inference/research
verifiedhigh · 0.90updated 2026-10-03

vLLM issue #47159: triton_w4a16_gemm asserts qzeros.shape == (K // group_size, N // 8) on GPTQ models

forum

Fields

id
sources/gotcha-vllm-issue-47159-triton-w4a16-qzeros
URL
https://github.com/vllm-project/vllm/issues/47159
Publisher
vLLM GitHub issues
Kind
forum
Published
2026-06-30
Accessed
2026-10-03
notes
ROCm path; reproduces on two independent GPTQ-Int4 Qwen3.6-27B repos with --attention-backend TRITON_ATTN.

Notes

ROCm path; reproduces on two independent GPTQ-Int4 Qwen3.6-27B repos with --attention-backend TRITON_ATTN.

Cross-references

This record references no other records.