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 4, 11:15 UTC
main
[ROCm] Use quant_dtype in per_token_quant instead of hardcoded FP8 (#39121)
Andrew Barnes committed 6 months ago
[Feature]: IndexCache support for DSA models (#37735)
Chauncey committed 6 months ago
Laith Sakka committed 6 months ago
Fix timeout when using LoRA adapters with Nemotron Super (#40916)
danisereb committed 6 months ago
[CI/Build] Enable FP8 on NVIDIA Thor (#39712)
Cyrus Leung committed 6 months ago
better logging for large uncachable items (#41145)
Avshalom Manevich committed 6 months ago
Terrence Zhao committed 6 months ago
[Bugfix] Report compile time for in-memory cache hit path (#41023)
Frederik Gossen committed 6 months ago
[Model Runner v2] Fix block table IMA issue (#40648)
Wentao Ye committed 6 months ago
Wentao Ye committed 6 months ago
[Perf] Enable FlashInfer top-k/top-p sampler by default (#40376)
Artem Perevedentsev committed 6 months ago
[Bugfix][CPU] Backport PT cpp codegen indirect_assert scalar-mask fix (#40973)
Lalithnarayan C committed 6 months ago
hf_name argument for vllm bench throughput CLI (#41012)
pmaybank committed 6 months ago
[Feat] CPU fp8 attn for AMX/AVX-512 (#39445)
Tianmu Li committed 6 months ago
[Bugfix] Fix repeated DSv4 RoPE cache initialization (#41148)
Jee Jee Li committed 6 months ago
[Frontend] Add `defer_loading` and `tool_reference` support for Anthropic and OpenAI APIs (#40190)
Jared Wen committed 6 months ago
[DSV4] Support `max` reasoning effort (#40982)
Bugen Zhao committed 6 months ago
[KV Offload] Tighten `keys` type from `Iterable` to `Sequence` in `OffloadingManager` (#41200)
Ronen Schaffer committed 6 months ago
Rohit Kumar Singh committed 6 months ago
[Bugfix] BailingMoeV2.5: rotate full qk_rope_head_dim in MLA RoPE (#41185)
Jiangyun Zhu committed 6 months ago