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 2, 23:15 UTC
main
Add `model` to `WeightTransferEngine.__init__` (#42922)
Sumanth R Hegde committed 5 months ago
[Attention] Add head_dim=512 support for FlashInfer trtllm attention backend (#38822)
Duncan Moss committed 5 months ago
elastic_ep: stage/commit MoE quant method on reconfigure (#40881)
Itay Alroy committed 5 months ago
[XPU]fix: add XPU platform guards to DeepSeek-V4 ops (#42950)
Ma Jian committed 5 months ago
[Perf] Optimize hidden state extraction logic (#37374)
Benjamin Chislett committed 5 months ago
DSv4 fused Q-norm kernel grid refactor (#42353)
gnovack committed 5 months ago
[Model Runner V2] Support sharing kv cache layers (#35045)
Nick Hill committed 5 months ago
[Quantization][ModelOpt] W4A16 NVFP4 fused MoE + mixed-precision dispatch (#42566)
Juhi Mittal committed 5 months ago
mhc_post - remove sts & add vectorized copies (#43437)
gnovack committed 5 months ago
[Bugfix] Clear error message for FP8 torchao quantization on unsupported GPUs (#36854)
haosdent committed 5 months ago
[Bugfix] Detect wrong libcute_dsl_runtime.so variant in FlashInfer GDN (#43427)
Artem Perevedentsev committed 5 months ago
[RFC][EPLB][#32028] Remove dead torch.accelerator.synchronize() from sync path (#40733)
SandishKumarHN committed 5 months ago
[Frontend] Simplify AuthenticationMiddleware path extraction (#43426)
Russell Bryant committed 5 months ago
[Refactor] Extract DeepSeek V4 sparse MLA impl into model folder (#43149)
Yongye Zhu committed 5 months ago
[KV Connector] MooncakeStore: don't co-queue save with load to avoid double delayed-free (#43371)
Dao007forever committed 5 months ago
[Bugfix] Source num_qo_heads from Attention layers in Flashinfer/Triton metadata builders (#42650)
Zhanda Zhu committed 5 months ago
[Rust Frontend] [Refactor] Extract a newtype for utility call ID (#43405)
Bugen Zhao committed 5 months ago
[Multimodal] Simplify ViT CUDA graph interfaces (#41234)
Isotr0py committed 5 months ago
Add NVFP4 MOE support for Deepseek V4. (#42209)
sychen52 committed 5 months ago
[CI] Fix AMD docker build tests (#43329)
haosdent committed 5 months ago