A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 09:45 UTC
[Bugfix] fix adding bias twice in ipex GPTQ quantization (#18363)
Signed-off-by: rand-fly <randfly@outlook.com>
Random Fly committed
May 20, 2025 at 15:54 UTC
bca55b556f6a0cc7f8d75d5e12205be44eacd381
1 file changed
-2
vllm/model_executor/layers/quantization/ipex_quant.py
-2
@@ -181,8 +181,6 @@ class IPEXGPTQLinearMethod(GPTQLinearMethod):
181
bias: Optional[torch.Tensor] = None) -> torch.Tensor:
182
reshaped_x = x.reshape(-1, x.shape[-1])
183
out = layer.ipex_qlinear(reshaped_x)
184
- if bias is not None:
185
- out.add_(bias)
184
return out.reshape(x.shape[:-1] + (layer.ipex_output_size, ))
185
186