EngineX-Cambricon/enginex-mlu370-vllm
10
0
Fork 1
Code Issues Pull Requests Actions Projects Releases Wiki Activity

add deepseekv3 and llama4

Browse Source
This commit is contained in:
Chranos
2026-02-11 14:56:40 +08:00
parent 5da783780d
commit a21eae79a1
2 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
vllm-v0.6.2/vllm/model_executor/models/registry.py
View File

@@ -483,6 +483,7 @@ class _ModelRegistry:
architectures = [architectures] architectures = [architectures]
if not architectures: if not architectures:
logger.warning("No model architectures are specified") logger.warning("No model architectures are specified")
return []
return architectures return architectures

8
vllm-v0.6.2/vllm/transformers_utils/config.py
View File

@@ -229,6 +229,14 @@ def get_config(
model_type = MODEL_FOR_CAUSAL_LM_MAPPING_NAMES[config.model_type] model_type = MODEL_FOR_CAUSAL_LM_MAPPING_NAMES[config.model_type]
config.update({"architectures": [model_type]}) config.update({"architectures": [model_type]})
# Some composite config classes (e.g. Llama4Config) may not preserve
# the 'architectures' field from config.json. Restore it from the
# raw config_dict if needed.
if config_format == ConfigFormat.HF:
raw_archs = config_dict.get("architectures")
if raw_archs and not getattr(config, "architectures", None):
config.architectures = raw_archs
patch_rope_scaling(config) patch_rope_scaling(config)
if trust_remote_code: if trust_remote_code: