vllm
https://github.com/vllm-project/vllm
Python
A high-throughput and memory-efficient inference and serving engine for LLMs
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported51 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [ROCm][Perf][MiniMax-M3] Optimize sparse GQA prefill attention
- [ROCm][MoE] Pack UE8M0 scales directly in Triton silu_mul_quant
- [ROCm][Perf] W4A16: magic-bias dequant + scale hoist for gfx90a, gated on BLOCK_M <= 16
- [ROCm][Perf] W4A16 gfx90a: extend the narrow-tile rung to M<=16
- [ROCm] Fused Triton W8A16 fp8 GEMM for gfx90a + aiter arch-gate error message fix
- [Bug]: Block-FP8 tensor-parallel shards smaller than a quantization block fail to load
- [ROCm][Perf] Replace aceorch.einsum with batched_gemm_bf16 for Tâ¤32
- [Bug]: NemotronParseForConditionalGeneration does not tie lm_head.weight to decoder.embed_tokens.weight, produces garbage output
- [Performance][Model] Fuse GLM indexer attention projections
- [Bug][Spec Decode] All-NaN logits row is laundered into an out-of-vocab token id by the rejection sampler (device assert far from origin); NaN-metric blind spot on the verify path; uninit-scratch audit
- Docs
- Python not yet supported