensors shards directly without loading the full model into memory. "
778
+ "Bypasses model loading, calibration, and standard quantization flow. Requires --model_export hf_format.",
779
+ )
780
+
781
+ parser.add_argument(
782
+ "--quant_scheme",
783
+ help="Quantization scheme to use. Supported schemes: all built-in schemes and custom schemes registered."
784
+ "For the built-in schemes and their detailed configuration, see https://quark.docs.amd.com/latest/pytorch/user_guide_config_for_llm.html. "
785
+ "To register custom schemes, please uncomment and modify the 'Custom Quantization Schemes' section at the top of this file.",
786
+ choices=LLMTemplate.get_supported_schemes(),
787
+ default=None,
788
+ type=str,
789
+ )
790
+
791
+ parser.add_argument(
792
+ "--layer_quant_scheme",
793
+ action="append",
794
+ nargs=2,
795
+ metavar=("PATTERN", "QUANT_SCHEME"),
796
+ help="Directly specify a quantization scheme for layers matching the given pattern. "
797
+ "Can be repeated for multiple patterns. "
798
+ "Example: --quant_scheme int4_wo_128 --layer_quant_scheme lm_head int8 "
799
+ "(results in lm_head using int8 while other layers use int4_wo_128). "
800
+ "Supports wildcards: --layer_quant_scheme '*down_proj' fp8",
801
+ )
802
+
803
+ parser.add_argument(
804
+ "--kv_cache_dtype", "--kv_cache_quant_scheme", help="KV Cache dtype.", default=None, choices=["fp8", None]
805
+ )
806
+
807
+ parser.add_argument("--min_kv_scale", help="Minimum value of KV Cache scale.", type=float, default=0.0)
808
+ parser.add_argument(
809
+ "--kv_cache_post_rope",
810
+ action="store_true",
811
+ help="If set, quantize KV cache after RoPE (inside cache) instead of at k_proj/v_proj outputs.",
812
+ )
813
+ parser.add_argument(
814
+ "--attention_dtype", help="The dtype of attention quantization.", type=str, default=None, choices=["fp8"]
815
+ )
816
+ parser.add_argument(
817
+ "--quant_algo",
818
+ default=None,
819
+ type=lambda s: s.split(","),
820
+ metavar="alg1,alg2",
821
+ help="Comma-separated list of algorithms. Options include awq, gptq, smoothquant, rotation.",
822
+ )
823
+ parser.add_argument(
824
+ "--quant_algo_config_file",
825
+ action="append",
826
+ nargs=2,
827
+ metavar=("ALGO_NAME", "CONFIG_FILE"),
828
+ help="Specify a configuration file for a specific quantization algorithm. "
829
+ "Can be repeated for multiple algorithms. "
830
+ "Example: --quant_algo_config_file awq ./awq_config.json --quant_algo_config_file gptq ./gptq_config.json "
831
+ "(provides custom config files for AWQ and GPTQ algorithms).",
832
+ )
833
+
834
+ parser.add_argument(
835
+ "--exclude_layers",
836
+ type=str,
837
+ nargs="*", # Allows to pass a list of strings
838
+ default=None, # Default is None to allow model-specific layer exclusion
839
+ help='List of layers to exclude from quantization. Default depends on model type. Usage: `--exclude_layers "*down_proj*" "*31.fc*" "*k_proj"`. To avoid excluding layers at all, simply use `--exclude_layers` without any argument.',
840
+ )
841
+ parser.add_argument(
842
+ "--enable_native_inference",
843
+ action="store_true",
844
+ help="Enable native inference layer conversion during freeze().",
845
+ )
846
+ parser.add_argument(
847
+ "--native_linear_mode",
848
+ type=str,
849
+ default="auto",
850
+ choices=["auto", "fp8_per_tensor"],
851
+ help="Native linear implementation mode used when native inference is enabled.",
852
+ )
853
+
854
+ # Argument for reloading
855
+ parser.add_argument("--model_reload", help="safetensors or pth model reload", action="store_true")
856
+ parser.add_argument(
857
+ "--import_model_dir",
858
+ help="[Deprecated: use --model_dir instead] directory of hf or quark model, override model directory for reload, if not provided, --model_dir is used.",
859
+ )
860
+ parser.add_argument("--params_load", help="Model parameters load", action="store_true")
861
+ parser.add_argument("--json_path", help="Specify the path of saved json file")
862
+ parser.add_argument("--safetensors_path", help="Specify the path of saved safetensors file")
863
+
864
+ # Argument for export
865
+ parser.add_argument(
866
+ "--model_export",
867
+ help="Model export format",
868
+ default=None,
869
+ action="append",
870
+ choices=[None, "onnx", "hf_format", "gguf"],
871
+ )
872
+ parser.add_argument(
873
+ "--custom_mode",
874
+ help="When selecting `--custom_mode awq` or `--custom_mode fp8`, this legacy argument allows to export FP8 and AWQ models in the custom format they were exported with with quark<1.0, with custom config saved in the config.json, and config checkpoint format (AWQ uses `qzeros`, `qweight`, transposed `scales`).",
875
+ default="quark",
876
+ type=str,
877
+ choices=["quark", "awq", "fp8"],
878
+ )
879
+ parser.add_argument("--torch_compile", help="Model torch compile", action="store_true")
880
+ parser.add_argument(
881
+ "--pack_method", type=str, help="Pack method for awq_export", default="reorder", choices=["order", "reorder"]
882
+ )
883
+ parser.add_argument("--output_dir", default="exported_model")
884
+ parser.add_argument(
885
+ "--export_weight_format",
886
+ type=str,
887
+ help="Whether to export weights compressed or uncompressed",
888
+ default="real_quantized",
889
+ choices=["fake_quantized", "real_quantized"],
890
+ )
891
+ parser.add_argument(
892
+ "--no_keep_prequantized_layers",
893
+ action="store_true",
894
+ help="Force dequantization of excluded pre-quantized layers to bf16/fp16 on export. "
895
+ "By default (flag omitted), such layers are preserved in their original quantized format "
896
+ "(converted to Quark format); unsupported formats fall back to dequantization with a warning.",
897
+ )
898
+ parser.add_argument(
899
+ "--keep_excluded_layers_as_original_model_state",
900
+ action="store_true",
901
+ help="File-to-file mode only: keep already-quantized excluded layers (e.g. FP8 attention "
902
+ "in the official DeepSeek-V4 checkpoint) in their original on-disk format instead of "
903
+ "dequantizing them to bf16/fp16. Off by default; only enable for source checkpoints whose "
904
+ "quantization_config declares the excluded layers' format.",
905
+ )
906
+
907
+ # Argument for saving
908
+ parser.add_argument("--params_save", help="Model parameters save", action="store_true")
909
+ parser.add_argument(
910
+ "--save_dir",
911
+ help="Directory to save model parameters as safetensors or pth, in the case when --params_save is used.",
912
+ default="model_params",
913
+ )
914
+
915
+ # Argument for evaluation
916
+ parser.add_argument("--skip_evaluation", action="store_true")
917
+ parser.add_argument(
918
+ "--evaluation_dataset",
919
+ help="Dataset for evaluation",
920
+ default="wikitext",
921
+ choices=["wikitext", "wikitext_gpt_oss_120b", "wikitext_gpt_oss_20b"],
922
+ )
923
+ parser.add_argument("--use_ppl_eval_model", action="store_true")
924
+ parser.add_argument("--save_metrics_to_csv", action="store_true")
925
+ parser.add_argument("--metrics_output_dir", default="metrics_output_dir", help="Output path of csv with metrics.")
926
+ parser.add_argument(
927
+ "--tasks",
928
+ default=None,
929
+ type=str,
930
+ metavar="task1,task2",
931
+ help="Comma-separated list of task names or task groupings to evaluate on.",
932
+ )
933
+ parser.add_argument("--use_ppl_eval_for_kv_cache", action="store_true")
934
+ parser.add_argument(
935
+ "--ppl_eval_for_kv_cache_context_size",
936
+ type=int,
937
+ help="Context size used in PPL evaluation for KV cache.",
938
+ default=1024,
939
+ )
940
+ parser.add_argument(
941
+ "--ppl_eval_for_kv_cache_sample_size",
942
+ type=int,
943
+ help="Sample size used in PPL evaluation for KV cache.",
944
+ default=512,
945
+ )
946
+ parser.add_argument(
947
+ "--ppl_eval_for_kv_cache_patch_size",
948
+ type=int,
949
+ help="Patch size used in PPL evaluation for KV cache.",
950
+ default=None,
951
+ )
952
+ parser.add_argument(
953
+ "--eval_batch_size",
954
+ type=str,
955
+ default=1,
956
+ metavar="auto|auto:N|N",
957
+ help="Batch size used for evaluation. Acceptable values are 'auto', 'auto:N' or N, where N is a positive integer. Default is `1`.",
958
+ )
959
+ parser.add_argument(
960
+ "--max_eval_batch_size",
961
+ type=int,
962
+ default=64,
963
+ metavar="P",
964
+ help="Maximal batch size to try with `--batch_size auto`.",
965
+ )
966
+ parser.add_argument(
967
+ "--num_eval_data",
968
+ help="Number of samples for evaluation. The default value is -1, which means the entire dataset is used for evaluation.",
969
+ type=int,
970
+ default=-1,
971
+ )
972
+ parser.add_argument(
973
+ "--num_fewshot", type=int, default=None, metavar="N", help="Number of examples in few-shot context"
974
+ )
975
+ parser.add_argument(
976
+ "--apply_chat_template",
977
+ action="store_true",
978
+ help="Providing `--apply_chat_template` without an argument will apply the default chat template to the prompt.",
979
+ )
980
+ parser.add_argument("--use_mlperf_rouge", action="store_true")
981
+ parser.add_argument("--eval_data_dir", help="Dataset for evaluation", type=str, default=None)
982
+ parser.add_argument(
983
+ "--use_tp", action="store_true", help="Enable tensor parallelism exclusively for model evaluation."
984
+ )
985
+ group = parser.add_mutually_exclusive_group()
986
+ group.add_argument(
987
+ "--trust_remote_code",
988
+ action="store_true",
989
+ dest="trust_remote_code",
990
+ help="Enable execution of custom model code from the Hub (use only with repositories you fully trust).",
991
+ )
992
+ group.add_argument(
993
+ "--no_trust_remote_code",
994
+ action="store_false",
995
+ dest="trust_remote_code",
996
+ help="Disable execution of custom model code from the Hub (safer, recommended if unsure).",
997
+ )
998
+ parser.set_defaults(trust_remote_code=True)
999
+ args = parser.parse_args()
1000
+
1001
+ os.makedirs(args.output_dir, exist_ok=True)
1002
+
1003
+ if args.layer_quant_scheme is not None:
1004
+ for layer_info in args.layer_quant_scheme:
1005
+ if len(layer_info) != 2:
1006
+ raise ValueError(
1007
+ f"Invalid --layer_quant_scheme argument: {layer_info}. "
1008
+ f"Expected exactly 2 values (PATTERN, QUANT_SCHEME), but got {len(layer_info)}."
1009
+ )
1010
+
1011
+ if args.quant_algo_config_file is not None:
1012
+ for algo_config in args.quant_algo_config_file:
1013
+ if len(algo_config) != 2:
1014
+ raise ValueError(
1015
+ f"Invalid --quant_algo_config_file argument: {algo_config}. "
1016
+ f"Expected exactly 2 values (ALGO_NAME, CONFIG_FILE), but got {len(algo_config)}."
1017
+ )
1018
+ algo_name, config_file = algo_config
1019
+ if not os.path.isfile(config_file):
1020
+ raise ValueError(
1021
+ f"Configuration file '{config_file}' for algorithm '{algo_name}' does not exist. "
1022
+ f"Please provide a valid config file path."
1023
+ )
1024
+
1025
+ main(args)
special_tokens_map.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|endoftext|>",
4
+ "<|message_user|>",
5
+ "<|message_model|>",
6
+ "<|message_system|>",
7
+ "<|message_tool|>",
8
+ "<|content_text|>",
9
+ "<|content_image|>",
10
+ "<|content_model_end_sampling|>",
11
+ "<|content_thinking|>",
12
+ "<|end_message|>",
13
+ "<|content_audio_input|>",
14
+ "<|content_tool_error|>",
15
+ "<|audio|>",
16
+ "<|content_xml|>",
17
+ "<|begin_of_text|>",
18
+ "<|audio_end|>",
19
+ "<|content_invoke_tool_json|>",
20
+ "<|content_invoke_tool_text|>"
21
+ ]
22
+ }
tiktoken/tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc253fd2b702f7a6da7105eaa8f3463b2f1247e83614f23e5323b921088bed2a
3
+ size 3615874
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9fb6333a7db8fe5da90728e741e4a3ee4ac2ae12c5dd4958cc6f31688787d3c2
3
+ size 27875797
tokenizer_config.json ADDED
@@ -0,0 +1,508 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "199998": {
4
+ "content": "<|unused|>",
5
+ "single_word": false,
6
+ "lstrip": false,
7
+ "rstrip": false,
8
+ "normalized": false,
9
+ "special": true
10
+ },
11
+ "199999": {
12
+ "content": "<|endoftext|>",
13
+ "single_word": false,
14
+ "lstrip": false,
15
+ "rstrip": false,
16
+ "normalized": false,
17
+ "special": true
18
+ },
19
+ "200000": {
20
+ "content": "<|message_user|>",
21
+ "single_word": false,
22
+ "lstrip": false,
23
+ "rstrip": false,
24
+ "normalized": false,
25
+ "special": true
26
+ },
27
+ "200001": {
28
+ "content": "<|message_model|>",
29
+ "single_word": false,
30
+ "lstrip": false,
31
+ "rstrip": false,
32
+ "normalized": false,
33
+ "special": true
34
+ },
35
+ "200002": {
36
+ "content": "<|message_system|>",
37
+ "single_word": false,
38
+ "lstrip": false,
39
+ "rstrip": false,
40
+ "normalized": false,
41
+ "special": true
42
+ },
43
+ "200003": {
44
+ "content": "<|message_tool|>",
45
+ "single_word": false,
46
+ "lstrip": false,
47
+ "rstrip": false,
48
+ "normalized": false,
49
+ "special": true
50
+ },
51
+ "200004": {
52
+ "content": "<|content_text|>",
53
+ "single_word": false,
54
+ "lstrip": false,
55
+ "rstrip": false,
56
+ "normalized": false,
57
+ "special": true
58
+ },
59
+ "200005": {
60
+ "content": "<|content_image|>",
61
+ "single_word": false,
62
+ "lstrip": false,
63
+ "rstrip": false,
64
+ "normalized": false,
65
+ "special": true
66
+ },
67
+ "200006": {
68
+ "content": "<|content_model_end_sampling|>",
69
+ "single_word": false,
70
+ "lstrip": false,
71
+ "rstrip": false,
72
+ "normalized": false,
73
+ "special": true
74
+ },
75
+ "200007": {
76
+ "content": "<|unused_200007|>",
77
+ "single_word": false,
78
+ "lstrip": false,
79
+ "rstrip": false,
80
+ "normalized": false,
81
+ "special": true
82
+ },
83
+ "200008": {
84
+ "content": "<|content_thinking|>",
85
+ "single_word": false,
86
+ "lstrip": false,
87
+ "rstrip": false,
88
+ "normalized": false,
89
+ "special": true
90
+ },
91
+ "200009": {
92
+ "content": "<|unused_200009|>",
93
+ "single_word": false,
94
+ "lstrip": false,
95
+ "rstrip": false,
96
+ "normalized": false,
97
+ "special": true
98
+ },
99
+ "200010": {
100
+ "content": "<|end_message|>",
101
+ "single_word": false,
102
+ "lstrip": false,
103
+ "rstrip": false,
104
+ "normalized": false,
105
+ "special": true
106
+ },
107
+ "200011": {
108
+ "content": "<|unused_200011|>",
109
+ "single_word": false,
110
+ "lstrip": false,
111
+ "rstrip": false,
112
+ "normalized": false,
113
+ "special": true
114
+ },
115
+ "200012": {
116
+ "content": "<|unused_200012|>",
117
+ "single_word": false,
118
+ "lstrip": false,
119
+ "rstrip": false,
120
+ "normalized": false,
121
+ "special": true
122
+ },
123
+ "200013": {
124
+ "content": "<|unused_200013|>",
125
+ "single_word": false,
126
+ "lstrip": false,
127
+ "rstrip": false,
128
+ "normalized": false,
129
+ "special": true
130
+ },
131
+ "200014": {
132
+ "content": "<|unused_200014|>",
133
+ "single_word": false,
134
+ "lstrip": false,
135
+ "rstrip": false,
136
+ "normalized": false,
137
+ "special": true
138
+ },
139
+ "200015": {
140
+ "content": "<|unused_200015|>",
141
+ "single_word": false,
142
+ "lstrip": false,
143
+ "rstrip": false,
144
+ "normalized": false,
145
+ "special": true
146
+ },
147
+ "200016": {
148
+ "content": "<|unused_200016|>",
149
+ "single_word": false,
150
+ "lstrip": false,
151
+ "rstrip": false,
152
+ "normalized": false,
153
+ "special": true
154
+ },
155
+ "200017": {
156
+ "content": "<|unused_200017|>",
157
+ "single_word": false,
158
+ "lstrip": false,
159
+ "rstrip": false,
160
+ "normalized": false,
161
+ "special": true
162
+ },
163
+ "200018": {
164
+ "content": "<|unused_200018|>",
165
+ "single_word": false,
166
+ "lstrip": false,
167
+ "rstrip": false,
168
+ "normalized": false,
169
+ "special": true
170
+ },
171
+ "200019": {
172
+ "content": "<|unused_200019|>",
173
+ "single_word": false,
174
+ "lstrip": false,
175
+ "rstrip": false,
176
+ "normalized": false,
177
+ "special": true
178
+ },
179
+ "200020": {
180
+ "content": "<|content_audio_input|>",
181
+ "single_word": false,
182
+ "lstrip": false,
183
+ "rstrip": false,
184
+ "normalized": false,
185
+ "special": true
186
+ },
187
+ "200021": {
188
+ "content": "<|unused_200021|>",
189
+ "single_word": false,
190
+ "lstrip": false,
191
+ "rstrip": false,
192
+ "normalized": false,
193
+ "special": true
194
+ },
195
+ "200022": {
196
+ "content": "<|content_tool_error|>",
197
+ "single_word": false,
198
+ "lstrip": false,
199
+ "rstrip": false,
200
+ "normalized": false,
201
+ "special": true
202
+ },
203
+ "200023": {
204
+ "content": "<|audio|>",
205
+ "single_word": false,
206
+ "lstrip": false,
207
+ "rstrip": false,
208
+ "normalized": false,
209
+ "special": true
210
+ },
211
+ "200024": {
212
+ "content": "<|content_xml|>",
213
+ "single_word": false,
214
+ "lstrip": false,
215
+ "rstrip": false,
216
+ "normalized": false,
217
+ "special": true
218
+ },
219
+ "200025": {
220
+ "content": "<|unused_200025|>",
221
+ "single_word": false,
222
+ "lstrip": false,
223
+ "rstrip": false,
224
+ "normalized": false,
225
+ "special": true
226
+ },
227
+ "200026": {
228
+ "content": "<|unused_200026|>",
229
+ "single_word": false,
230
+ "lstrip": false,
231
+ "rstrip": false,
232
+ "normalized": false,
233
+ "special": true
234
+ },
235
+ "200027": {
236
+ "content": "<|unused_200027|>",
237
+ "single_word": false,
238
+ "lstrip": false,
239
+ "rstrip": false,
240
+ "normalized": false,
241
+ "special": true
242
+ },
243
+ "200028": {
244
+ "content": "<|begin_of_text|>",
245
+ "single_word": false,
246
+ "lstrip": false,
247
+ "rstrip": false,
248
+ "normalized": false,
249
+ "special": true
250
+ },
251
+ "200029": {
252
+ "content": "<|unused_200029|>",
253
+ "single_word": false,
254
+ "lstrip": false,
255
+ "rstrip": false,
256
+ "normalized": false,
257
+ "special": true
258
+ },
259
+ "200030": {
260
+ "content": "<|unused_200030|>",
261
+ "single_word": false,
262
+ "lstrip": false,
263
+ "rstrip": false,
264
+ "normalized": false,
265
+ "special": true
266
+ },
267
+ "200031": {
268
+ "content": "<|unused_200031|>",
269
+ "single_word": false,
270
+ "lstrip": false,
271
+ "rstrip": false,
272
+ "normalized": false,
273
+ "special": true
274
+ },
275
+ "200032": {
276
+ "content": "<|unused_200032|>",
277
+ "single_word": false,
278
+ "lstrip": false,
279
+ "rstrip": false,
280
+ "normalized": false,
281
+ "special": true
282
+ },
283
+ "200033": {
284
+ "content": "<|unused_200033|>",
285
+ "single_word": false,
286
+ "lstrip": false,
287
+ "rstrip": false,
288
+ "normalized": false,
289
+ "special": true
290
+ },
291
+ "200034": {
292
+ "content": "<|unused_200034|>",
293
+ "single_word": false,
294
+ "lstrip": false,
295
+ "rstrip": false,
296
+ "normalized": false,
297
+ "special": true
298
+ },
299
+ "200035": {
300
+ "content": "<|unused_200035|>",
301
+ "single_word": false,
302
+ "lstrip": false,
303
+ "rstrip": false,
304
+ "normalized": false,
305
+ "special": true
306
+ },
307
+ "200036": {
308
+ "content": "<|unused_200036|>",
309
+ "single_word": false,
310
+ "lstrip": false,
311
+ "rstrip": false,
312
+ "normalized": false,
313
+ "special": true
314
+ },
315
+ "200037": {
316
+ "content": "<|unused_200037|>",
317
+ "single_word": false,
318
+ "lstrip": false,
319
+ "rstrip": false,
320
+ "normalized": false,
321
+ "special": true
322
+ },
323
+ "200038": {
324
+ "content": "<|unused_200038|>",
325
+ "single_word": false,
326
+ "lstrip": false,
327
+ "rstrip": false,
328
+ "normalized": false,
329
+ "special": true
330
+ },
331
+ "200039": {
332
+ "content": "<|unused_200039|>",
333
+ "single_word": false,
334
+ "lstrip": false,
335
+ "rstrip": false,
336
+ "normalized": false,
337
+ "special": true
338
+ },
339
+ "200040": {
340
+ "content": "<|unused_200040|>",
341
+ "single_word": false,
342
+ "lstrip": false,
343
+ "rstrip": false,
344
+ "normalized": false,
345
+ "special": true
346
+ },
347
+ "200041": {
348
+ "content": "<|unused_200041|>",
349
+ "single_word": false,
350
+ "lstrip": false,
351
+ "rstrip": false,
352
+ "normalized": false,
353
+ "special": true
354
+ },
355
+ "200042": {
356
+ "content": "<|unused_200042|>",
357
+ "single_word": false,
358
+ "lstrip": false,
359
+ "rstrip": false,
360
+ "normalized": false,
361
+ "special": true
362
+ },
363
+ "200043": {
364
+ "content": "<|audio_end|>",
365
+ "single_word": false,
366
+ "lstrip": false,
367
+ "rstrip": false,
368
+ "normalized": false,
369
+ "special": true
370
+ },
371
+ "200044": {
372
+ "content": "<|unused_200044|>",
373
+ "single_word": false,
374
+ "lstrip": false,
375
+ "rstrip": false,
376
+ "normalized": false,
377
+ "special": true
378
+ },
379
+ "200045": {
380
+ "content": "<|unused_200045|>",
381
+ "single_word": false,
382
+ "lstrip": false,
383
+ "rstrip": false,
384
+ "normalized": false,
385
+ "special": true
386
+ },
387
+ "200046": {
388
+ "content": "<|unused_200046|>",
389
+ "single_word": false,
390
+ "lstrip": false,
391
+ "rstrip": false,
392
+ "normalized": false,
393
+ "special": true
394
+ },
395
+ "200047": {
396
+ "content": "<|unused_200047|>",
397
+ "single_word": false,
398
+ "lstrip": false,
399
+ "rstrip": false,
400
+ "normalized": false,
401
+ "special": true
402
+ },
403
+ "200048": {
404
+ "content": "<|unused_200048|>",
405
+ "single_word": false,
406
+ "lstrip": false,
407
+ "rstrip": false,
408
+ "normalized": false,
409
+ "special": true
410
+ },
411
+ "200049": {
412
+ "content": "<|content_invoke_tool_json|>",
413
+ "single_word": false,
414
+ "lstrip": false,
415
+ "rstrip": false,
416
+ "normalized": false,
417
+ "special": true
418
+ },
419
+ "200050": {
420
+ "content": "<|unused_200050|>",
421
+ "single_word": false,
422
+ "lstrip": false,
423
+ "rstrip": false,
424
+ "normalized": false,
425
+ "special": true
426
+ },
427
+ "200051": {
428
+ "content": "<|unused_200051|>",
429
+ "single_word": false,
430
+ "lstrip": false,
431
+ "rstrip": false,
432
+ "normalized": false,
433
+ "special": true
434
+ },
435
+ "200052": {
436
+ "content": "<|unused_200052|>",
437
+ "single_word": false,
438
+ "lstrip": false,
439
+ "rstrip": false,
440
+ "normalized": false,
441
+ "special": true
442
+ },
443
+ "200053": {
444
+ "content": "<|unused_200053|>",
445
+ "single_word": false,
446
+ "lstrip": false,
447
+ "rstrip": false,
448
+ "normalized": false,
449
+ "special": true
450
+ },
451
+ "200054": {
452
+ "content": "<|unused_200054|>",
453
+ "single_word": false,
454
+ "lstrip": false,
455
+ "rstrip": false,
456
+ "normalized": false,
457
+ "special": true
458
+ },
459
+ "200055": {
460
+ "content": "<|unused_200055|>",
461
+ "single_word": false,
462
+ "lstrip": false,
463
+ "rstrip": false,
464
+ "normalized": false,
465
+ "special": true
466
+ },
467
+ "200056": {
468
+ "content": "<|unused_200056|>",
469
+ "single_word": false,
470
+ "lstrip": false,
471
+ "rstrip": false,
472
+ "normalized": false,
473
+ "special": true
474
+ },
475
+ "200057": {
476
+ "content": "<|content_invoke_tool_text|>",
477
+ "single_word": false,
478
+ "lstrip": false,
479
+ "rstrip": false,
480
+ "normalized": false,
481
+ "special": true
482
+ }
483
+ },
484
+ "clean_up_tokenization_spaces": false,
485
+ "fix_mistral_regex": false,
486
+ "extra_special_tokens": {
487
+ "endoftext": "<|endoftext|>",
488
+ "message_user": "<|message_user|>",
489
+ "message_model": "<|message_model|>",
490
+ "message_system": "<|message_system|>",
491
+ "message_tool": "<|message_tool|>",
492
+ "content_text": "<|content_text|>",
493
+ "content_image": "<|content_image|>",
494
+ "content_model_end_sampling": "<|content_model_end_sampling|>",
495
+ "content_thinking": "<|content_thinking|>",
496
+ "end_message": "<|end_message|>",
497
+ "content_audio_input": "<|content_audio_input|>",
498
+ "content_tool_error": "<|content_tool_error|>",
499
+ "audio": "<|audio|>",
500
+ "content_xml": "<|content_xml|>",
501
+ "begin_of_text": "<|begin_of_text|>",
502
+ "audio_end": "<|audio_end|>",
503
+ "content_invoke_tool_json": "<|content_invoke_tool_json|>",
504
+ "content_invoke_tool_text": "<|content_invoke_tool_text|>"
505
+ },
506
+ "model_max_length": 1000000000000000019884624838656,
507
+ "tokenizer_class": "PreTrainedTokenizerFast"
508
+ }