| { | |
| "_name_or_path": "/mnt/data/zhaoliang/run_output/MoE-13bx16/base-0528/0000127_hf", | |
| "architectures": [ | |
| "SkyworkForCausalLM" | |
| ], | |
| "auto_map": { | |
| "AutoConfig": "configuration_skywork_moe.SkyworkMoeConfig", | |
| "AutoModelForCausalLM": "modeling_skywork_moe.SkyworkForCausalLM" | |
| }, | |
| "bos_token_id": 1, | |
| "eos_token_id": 2, | |
| "hidden_act": "silu", | |
| "hidden_size": 4608, | |
| "initializer_range": 0.01, | |
| "intermediate_size": 12288, | |
| "max_position_embeddings": 8192, | |
| "model_type": "skywork_moe", | |
| "moe_2layer_gate": false, | |
| "moe_expert_interval": 1, | |
| "moe_feature_no_mul_topk": true, | |
| "moe_gate_norm_std": 1.0, | |
| "moe_use_logits_norm": true, | |
| "moe_use_skywork_gating": false, | |
| "num_attention_heads": 36, | |
| "num_experts": [ | |
| 16 | |
| ], | |
| "num_hidden_layers": 52, | |
| "num_key_value_heads": 36, | |
| "pad_token_id": 0, | |
| "pretraining_tp": 1, | |
| "quantization_config": { | |
| "activation_scheme": "dynamic", | |
| "quant_method": "fp8" | |
| }, | |
| "rms_norm_eps": 1e-06, | |
| "rope_scaling": null, | |
| "rope_theta": 10000, | |
| "rotary_percent": 1.0, | |
| "sliding_window": null, | |
| "tie_word_embeddings": false, | |
| "torch_dtype": "bfloat16", | |
| "transformers_version": "4.41.1", | |
| "use_cache": true, | |
| "vocab_size": 65532 | |
| } | |