Phase 2 full GPTQ + BF16 MTP: 89.1% MTP acceptance
Browse files- config.json +434 -0
- generation_config.json +9 -0
- model-00001-of-00004.safetensors +3 -0
- model-00002-of-00004.safetensors +3 -0
- model-00003-of-00004.safetensors +3 -0
- model-00004-of-00004.safetensors +3 -0
- model.safetensors.index.json +0 -0
- recipe.yaml +61 -0
- tokenizer.json +0 -0
- tokenizer_config.json +14 -0
config.json
ADDED
|
@@ -0,0 +1,434 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"DeepseekV4ForCausalLM"
|
| 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,
|
| 13 |
+
"dtype": "bfloat16",
|
| 14 |
+
"eos_token_id": 1,
|
| 15 |
+
"expert_dtype": "bf16",
|
| 16 |
+
"hc_eps": 1e-06,
|
| 17 |
+
"hc_mult": 4,
|
| 18 |
+
"hc_sinkhorn_iters": 20,
|
| 19 |
+
"head_dim": 512,
|
| 20 |
+
"hidden_act": "silu",
|
| 21 |
+
"hidden_size": 4096,
|
| 22 |
+
"index_head_dim": 128,
|
| 23 |
+
"index_n_heads": 64,
|
| 24 |
+
"index_topk": 512,
|
| 25 |
+
"initializer_range": 0.02,
|
| 26 |
+
"layer_types": [
|
| 27 |
+
"sliding_attention",
|
| 28 |
+
"sliding_attention",
|
| 29 |
+
"compressed_sparse_attention",
|
| 30 |
+
"heavily_compressed_attention",
|
| 31 |
+
"compressed_sparse_attention",
|
| 32 |
+
"heavily_compressed_attention",
|
| 33 |
+
"compressed_sparse_attention",
|
| 34 |
+
"heavily_compressed_attention",
|
| 35 |
+
"compressed_sparse_attention",
|
| 36 |
+
"heavily_compressed_attention",
|
| 37 |
+
"compressed_sparse_attention",
|
| 38 |
+
"heavily_compressed_attention",
|
| 39 |
+
"compressed_sparse_attention",
|
| 40 |
+
"heavily_compressed_attention",
|
| 41 |
+
"compressed_sparse_attention",
|
| 42 |
+
"heavily_compressed_attention",
|
| 43 |
+
"compressed_sparse_attention",
|
| 44 |
+
"heavily_compressed_attention",
|
| 45 |
+
"compressed_sparse_attention",
|
| 46 |
+
"heavily_compressed_attention",
|
| 47 |
+
"compressed_sparse_attention",
|
| 48 |
+
"heavily_compressed_attention",
|
| 49 |
+
"compressed_sparse_attention",
|
| 50 |
+
"heavily_compressed_attention",
|
| 51 |
+
"compressed_sparse_attention",
|
| 52 |
+
"heavily_compressed_attention",
|
| 53 |
+
"compressed_sparse_attention",
|
| 54 |
+
"heavily_compressed_attention",
|
| 55 |
+
"compressed_sparse_attention",
|
| 56 |
+
"heavily_compressed_attention",
|
| 57 |
+
"compressed_sparse_attention",
|
| 58 |
+
"heavily_compressed_attention",
|
| 59 |
+
"compressed_sparse_attention",
|
| 60 |
+
"heavily_compressed_attention",
|
| 61 |
+
"compressed_sparse_attention",
|
| 62 |
+
"heavily_compressed_attention",
|
| 63 |
+
"compressed_sparse_attention",
|
| 64 |
+
"heavily_compressed_attention",
|
| 65 |
+
"compressed_sparse_attention",
|
| 66 |
+
"heavily_compressed_attention",
|
| 67 |
+
"compressed_sparse_attention",
|
| 68 |
+
"heavily_compressed_attention",
|
| 69 |
+
"compressed_sparse_attention"
|
| 70 |
+
],
|
| 71 |
+
"max_position_embeddings": 1048576,
|
| 72 |
+
"mlp_bias": false,
|
| 73 |
+
"mlp_layer_types": [
|
| 74 |
+
"hash_moe",
|
| 75 |
+
"hash_moe",
|
| 76 |
+
"hash_moe",
|
| 77 |
+
"moe",
|
| 78 |
+
"moe",
|
| 79 |
+
"moe",
|
| 80 |
+
"moe",
|
| 81 |
+
"moe",
|
| 82 |
+
"moe",
|
| 83 |
+
"moe",
|
| 84 |
+
"moe",
|
| 85 |
+
"moe",
|
| 86 |
+
"moe",
|
| 87 |
+
"moe",
|
| 88 |
+
"moe",
|
| 89 |
+
"moe",
|
| 90 |
+
"moe",
|
| 91 |
+
"moe",
|
| 92 |
+
"moe",
|
| 93 |
+
"moe",
|
| 94 |
+
"moe",
|
| 95 |
+
"moe",
|
| 96 |
+
"moe",
|
| 97 |
+
"moe",
|
| 98 |
+
"moe",
|
| 99 |
+
"moe",
|
| 100 |
+
"moe",
|
| 101 |
+
"moe",
|
| 102 |
+
"moe",
|
| 103 |
+
"moe",
|
| 104 |
+
"moe",
|
| 105 |
+
"moe",
|
| 106 |
+
"moe",
|
| 107 |
+
"moe",
|
| 108 |
+
"moe",
|
| 109 |
+
"moe",
|
| 110 |
+
"moe",
|
| 111 |
+
"moe",
|
| 112 |
+
"moe",
|
| 113 |
+
"moe",
|
| 114 |
+
"moe",
|
| 115 |
+
"moe",
|
| 116 |
+
"moe"
|
| 117 |
+
],
|
| 118 |
+
"model_type": "deepseek_v4",
|
| 119 |
+
"moe_intermediate_size": 2048,
|
| 120 |
+
"n_routed_experts": 256,
|
| 121 |
+
"n_shared_experts": 1,
|
| 122 |
+
"norm_topk_prob": true,
|
| 123 |
+
"num_attention_heads": 64,
|
| 124 |
+
"num_experts_per_tok": 6,
|
| 125 |
+
"num_hidden_layers": 43,
|
| 126 |
+
"num_key_value_heads": 1,
|
| 127 |
+
"num_nextn_predict_layers": 1,
|
| 128 |
+
"o_groups": 8,
|
| 129 |
+
"o_lora_rank": 1024,
|
| 130 |
+
"output_router_logits": false,
|
| 131 |
+
"pad_token_id": null,
|
| 132 |
+
"partial_rotary_factor": 0.125,
|
| 133 |
+
"q_lora_rank": 1024,
|
| 134 |
+
"qk_rope_head_dim": 64,
|
| 135 |
+
"quantization_config": {
|
| 136 |
+
"config_groups": {
|
| 137 |
+
"group_0": {
|
| 138 |
+
"format": "float-quantized",
|
| 139 |
+
"input_activations": {
|
| 140 |
+
"actorder": null,
|
| 141 |
+
"block_structure": null,
|
| 142 |
+
"dynamic": true,
|
| 143 |
+
"group_size": 128,
|
| 144 |
+
"num_bits": 8,
|
| 145 |
+
"observer": null,
|
| 146 |
+
"observer_kwargs": {},
|
| 147 |
+
"scale_dtype": null,
|
| 148 |
+
"strategy": "group",
|
| 149 |
+
"symmetric": true,
|
| 150 |
+
"type": "float",
|
| 151 |
+
"zp_dtype": null
|
| 152 |
+
},
|
| 153 |
+
"output_activations": null,
|
| 154 |
+
"targets": [
|
| 155 |
+
"re:.*attn\\.(wq_a|wq_b|wkv|wo_a|wo_b|fused_wqa_wkv|q_a_proj|q_b_proj|kv_proj|o_a_proj|o_b_proj)$",
|
| 156 |
+
"re:.*attn\\.compressor\\.(wgate|wkv|fused_wkv_wgate|gate_proj|kv_proj)$",
|
| 157 |
+
"re:.*attn\\.indexer\\.(weights_proj|wq_b|q_b_proj)$",
|
| 158 |
+
"re:.*attn\\.indexer\\.compressor\\.(wgate|wkv|gate_proj|kv_proj)$"
|
| 159 |
+
],
|
| 160 |
+
"weights": {
|
| 161 |
+
"actorder": null,
|
| 162 |
+
"block_structure": [
|
| 163 |
+
128,
|
| 164 |
+
128
|
| 165 |
+
],
|
| 166 |
+
"dynamic": false,
|
| 167 |
+
"group_size": null,
|
| 168 |
+
"num_bits": 8,
|
| 169 |
+
"observer": "memoryless_minmax",
|
| 170 |
+
"observer_kwargs": {},
|
| 171 |
+
"scale_dtype": null,
|
| 172 |
+
"strategy": "block",
|
| 173 |
+
"symmetric": true,
|
| 174 |
+
"type": "float",
|
| 175 |
+
"zp_dtype": null
|
| 176 |
+
}
|
| 177 |
+
},
|
| 178 |
+
"group_1": {
|
| 179 |
+
"format": "pack-quantized",
|
| 180 |
+
"input_activations": null,
|
| 181 |
+
"output_activations": null,
|
| 182 |
+
"targets": [
|
| 183 |
+
"re:.*experts\\.\\d+\\.(w1|w2|w3|gate_proj|up_proj|down_proj|gate_up_proj)$"
|
| 184 |
+
],
|
| 185 |
+
"weights": {
|
| 186 |
+
"actorder": "static",
|
| 187 |
+
"block_structure": null,
|
| 188 |
+
"dynamic": false,
|
| 189 |
+
"group_size": 128,
|
| 190 |
+
"num_bits": 4,
|
| 191 |
+
"observer": "memoryless_minmax",
|
| 192 |
+
"observer_kwargs": {},
|
| 193 |
+
"scale_dtype": null,
|
| 194 |
+
"strategy": "group",
|
| 195 |
+
"symmetric": true,
|
| 196 |
+
"type": "int",
|
| 197 |
+
"zp_dtype": null
|
| 198 |
+
}
|
| 199 |
+
}
|
| 200 |
+
},
|
| 201 |
+
"format": "mixed-precision",
|
| 202 |
+
"global_compression_ratio": null,
|
| 203 |
+
"ignore": [
|
| 204 |
+
"layers.0.ffn.shared_experts.w1",
|
| 205 |
+
"layers.0.ffn.shared_experts.w2",
|
| 206 |
+
"layers.0.ffn.shared_experts.w3",
|
| 207 |
+
"layers.1.ffn.shared_experts.w1",
|
| 208 |
+
"layers.1.ffn.shared_experts.w2",
|
| 209 |
+
"layers.1.ffn.shared_experts.w3",
|
| 210 |
+
"layers.2.ffn.shared_experts.w1",
|
| 211 |
+
"layers.2.ffn.shared_experts.w2",
|
| 212 |
+
"layers.2.ffn.shared_experts.w3",
|
| 213 |
+
"layers.3.ffn.shared_experts.w1",
|
| 214 |
+
"layers.3.ffn.shared_experts.w2",
|
| 215 |
+
"layers.3.ffn.shared_experts.w3",
|
| 216 |
+
"layers.4.ffn.shared_experts.w1",
|
| 217 |
+
"layers.4.ffn.shared_experts.w2",
|
| 218 |
+
"layers.4.ffn.shared_experts.w3",
|
| 219 |
+
"layers.5.ffn.shared_experts.w1",
|
| 220 |
+
"layers.5.ffn.shared_experts.w2",
|
| 221 |
+
"layers.5.ffn.shared_experts.w3",
|
| 222 |
+
"layers.6.ffn.shared_experts.w1",
|
| 223 |
+
"layers.6.ffn.shared_experts.w2",
|
| 224 |
+
"layers.6.ffn.shared_experts.w3",
|
| 225 |
+
"layers.7.ffn.shared_experts.w1",
|
| 226 |
+
"layers.7.ffn.shared_experts.w2",
|
| 227 |
+
"layers.7.ffn.shared_experts.w3",
|
| 228 |
+
"layers.8.ffn.shared_experts.w1",
|
| 229 |
+
"layers.8.ffn.shared_experts.w2",
|
| 230 |
+
"layers.8.ffn.shared_experts.w3",
|
| 231 |
+
"layers.9.ffn.shared_experts.w1",
|
| 232 |
+
"layers.9.ffn.shared_experts.w2",
|
| 233 |
+
"layers.9.ffn.shared_experts.w3",
|
| 234 |
+
"layers.10.ffn.shared_experts.w1",
|
| 235 |
+
"layers.10.ffn.shared_experts.w2",
|
| 236 |
+
"layers.10.ffn.shared_experts.w3",
|
| 237 |
+
"layers.11.ffn.shared_experts.w1",
|
| 238 |
+
"layers.11.ffn.shared_experts.w2",
|
| 239 |
+
"layers.11.ffn.shared_experts.w3",
|
| 240 |
+
"layers.12.ffn.shared_experts.w1",
|
| 241 |
+
"layers.12.ffn.shared_experts.w2",
|
| 242 |
+
"layers.12.ffn.shared_experts.w3",
|
| 243 |
+
"layers.13.ffn.shared_experts.w1",
|
| 244 |
+
"layers.13.ffn.shared_experts.w2",
|
| 245 |
+
"layers.13.ffn.shared_experts.w3",
|
| 246 |
+
"layers.14.ffn.shared_experts.w1",
|
| 247 |
+
"layers.14.ffn.shared_experts.w2",
|
| 248 |
+
"layers.14.ffn.shared_experts.w3",
|
| 249 |
+
"layers.15.ffn.shared_experts.w1",
|
| 250 |
+
"layers.15.ffn.shared_experts.w2",
|
| 251 |
+
"layers.15.ffn.shared_experts.w3",
|
| 252 |
+
"layers.16.ffn.shared_experts.w1",
|
| 253 |
+
"layers.16.ffn.shared_experts.w2",
|
| 254 |
+
"layers.16.ffn.shared_experts.w3",
|
| 255 |
+
"layers.17.ffn.shared_experts.w1",
|
| 256 |
+
"layers.17.ffn.shared_experts.w2",
|
| 257 |
+
"layers.17.ffn.shared_experts.w3",
|
| 258 |
+
"layers.18.ffn.shared_experts.w1",
|
| 259 |
+
"layers.18.ffn.shared_experts.w2",
|
| 260 |
+
"layers.18.ffn.shared_experts.w3",
|
| 261 |
+
"layers.19.ffn.shared_experts.w1",
|
| 262 |
+
"layers.19.ffn.shared_experts.w2",
|
| 263 |
+
"layers.19.ffn.shared_experts.w3",
|
| 264 |
+
"layers.20.ffn.shared_experts.w1",
|
| 265 |
+
"layers.20.ffn.shared_experts.w2",
|
| 266 |
+
"layers.20.ffn.shared_experts.w3",
|
| 267 |
+
"layers.21.ffn.shared_experts.w1",
|
| 268 |
+
"layers.21.ffn.shared_experts.w2",
|
| 269 |
+
"layers.21.ffn.shared_experts.w3",
|
| 270 |
+
"layers.22.ffn.shared_experts.w1",
|
| 271 |
+
"layers.22.ffn.shared_experts.w2",
|
| 272 |
+
"layers.22.ffn.shared_experts.w3",
|
| 273 |
+
"layers.23.ffn.shared_experts.w1",
|
| 274 |
+
"layers.23.ffn.shared_experts.w2",
|
| 275 |
+
"layers.23.ffn.shared_experts.w3",
|
| 276 |
+
"layers.24.ffn.shared_experts.w1",
|
| 277 |
+
"layers.24.ffn.shared_experts.w2",
|
| 278 |
+
"layers.24.ffn.shared_experts.w3",
|
| 279 |
+
"layers.25.ffn.shared_experts.w1",
|
| 280 |
+
"layers.25.ffn.shared_experts.w2",
|
| 281 |
+
"layers.25.ffn.shared_experts.w3",
|
| 282 |
+
"layers.26.ffn.shared_experts.w1",
|
| 283 |
+
"layers.26.ffn.shared_experts.w2",
|
| 284 |
+
"layers.26.ffn.shared_experts.w3",
|
| 285 |
+
"layers.27.ffn.shared_experts.w1",
|
| 286 |
+
"layers.27.ffn.shared_experts.w2",
|
| 287 |
+
"layers.27.ffn.shared_experts.w3",
|
| 288 |
+
"layers.28.ffn.shared_experts.w1",
|
| 289 |
+
"layers.28.ffn.shared_experts.w2",
|
| 290 |
+
"layers.28.ffn.shared_experts.w3",
|
| 291 |
+
"layers.29.ffn.shared_experts.w1",
|
| 292 |
+
"layers.29.ffn.shared_experts.w2",
|
| 293 |
+
"layers.29.ffn.shared_experts.w3",
|
| 294 |
+
"layers.30.ffn.shared_experts.w1",
|
| 295 |
+
"layers.30.ffn.shared_experts.w2",
|
| 296 |
+
"layers.30.ffn.shared_experts.w3",
|
| 297 |
+
"layers.31.ffn.shared_experts.w1",
|
| 298 |
+
"layers.31.ffn.shared_experts.w2",
|
| 299 |
+
"layers.31.ffn.shared_experts.w3",
|
| 300 |
+
"layers.32.ffn.shared_experts.w1",
|
| 301 |
+
"layers.32.ffn.shared_experts.w2",
|
| 302 |
+
"layers.32.ffn.shared_experts.w3",
|
| 303 |
+
"layers.33.ffn.shared_experts.w1",
|
| 304 |
+
"layers.33.ffn.shared_experts.w2",
|
| 305 |
+
"layers.33.ffn.shared_experts.w3",
|
| 306 |
+
"layers.34.ffn.shared_experts.w1",
|
| 307 |
+
"layers.34.ffn.shared_experts.w2",
|
| 308 |
+
"layers.34.ffn.shared_experts.w3",
|
| 309 |
+
"layers.35.ffn.shared_experts.w1",
|
| 310 |
+
"layers.35.ffn.shared_experts.w2",
|
| 311 |
+
"layers.35.ffn.shared_experts.w3",
|
| 312 |
+
"layers.36.ffn.shared_experts.w1",
|
| 313 |
+
"layers.36.ffn.shared_experts.w2",
|
| 314 |
+
"layers.36.ffn.shared_experts.w3",
|
| 315 |
+
"layers.37.ffn.shared_experts.w1",
|
| 316 |
+
"layers.37.ffn.shared_experts.w2",
|
| 317 |
+
"layers.37.ffn.shared_experts.w3",
|
| 318 |
+
"layers.38.ffn.shared_experts.w1",
|
| 319 |
+
"layers.38.ffn.shared_experts.w2",
|
| 320 |
+
"layers.38.ffn.shared_experts.w3",
|
| 321 |
+
"layers.39.ffn.shared_experts.w1",
|
| 322 |
+
"layers.39.ffn.shared_experts.w2",
|
| 323 |
+
"layers.39.ffn.shared_experts.w3",
|
| 324 |
+
"layers.40.ffn.shared_experts.w1",
|
| 325 |
+
"layers.40.ffn.shared_experts.w2",
|
| 326 |
+
"layers.40.ffn.shared_experts.w3",
|
| 327 |
+
"layers.41.ffn.shared_experts.w1",
|
| 328 |
+
"layers.41.ffn.shared_experts.w2",
|
| 329 |
+
"layers.41.ffn.shared_experts.w3",
|
| 330 |
+
"layers.42.ffn.shared_experts.w1",
|
| 331 |
+
"layers.42.ffn.shared_experts.w2",
|
| 332 |
+
"layers.42.ffn.shared_experts.w3",
|
| 333 |
+
"re:^layers\\.43\\.",
|
| 334 |
+
"re:^model\\.layers\\.43\\.",
|
| 335 |
+
"lm_head"
|
| 336 |
+
],
|
| 337 |
+
"kv_cache_scheme": null,
|
| 338 |
+
"quant_method": "compressed-tensors",
|
| 339 |
+
"quantization_status": "compressed",
|
| 340 |
+
"sparsity_config": {},
|
| 341 |
+
"transform_config": {},
|
| 342 |
+
"version": "0.15.1.a20260515",
|
| 343 |
+
"scale_fmt": "ue8m0"
|
| 344 |
+
},
|
| 345 |
+
"rms_norm_eps": 1e-06,
|
| 346 |
+
"rope_parameters": {
|
| 347 |
+
"compress": {
|
| 348 |
+
"attention_factor": 1.0,
|
| 349 |
+
"beta_fast": 32,
|
| 350 |
+
"beta_slow": 1,
|
| 351 |
+
"factor": 16,
|
| 352 |
+
"original_max_position_embeddings": 65536,
|
| 353 |
+
"partial_rotary_factor": 0.125,
|
| 354 |
+
"rope_theta": 160000,
|
| 355 |
+
"rope_type": "yarn",
|
| 356 |
+
"type": "yarn"
|
| 357 |
+
},
|
| 358 |
+
"main": {
|
| 359 |
+
"partial_rotary_factor": 0.125,
|
| 360 |
+
"rope_theta": 10000,
|
| 361 |
+
"rope_type": "default"
|
| 362 |
+
},
|
| 363 |
+
"partial_rotary_factor": 0.125,
|
| 364 |
+
"rope_theta": 10000,
|
| 365 |
+
"rope_type": "default"
|
| 366 |
+
},
|
| 367 |
+
"rope_theta": 10000,
|
| 368 |
+
"routed_scaling_factor": 1.5,
|
| 369 |
+
"router_aux_loss_coef": 0.001,
|
| 370 |
+
"router_jitter_noise": 0.0,
|
| 371 |
+
"scoring_func": "sqrtsoftplus",
|
| 372 |
+
"sliding_window": 128,
|
| 373 |
+
"swiglu_limit": 10.0,
|
| 374 |
+
"tie_word_embeddings": false,
|
| 375 |
+
"topk_method": "noaux_tc",
|
| 376 |
+
"transformers_version": "5.8.1",
|
| 377 |
+
"use_cache": true,
|
| 378 |
+
"vocab_size": 129280,
|
| 379 |
+
"compress_ratios": [
|
| 380 |
+
0,
|
| 381 |
+
0,
|
| 382 |
+
4,
|
| 383 |
+
128,
|
| 384 |
+
4,
|
| 385 |
+
128,
|
| 386 |
+
4,
|
| 387 |
+
128,
|
| 388 |
+
4,
|
| 389 |
+
128,
|
| 390 |
+
4,
|
| 391 |
+
128,
|
| 392 |
+
4,
|
| 393 |
+
128,
|
| 394 |
+
4,
|
| 395 |
+
128,
|
| 396 |
+
4,
|
| 397 |
+
128,
|
| 398 |
+
4,
|
| 399 |
+
128,
|
| 400 |
+
4,
|
| 401 |
+
128,
|
| 402 |
+
4,
|
| 403 |
+
128,
|
| 404 |
+
4,
|
| 405 |
+
128,
|
| 406 |
+
4,
|
| 407 |
+
128,
|
| 408 |
+
4,
|
| 409 |
+
128,
|
| 410 |
+
4,
|
| 411 |
+
128,
|
| 412 |
+
4,
|
| 413 |
+
128,
|
| 414 |
+
4,
|
| 415 |
+
128,
|
| 416 |
+
4,
|
| 417 |
+
128,
|
| 418 |
+
4,
|
| 419 |
+
128,
|
| 420 |
+
4,
|
| 421 |
+
128,
|
| 422 |
+
4,
|
| 423 |
+
0
|
| 424 |
+
],
|
| 425 |
+
"num_hash_layers": 3,
|
| 426 |
+
"rope_scaling": {
|
| 427 |
+
"beta_fast": 32,
|
| 428 |
+
"beta_slow": 1,
|
| 429 |
+
"factor": 16,
|
| 430 |
+
"original_max_position_embeddings": 65536,
|
| 431 |
+
"type": "yarn"
|
| 432 |
+
},
|
| 433 |
+
"torch_dtype": "bfloat16"
|
| 434 |
+
}
|
generation_config.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 0,
|
| 4 |
+
"do_sample": true,
|
| 5 |
+
"eos_token_id": 1,
|
| 6 |
+
"temperature": 1.0,
|
| 7 |
+
"top_p": 1.0,
|
| 8 |
+
"transformers_version": "5.8.1"
|
| 9 |
+
}
|
model-00001-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d986f996385b30f897e047fc54b755f32c6998133add79bafc02215714de3e3a
|
| 3 |
+
size 51059154690
|
model-00002-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cdf611410ef972e28db7d91d591a31d846751bb41c765f6f679d473db6a3903a
|
| 3 |
+
size 50002265884
|
model-00003-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6b8e17840571d8fc6ce0703c7d81d50f113d60313c455d1c0795e8b0673e6562
|
| 3 |
+
size 50001322552
|
model-00004-of-00004.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bcd520418011b6c24a01786e3c57176d3ed00fca736ff90d3d35c5760f9c749d
|
| 3 |
+
size 18795041598
|
model.safetensors.index.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
recipe.yaml
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
default_stage:
|
| 2 |
+
default_modifiers:
|
| 3 |
+
GPTQModifier:
|
| 4 |
+
config_groups:
|
| 5 |
+
attention:
|
| 6 |
+
targets: ['re:^model\.layers\.\d+\.self_attn\.(q_a_proj|q_b_proj|kv_proj|o_a_proj|o_b_proj)$',
|
| 7 |
+
're:^model\.layers\.\d+\.self_attn\.compressor\.(gate_proj|kv_proj)$', 're:^model\.layers\.\d+\.self_attn\.compressor\.indexer\.(gate_proj|kv_proj|q_b_proj|weights_proj)$']
|
| 8 |
+
weights:
|
| 9 |
+
num_bits: 8
|
| 10 |
+
type: float
|
| 11 |
+
symmetric: true
|
| 12 |
+
group_size: null
|
| 13 |
+
strategy: block
|
| 14 |
+
block_structure: [128, 128]
|
| 15 |
+
dynamic: false
|
| 16 |
+
actorder: null
|
| 17 |
+
scale_dtype: null
|
| 18 |
+
zp_dtype: null
|
| 19 |
+
observer: memoryless_minmax
|
| 20 |
+
observer_kwargs: {}
|
| 21 |
+
input_activations:
|
| 22 |
+
num_bits: 8
|
| 23 |
+
type: float
|
| 24 |
+
symmetric: true
|
| 25 |
+
group_size: 128
|
| 26 |
+
strategy: group
|
| 27 |
+
block_structure: null
|
| 28 |
+
dynamic: true
|
| 29 |
+
actorder: null
|
| 30 |
+
scale_dtype: null
|
| 31 |
+
zp_dtype: null
|
| 32 |
+
observer: null
|
| 33 |
+
observer_kwargs: {}
|
| 34 |
+
output_activations: null
|
| 35 |
+
format: null
|
| 36 |
+
experts:
|
| 37 |
+
targets: ['re:^model\.layers\.\d+\.mlp\.experts\.\d+\.(gate_proj|up_proj|down_proj)$']
|
| 38 |
+
weights:
|
| 39 |
+
num_bits: 4
|
| 40 |
+
type: int
|
| 41 |
+
symmetric: true
|
| 42 |
+
group_size: 128
|
| 43 |
+
strategy: group
|
| 44 |
+
block_structure: null
|
| 45 |
+
dynamic: false
|
| 46 |
+
actorder: !!python/object/apply:compressed_tensors.quantization.quant_args.ActivationOrdering [
|
| 47 |
+
static]
|
| 48 |
+
scale_dtype: null
|
| 49 |
+
zp_dtype: null
|
| 50 |
+
observer: memoryless_minmax
|
| 51 |
+
observer_kwargs: {}
|
| 52 |
+
input_activations: null
|
| 53 |
+
output_activations: null
|
| 54 |
+
format: null
|
| 55 |
+
targets: [Linear]
|
| 56 |
+
ignore: [lm_head, 're:.*mtp\..*']
|
| 57 |
+
bypass_divisibility_checks: false
|
| 58 |
+
block_size: 128
|
| 59 |
+
dampening_frac: 0.1
|
| 60 |
+
actorder: static
|
| 61 |
+
offload_hessians: true
|
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": 1048576,
|
| 10 |
+
"pad_token": "<｜end▁of▁sentence｜>",
|
| 11 |
+
"sp_model_kwargs": {},
|
| 12 |
+
"tokenizer_class": "TokenizersBackend",
|
| 13 |
+
"unk_token": null
|
| 14 |
+
}
|