Update tiny models for DeepseekV4Model
Browse files- config.json +76 -0
- model.safetensors +3 -0
- tokenizer.json +0 -0
- tokenizer_config.json +14 -0
config.json
ADDED
|
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"DeepseekV4Model"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 0,
|
| 8 |
+
"compress_rates": {
|
| 9 |
+
"compressed_sparse_attention": 4,
|
| 10 |
+
"heavily_compressed_attention": 128
|
| 11 |
+
},
|
| 12 |
+
"compress_rope_theta": 160000.0,
|
| 13 |
+
"dtype": "float32",
|
| 14 |
+
"eos_token_id": 1,
|
| 15 |
+
"hc_eps": 1e-06,
|
| 16 |
+
"hc_mult": 2,
|
| 17 |
+
"hc_sinkhorn_iters": 3,
|
| 18 |
+
"head_dim": 32,
|
| 19 |
+
"hidden_act": "gelu",
|
| 20 |
+
"hidden_size": 64,
|
| 21 |
+
"index_head_dim": 16,
|
| 22 |
+
"index_n_heads": 2,
|
| 23 |
+
"index_topk": 2,
|
| 24 |
+
"initializer_range": 0.02,
|
| 25 |
+
"layer_types": [
|
| 26 |
+
"heavily_compressed_attention",
|
| 27 |
+
"compressed_sparse_attention"
|
| 28 |
+
],
|
| 29 |
+
"max_position_embeddings": 200,
|
| 30 |
+
"mlp_bias": false,
|
| 31 |
+
"mlp_layer_types": [
|
| 32 |
+
"moe",
|
| 33 |
+
"moe"
|
| 34 |
+
],
|
| 35 |
+
"model_type": "deepseek_v4",
|
| 36 |
+
"moe_intermediate_size": 64,
|
| 37 |
+
"n_routed_experts": 4,
|
| 38 |
+
"n_shared_experts": 1,
|
| 39 |
+
"norm_topk_prob": true,
|
| 40 |
+
"num_attention_heads": 4,
|
| 41 |
+
"num_experts_per_tok": 2,
|
| 42 |
+
"num_hidden_layers": 2,
|
| 43 |
+
"num_key_value_heads": 1,
|
| 44 |
+
"num_nextn_predict_layers": 0,
|
| 45 |
+
"o_groups": 2,
|
| 46 |
+
"o_lora_rank": 16,
|
| 47 |
+
"output_router_logits": false,
|
| 48 |
+
"pad_token_id": 1,
|
| 49 |
+
"partial_rotary_factor": 0.25,
|
| 50 |
+
"q_lora_rank": 32,
|
| 51 |
+
"qk_rope_head_dim": 8,
|
| 52 |
+
"rms_norm_eps": 1e-06,
|
| 53 |
+
"rope_parameters": {
|
| 54 |
+
"compress": {
|
| 55 |
+
"partial_rotary_factor": 0.25,
|
| 56 |
+
"rope_theta": 160000.0,
|
| 57 |
+
"rope_type": "default"
|
| 58 |
+
},
|
| 59 |
+
"main": {
|
| 60 |
+
"partial_rotary_factor": 0.25,
|
| 61 |
+
"rope_theta": 10000.0,
|
| 62 |
+
"rope_type": "default"
|
| 63 |
+
}
|
| 64 |
+
},
|
| 65 |
+
"rope_theta": 10000.0,
|
| 66 |
+
"routed_scaling_factor": 1.5,
|
| 67 |
+
"router_aux_loss_coef": 0.001,
|
| 68 |
+
"router_jitter_noise": 0.0,
|
| 69 |
+
"scoring_func": "sqrtsoftplus",
|
| 70 |
+
"sliding_window": 8,
|
| 71 |
+
"swiglu_limit": 10.0,
|
| 72 |
+
"tie_word_embeddings": false,
|
| 73 |
+
"transformers_version": "5.16.0.dev0",
|
| 74 |
+
"use_cache": true,
|
| 75 |
+
"vocab_size": 129280
|
| 76 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:33c816d3152785b0ca27568da3f5f3af1513b74b386438c2e7cad831dcbe4a2d
|
| 3 |
+
size 33803676
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"backend": "tokenizers",
|
| 3 |
+
"bos_token": "<｜begin▁of▁sentence｜>",
|
| 4 |
+
"clean_up_tokenization_spaces": false,
|
| 5 |
+
"eos_token": "<｜end▁of▁sentence｜>",
|
| 6 |
+
"is_local": true,
|
| 7 |
+
"legacy": true,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 200,
|
| 10 |
+
"pad_token": "<｜end▁of▁sentence｜>",
|
| 11 |
+
"sp_model_kwargs": {},
|
| 12 |
+
"tokenizer_class": "TokenizersBackend",
|
| 13 |
+
"unk_token": null
|
| 14 |
+
}
|