mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2026-03-19 09:27:13 +08:00
ef89079712
Use bytearray to create a mutable copy of the binary data before passing to np.frombuffer. This ensures the numpy array is writable, avoiding UserWarning from torch.from_numpy on read-only arrays. The warning occurred because base64.b64decode returns immutable bytes, and np.frombuffer on immutable bytes returns a read-only array. When converted to a torch tensor via torch.from_numpy, PyTorch would emit: "UserWarning: The given buffer is not writable..." This fix maintains the efficient numpy-based conversion while ensuring compatibility with all embed_dtype formats (float32, float16, bfloat16, fp8_e4m3, fp8_e5m2) that numpy doesn't natively support. Fixes #26781 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> Signed-off-by: yurekami <yurekami@users.noreply.github.com>