Error Quantize Mistral-Small-4-119B-2603-bf16 #446
Description
Describe the bug
Model does not load when Quantized to oq2
The Quantizing works very well but i cant load the Model after it quantized, tryed it in the Chat, got this error:
2026-03-28 12:37:49,180 - omlx.server - ERROR - [-] - POST /v1/chat/completions → 500 (unhandled): Received 936 parameters not in model: language_model.model.layers.0.mlp.gate.weight, language_model.model.layers.0.mlp.shared_experts.down_proj.scales, language_model.model.layers.0.mlp.shared_experts.down_proj.weight, language_model.model.layers.0.mlp.shared_experts.gate_proj.scales, language_model.model.layers.0.mlp.shared_experts.gate_proj.weight, language_model.model.layers.0.mlp.shared_experts.up_proj.scales, language_model.model.layers.0.mlp.shared_experts.up_proj.weight, language_model.model.layers.0.mlp.switch_mlp.down_proj.biases, language_model.model.layers.0.mlp.switch_mlp.down_proj.scales, language_model.model.layers.0.mlp.switch_mlp.down_proj.weight, language_model.model.layers.0.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.0.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.0.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.0.mlp.switch_mlp.up_proj.biases, language_model.model.layers.0.mlp.switch_mlp.up_proj.scales, language_model.model.layers.0.mlp.switch_mlp.up_proj.weight, language_model.model.layers.0.self_attn.kv_a_layernorm.weight, language_model.model.layers.0.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.0.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.0.self_attn.kv_b_proj.scales, language_model.model.layers.0.self_attn.kv_b_proj.weight, language_model.model.layers.0.self_attn.q_a_layernorm.weight, language_model.model.layers.0.self_attn.q_a_proj.scales, language_model.model.layers.0.self_attn.q_a_proj.weight, language_model.model.layers.0.self_attn.q_b_proj.scales, language_model.model.layers.0.self_attn.q_b_proj.weight, language_model.model.layers.1.mlp.gate.weight, language_model.model.layers.1.mlp.shared_experts.down_proj.scales, language_model.model.layers.1.mlp.shared_experts.down_proj.weight, language_model.model.layers.1.mlp.shared_experts.gate_proj.scales, language_model.model.layers.1.mlp.shared_experts.gate_proj.weight, language_model.model.layers.1.mlp.shared_experts.up_proj.scales, language_model.model.layers.1.mlp.shared_experts.up_proj.weight, language_model.model.layers.1.mlp.switch_mlp.down_proj.biases, language_model.model.layers.1.mlp.switch_mlp.down_proj.scales, language_model.model.layers.1.mlp.switch_mlp.down_proj.weight, language_model.model.layers.1.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.1.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.1.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.1.mlp.switch_mlp.up_proj.biases, language_model.model.layers.1.mlp.switch_mlp.up_proj.scales, language_model.model.layers.1.mlp.switch_mlp.up_proj.weight, language_model.model.layers.1.self_attn.kv_a_layernorm.weight, language_model.model.layers.1.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.1.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.1.self_attn.kv_b_proj.scales, language_model.model.layers.1.self_attn.kv_b_proj.weight, language_model.model.layers.1.self_attn.q_a_layernorm.weight, language_model.model.layers.1.self_attn.q_a_proj.scales, language_model.model.layers.1.self_attn.q_a_proj.weight, language_model.model.layers.1.self_attn.q_b_proj.scales, language_model.model.layers.1.self_attn.q_b_proj.weight, language_model.model.layers.10.mlp.gate.weight, language_model.model.layers.10.mlp.shared_experts.down_proj.scales, language_model.model.layers.10.mlp.shared_experts.down_proj.weight, language_model.model.layers.10.mlp.shared_experts.gate_proj.scales, language_model.model.layers.10.mlp.shared_experts.gate_proj.weight, language_model.model.layers.10.mlp.shared_experts.up_proj.scales, language_model.model.layers.10.mlp.shared_experts.up_proj.weight, language_model.model.layers.10.mlp.switch_mlp.down_proj.biases, language_model.model.layers.10.mlp.switch_mlp.down_proj.scales, language_model.model.layers.10.mlp.switch_mlp.down_proj.weight, language_model.model.layers.10.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.10.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.10.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.10.mlp.switch_mlp.up_proj.biases, language_model.model.layers.10.mlp.switch_mlp.up_proj.scales, language_model.model.layers.10.mlp.switch_mlp.up_proj.weight, language_model.model.layers.10.self_attn.kv_a_layernorm.weight, language_model.model.layers.10.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.10.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.10.self_attn.kv_b_proj.scales, language_model.model.layers.10.self_attn.kv_b_proj.weight, language_model.model.layers.10.self_attn.q_a_layernorm.weight, language_model.model.layers.10.self_attn.q_a_proj.scales, language_model.model.layers.10.self_attn.q_a_proj.weight, language_model.model.layers.10.self_attn.q_b_proj.scales, language_model.model.layers.10.self_attn.q_b_proj.weight, language_model.model.layers.11.mlp.gate.weight, language_model.model.layers.11.mlp.shared_experts.down_proj.scales, language_model.model.layers.11.mlp.shared_experts.down_proj.weight, language_model.model.layers.11.mlp.shared_experts.gate_proj.scales, language_model.model.layers.11.mlp.shared_experts.gate_proj.weight, language_model.model.layers.11.mlp.shared_experts.up_proj.scales, language_model.model.layers.11.mlp.shared_experts.up_proj.weight, language_model.model.layers.11.mlp.switch_mlp.down_proj.biases, language_model.model.layers.11.mlp.switch_mlp.down_proj.scales, language_model.model.layers.11.mlp.switch_mlp.down_proj.weight, language_model.model.layers.11.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.11.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.11.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.11.mlp.switch_mlp.up_proj.biases, language_model.model.layers.11.mlp.switch_mlp.up_proj.scales, language_model.model.layers.11.mlp.switch_mlp.up_proj.weight, language_model.model.layers.11.self_attn.kv_a_layernorm.weight, language_model.model.layers.11.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.11.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.11.self_attn.kv_b_proj.scales, language_model.model.layers.11.self_attn.kv_b_proj.weight, language_model.model.layers.11.self_attn.q_a_layernorm.weight, language_model.model.layers.11.self_attn.q_a_proj.scales, language_model.model.layers.11.self_attn.q_a_proj.weight, language_model.model.layers.11.self_attn.q_b_proj.scales, language_model.model.layers.11.self_attn.q_b_proj.weight, language_model.model.layers.12.mlp.gate.weight, language_model.model.layers.12.mlp.shared_experts.down_proj.scales, language_model.model.layers.12.mlp.shared_experts.down_proj.weight, language_model.model.layers.12.mlp.shared_experts.gate_proj.scales, language_model.model.layers.12.mlp.shared_experts.gate_proj.weight, language_model.model.layers.12.mlp.shared_experts.up_proj.scales, language_model.model.layers.12.mlp.shared_experts.up_proj.weight, language_model.model.layers.12.mlp.switch_mlp.down_proj.biases, language_model.model.layers.12.mlp.switch_mlp.down_proj.scales, language_model.model.layers.12.mlp.switch_mlp.down_proj.weight, language_model.model.layers.12.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.12.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.12.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.12.mlp.switch_mlp.up_proj.biases, language_model.model.layers.12.mlp.switch_mlp.up_proj.scales, language_model.model.layers.12.mlp.switch_mlp.up_proj.weight, language_model.model.layers.12.self_attn.kv_a_layernorm.weight, language_model.model.layers.12.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.12.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.12.self_attn.kv_b_proj.scales, language_model.model.layers.12.self_attn.kv_b_proj.weight, language_model.model.layers.12.self_attn.q_a_layernorm.weight, language_model.model.layers.12.self_attn.q_a_proj.scales, language_model.model.layers.12.self_attn.q_a_proj.weight, language_model.model.layers.12.self_attn.q_b_proj.scales, language_model.model.layers.12.self_attn.q_b_proj.weight, language_model.model.layers.13.mlp.gate.weight, language_model.model.layers.13.mlp.shared_experts.down_proj.scales, language_model.model.layers.13.mlp.shared_experts.down_proj.weight, language_model.model.layers.13.mlp.shared_experts.gate_proj.scales, language_model.model.layers.13.mlp.shared_experts.gate_proj.weight, language_model.model.layers.13.mlp.shared_experts.up_proj.scales, language_model.model.layers.13.mlp.shared_experts.up_proj.weight, language_model.model.layers.13.mlp.switch_mlp.down_proj.biases, language_model.model.layers.13.mlp.switch_mlp.down_proj.scales, language_model.model.layers.13.mlp.switch_mlp.down_proj.weight, language_model.model.layers.13.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.13.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.13.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.13.mlp.switch_mlp.up_proj.biases, language_model.model.layers.13.mlp.switch_mlp.up_proj.scales, language_model.model.layers.13.mlp.switch_mlp.up_proj.weight, language_model.model.layers.13.self_attn.kv_a_layernorm.weight, language_model.model.layers.13.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.13.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.13.self_attn.kv_b_proj.scales, language_model.model.layers.13.self_attn.kv_b_proj.weight, language_model.model.layers.13.self_attn.q_a_layernorm.weight, language_model.model.layers.13.self_attn.q_a_proj.scales, language_model.model.layers.13.self_attn.q_a_proj.weight, language_model.model.layers.13.self_attn.q_b_proj.scales, language_model.model.layers.13.self_attn.q_b_proj.weight, language_model.model.layers.14.mlp.gate.weight, language_model.model.layers.14.mlp.shared_experts.down_proj.scales, language_model.model.layers.14.mlp.shared_experts.down_proj.weight, language_model.model.layers.14.mlp.shared_experts.gate_proj.scales, language_model.model.layers.14.mlp.shared_experts.gate_proj.weight, language_model.model.layers.14.mlp.shared_experts.up_proj.scales, language_model.model.layers.14.mlp.shared_experts.up_proj.weight, language_model.model.layers.14.mlp.switch_mlp.down_proj.biases, language_model.model.layers.14.mlp.switch_mlp.down_proj.scales, language_model.model.layers.14.mlp.switch_mlp.down_proj.weight, language_model.model.layers.14.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.14.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.14.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.14.mlp.switch_mlp.up_proj.biases, language_model.model.layers.14.mlp.switch_mlp.up_proj.scales, language_model.model.layers.14.mlp.switch_mlp.up_proj.weight, language_model.model.layers.14.self_attn.kv_a_layernorm.weight, language_model.model.layers.14.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.14.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.14.self_attn.kv_b_proj.scales, language_model.model.layers.14.self_attn.kv_b_proj.weight, language_model.model.layers.14.self_attn.q_a_layernorm.weight, language_model.model.layers.14.self_attn.q_a_proj.scales, language_model.model.layers.14.self_attn.q_a_proj.weight, language_model.model.layers.14.self_attn.q_b_proj.scales, language_model.model.layers.14.self_attn.q_b_proj.weight, language_model.model.layers.15.mlp.gate.weight, language_model.model.layers.15.mlp.shared_experts.down_proj.scales, language_model.model.layers.15.mlp.shared_experts.down_proj.weight, language_model.model.layers.15.mlp.shared_experts.gate_proj.scales, language_model.model.layers.15.mlp.shared_experts.gate_proj.weight, language_model.model.layers.15.mlp.shared_experts.up_proj.scales, language_model.model.layers.15.mlp.shared_experts.up_proj.weight, language_model.model.layers.15.mlp.switch_mlp.down_proj.biases, language_model.model.layers.15.mlp.switch_mlp.down_proj.scales, language_model.model.layers.15.mlp.switch_mlp.down_proj.weight, language_model.model.layers.15.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.15.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.15.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.15.mlp.switch_mlp.up_proj.biases, language_model.model.layers.15.mlp.switch_mlp.up_proj.scales, language_model.model.layers.15.mlp.switch_mlp.up_proj.weight, language_model.model.layers.15.self_attn.kv_a_layernorm.weight, language_model.model.layers.15.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.15.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.15.self_attn.kv_b_proj.scales, language_model.model.layers.15.self_attn.kv_b_proj.weight, language_model.model.layers.15.self_attn.q_a_layernorm.weight, language_model.model.layers.15.self_attn.q_a_proj.scales, language_model.model.layers.15.self_attn.q_a_proj.weight, language_model.model.layers.15.self_attn.q_b_proj.scales, language_model.model.layers.15.self_attn.q_b_proj.weight, language_model.model.layers.16.mlp.gate.weight, language_model.model.layers.16.mlp.shared_experts.down_proj.scales, language_model.model.layers.16.mlp.shared_experts.down_proj.weight, language_model.model.layers.16.mlp.shared_experts.gate_proj.scales, language_model.model.layers.16.mlp.shared_experts.gate_proj.weight, language_model.model.layers.16.mlp.shared_experts.up_proj.scales, language_model.model.layers.16.mlp.shared_experts.up_proj.weight, language_model.model.layers.16.mlp.switch_mlp.down_proj.biases, language_model.model.layers.16.mlp.switch_mlp.down_proj.scales, language_model.model.layers.16.mlp.switch_mlp.down_proj.weight, language_model.model.layers.16.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.16.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.16.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.16.mlp.switch_mlp.up_proj.biases, language_model.model.layers.16.mlp.switch_mlp.up_proj.scales, language_model.model.layers.16.mlp.switch_mlp.up_proj.weight, language_model.model.layers.16.self_attn.kv_a_layernorm.weight, language_model.model.layers.16.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.16.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.16.self_attn.kv_b_proj.scales, language_model.model.layers.16.self_attn.kv_b_proj.weight, language_model.model.layers.16.self_attn.q_a_layernorm.weight, language_model.model.layers.16.self_attn.q_a_proj.scales, language_model.model.layers.16.self_attn.q_a_proj.weight, language_model.model.layers.16.self_attn.q_b_proj.scales, language_model.model.layers.16.self_attn.q_b_proj.weight, language_model.model.layers.17.mlp.gate.weight, language_model.model.layers.17.mlp.shared_experts.down_proj.scales, language_model.model.layers.17.mlp.shared_experts.down_proj.weight, language_model.model.layers.17.mlp.shared_experts.gate_proj.scales, language_model.model.layers.17.mlp.shared_experts.gate_proj.weight, language_model.model.layers.17.mlp.shared_experts.up_proj.scales, language_model.model.layers.17.mlp.shared_experts.up_proj.weight, language_model.model.layers.17.mlp.switch_mlp.down_proj.biases, language_model.model.layers.17.mlp.switch_mlp.down_proj.scales, language_model.model.layers.17.mlp.switch_mlp.down_proj.weight, language_model.model.layers.17.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.17.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.17.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.17.mlp.switch_mlp.up_proj.biases, language_model.model.layers.17.mlp.switch_mlp.up_proj.scales, language_model.model.layers.17.mlp.switch_mlp.up_proj.weight, language_model.model.layers.17.self_attn.kv_a_layernorm.weight, language_model.model.layers.17.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.17.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.17.self_attn.kv_b_proj.scales, language_model.model.layers.17.self_attn.kv_b_proj.weight, language_model.model.layers.17.self_attn.q_a_layernorm.weight, language_model.model.layers.17.self_attn.q_a_proj.scales, language_model.model.layers.17.self_attn.q_a_proj.weight, language_model.model.layers.17.self_attn.q_b_proj.scales, language_model.model.layers.17.self_attn.q_b_proj.weight, language_model.model.layers.18.mlp.gate.weight, language_model.model.layers.18.mlp.shared_experts.down_proj.scales, language_model.model.layers.18.mlp.shared_experts.down_proj.weight, language_model.model.layers.18.mlp.shared_experts.gate_proj.scales, language_model.model.layers.18.mlp.shared_experts.gate_proj.weight, language_model.model.layers.18.mlp.shared_experts.up_proj.scales, language_model.model.layers.18.mlp.shared_experts.up_proj.weight, language_model.model.layers.18.mlp.switch_mlp.down_proj.biases, language_model.model.layers.18.mlp.switch_mlp.down_proj.scales, language_model.model.layers.18.mlp.switch_mlp.down_proj.weight, language_model.model.layers.18.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.18.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.18.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.18.mlp.switch_mlp.up_proj.biases, language_model.model.layers.18.mlp.switch_mlp.up_proj.scales, language_model.model.layers.18.mlp.switch_mlp.up_proj.weight, language_model.model.layers.18.self_attn.kv_a_layernorm.weight, language_model.model.layers.18.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.18.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.18.self_attn.kv_b_proj.scales, language_model.model.layers.18.self_attn.kv_b_proj.weight, language_model.model.layers.18.self_attn.q_a_layernorm.weight, language_model.model.layers.18.self_attn.q_a_proj.scales, language_model.model.layers.18.self_attn.q_a_proj.weight, language_model.model.layers.18.self_attn.q_b_proj.scales, language_model.model.layers.18.self_attn.q_b_proj.weight, language_model.model.layers.19.mlp.gate.weight, language_model.model.layers.19.mlp.shared_experts.down_proj.scales, language_model.model.layers.19.mlp.shared_experts.down_proj.weight, language_model.model.layers.19.mlp.shared_experts.gate_proj.scales, language_model.model.layers.19.mlp.shared_experts.gate_proj.weight, language_model.model.layers.19.mlp.shared_experts.up_proj.scales, language_model.model.layers.19.mlp.shared_experts.up_proj.weight, language_model.model.layers.19.mlp.switch_mlp.down_proj.biases, language_model.model.layers.19.mlp.switch_mlp.down_proj.scales, language_model.model.layers.19.mlp.switch_mlp.down_proj.weight, language_model.model.layers.19.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.19.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.19.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.19.mlp.switch_mlp.up_proj.biases, language_model.model.layers.19.mlp.switch_mlp.up_proj.scales, language_model.model.layers.19.mlp.switch_mlp.up_proj.weight, language_model.model.layers.19.self_attn.kv_a_layernorm.weight, language_model.model.layers.19.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.19.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.19.self_attn.kv_b_proj.scales, language_model.model.layers.19.self_attn.kv_b_proj.weight, language_model.model.layers.19.self_attn.q_a_layernorm.weight, language_model.model.layers.19.self_attn.q_a_proj.scales, language_model.model.layers.19.self_attn.q_a_proj.weight, language_model.model.layers.19.self_attn.q_b_proj.scales, language_model.model.layers.19.self_attn.q_b_proj.weight, language_model.model.layers.2.mlp.gate.weight, language_model.model.layers.2.mlp.shared_experts.down_proj.scales, language_model.model.layers.2.mlp.shared_experts.down_proj.weight, language_model.model.layers.2.mlp.shared_experts.gate_proj.scales, language_model.model.layers.2.mlp.shared_experts.gate_proj.weight, language_model.model.layers.2.mlp.shared_experts.up_proj.scales, language_model.model.layers.2.mlp.shared_experts.up_proj.weight, language_model.model.layers.2.mlp.switch_mlp.down_proj.biases, language_model.model.layers.2.mlp.switch_mlp.down_proj.scales, language_model.model.layers.2.mlp.switch_mlp.down_proj.weight, language_model.model.layers.2.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.2.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.2.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.2.mlp.switch_mlp.up_proj.biases, language_model.model.layers.2.mlp.switch_mlp.up_proj.scales, language_model.model.layers.2.mlp.switch_mlp.up_proj.weight, language_model.model.layers.2.self_attn.kv_a_layernorm.weight, language_model.model.layers.2.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.2.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.2.self_attn.kv_b_proj.scales, language_model.model.layers.2.self_attn.kv_b_proj.weight, language_model.model.layers.2.self_attn.q_a_layernorm.weight, language_model.model.layers.2.self_attn.q_a_proj.scales, language_model.model.layers.2.self_attn.q_a_proj.weight, language_model.model.layers.2.self_attn.q_b_proj.scales, language_model.model.layers.2.self_attn.q_b_proj.weight, language_model.model.layers.20.mlp.gate.weight, language_model.model.layers.20.mlp.shared_experts.down_proj.scales, language_model.model.layers.20.mlp.shared_experts.down_proj.weight, language_model.model.layers.20.mlp.shared_experts.gate_proj.scales, language_model.model.layers.20.mlp.shared_experts.gate_proj.weight, language_model.model.layers.20.mlp.shared_experts.up_proj.scales, language_model.model.layers.20.mlp.shared_experts.up_proj.weight, language_model.model.layers.20.mlp.switch_mlp.down_proj.biases, language_model.model.layers.20.mlp.switch_mlp.down_proj.scales, language_model.model.layers.20.mlp.switch_mlp.down_proj.weight, language_model.model.layers.20.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.20.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.20.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.20.mlp.switch_mlp.up_proj.biases, language_model.model.layers.20.mlp.switch_mlp.up_proj.scales, language_model.model.layers.20.mlp.switch_mlp.up_proj.weight, language_model.model.layers.20.self_attn.kv_a_layernorm.weight, language_model.model.layers.20.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.20.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.20.self_attn.kv_b_proj.scales, language_model.model.layers.20.self_attn.kv_b_proj.weight, language_model.model.layers.20.self_attn.q_a_layernorm.weight, language_model.model.layers.20.self_attn.q_a_proj.scales, language_model.model.layers.20.self_attn.q_a_proj.weight, language_model.model.layers.20.self_attn.q_b_proj.scales, language_model.model.layers.20.self_attn.q_b_proj.weight, language_model.model.layers.21.mlp.gate.weight, language_model.model.layers.21.mlp.shared_experts.down_proj.scales, language_model.model.layers.21.mlp.shared_experts.down_proj.weight, language_model.model.layers.21.mlp.shared_experts.gate_proj.scales, language_model.model.layers.21.mlp.shared_experts.gate_proj.weight, language_model.model.layers.21.mlp.shared_experts.up_proj.scales, language_model.model.layers.21.mlp.shared_experts.up_proj.weight, language_model.model.layers.21.mlp.switch_mlp.down_proj.biases, language_model.model.layers.21.mlp.switch_mlp.down_proj.scales, language_model.model.layers.21.mlp.switch_mlp.down_proj.weight, language_model.model.layers.21.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.21.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.21.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.21.mlp.switch_mlp.up_proj.biases, language_model.model.layers.21.mlp.switch_mlp.up_proj.scales, language_model.model.layers.21.mlp.switch_mlp.up_proj.weight, language_model.model.layers.21.self_attn.kv_a_layernorm.weight, language_model.model.layers.21.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.21.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.21.self_attn.kv_b_proj.scales, language_model.model.layers.21.self_attn.kv_b_proj.weight, language_model.model.layers.21.self_attn.q_a_layernorm.weight, language_model.model.layers.21.self_attn.q_a_proj.scales, language_model.model.layers.21.self_attn.q_a_proj.weight, language_model.model.layers.21.self_attn.q_b_proj.scales, language_model.model.layers.21.self_attn.q_b_proj.weight, language_model.model.layers.22.mlp.gate.weight, language_model.model.layers.22.mlp.shared_experts.down_proj.scales, language_model.model.layers.22.mlp.shared_experts.down_proj.weight, language_model.model.layers.22.mlp.shared_experts.gate_proj.scales, language_model.model.layers.22.mlp.shared_experts.gate_proj.weight, language_model.model.layers.22.mlp.shared_experts.up_proj.scales, language_model.model.layers.22.mlp.shared_experts.up_proj.weight, language_model.model.layers.22.mlp.switch_mlp.down_proj.biases, language_model.model.layers.22.mlp.switch_mlp.down_proj.scales, language_model.model.layers.22.mlp.switch_mlp.down_proj.weight, language_model.model.layers.22.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.22.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.22.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.22.mlp.switch_mlp.up_proj.biases, language_model.model.layers.22.mlp.switch_mlp.up_proj.scales, language_model.model.layers.22.mlp.switch_mlp.up_proj.weight, language_model.model.layers.22.self_attn.kv_a_layernorm.weight, language_model.model.layers.22.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.22.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.22.self_attn.kv_b_proj.scales, language_model.model.layers.22.self_attn.kv_b_proj.weight, language_model.model.layers.22.self_attn.q_a_layernorm.weight, language_model.model.layers.22.self_attn.q_a_proj.scales, language_model.model.layers.22.self_attn.q_a_proj.weight, language_model.model.layers.22.self_attn.q_b_proj.scales, language_model.model.layers.22.self_attn.q_b_proj.weight, language_model.model.layers.23.mlp.gate.weight, language_model.model.layers.23.mlp.shared_experts.down_proj.scales, language_model.model.layers.23.mlp.shared_experts.down_proj.weight, language_model.model.layers.23.mlp.shared_experts.gate_proj.scales, language_model.model.layers.23.mlp.shared_experts.gate_proj.weight, language_model.model.layers.23.mlp.shared_experts.up_proj.scales, language_model.model.layers.23.mlp.shared_experts.up_proj.weight, language_model.model.layers.23.mlp.switch_mlp.down_proj.biases, language_model.model.layers.23.mlp.switch_mlp.down_proj.scales, language_model.model.layers.23.mlp.switch_mlp.down_proj.weight, language_model.model.layers.23.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.23.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.23.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.23.mlp.switch_mlp.up_proj.biases, language_model.model.layers.23.mlp.switch_mlp.up_proj.scales, language_model.model.layers.23.mlp.switch_mlp.up_proj.weight, language_model.model.layers.23.self_attn.kv_a_layernorm.weight, language_model.model.layers.23.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.23.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.23.self_attn.kv_b_proj.scales, language_model.model.layers.23.self_attn.kv_b_proj.weight, language_model.model.layers.23.self_attn.q_a_layernorm.weight, language_model.model.layers.23.self_attn.q_a_proj.scales, language_model.model.layers.23.self_attn.q_a_proj.weight, language_model.model.layers.23.self_attn.q_b_proj.scales, language_model.model.layers.23.self_attn.q_b_proj.weight, language_model.model.layers.24.mlp.gate.weight, language_model.model.layers.24.mlp.shared_experts.down_proj.scales, language_model.model.layers.24.mlp.shared_experts.down_proj.weight, language_model.model.layers.24.mlp.shared_experts.gate_proj.scales, language_model.model.layers.24.mlp.shared_experts.gate_proj.weight, language_model.model.layers.24.mlp.shared_experts.up_proj.scales, language_model.model.layers.24.mlp.shared_experts.up_proj.weight, language_model.model.layers.24.mlp.switch_mlp.down_proj.biases, language_model.model.layers.24.mlp.switch_mlp.down_proj.scales, language_model.model.layers.24.mlp.switch_mlp.down_proj.weight, language_model.model.layers.24.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.24.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.24.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.24.mlp.switch_mlp.up_proj.biases, language_model.model.layers.24.mlp.switch_mlp.up_proj.scales, language_model.model.layers.24.mlp.switch_mlp.up_proj.weight, language_model.model.layers.24.self_attn.kv_a_layernorm.weight, language_model.model.layers.24.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.24.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.24.self_attn.kv_b_proj.scales, language_model.model.layers.24.self_attn.kv_b_proj.weight, language_model.model.layers.24.self_attn.q_a_layernorm.weight, language_model.model.layers.24.self_attn.q_a_proj.scales, language_model.model.layers.24.self_attn.q_a_proj.weight, language_model.model.layers.24.self_attn.q_b_proj.scales, language_model.model.layers.24.self_attn.q_b_proj.weight, language_model.model.layers.25.mlp.gate.weight, language_model.model.layers.25.mlp.shared_experts.down_proj.scales, language_model.model.layers.25.mlp.shared_experts.down_proj.weight, language_model.model.layers.25.mlp.shared_experts.gate_proj.scales, language_model.model.layers.25.mlp.shared_experts.gate_proj.weight, language_model.model.layers.25.mlp.shared_experts.up_proj.scales, language_model.model.layers.25.mlp.shared_experts.up_proj.weight, language_model.model.layers.25.mlp.switch_mlp.down_proj.biases, language_model.model.layers.25.mlp.switch_mlp.down_proj.scales, language_model.model.layers.25.mlp.switch_mlp.down_proj.weight, language_model.model.layers.25.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.25.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.25.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.25.mlp.switch_mlp.up_proj.biases, language_model.model.layers.25.mlp.switch_mlp.up_proj.scales, language_model.model.layers.25.mlp.switch_mlp.up_proj.weight, language_model.model.layers.25.self_attn.kv_a_layernorm.weight, language_model.model.layers.25.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.25.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.25.self_attn.kv_b_proj.scales, language_model.model.layers.25.self_attn.kv_b_proj.weight, language_model.model.layers.25.self_attn.q_a_layernorm.weight, language_model.model.layers.25.self_attn.q_a_proj.scales, language_model.model.layers.25.self_attn.q_a_proj.weight, language_model.model.layers.25.self_attn.q_b_proj.scales, language_model.model.layers.25.self_attn.q_b_proj.weight, language_model.model.layers.26.mlp.gate.weight, language_model.model.layers.26.mlp.shared_experts.down_proj.scales, language_model.model.layers.26.mlp.shared_experts.down_proj.weight, language_model.model.layers.26.mlp.shared_experts.gate_proj.scales, language_model.model.layers.26.mlp.shared_experts.gate_proj.weight, language_model.model.layers.26.mlp.shared_experts.up_proj.scales, language_model.model.layers.26.mlp.shared_experts.up_proj.weight, language_model.model.layers.26.mlp.switch_mlp.down_proj.biases, language_model.model.layers.26.mlp.switch_mlp.down_proj.scales, language_model.model.layers.26.mlp.switch_mlp.down_proj.weight, language_model.model.layers.26.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.26.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.26.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.26.mlp.switch_mlp.up_proj.biases, language_model.model.layers.26.mlp.switch_mlp.up_proj.scales, language_model.model.layers.26.mlp.switch_mlp.up_proj.weight, language_model.model.layers.26.self_attn.kv_a_layernorm.weight, language_model.model.layers.26.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.26.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.26.self_attn.kv_b_proj.scales, language_model.model.layers.26.self_attn.kv_b_proj.weight, language_model.model.layers.26.self_attn.q_a_layernorm.weight, language_model.model.layers.26.self_attn.q_a_proj.scales, language_model.model.layers.26.self_attn.q_a_proj.weight, language_model.model.layers.26.self_attn.q_b_proj.scales, language_model.model.layers.26.self_attn.q_b_proj.weight, language_model.model.layers.27.mlp.gate.weight, language_model.model.layers.27.mlp.shared_experts.down_proj.scales, language_model.model.layers.27.mlp.shared_experts.down_proj.weight, language_model.model.layers.27.mlp.shared_experts.gate_proj.scales, language_model.model.layers.27.mlp.shared_experts.gate_proj.weight, language_model.model.layers.27.mlp.shared_experts.up_proj.scales, language_model.model.layers.27.mlp.shared_experts.up_proj.weight, language_model.model.layers.27.mlp.switch_mlp.down_proj.biases, language_model.model.layers.27.mlp.switch_mlp.down_proj.scales, language_model.model.layers.27.mlp.switch_mlp.down_proj.weight, language_model.model.layers.27.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.27.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.27.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.27.mlp.switch_mlp.up_proj.biases, language_model.model.layers.27.mlp.switch_mlp.up_proj.scales, language_model.model.layers.27.mlp.switch_mlp.up_proj.weight, language_model.model.layers.27.self_attn.kv_a_layernorm.weight, language_model.model.layers.27.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.27.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.27.self_attn.kv_b_proj.scales, language_model.model.layers.27.self_attn.kv_b_proj.weight, language_model.model.layers.27.self_attn.q_a_layernorm.weight, language_model.model.layers.27.self_attn.q_a_proj.scales, language_model.model.layers.27.self_attn.q_a_proj.weight, language_model.model.layers.27.self_attn.q_b_proj.scales, language_model.model.layers.27.self_attn.q_b_proj.weight, language_model.model.layers.28.mlp.gate.weight, language_model.model.layers.28.mlp.shared_experts.down_proj.scales, language_model.model.layers.28.mlp.shared_experts.down_proj.weight, language_model.model.layers.28.mlp.shared_experts.gate_proj.scales, language_model.model.layers.28.mlp.shared_experts.gate_proj.weight, language_model.model.layers.28.mlp.shared_experts.up_proj.scales, language_model.model.layers.28.mlp.shared_experts.up_proj.weight, language_model.model.layers.28.mlp.switch_mlp.down_proj.biases, language_model.model.layers.28.mlp.switch_mlp.down_proj.scales, language_model.model.layers.28.mlp.switch_mlp.down_proj.weight, language_model.model.layers.28.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.28.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.28.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.28.mlp.switch_mlp.up_proj.biases, language_model.model.layers.28.mlp.switch_mlp.up_proj.scales, language_model.model.layers.28.mlp.switch_mlp.up_proj.weight, language_model.model.layers.28.self_attn.kv_a_layernorm.weight, language_model.model.layers.28.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.28.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.28.self_attn.kv_b_proj.scales, language_model.model.layers.28.self_attn.kv_b_proj.weight, language_model.model.layers.28.self_attn.q_a_layernorm.weight, language_model.model.layers.28.self_attn.q_a_proj.scales, language_model.model.layers.28.self_attn.q_a_proj.weight, language_model.model.layers.28.self_attn.q_b_proj.scales, language_model.model.layers.28.self_attn.q_b_proj.weight, language_model.model.layers.29.mlp.gate.weight, language_model.model.layers.29.mlp.shared_experts.down_proj.scales, language_model.model.layers.29.mlp.shared_experts.down_proj.weight, language_model.model.layers.29.mlp.shared_experts.gate_proj.scales, language_model.model.layers.29.mlp.shared_experts.gate_proj.weight, language_model.model.layers.29.mlp.shared_experts.up_proj.scales, language_model.model.layers.29.mlp.shared_experts.up_proj.weight, language_model.model.layers.29.mlp.switch_mlp.down_proj.biases, language_model.model.layers.29.mlp.switch_mlp.down_proj.scales, language_model.model.layers.29.mlp.switch_mlp.down_proj.weight, language_model.model.layers.29.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.29.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.29.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.29.mlp.switch_mlp.up_proj.biases, language_model.model.layers.29.mlp.switch_mlp.up_proj.scales, language_model.model.layers.29.mlp.switch_mlp.up_proj.weight, language_model.model.layers.29.self_attn.kv_a_layernorm.weight, language_model.model.layers.29.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.29.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.29.self_attn.kv_b_proj.scales, language_model.model.layers.29.self_attn.kv_b_proj.weight, language_model.model.layers.29.self_attn.q_a_layernorm.weight, language_model.model.layers.29.self_attn.q_a_proj.scales, language_model.model.layers.29.self_attn.q_a_proj.weight, language_model.model.layers.29.self_attn.q_b_proj.scales, language_model.model.layers.29.self_attn.q_b_proj.weight, language_model.model.layers.3.mlp.gate.weight, language_model.model.layers.3.mlp.shared_experts.down_proj.scales, language_model.model.layers.3.mlp.shared_experts.down_proj.weight, language_model.model.layers.3.mlp.shared_experts.gate_proj.scales, language_model.model.layers.3.mlp.shared_experts.gate_proj.weight, language_model.model.layers.3.mlp.shared_experts.up_proj.scales, language_model.model.layers.3.mlp.shared_experts.up_proj.weight, language_model.model.layers.3.mlp.switch_mlp.down_proj.biases, language_model.model.layers.3.mlp.switch_mlp.down_proj.scales, language_model.model.layers.3.mlp.switch_mlp.down_proj.weight, language_model.model.layers.3.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.3.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.3.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.3.mlp.switch_mlp.up_proj.biases, language_model.model.layers.3.mlp.switch_mlp.up_proj.scales, language_model.model.layers.3.mlp.switch_mlp.up_proj.weight, language_model.model.layers.3.self_attn.kv_a_layernorm.weight, language_model.model.layers.3.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.3.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.3.self_attn.kv_b_proj.scales, language_model.model.layers.3.self_attn.kv_b_proj.weight, language_model.model.layers.3.self_attn.q_a_layernorm.weight, language_model.model.layers.3.self_attn.q_a_proj.scales, language_model.model.layers.3.self_attn.q_a_proj.weight, language_model.model.layers.3.self_attn.q_b_proj.scales, language_model.model.layers.3.self_attn.q_b_proj.weight, language_model.model.layers.30.mlp.gate.weight, language_model.model.layers.30.mlp.shared_experts.down_proj.scales, language_model.model.layers.30.mlp.shared_experts.down_proj.weight, language_model.model.layers.30.mlp.shared_experts.gate_proj.scales, language_model.model.layers.30.mlp.shared_experts.gate_proj.weight, language_model.model.layers.30.mlp.shared_experts.up_proj.scales, language_model.model.layers.30.mlp.shared_experts.up_proj.weight, language_model.model.layers.30.mlp.switch_mlp.down_proj.biases, language_model.model.layers.30.mlp.switch_mlp.down_proj.scales, language_model.model.layers.30.mlp.switch_mlp.down_proj.weight, language_model.model.layers.30.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.30.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.30.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.30.mlp.switch_mlp.up_proj.biases, language_model.model.layers.30.mlp.switch_mlp.up_proj.scales, language_model.model.layers.30.mlp.switch_mlp.up_proj.weight, language_model.model.layers.30.self_attn.kv_a_layernorm.weight, language_model.model.layers.30.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.30.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.30.self_attn.kv_b_proj.scales, language_model.model.layers.30.self_attn.kv_b_proj.weight, language_model.model.layers.30.self_attn.q_a_layernorm.weight, language_model.model.layers.30.self_attn.q_a_proj.scales, language_model.model.layers.30.self_attn.q_a_proj.weight, language_model.model.layers.30.self_attn.q_b_proj.scales, language_model.model.layers.30.self_attn.q_b_proj.weight, language_model.model.layers.31.mlp.gate.weight, language_model.model.layers.31.mlp.shared_experts.down_proj.scales, language_model.model.layers.31.mlp.shared_experts.down_proj.weight, language_model.model.layers.31.mlp.shared_experts.gate_proj.scales, language_model.model.layers.31.mlp.shared_experts.gate_proj.weight, language_model.model.layers.31.mlp.shared_experts.up_proj.scales, language_model.model.layers.31.mlp.shared_experts.up_proj.weight, language_model.model.layers.31.mlp.switch_mlp.down_proj.biases, language_model.model.layers.31.mlp.switch_mlp.down_proj.scales, language_model.model.layers.31.mlp.switch_mlp.down_proj.weight, language_model.model.layers.31.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.31.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.31.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.31.mlp.switch_mlp.up_proj.biases, language_model.model.layers.31.mlp.switch_mlp.up_proj.scales, language_model.model.layers.31.mlp.switch_mlp.up_proj.weight, language_model.model.layers.31.self_attn.kv_a_layernorm.weight, language_model.model.layers.31.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.31.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.31.self_attn.kv_b_proj.scales, language_model.model.layers.31.self_attn.kv_b_proj.weight, language_model.model.layers.31.self_attn.q_a_layernorm.weight, language_model.model.layers.31.self_attn.q_a_proj.scales, language_model.model.layers.31.self_attn.q_a_proj.weight, language_model.model.layers.31.self_attn.q_b_proj.scales, language_model.model.layers.31.self_attn.q_b_proj.weight, language_model.model.layers.32.mlp.gate.weight, language_model.model.layers.32.mlp.shared_experts.down_proj.scales, language_model.model.layers.32.mlp.shared_experts.down_proj.weight, language_model.model.layers.32.mlp.shared_experts.gate_proj.scales, language_model.model.layers.32.mlp.shared_experts.gate_proj.weight, language_model.model.layers.32.mlp.shared_experts.up_proj.scales, language_model.model.layers.32.mlp.shared_experts.up_proj.weight, language_model.model.layers.32.mlp.switch_mlp.down_proj.biases, language_model.model.layers.32.mlp.switch_mlp.down_proj.scales, language_model.model.layers.32.mlp.switch_mlp.down_proj.weight, language_model.model.layers.32.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.32.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.32.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.32.mlp.switch_mlp.up_proj.biases, language_model.model.layers.32.mlp.switch_mlp.up_proj.scales, language_model.model.layers.32.mlp.switch_mlp.up_proj.weight, language_model.model.layers.32.self_attn.kv_a_layernorm.weight, language_model.model.layers.32.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.32.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.32.self_attn.kv_b_proj.scales, language_model.model.layers.32.self_attn.kv_b_proj.weight, language_model.model.layers.32.self_attn.q_a_layernorm.weight, language_model.model.layers.32.self_attn.q_a_proj.scales, language_model.model.layers.32.self_attn.q_a_proj.weight, language_model.model.layers.32.self_attn.q_b_proj.scales, language_model.model.layers.32.self_attn.q_b_proj.weight, language_model.model.layers.33.mlp.gate.weight, language_model.model.layers.33.mlp.shared_experts.down_proj.scales, language_model.model.layers.33.mlp.shared_experts.down_proj.weight, language_model.model.layers.33.mlp.shared_experts.gate_proj.scales, language_model.model.layers.33.mlp.shared_experts.gate_proj.weight, language_model.model.layers.33.mlp.shared_experts.up_proj.scales, language_model.model.layers.33.mlp.shared_experts.up_proj.weight, language_model.model.layers.33.mlp.switch_mlp.down_proj.biases, language_model.model.layers.33.mlp.switch_mlp.down_proj.scales, language_model.model.layers.33.mlp.switch_mlp.down_proj.weight, language_model.model.layers.33.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.33.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.33.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.33.mlp.switch_mlp.up_proj.biases, language_model.model.layers.33.mlp.switch_mlp.up_proj.scales, language_model.model.layers.33.mlp.switch_mlp.up_proj.weight, language_model.model.layers.33.self_attn.kv_a_layernorm.weight, language_model.model.layers.33.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.33.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.33.self_attn.kv_b_proj.scales, language_model.model.layers.33.self_attn.kv_b_proj.weight, language_model.model.layers.33.self_attn.q_a_layernorm.weight, language_model.model.layers.33.self_attn.q_a_proj.scales, language_model.model.layers.33.self_attn.q_a_proj.weight, language_model.model.layers.33.self_attn.q_b_proj.scales, language_model.model.layers.33.self_attn.q_b_proj.weight, language_model.model.layers.34.mlp.gate.weight, language_model.model.layers.34.mlp.shared_experts.down_proj.scales, language_model.model.layers.34.mlp.shared_experts.down_proj.weight, language_model.model.layers.34.mlp.shared_experts.gate_proj.scales, language_model.model.layers.34.mlp.shared_experts.gate_proj.weight, language_model.model.layers.34.mlp.shared_experts.up_proj.scales, language_model.model.layers.34.mlp.shared_experts.up_proj.weight, language_model.model.layers.34.mlp.switch_mlp.down_proj.biases, language_model.model.layers.34.mlp.switch_mlp.down_proj.scales, language_model.model.layers.34.mlp.switch_mlp.down_proj.weight, language_model.model.layers.34.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.34.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.34.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.34.mlp.switch_mlp.up_proj.biases, language_model.model.layers.34.mlp.switch_mlp.up_proj.scales, language_model.model.layers.34.mlp.switch_mlp.up_proj.weight, language_model.model.layers.34.self_attn.kv_a_layernorm.weight, language_model.model.layers.34.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.34.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.34.self_attn.kv_b_proj.scales, language_model.model.layers.34.self_attn.kv_b_proj.weight, language_model.model.layers.34.self_attn.q_a_layernorm.weight, language_model.model.layers.34.self_attn.q_a_proj.scales, language_model.model.layers.34.self_attn.q_a_proj.weight, language_model.model.layers.34.self_attn.q_b_proj.scales, language_model.model.layers.34.self_attn.q_b_proj.weight, language_model.model.layers.35.mlp.gate.weight, language_model.model.layers.35.mlp.shared_experts.down_proj.scales, language_model.model.layers.35.mlp.shared_experts.down_proj.weight, language_model.model.layers.35.mlp.shared_experts.gate_proj.scales, language_model.model.layers.35.mlp.shared_experts.gate_proj.weight, language_model.model.layers.35.mlp.shared_experts.up_proj.scales, language_model.model.layers.35.mlp.shared_experts.up_proj.weight, language_model.model.layers.35.mlp.switch_mlp.down_proj.biases, language_model.model.layers.35.mlp.switch_mlp.down_proj.scales, language_model.model.layers.35.mlp.switch_mlp.down_proj.weight, language_model.model.layers.35.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.35.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.35.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.35.mlp.switch_mlp.up_proj.biases, language_model.model.layers.35.mlp.switch_mlp.up_proj.scales, language_model.model.layers.35.mlp.switch_mlp.up_proj.weight, language_model.model.layers.35.self_attn.kv_a_layernorm.weight, language_model.model.layers.35.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.35.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.35.self_attn.kv_b_proj.scales, language_model.model.layers.35.self_attn.kv_b_proj.weight, language_model.model.layers.35.self_attn.q_a_layernorm.weight, language_model.model.layers.35.self_attn.q_a_proj.scales, language_model.model.layers.35.self_attn.q_a_proj.weight, language_model.model.layers.35.self_attn.q_b_proj.scales, language_model.model.layers.35.self_attn.q_b_proj.weight, language_model.model.layers.4.mlp.gate.weight, language_model.model.layers.4.mlp.shared_experts.down_proj.scales, language_model.model.layers.4.mlp.shared_experts.down_proj.weight, language_model.model.layers.4.mlp.shared_experts.gate_proj.scales, language_model.model.layers.4.mlp.shared_experts.gate_proj.weight, language_model.model.layers.4.mlp.shared_experts.up_proj.scales, language_model.model.layers.4.mlp.shared_experts.up_proj.weight, language_model.model.layers.4.mlp.switch_mlp.down_proj.biases, language_model.model.layers.4.mlp.switch_mlp.down_proj.scales, language_model.model.layers.4.mlp.switch_mlp.down_proj.weight, language_model.model.layers.4.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.4.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.4.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.4.mlp.switch_mlp.up_proj.biases, language_model.model.layers.4.mlp.switch_mlp.up_proj.scales, language_model.model.layers.4.mlp.switch_mlp.up_proj.weight, language_model.model.layers.4.self_attn.kv_a_layernorm.weight, language_model.model.layers.4.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.4.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.4.self_attn.kv_b_proj.scales, language_model.model.layers.4.self_attn.kv_b_proj.weight, language_model.model.layers.4.self_attn.q_a_layernorm.weight, language_model.model.layers.4.self_attn.q_a_proj.scales, language_model.model.layers.4.self_attn.q_a_proj.weight, language_model.model.layers.4.self_attn.q_b_proj.scales, language_model.model.layers.4.self_attn.q_b_proj.weight, language_model.model.layers.5.mlp.gate.weight, language_model.model.layers.5.mlp.shared_experts.down_proj.scales, language_model.model.layers.5.mlp.shared_experts.down_proj.weight, language_model.model.layers.5.mlp.shared_experts.gate_proj.scales, language_model.model.layers.5.mlp.shared_experts.gate_proj.weight, language_model.model.layers.5.mlp.shared_experts.up_proj.scales, language_model.model.layers.5.mlp.shared_experts.up_proj.weight, language_model.model.layers.5.mlp.switch_mlp.down_proj.biases, language_model.model.layers.5.mlp.switch_mlp.down_proj.scales, language_model.model.layers.5.mlp.switch_mlp.down_proj.weight, language_model.model.layers.5.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.5.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.5.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.5.mlp.switch_mlp.up_proj.biases, language_model.model.layers.5.mlp.switch_mlp.up_proj.scales, language_model.model.layers.5.mlp.switch_mlp.up_proj.weight, language_model.model.layers.5.self_attn.kv_a_layernorm.weight, language_model.model.layers.5.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.5.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.5.self_attn.kv_b_proj.scales, language_model.model.layers.5.self_attn.kv_b_proj.weight, language_model.model.layers.5.self_attn.q_a_layernorm.weight, language_model.model.layers.5.self_attn.q_a_proj.scales, language_model.model.layers.5.self_attn.q_a_proj.weight, language_model.model.layers.5.self_attn.q_b_proj.scales, language_model.model.layers.5.self_attn.q_b_proj.weight, language_model.model.layers.6.mlp.gate.weight, language_model.model.layers.6.mlp.shared_experts.down_proj.scales, language_model.model.layers.6.mlp.shared_experts.down_proj.weight, language_model.model.layers.6.mlp.shared_experts.gate_proj.scales, language_model.model.layers.6.mlp.shared_experts.gate_proj.weight, language_model.model.layers.6.mlp.shared_experts.up_proj.scales, language_model.model.layers.6.mlp.shared_experts.up_proj.weight, language_model.model.layers.6.mlp.switch_mlp.down_proj.biases, language_model.model.layers.6.mlp.switch_mlp.down_proj.scales, language_model.model.layers.6.mlp.switch_mlp.down_proj.weight, language_model.model.layers.6.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.6.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.6.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.6.mlp.switch_mlp.up_proj.biases, language_model.model.layers.6.mlp.switch_mlp.up_proj.scales, language_model.model.layers.6.mlp.switch_mlp.up_proj.weight, language_model.model.layers.6.self_attn.kv_a_layernorm.weight, language_model.model.layers.6.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.6.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.6.self_attn.kv_b_proj.scales, language_model.model.layers.6.self_attn.kv_b_proj.weight, language_model.model.layers.6.self_attn.q_a_layernorm.weight, language_model.model.layers.6.self_attn.q_a_proj.scales, language_model.model.layers.6.self_attn.q_a_proj.weight, language_model.model.layers.6.self_attn.q_b_proj.scales, language_model.model.layers.6.self_attn.q_b_proj.weight, language_model.model.layers.7.mlp.gate.weight, language_model.model.layers.7.mlp.shared_experts.down_proj.scales, language_model.model.layers.7.mlp.shared_experts.down_proj.weight, language_model.model.layers.7.mlp.shared_experts.gate_proj.scales, language_model.model.layers.7.mlp.shared_experts.gate_proj.weight, language_model.model.layers.7.mlp.shared_experts.up_proj.scales, language_model.model.layers.7.mlp.shared_experts.up_proj.weight, language_model.model.layers.7.mlp.switch_mlp.down_proj.biases, language_model.model.layers.7.mlp.switch_mlp.down_proj.scales, language_model.model.layers.7.mlp.switch_mlp.down_proj.weight, language_model.model.layers.7.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.7.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.7.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.7.mlp.switch_mlp.up_proj.biases, language_model.model.layers.7.mlp.switch_mlp.up_proj.scales, language_model.model.layers.7.mlp.switch_mlp.up_proj.weight, language_model.model.layers.7.self_attn.kv_a_layernorm.weight, language_model.model.layers.7.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.7.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.7.self_attn.kv_b_proj.scales, language_model.model.layers.7.self_attn.kv_b_proj.weight, language_model.model.layers.7.self_attn.q_a_layernorm.weight, language_model.model.layers.7.self_attn.q_a_proj.scales, language_model.model.layers.7.self_attn.q_a_proj.weight, language_model.model.layers.7.self_attn.q_b_proj.scales, language_model.model.layers.7.self_attn.q_b_proj.weight, language_model.model.layers.8.mlp.gate.weight, language_model.model.layers.8.mlp.shared_experts.down_proj.scales, language_model.model.layers.8.mlp.shared_experts.down_proj.weight, language_model.model.layers.8.mlp.shared_experts.gate_proj.scales, language_model.model.layers.8.mlp.shared_experts.gate_proj.weight, language_model.model.layers.8.mlp.shared_experts.up_proj.scales, language_model.model.layers.8.mlp.shared_experts.up_proj.weight, language_model.model.layers.8.mlp.switch_mlp.down_proj.biases, language_model.model.layers.8.mlp.switch_mlp.down_proj.scales, language_model.model.layers.8.mlp.switch_mlp.down_proj.weight, language_model.model.layers.8.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.8.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.8.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.8.mlp.switch_mlp.up_proj.biases, language_model.model.layers.8.mlp.switch_mlp.up_proj.scales, language_model.model.layers.8.mlp.switch_mlp.up_proj.weight, language_model.model.layers.8.self_attn.kv_a_layernorm.weight, language_model.model.layers.8.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.8.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.8.self_attn.kv_b_proj.scales, language_model.model.layers.8.self_attn.kv_b_proj.weight, language_model.model.layers.8.self_attn.q_a_layernorm.weight, language_model.model.layers.8.self_attn.q_a_proj.scales, language_model.model.layers.8.self_attn.q_a_proj.weight, language_model.model.layers.8.self_attn.q_b_proj.scales, language_model.model.layers.8.self_attn.q_b_proj.weight, language_model.model.layers.9.mlp.gate.weight, language_model.model.layers.9.mlp.shared_experts.down_proj.scales, language_model.model.layers.9.mlp.shared_experts.down_proj.weight, language_model.model.layers.9.mlp.shared_experts.gate_proj.scales, language_model.model.layers.9.mlp.shared_experts.gate_proj.weight, language_model.model.layers.9.mlp.shared_experts.up_proj.scales, language_model.model.layers.9.mlp.shared_experts.up_proj.weight, language_model.model.layers.9.mlp.switch_mlp.down_proj.biases, language_model.model.layers.9.mlp.switch_mlp.down_proj.scales, language_model.model.layers.9.mlp.switch_mlp.down_proj.weight, language_model.model.layers.9.mlp.switch_mlp.gate_proj.biases, language_model.model.layers.9.mlp.switch_mlp.gate_proj.scales, language_model.model.layers.9.mlp.switch_mlp.gate_proj.weight, language_model.model.layers.9.mlp.switch_mlp.up_proj.biases, language_model.model.layers.9.mlp.switch_mlp.up_proj.scales, language_model.model.layers.9.mlp.switch_mlp.up_proj.weight, language_model.model.layers.9.self_attn.kv_a_layernorm.weight, language_model.model.layers.9.self_attn.kv_a_proj_with_mqa.scales, language_model.model.layers.9.self_attn.kv_a_proj_with_mqa.weight, language_model.model.layers.9.self_attn.kv_b_proj.scales, language_model.model.layers.9.self_attn.kv_b_proj.weight, language_model.model.layers.9.self_attn.q_a_layernorm.weight, language_model.model.layers.9.self_attn.q_a_proj.scales, language_model.model.layers.9.self_attn.q_a_proj.weight, language_model.model.layers.9.self_attn.q_b_proj.scales, language_model.model.layers.9.self_attn.q_b_proj.weight.