Commit Graph

Select branches

Hide Pull Requests

main

patches

sm86

Mono Color

• ceb340e2eb fix: prevent PyNvVideoCodec decoder slot limit bypass via ClassVar shadowing (#52126) Juan Pérez de Algaba 2026-08-17 19:45:40 +02:00
• cfbc5afbf7 [BugFix] lora_base_layer / routed_experts order in expert param mapping (#52552) ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟 2026-08-17 20:40:51 +03:00
• 49905ad94d [3/N][Feat][Perf] Add new warmup infrastructure for JITs. Add provider registry and orchestration for JIT warmup (#50174) Roberto L. CastroandCodex 2026-08-17 19:11:14 +02:00
• 7075ddac28 Support DSpark configs with architectures=DSparkDraftModel + model_type=qwen3 (#52197) Misha Goin 2026-08-17 12:48:30 -04:00
• 4ab5e5012a [Refactor] Simplify B12X linear kernels and warmup (#52368) Misha Goin 2026-08-17 12:45:45 -04:00
• 017e9f4448 Promote prefix_cache_retention_interval to an argument and change the default to 0 (#52216) Tyler Michael SmithandOpenAI Codex 2026-08-17 09:20:50 -04:00
• 1d3a8b9e22 [ROCm][Bugfix] Fix Triton W4A16 bug in determining if transpose is required for GPTQ/AutoGPTQ (#48998) qli88 2026-08-17 07:26:05 -05:00
• 70afdedc10 [K3] support recoverssm for K3 (#51855) 4 people 2026-08-17 19:34:14 +08:00
• f27ae25473 [Bugfix][CPU] Take an attention group's query head count from its layers (#51852) 3 people 2026-08-17 16:15:59 +05:30
• 95901ce70a fix(pooling): validate BGE-M3 combined task ownership (#51823) OlaandOpenAI Codex 2026-08-17 18:30:35 +08:00
• cc7cf71fc8 [XPU] Enable Kimi K3 KDA kernel tests on XPU (#51809) pmanczak 2026-08-17 11:21:34 +02:00
• bb233626ca [CI] Shard Humming A100 eval (#52325) Kevin H. Luu 2026-08-17 02:03:33 -07:00
• c05d923f18 [Doc] [ROCm] Update installation documentation (#52303) TJian 2026-08-17 16:56:44 +08:00
• 0ff370b51c docs: fix incorrect --custom-skip-chat-template flag reference (#52588) Amal Sebastian 2026-08-17 14:18:11 +05:30
• 5fd7a88838 [CI/Build] Fix accident pre-commit breakage due to concurrent merge (#52578) Isotr0py 2026-08-17 14:40:36 +08:00
• a02cfccbc6 [Bugfix][Mamba] Fix overlapping state copy race (#50729) Andreas KaratzasandOpenAI Codex 2026-08-17 01:29:41 -05:00
• 93550cc4cd [Frontend] Consolidate entrypoint middleware (#52309) wang.yuqi 2026-08-17 14:17:15 +08:00
• 53e211d292 [CI/Build] Reduce more duplicate runner startup in tests (#52570) Isotr0py 2026-08-17 13:19:07 +08:00
• 311b3513af [ROCm][CI] Avoid forcing FlashAttention in the ColPali pooling test (#52565) Andreas Karatzas 2026-08-16 23:08:35 -05:00
• 71b578b9cc [ROCm][CI] Use the same-build wheel in Python-only CI (#49514) Andreas Karatzas 2026-08-16 22:52:24 -05:00
• 7ea4b40954 [Hardware][NVIDIA] Add GB10 fused-MoE fp8 tuning configs (E=256, E=512) (#52502) pavelzak 2026-08-16 20:04:15 -07:00
• 292187dd8c [Bugfix][DSv4] Keep indexer scoring in breakable graphs (#52492) 3 people 2026-08-16 23:01:13 -04:00
• 6664d397bf [Performance][MRV2] Cache logits-processing request state (#52329) Cheng RuiandRui Rui4 Cheng 2026-08-16 19:36:41 -07:00
• 502af5ed00 [Doc] Add MatrixHub as a model loading source (#50492) Yiting Jiang 2026-08-17 10:32:23 +08:00
• a6a2a93f9b [Bugfix][Frontend] Guard remaining before-validators against non-object JSON bodies (#52528) Kaif Kohari 2026-08-17 03:27:59 +01:00
• 967e104fad [Config] Unify indexer cache dtype under attention_config.indexer_kv_dtype (#52550) Yongye ZhuandClaude Opus 5 2026-08-16 19:20:04 -07:00
• 0ad04cff1b [ROCm][CI] Enable ViT CUDA graph tests on AMD gfx950 GPUs (#52256) Shanshan ShenandAndreas Karatzas 2026-08-17 09:52:02 +08:00
• a18c9b56ff [Kimi-K3][Perf] Update FlashKDA for automatic K2 V-split (#52458) kiroxu 2026-08-17 09:04:38 +08:00
• dc9ae4b8ac [Bugfix][Mooncake] Reference GPU blocks for in-flight store jobs and key the store ledger by store_job_id (#52372) Guanyi Chen 2026-08-17 07:09:46 +08:00
• eee538d5da [Bugfix][V1][Multimodal] Ignore stale same-step encoder cache evictions (#52482) Tianyu Guo 2026-08-17 06:21:55 +08:00
• e3c1cb54fc [CI/Build] Avoid duplicate runner startup for multimodal test (#52417) Isotr0py 2026-08-17 06:17:25 +08:00
• 6b0b850a8b [CI] Fit small KV-offload evals within shared memory (#52496) Taneem Ibrahim 2026-08-16 15:16:51 -05:00
• fdab2b10bc [ModelRunner v2] Support Transformers pooling model (#52425) Taneem Ibrahim 2026-08-16 14:25:49 -05:00
• ef43e3101b [ROCm][DSV4][Perf] Optimize Triton sparse-MLA decode on gfx950 (#52212) 3 people 2026-08-16 15:16:46 -04:00
• 1f0e0bf612 [Bugfix][Attention] Temporarily disable FA4 head-dim 256 (#52050) Taneem IbrahimandWentao Ye 2026-08-16 12:13:02 -05:00
• 7d7b6f26f4 [Refactor] Remove dead code for quantization (#52221) Wentao Ye 2026-08-16 13:10:16 -04:00
• 6914d60b1e [ROCm][Perf] gfx942: use FlyDSL fp8 MQA logits kernel (ROCm/aiter#3913) (#49544) akii96 2026-08-16 19:50:31 +03:00
• 83f591d7f6 [Perf][DSV4] Optimize global top-k index kernel with compile-time constants (#51967) Chaunceyandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-16 23:24:14 +08:00
• 9409f59e09 [Core] Add CuMemAllocator.discard() for tag-selective GPU memory release (#52514) 3 people 2026-08-16 23:15:14 +08:00
• fe1c317157 [Bugfix][ROCm] Skip FP8 MLA prefill PS-metadata build for chunked-context batches (#52356) Shantipriya Parida 2026-08-16 16:21:31 +03:00
• 836aac92ff [Perf][DSV4] Optimize sparse top-k metadata kernels for higher prefill throughput (#52084) Chauncey 2026-08-16 21:15:47 +08:00
• 4d2a68d64d [Bugfix][Spec Decode][Structured Output] DSpark: fix the grammar bitmask mapping when the draft budget is zero (#52436) 3 people 2026-08-16 19:09:23 +08:00
• 84530eb235 [Bugfix][Multimodal] Keep Gemma 4 video frame counts on CPU (#52441) Chauncey 2026-08-16 17:17:04 +08:00
• 1b079c40ff [Bugfix][Model Runner V2][Spec Decode] Fix off-by-one in bad_words draft-prefix matching (#52311) 4 people 2026-08-16 17:02:20 +08:00
• 70aaec832b [Bugfix][Anthropic] Return 4xx for client-caused errors in /v1/messages (#52246) Do_it_now_! 2026-08-16 16:05:17 +08:00
• 41f12a0daf [Bugfix] Raise VLLMValidationError from structured output validators (#52394) Jeffrey WangandClaude Opus 5 2026-08-16 01:04:51 -07:00
• 8efa13b700 [Bugfix] Pick the DeepSeek V4 eager cudagraph region per model runner (#52401) Nick Hill 2026-08-15 21:04:08 -07:00
• 6593754e61 [Bugfix][Spec Decode] Keep EAGLE cache registration on the partial-hash-hit path (#52419) mispa-ms 2026-08-15 21:03:53 -07:00
• edd4c8176c [Bugfix][DSv4] Revert adaptive C128A metadata packing (#51318) 5 people 2026-08-15 18:39:23 -07:00
• fa9d67f782 [EC Connector] Added Build Connector Worker Meta for EC Connector (#49585) 3 people 2026-08-16 01:40:38 +03:00
• c94cdd0ae0 [Bugfix][Sampling] Clear empty side on thinking-budget asymmetric SWAP (#49613) Henry Su 2026-08-15 16:13:02 -05:00
• ed0f4750f8 [Bugfix][Model] Kimi-K3 MegaMoE: pass situ_beta/situ_linear_beta to fp8_fp4_mega_moe (#52445) UranusandClaude 2026-08-16 02:59:22 +08:00
• d4801990a4 [CI/Build] Add warning for unsupported global PTX architecture requests in... (#51901) Shane Widanagama 2026-08-15 10:01:52 -04:00
• 97388c44f9 [Bugfix] Make DSV4 sparse MLA work end-to-end for plain decode, MTP, and DSpark (#51538) Gabriel Wu 2026-08-15 22:01:32 +08:00
• ac2ae8798c [Frontend] Support count_reasoning_tokens in the Streaming Parser Engine (#45802) Chaunceyandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-15 21:51:14 +08:00
• 5cecfc0137 [Bugfix] Fix modelscope usage (#52431) Cyrus Leung 2026-08-15 17:40:24 +08:00
• 42156466db [Rust Frontend][gRPC] Preserve skip_special_tokens decoding option (#52384) Biswa Panda 2026-08-15 01:54:58 -07:00
• 44fc57d7b7 [ROCm][CI] Select CPU platform for native no-GPU jobs (#49515) Andreas Karatzas 2026-08-15 01:30:56 -05:00
• acb0f1dcdb [Bugfix][Spec Decode] DSpark: inherit the target's attention backend when the speculative config names none (#52288) Yongye Zhu 2026-08-14 18:32:59 -07:00
• 615d4cfade [Core] Check for GPU<->CPU syncs during CI (#43107) Nick HillandClaude Opus 4.8 2026-08-14 17:42:05 -07:00
• 7b544ecb52 [Spec Decode][Perf] Fuse the MTP trailing all-reduce; local-argmax draft tokens (#49793) 3 people 2026-08-15 08:07:07 +08:00
• d6f17f3d53 [MRV2] Support attention-free models (#52374) Nick Hill 2026-08-14 16:54:50 -07:00
• 9df9b0b0a1 [ROCm]: Drop pybind11 from Dockerfile.rocm to prevent version mismatch (#52400) Rohan Potdar 2026-08-14 18:50:23 -05:00
• bb4b448f98 [CI] Shard Humming H100 eval (#52326) Kevin H. Luu 2026-08-14 15:12:02 -07:00
• 81e81dac1f [CI] Shard MoE refactor B200 eval (#52327) Kevin H. Luu 2026-08-14 15:11:55 -07:00
• 694db075f8 [CI] Shard multimodal extended generation 2 (#52323) Kevin H. Luu 2026-08-14 15:11:48 -07:00
• 549cef0b0b [CI] Shard extended pooling model tests (#52322) Kevin H. Luu 2026-08-14 15:11:41 -07:00
• d87ef456ab [CI] Shard Quantization job into 4 parallel shards (≤30 min target) (#52328) Kevin H. Luu 2026-08-14 15:11:32 -07:00
• 925ea7e60f [CI][Test] Seed the DeepEP v2 MoE workers, not just the parent (#50589) Guanxin LiandClaude Opus 5 2026-08-14 13:01:24 -07:00
• c794754062 [ROCm]Remove special-case SiTU support model-specific gating (#50597) sroberts-amdandStacy Roberts 2026-08-14 13:38:04 -05:00
• f473870ecf [Rust Frontend][gRPC] Add RL lifecycle control (#51316) Connor CarpenterandNick Hill 2026-08-14 11:32:39 -07:00
• 3e3ceb1961 [Perf] Avoid more GPU<->CPU syncs in multimodal encoders (#52369) Nick Hill 2026-08-14 10:40:14 -07:00
• 9b0ab5dd53 [ROCm][AMD] Enable preshuffled sparse indexing for 16-token blocks (#51216) James E T SmithandAndreas Karatzas 2026-08-14 13:38:14 -04:00
• 83ded8d839 [Test][LoRA] Speed up the LoRA test job (#52331) stefankoncarevic 2026-08-14 19:08:41 +02:00
• e078a2238a [Bugfix] Widen flashinfer.comm import guard so a failed import doesn't abort engine startup (#52241) shanjiaz 2026-08-14 12:27:24 -04:00
• 03a8d0b1ed [Model][Spec Decode] Tap the pre-norm AttnRes mixture as the Kimi K3 DFlash aux state (#50487) Rahul ChalamalaandJanelle Cai 2026-08-14 12:03:42 -04:00
• cdc4824a21 [Misc] Remove override_attention_dtype (#48684) wangxiyuanandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 2026-08-14 23:15:43 +08:00
• 1f7427bc0a [UT][XPU] fix b12x UT (#52265) Qiming ZhangandAndreas Karatzas 2026-08-14 06:49:47 -07:00
• 57bd0ed441 [5/N][KV-Cache Layout Refactor] Backend-published KV packing via customize_spec (#51704) Lucas WilkinsonandClaude 2026-08-14 09:47:55 -04:00
• 63a9a5010a [Attention][DSA] Take the native decode path for MTP=3 on SM90 (#52164) Zhuobin Huangandnodeeeeee 2026-08-14 18:09:51 +08:00
• aa31003573 [Bugfix][Helm] Fix chart resource references (#51664) iwannagotobedandChauncey 2026-08-14 18:47:42 +09:00
• 69e0e58da1 [Doc] Update model support information (#52289) Jee Jee Li 2026-08-14 17:47:30 +08:00
• 20405bfb15 [Bugfix] Fix Cosmos3-Edge processor after transformers 5.15 release (#51989) bastefaniak 2026-08-14 11:46:46 +02:00
• 66728feb1f [MRV2][Multimodal] Enable encoder cuda graph for model runner v2 (#49852) Isotr0pyandOpenAI Codex 2026-08-14 17:33:21 +08:00
• b8165e5e58 [Frontend] Consolidate entrypoint exception handler (#52261) wang.yuqi 2026-08-14 17:27:24 +08:00
• 103c419f04 [Perf][Frontend] Vectorize Cohere binary embedding bit-packing (#52277) Fangchen LiandClaude Opus 5 2026-08-14 02:15:59 -07:00
• 624999aae5 [XPU]bump up vllm_xpu_kernels to 0.1.13.2 (#52138) Kunshang Ji 2026-08-14 17:01:37 +08:00
• 3c8676aebb [PP][XPU]Overlap async-scheduling PP sampled-token broadcast with compute (#51650) YiSheng5andKunshang Ji 2026-08-14 17:00:23 +08:00
• d4c24e6f5d [CI] Increase extended generation test timeout (#52252) Lucas Wilkinson 2026-08-14 04:23:51 -04:00
• bda4c3e8ee [CPU] Fold the MXFP4 block scale in 2 instructions instead of 4 (#51583) ccaadaroandLi, Jiang <jiang1.li@intel.com> 2026-08-14 10:19:21 +02:00
• 8e6d8e4f6a [XPU][CI/Release][3/N] Add xpu wheel release to release pipeline (#52108) Kunshang Ji 2026-08-14 14:00:17 +08:00
• ac7509e2b1 [Bugfix][CPU][RISC-V] Fix build: make FP32Vec copy constructors non-explicit (#51099) velonica0andLi, Jiang <jiang1.li@intel.com> 2026-08-14 12:59:01 +08:00
• 6adad08767 Add Muse Glimmer model support (#51655) 7 people 2026-08-14 12:28:56 +08:00
• 59d1af5698 Detect ROCm wheel variant from environment for precompiled wheels. (#49365) 4 people 2026-08-13 23:19:20 -05:00
• 653cc6faca [Bugfix][NIXL] Include transfer mode (push/pull) in the compatibility hash (#50620) Tzu-Ling KanandClaude Opus 4.8 2026-08-13 23:11:22 -05:00
• c05d75aaa9 [Bugfix][Refactor] Keep Qwen3Next layer boundaries sequence parallel (#50685) 3 people 2026-08-14 11:58:32 +08:00
• 1be3628367 [Kernel][Perf] Add fused CUDA post-conv MTP decode kernel for Qwen3.5 GDN (#51674) 3 people 2026-08-14 11:44:45 +08:00
• d18bb7b390 [UT] fix device of test_outputs.py (#52237) Qiming Zhang 2026-08-13 20:29:54 -07:00
• 3c79b1a8bf [Kernel] Add B12X dense linear backends (#52016) Luke Alonso 2026-08-13 19:48:33 -07:00
• fe4c5dcd4c [XPU] [Bugfix] process ragged weights in xpu linear backend (#52118) zofia 2026-08-14 09:11:43 +08:00