| { |
| "architectures": [ |
| "LagunaForCausalLM" |
| ], |
| "auto_map": { |
| "AutoConfig": "configuration_laguna.LagunaConfig", |
| "AutoModelForCausalLM": "modeling_laguna.LagunaForCausalLM" |
| }, |
| "model_type": "laguna", |
| "vocab_size": 100352, |
| "hidden_size": 2048, |
| "intermediate_size": 8192, |
| "num_hidden_layers": 40, |
| "num_attention_heads": 48, |
| "num_key_value_heads": 8, |
| "head_dim": 128, |
| "max_position_embeddings": 262144, |
| "attention_bias": false, |
| "attention_dropout": 0.0, |
| "rms_norm_eps": 1e-06, |
| "num_experts": 256, |
| "num_experts_per_tok": 8, |
| "moe_intermediate_size": 512, |
| "shared_expert_intermediate_size": 512, |
| "norm_topk_prob": true, |
| "router_aux_loss_coef": 0.0, |
| "decoder_sparse_step": 1, |
| "mlp_only_layers": [ |
| 0 |
| ], |
| "bos_token_id": 2, |
| "eos_token_id": [ |
| 2, |
| 24 |
| ], |
| "pad_token_id": 9, |
| "tie_word_embeddings": false, |
| "use_cache": true, |
| "torch_dtype": "bfloat16", |
| "gating": "per-head", |
| "sliding_window": 512, |
| "rope_parameters": { |
| "full_attention": { |
| "rope_theta": 500000.0, |
| "rope_type": "yarn", |
| "factor": 32.0, |
| "original_max_position_embeddings": 8192, |
| "beta_slow": 1.0, |
| "beta_fast": 64.0, |
| "attention_factor": 1.3465735902799727, |
| "partial_rotary_factor": 0.5 |
| }, |
| "sliding_attention": { |
| "rope_type": "default", |
| "rope_theta": 10000.0, |
| "partial_rotary_factor": 1.0 |
| } |
| }, |
| "layer_types": [ |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "full_attention", |
| "sliding_attention", |
| "sliding_attention", |
| "sliding_attention" |
| ], |
| "moe_apply_router_weight_on_input": false, |
| "mlp_layer_types": [ |
| "dense", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse", |
| "sparse" |
| ], |
| "gating_types": [ |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head", |
| "per_head" |
| ], |
| "moe_routed_scaling_factor": 2.5, |
| "num_attention_heads_per_layer": [ |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64, |
| 48, |
| 64, |
| 64, |
| 64 |
| ], |
| "quantization": { |
| "group_size": 64, |
| "bits": 4, |
| "mode": "affine", |
| "lm_head": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.embed_tokens": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.mlp.down_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.mlp.gate_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.mlp.up_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.1.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.1.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.10.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.10.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.11.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.11.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.12.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.12.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.13.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.13.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.14.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.14.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.15.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.15.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.16.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.16.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.17.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.17.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.18.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.18.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.19.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.19.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.2.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.2.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.20.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.20.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.21.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.21.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.22.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.22.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.23.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.23.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.24.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.24.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.25.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.25.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.26.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.26.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.27.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.27.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.28.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.28.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.29.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.29.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.3.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.3.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.30.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.30.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.31.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.31.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.32.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.32.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.33.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.33.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.34.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.34.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.35.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.35.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.36.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.36.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.37.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.37.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.38.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.38.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.39.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.39.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.4.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.4.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.5.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.5.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.6.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.6.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.7.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.7.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.8.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.8.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.9.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.9.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| } |
| }, |
| "quantization_config": { |
| "group_size": 64, |
| "bits": 4, |
| "mode": "affine", |
| "lm_head": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.embed_tokens": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.mlp.down_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.mlp.gate_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.mlp.up_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.0.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.1.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.1.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.1.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.10.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.10.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.10.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.11.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.11.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.11.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.12.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.12.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.12.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.13.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.13.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.13.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.14.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.14.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.14.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.15.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.15.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.15.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.16.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.16.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.16.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.17.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.17.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.17.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.18.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.18.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.18.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.19.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.19.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.19.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.2.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.2.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.2.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.20.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.20.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.20.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.21.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.21.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.21.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.22.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.22.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.22.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.23.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.23.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.23.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.24.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.24.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.24.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.25.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.25.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.25.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.26.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.26.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.26.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.27.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.27.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.27.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.28.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.28.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.28.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.29.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.29.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.29.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.3.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.3.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.3.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.30.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.30.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.30.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.31.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.31.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.31.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.32.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.32.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.32.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.33.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.33.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.33.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.34.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.34.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.34.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.35.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.35.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.35.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.36.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.36.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.g_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.k_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.o_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.q_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.36.self_attn.v_proj": { |
| "bits": 8, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.37.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.37.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.37.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.38.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.38.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.38.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.39.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.39.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.39.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.4.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.4.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.g_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.k_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.o_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.q_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.4.self_attn.v_proj": { |
| "bits": 5, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.5.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.5.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.5.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.6.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.6.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.6.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.7.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.7.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.7.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.8.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.8.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.8.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.mlp.shared_expert.down_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.9.mlp.shared_expert.gate_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.9.mlp.shared_expert.up_proj": { |
| "bits": 8, |
| "group_size": 128, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.g_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.k_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.o_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.q_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| }, |
| "model.layers.9.self_attn.v_proj": { |
| "bits": 6, |
| "group_size": 64, |
| "mode": "affine" |
| } |
| } |
| } |
