Update tiny models for CsmForConditionalGeneration

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

chat_template.jinja

@@ -0,0 +1,45 @@
+
+{%- for message in messages %}
+    {#-- Validate role is a stringified integer --#}
+    {%- if not message['role'] is string or not message['role'].isdigit() %}
+        {{- raise_exception("The role must be an integer or a stringified integer (e.g. '0') designating the speaker id") }}
+    {%- endif %}
+
+    {#-- Validate content is a list --#}
+    {%- set content = message['content'] %}
+    {%- if content is not iterable or content is string %}
+        {{- raise_exception("The content must be a list") }}
+    {%- endif %}
+
+    {#-- Collect content types --#}
+    {%- set content_types = content | map(attribute='type') | list %}
+    {%- set is_last = loop.last %}
+
+    {#-- Last message validation --#}
+    {%- if is_last %}
+        {%- if 'text' not in content_types %}
+            {{- raise_exception("The last message must include one item of type 'text'") }}
+        {%- elif (content_types | select('equalto', 'text') | list | length > 1) or (content_types | select('equalto', 'audio') | list | length > 1) %}
+            {{- raise_exception("At most two items are allowed in the last message: one 'text' and one 'audio'") }}
+        {%- endif %}
+
+    {#-- All other messages validation --#}
+    {%- else %}
+        {%- if content_types | select('equalto', 'text') | list | length != 1
+              or content_types | select('equalto', 'audio') | list | length != 1 %}
+            {{- raise_exception("Each message (except the last) must contain exactly one 'text' and one 'audio' item") }}
+        {%- elif content_types | reject('in', ['text', 'audio']) | list | length > 0 %}
+            {{- raise_exception("Only 'text' and 'audio' types are allowed in content") }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+
+{%- for message in messages %}
+    {{- bos_token }}
+    {{- '[' + message['role'] + ']' }}
+    {{- message['content'][0]['text'] }}
+    {{- eos_token }}
+    {%- if message['content']|length > 1 %}
+        {{- '<|AUDIO|><|audio_eos|>' }}
+    {%- endif %}
+{%- endfor %}

config.json

@@ -0,0 +1,114 @@
+{
+  "architectures": [
+    "CsmForConditionalGeneration"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "audio_eos_token_id": 128003,
+  "audio_token_id": 128002,
+  "bos_token_id": 128000,
+  "codebook_eos_token_id": 3,
+  "codebook_pad_token_id": 2,
+  "codec_config": {
+    "_frame_rate": null,
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "audio_channels": 1,
+    "chunk_in_sec": null,
+    "codebook_dim": 64,
+    "codebook_size": 64,
+    "compress": 2,
+    "dilation_growth_rate": 2,
+    "head_dim": 16,
+    "hidden_act": "gelu",
+    "hidden_size": 32,
+    "initializer_range": 0.02,
+    "intermediate_size": 2048,
+    "kernel_size": 7,
+    "last_kernel_size": 3,
+    "layer_scale_initial_scale": 0.01,
+    "max_position_embeddings": 8000,
+    "model_type": "mimi",
+    "norm_eps": 1e-05,
+    "num_attention_heads": 2,
+    "num_filters": 8,
+    "num_hidden_layers": 2,
+    "num_key_value_heads": 2,
+    "num_quantizers": 32,
+    "num_residual_layers": 1,
+    "num_semantic_quantizers": 1,
+    "pad_mode": "constant",
+    "residual_kernel_size": 3,
+    "rope_parameters": {
+      "rope_theta": 10000.0,
+      "rope_type": "default"
+    },
+    "sampling_rate": 24000,
+    "sliding_window": 4,
+    "tie_word_embeddings": true,
+    "trim_right_ratio": 1.0,
+    "upsample_groups": 32,
+    "upsampling_ratios": [
+      8,
+      4
+    ],
+    "use_cache": false,
+    "use_causal_conv": true,
+    "use_conv_shortcut": false,
+    "use_streaming": false,
+    "vector_quantization_hidden_dimension": 64
+  },
+  "depth_decoder_config": {
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "backbone_hidden_size": 64,
+    "bos_token_id": null,
+    "eos_token_id": null,
+    "head_dim": 16,
+    "hidden_act": "silu",
+    "hidden_size": 64,
+    "initializer_range": 0.02,
+    "intermediate_size": 128,
+    "max_position_embeddings": 10,
+    "mlp_bias": false,
+    "model_type": "csm_depth_decoder_model",
+    "num_attention_heads": 4,
+    "num_codebooks": 10,
+    "num_hidden_layers": 2,
+    "num_key_value_heads": 2,
+    "pad_token_id": null,
+    "rms_norm_eps": 1e-05,
+    "rope_parameters": {
+      "rope_theta": 500000.0,
+      "rope_type": "default"
+    },
+    "use_cache": true,
+    "vocab_size": 6
+  },
+  "dtype": "float32",
+  "eos_token_id": 128001,
+  "head_dim": 16,
+  "hidden_act": "silu",
+  "hidden_size": 64,
+  "initializer_range": 0.02,
+  "intermediate_size": 64,
+  "max_position_embeddings": 200,
+  "mlp_bias": false,
+  "model_type": "csm",
+  "num_attention_heads": 4,
+  "num_codebooks": 10,
+  "num_hidden_layers": 2,
+  "num_key_value_heads": 2,
+  "pad_token_id": 128001,
+  "rms_norm_eps": 1e-05,
+  "rope_parameters": {
+    "rope_theta": 500000.0,
+    "rope_type": "default"
+  },
+  "text_vocab_size": 99,
+  "tie_codebooks_embeddings": true,
+  "tie_word_embeddings": false,
+  "transformers_version": "5.16.0.dev0",
+  "use_cache": true,
+  "vocab_size": 128256
+}

generation_config.json

@@ -0,0 +1,14 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 128000,
+  "depth_decoder__from_model_config": true,
+  "depth_decoder_output_attentions": false,
+  "depth_decoder_output_hidden_states": false,
+  "depth_decoder_use_cache": true,
+  "eos_token_id": 128001,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "pad_token_id": 128001,
+  "transformers_version": "5.16.0.dev0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a50f20aec685c2520cddcf68d4df182a960d49f52cff183766a22bfd5a41d2b
+size 364634044

preprocessor_config.json

@@ -0,0 +1,10 @@
+{
+  "chunk_length_s": null,
+  "feature_extractor_type": "EncodecFeatureExtractor",
+  "feature_size": 1,
+  "overlap": null,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": true,
+  "sampling_rate": 24000
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80f511469b4aceda8c17ccafef2310eceeda9dd4589cf2c062bd0279eef5646d
+size 17209980

tokenizer_config.json

@@ -0,0 +1,16 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "is_local": true,
+  "local_files_only": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 200,
+  "pad_token": "<|end_of_text|>",
+  "processor_class": "CsmProcessor",
+  "tokenizer_class": "TokenizersBackend"
+}