@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 2a2c135b41
Fork 0

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 17, 06:45 UTC

Fix loading error when safetensors contains empty tensor (#1687)

twaka committed Nov 17, 2023 at 03:38 UTC 2a2c135b41e8894a8feb1a2b8a434bb23d3eb623
1 file changed +6 -1
vllm/model_executor/weight_utils.py
+6 -1
@@ -258,7 +258,12 @@ def convert_pyslice_to_tensor(x: Any) -> torch.Tensor:
258 tensor first.
259 """
260 if not isinstance(x, torch.Tensor):
261 - x = x[:]
261 + try:
262 + x = x[:]
263 + except IndexError:
264 + # IndexError happens when the tensor is empty.
265 + # transformer.h.0.attn.masked_bias is empty in some gpt2 models.
266 + return torch.Tensor()
267 return x
268
269