Add config.json from PathFinderKR/gpt-oss-20b-expert0
Browse files- config.json +78 -0
config.json
ADDED
|
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"GptOssForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": true,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"dtype": "bfloat16",
|
| 8 |
+
"eos_token_id": 200002,
|
| 9 |
+
"experts_per_token": 4,
|
| 10 |
+
"head_dim": 64,
|
| 11 |
+
"hidden_act": "silu",
|
| 12 |
+
"hidden_size": 2880,
|
| 13 |
+
"initial_context_length": 4096,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 2880,
|
| 16 |
+
"layer_types": [
|
| 17 |
+
"sliding_attention",
|
| 18 |
+
"full_attention",
|
| 19 |
+
"sliding_attention",
|
| 20 |
+
"full_attention",
|
| 21 |
+
"sliding_attention",
|
| 22 |
+
"full_attention",
|
| 23 |
+
"sliding_attention",
|
| 24 |
+
"full_attention",
|
| 25 |
+
"sliding_attention",
|
| 26 |
+
"full_attention",
|
| 27 |
+
"sliding_attention",
|
| 28 |
+
"full_attention",
|
| 29 |
+
"sliding_attention",
|
| 30 |
+
"full_attention",
|
| 31 |
+
"sliding_attention",
|
| 32 |
+
"full_attention",
|
| 33 |
+
"sliding_attention",
|
| 34 |
+
"full_attention",
|
| 35 |
+
"sliding_attention",
|
| 36 |
+
"full_attention",
|
| 37 |
+
"sliding_attention",
|
| 38 |
+
"full_attention",
|
| 39 |
+
"sliding_attention",
|
| 40 |
+
"full_attention"
|
| 41 |
+
],
|
| 42 |
+
"max_position_embeddings": 131072,
|
| 43 |
+
"model_type": "gpt_oss",
|
| 44 |
+
"num_attention_heads": 64,
|
| 45 |
+
"num_experts_per_tok": 1,
|
| 46 |
+
"num_hidden_layers": 24,
|
| 47 |
+
"num_key_value_heads": 8,
|
| 48 |
+
"num_local_experts": 1,
|
| 49 |
+
"output_router_logits": false,
|
| 50 |
+
"pad_token_id": 199999,
|
| 51 |
+
"quantization_config": {
|
| 52 |
+
"modules_to_not_convert": [
|
| 53 |
+
"model.layers.*.self_attn",
|
| 54 |
+
"model.layers.*.mlp.router",
|
| 55 |
+
"model.embed_tokens",
|
| 56 |
+
"lm_head"
|
| 57 |
+
],
|
| 58 |
+
"quant_method": "mxfp4"
|
| 59 |
+
},
|
| 60 |
+
"rms_norm_eps": 1e-05,
|
| 61 |
+
"rope_scaling": {
|
| 62 |
+
"beta_fast": 32.0,
|
| 63 |
+
"beta_slow": 1.0,
|
| 64 |
+
"factor": 32.0,
|
| 65 |
+
"original_max_position_embeddings": 4096,
|
| 66 |
+
"rope_type": "yarn",
|
| 67 |
+
"truncate": false
|
| 68 |
+
},
|
| 69 |
+
"rope_theta": 150000,
|
| 70 |
+
"router_aux_loss_coef": 0.9,
|
| 71 |
+
"sliding_window": 128,
|
| 72 |
+
"swiglu_limit": 7.0,
|
| 73 |
+
"tie_word_embeddings": false,
|
| 74 |
+
"top_k": null,
|
| 75 |
+
"transformers_version": "4.56.0",
|
| 76 |
+
"use_cache": true,
|
| 77 |
+
"vocab_size": 201088
|
| 78 |
+
}
|