Unverified50% confidenceFactExact time
vLLM v0.33.0rc2 includes a bugfix commit (#64567) from AMD engineer Micah Williamson that skips the AITER-tuned GEMM path on the gfx950 architecture when M*N>=2^31.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/30/2026
First Seen
Valid until: 1/28/2027
Sources
Related Entities
Related Claims
UnverifiedvLLM v0.28.0rc2 introduces a new speculative decoding feature called DFlash2, which combines two major technical components: local convolution and a candidate selector (PR #52816).72% similarUnverifiedvLLM released the v0.30.1rc0 pre-release version, introducing dense NVFP4 kernel and MoRI kernel image support for AMD MI355 accelerator cards (PR #58281).70% similarUnverifiedvLLM maintainer Bugen Zhao tagged the vllm-proto component with version proto-v0.4.0, corresponding to commit hash f28a508.69% similarUnverifiedvLLM v0.27.0 enables the sm_107 target for the NVIDIA Rubin architecture and provides an NVLink all-reduce path on SM107.68% similarUnverifiedThe deployment method officially recommended by vLLM does not support running FP8 models on Turing architecture cards such as the 2080 Ti.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1115073API
curl https://kongchang.com/api/v1/knowledge/claims/1115073MCP
get_claim(id=1115073)