diff --git a/vllm/model_executor/models/minimax_m2.py b/vllm/model_executor/models/minimax_m2.py index 2dc0f33cc..426caea1f 100644 --- a/vllm/model_executor/models/minimax_m2.py +++ b/vllm/model_executor/models/minimax_m2.py @@ -439,6 +439,17 @@ class MiniMaxM2Model(nn.Module): if is_pp_missing_parameter(name, self): continue + # Remap qkv_proj.[kv]_scale to attn.[kv]_scale + if name.endswith((".k_scale", ".v_scale")): + remapped_name = maybe_remap_kv_scale_name(name, params_dict) + if remapped_name is not None and remapped_name in params_dict: + param = params_dict[remapped_name] + weight_loader = getattr( + param, "weight_loader", default_weight_loader + ) + weight_loader(param, loaded_weight) + break + param = params_dict[name] weight_loader = param.weight_loader weight_loader(param, loaded_weight, shard_id)