biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V0 deprecation] Remove V0 HPU backend (#21131)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon
2025-07-17 16:37:36 -07:00
committed by GitHub
parent ac9fb732a5
commit 4de7146351
27 changed files with 10 additions and 3926 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/model_executor/model_loader/bitsandbytes_loader.py
View File

@@ -199,10 +199,6 @@ class BitsAndBytesModelLoader(BaseModelLoader):
if self.pre_quant:
if self.load_8bit:
if current_platform.is_hpu():
raise ValueError(
"currently hpu supports 4bit quantization only")
return self._quantized_8bit_generator(
hf_weights_files, use_safetensors,
quant_state_dict), quant_state_dict
@@ -306,10 +302,6 @@ class BitsAndBytesModelLoader(BaseModelLoader):
in temp_state_dict):
quant_state = _parse_quant_state(mapped_weight_name,
temp_state_dict)
if current_platform.is_hpu():
assert quant_state.quant_type == "nf4", (
"currently hpu supports nf4 quant_type only")
quant_state_dict[mapped_weight_name] = quant_state
yield org_weight_name, weight_tensor
else:
@@ -380,8 +372,7 @@ class BitsAndBytesModelLoader(BaseModelLoader):
...]
# bitsandbytes requires data in GPU
if (weight_sub_tensor.is_cuda
or weight_sub_tensor.device.type == "hpu"):
if weight_sub_tensor.is_cuda:
loaded_weight = weight_sub_tensor
else:
loaded_weight = weight_sub_tensor.to(