xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-16 04:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Enable 4bit bnb prequant MOE (#21548)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Andy Chen 2025-08-11 19:02:14 -07:00 committed by GitHub
parent 1891a265d3
commit 9b94d6ec8f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 4 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/model_loader/bitsandbytes_loader.py
View File

@ -427,14 +427,10 @@ class BitsAndBytesModelLoader(BaseModelLoader):
elif isinstance(module, FusedMoE) and hasattr( elif isinstance(module, FusedMoE) and hasattr(
module.quant_method, "quant_config"): module.quant_method, "quant_config"):
# TODO: support FusedMoE with prequant and 8bit. # TODO: support FusedMoE with prequant and 8bit.
if self.pre_quant: if self.pre_quant and self.load_8bit:
raise ValueError( raise ValueError(
"Prequant BitsAndBytes models with FusedMoE is not " "Prequant BitsAndBytes 8bit models with FusedMoE "
"supported yet.") "is not supported yet.")
if self.load_8bit:
raise ValueError(
"BitsAndBytes 8bit quantization with FusedMoE is not "
"supported yet.")
# Get the corresponding weight name using module name and # Get the corresponding weight name using module name and
# expert_params_mapping. # expert_params_mapping.