Chranos/enginex-mlu370-vllm
1
0
Fork 0
forked from EngineX-Cambricon/enginex-mlu370-vllm
Code Pull Requests Activity

add deepseekv3 and llama4

Browse Source
This commit is contained in:
Chranos
2026-02-11 15:13:14 +08:00
parent eaeb5169e0
commit 86fd3b5a92
3 changed files with 3 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm-v0.6.2/vllm/model_executor/model_loader/utils.py
View File

@@ -21,10 +21,6 @@ def set_default_torch_dtype(dtype: torch.dtype):
def get_model_architecture(
model_config: ModelConfig) -> Tuple[Type[nn.Module], str]:
architectures = getattr(model_config.hf_config, "architectures", None) or []
print(f"[DEBUG-ARCH] get_model_architecture: "
f"type(hf_config)={type(model_config.hf_config).__name__}, "
f"architectures={getattr(model_config.hf_config, 'architectures', 'MISSING')}, "
f"id(hf_config)={id(model_config.hf_config)}")
# Special handling for quantized Mixtral.
# FIXME(woosuk): This is a temporary hack.
mixtral_supported = [

3
vllm-v0.6.2/vllm/model_executor/models/llama4.py
View File

@@ -450,7 +450,10 @@ class Llama4ForCausalLM(nn.Module, SupportsPP):
# which has text_config sub-config. Extract it for text model.
text_config = getattr(config, "text_config", None)
if text_config is not None:
orig_archs = getattr(config, "architectures", None)
vllm_config.model_config.hf_config = text_config
if orig_archs and not getattr(text_config, "architectures", None):
text_config.architectures = orig_archs
config = text_config
quant_config = vllm_config.quant_config
lora_config = vllm_config.lora_config