diff --git "a/config.json" "b/config.json"
new file mode 100644--- /dev/null
+++ "b/config.json"
@@ -0,0 +1,5486 @@
+{
+    "_attn_implementation_autoset": false,
+    "add_cross_attention": false,
+    "architectures": [
+        "DeepseekV3ForCausalLM"
+    ],
+    "attention_bias": false,
+    "attention_dropout": 0.0,
+    "auto_map": {
+        "AutoConfig": "configuration_deepseek.DeepseekV3Config",
+        "AutoModel": "modeling_deepseek.DeepseekV3Model",
+        "AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM"
+    },
+    "aux_loss_alpha": 0.001,
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "bos_token_id": 163584,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "early_stopping": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": 163585,
+    "ep_size": 1,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "first_k_dense_replace": 1,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "hidden_act": "silu",
+    "hidden_size": 7168,
+    "id2label": {
+        "0": "LABEL_0",
+        "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 18432,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "kv_lora_rank": 512,
+    "label2id": {
+        "LABEL_0": 0,
+        "LABEL_1": 1
+    },
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "max_position_embeddings": 262144,
+    "min_length": 0,
+    "model_type": "kimi_k2",
+    "moe_intermediate_size": 2048,
+    "moe_layer_freq": 1,
+    "n_group": 1,
+    "n_routed_experts": 384,
+    "n_shared_experts": 1,
+    "no_repeat_ngram_size": 0,
+    "norm_topk_prob": true,
+    "num_attention_heads": 64,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_experts_per_tok": 8,
+    "num_hidden_layers": 61,
+    "num_key_value_heads": 64,
+    "num_nextn_predict_layers": 0,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": 163839,
+    "prefix": null,
+    "pretraining_tp": 1,
+    "problem_type": null,
+    "pruned_heads": {},
+    "q_lora_rank": 1536,
+    "qk_nope_head_dim": 128,
+    "qk_rope_head_dim": 64,
+    "quantization": {
+        "group_size": 64,
+        "bits": 4,
+        "mode": "affine",
+        "model.embed_tokens": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.1.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.1.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.2.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.2.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.3.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.3.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.4.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.4.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.5.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.5.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.6.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.6.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.7.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.9.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.10.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.12.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.13.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.15.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.16.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.18.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.19.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.21.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.22.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.24.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.25.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.27.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.28.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.30.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.31.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.33.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.34.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.36.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.37.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.39.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.40.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.42.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.43.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.45.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.46.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.48.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.49.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.51.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.52.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.53.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.54.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.54.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.55.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.55.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.56.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.56.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.57.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.57.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.58.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.58.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.59.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.59.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.60.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.60.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "lm_head": {
+            "group_size": 64,
+            "bits": 6
+        }
+    },
+    "quantization_config": {
+        "group_size": 64,
+        "bits": 4,
+        "mode": "affine",
+        "model.embed_tokens": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.0.mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.1.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.1.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.1.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.2.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.2.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.2.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.3.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.3.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.3.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.4.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.4.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.4.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.5.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.5.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.5.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.6.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.6.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.6.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.7.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.7.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.8.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.9.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.9.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.10.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.10.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.11.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.12.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.12.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.13.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.13.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.14.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.15.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.15.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.16.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.16.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.17.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.18.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.18.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.19.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.19.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.20.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.21.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.21.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.22.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.22.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.23.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.24.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.24.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.25.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.25.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.26.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.27.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.27.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.28.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.28.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.29.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.30.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.30.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.31.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.31.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.32.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.33.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.33.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.34.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.34.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.35.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.36.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.36.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.37.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.37.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.38.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.39.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.39.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.40.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.40.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.41.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.42.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.42.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.43.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.43.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.44.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.45.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.45.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.46.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.46.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.47.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.48.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.48.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.49.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.49.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.50.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.51.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.51.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.52.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.52.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.53.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.53.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.54.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.54.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.54.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.55.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.55.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.55.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.56.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.56.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.56.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.57.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.57.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.57.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.58.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.58.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.58.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.59.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.59.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.59.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.60.self_attn.q_a_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.q_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.kv_a_proj_with_mqa": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.kv_b_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.self_attn.o_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.switch_mlp.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.switch_mlp.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.switch_mlp.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "model.layers.60.mlp.shared_experts.gate_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.shared_experts.up_proj": {
+            "group_size": 64,
+            "bits": 2
+        },
+        "model.layers.60.mlp.shared_experts.down_proj": {
+            "group_size": 64,
+            "bits": 6
+        },
+        "lm_head": {
+            "group_size": 64,
+            "bits": 6
+        }
+    },
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "rms_norm_eps": 1e-05,
+    "rope_scaling": {
+        "beta_fast": 1.0,
+        "beta_slow": 1.0,
+        "factor": 64.0,
+        "mscale": 1.0,
+        "mscale_all_dim": 1.0,
+        "original_max_position_embeddings": 4096,
+        "type": "yarn"
+    },
+    "rope_theta": 50000.0,
+    "routed_scaling_factor": 2.827,
+    "scoring_func": "sigmoid",
+    "sep_token_id": null,
+    "seq_aux": true,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": false,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "topk_group": 1,
+    "topk_method": "noaux_tc",
+    "torch_dtype": "bfloat16",
+    "torchscript": false,
+    "transformers_version": "4.51.3",
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "use_cache": true,
+    "v_head_dim": 128,
+    "vocab_size": 163840
+}
\ No newline at end of file