[feature] MiniMax-M3-MXFP4 support added (#45896)
Signed-off-by: Qiang Li <qiang.li2@amd.com>
This commit is contained in:
@@ -924,6 +924,11 @@ class MiniMaxM3Model(nn.Module, EagleModelMixin):
|
||||
class MiniMaxM3SparseForCausalLM(nn.Module, SupportsEagle3):
|
||||
"""MiniMax M3 (sparse/dense backbone) for causal language modeling."""
|
||||
|
||||
packed_modules_mapping = {
|
||||
"qkv_proj": ["q_proj", "k_proj", "v_proj"],
|
||||
"gate_up_proj": ["gate_proj", "up_proj"],
|
||||
}
|
||||
|
||||
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
|
||||
super().__init__()
|
||||
config = vllm_config.model_config.hf_text_config
|
||||
@@ -987,14 +992,19 @@ class MiniMaxM3SparseForConditionalGeneration(
|
||||
# ranks (see ``_process_image_input`` / ``_process_video_input``).
|
||||
supports_encoder_tp_data = True
|
||||
|
||||
packed_modules_mapping = {
|
||||
"qkv_proj": ["q_proj", "k_proj", "v_proj"],
|
||||
"gate_up_proj": ["gate_proj", "up_proj"],
|
||||
}
|
||||
|
||||
hf_to_vllm_mapper = WeightsMapper(
|
||||
orig_to_new_prefix={
|
||||
"multi_modal_projector.": "vision_tower.multi_modal_projector.",
|
||||
"patch_merge_mlp.": "vision_tower.patch_merge_mlp.",
|
||||
},
|
||||
orig_to_new_substr={
|
||||
".mlp.fc1.": ".fc1.",
|
||||
".mlp.fc2.": ".fc2.",
|
||||
".mlp.fc1": ".fc1",
|
||||
".mlp.fc2": ".fc2",
|
||||
},
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user