vllm-project/vllm

Pull Requests Commits

[CI] Use uv-installed standalone Python in vllm-base (drop deadsnakes)

khluu committed 67 days ago

797cdc15

[CI] Use deadsnakes mirror for non-Ubuntu24.04 release images

khluu committed 67 days ago

438f6548

[DSV4] Add knob to enable pre-attn gemm (#41443)

zyongye committed 67 days ago

513b5f82

khluu committed 67 days ago

135722d5

[ROCm][Bugfix][GPTOSS]: fix input_ids and expert_map args for quark w4a8 gptoss (#41165)

Rohan138 committed 67 days ago

7758daca

[Perf] Intergrate Tile Kernels `head_compute_mix_kernel` for Deepseek-V4 (#41255)

Isotr0py committed 67 days ago

5f79634b

[Bugfix] Pass reasoning parser kwargs to structured output (#41199)

BugenZhao committed 67 days ago

05adde99

[Bugfix] BailingMoeV2.5: rotate full qk_rope_head_dim in MLA RoPE (#41185)

ZJY0516 committed 67 days ago

9cdcda5f

[Bugfix ] fix bailing_moe_linear (#40859)

ghphotoframe committed 67 days ago

a5d4a264

zyongye committed 68 days ago

a5e0ea09

[DSV4] Add BF16 and MXFP8 A2A support for flashinfer a2a one sided (#40960)

zyongye committed 68 days ago

28a1a85d

[Bugfix] Fix persistent_topk cooperative deadlock at TopK=1024 (#41189)

zyongye committed 68 days ago

82ccd9f3

[DSv4] Use `cvt` PTX for FP32->FP4 conversion (#41015)

gau-nernst committed 68 days ago

d9f3961a

[Bugfix] Fix repeated DSv4 RoPE cache initialization (#41148)

jeejeelee committed 68 days ago

4fee1390

[Bugfix] DSV32/V4 add missing type conversion for non-streaming tool calls (#41198)

chaunceyjiang committed 68 days ago

bf590b6a

[Bugfix] Fix max_num_batched_token not captured in cuda graph (#40734)

wzhao18 committed 68 days ago

66df2c03

[Bugfix] fix inductor error for dpsk v4 (#41135)

ZJY0516 committed 68 days ago

17896d81

[Bugfix] Fix Deepseek V4 import error due to AOT compile cache loading (#41090)

wzhao18 committed 68 days ago

1a131a06

[DSV4] Align aux stream API with DeepseekV4DecoderLayer (#41171)

zixi-qi committed 68 days ago

6fffb0b8

[Core] Account for `num_gpu_blocks_override` in `max_model_len` checks (#41069)

njhill committed 68 days ago

a076bd1e

[DSV4] Enable Multi-stream for Pre-Attn GEMM (#41061)

zyongye committed 68 days ago

c244b6fc

[Model][DSV4] Support base model (#41006)

jeejeelee committed 68 days ago

15cc87d2

Auto-disable expandable_segments around cumem memory pool (#40812)

youkaichao committed 68 days ago

c68fa39c

Temporary disable persistent topk (#41442)

zyongye committed 68 days ago

Verified 05ebca52

[Docker] Install numactl CLI in CUDA runtime image (#41032)

zhewenl committed 71 days ago

88d34c64

[DSV4] Add silu clamp limit to shared expert (#40950)

zyongye committed 72 days ago

b8160878

[Bugfix] Cap SWA/chunked-local runtime admission to startup pool-sizing bound (#40946)

Dao007forever committed 72 days ago

84c276d7

Revert "[Frontend] Remove frontend pooling multi task support. (#37861)"

khluu committed 72 days ago

5eb36575

[Feat] DeepSeek V4 Rebased (#40860)

ivanium committed 72 days ago

Verified 4d51588e

[torch.compile]: Disable Sequence Parallelism (SP) for piecewise compilation (#38373)

SouthWest7 committed 72 days ago

Verified 32e45636

Older