 ilyasher-harmonic
|
9ced8d0981
|
Optimize FP32 LM head for bf16/fp16 (#32370)
|
2026-08-11 15:14:50 -07:00 |
|
|
gongwei1027
|
2c07ca5e8d
|
[Fix] Allow flashinfer_sparse_mla DSA backend for HiSparse on SM120 FP8 KV (#33075)
|
2026-08-11 15:05:04 -07:00 |
|
|
|
93c1bff1d4
|
Fix default dtype restoration after model loader errors (#34440)
Co-authored-by: zhisbug <1654062+zhisbug@users.noreply.github.com>
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
|
2026-08-11 11:33:04 -07:00 |
|
|
Dmitrii SergeevandZhiqiang Xie
|
c58953d90a
|
O(1) slot allocation in ReqToTokenPool.alloc() (#32208)
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
|
2026-08-11 11:26:05 -07:00 |
|
|
Jeremy Zhang
|
aadb9720fe
|
fix: route scheduler aborts to multi-tokenizer workers (#33940)
|
2026-08-11 11:15:16 -07:00 |
|
|
huangtingweiandHanming Lu
|
8f3d3a31f4
|
[HiCache] Fix Mamba track-boundary bookkeeping under overlap scheduling (#29792)
Co-authored-by: Hanming Lu <hanminglu@meta.com>
|
2026-08-12 00:36:48 +08:00 |
|
|
Mohammad Miadh Angkad
|
a3bd7d9401
|
Bump CuTeDSL to 4.6.2 (#34372)
|
2026-08-12 00:35:35 +08:00 |
|
|
Mick
|
8267d76c2c
|
[VLM] replace deprecated image processor use_fast (#34175)
|
2026-08-12 00:14:07 +08:00 |
|
|
Ke Bao
|
b20c375c10
|
Fix flaky decode cache-hit check in Inkling test (#34405)
|
2026-08-11 21:47:01 +08:00 |
|
|
gjsheu
|
a50ab9cec7
|
[npu] [bugfix] Fix HiCache MHA backup for NPU (#34341)
|
2026-08-11 21:09:45 +08:00 |
|
|
Mohammad Miadh Angkad
|
2d193077f7
|
[JIT Kernel] Migrate per-token FP8 quantization from AOT to JIT (#34257)
|
2026-08-11 20:40:40 +08:00 |
|
|
|
a0a76e4485
|
[DSV4] perf: Enable alt stream during BCG prefill (#29070)
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
|
2026-08-11 05:10:19 -07:00 |
|
|
Xiaoyu ZhangandClaude Fable 5
|
546965fc72
|
[diffusion] LTX-2: mount the bit-exact fused modulate at the 8 bare adaLN sites (ltx23-one-stage denoise -2.8% H100 / -2.6% H200) (#34315)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
|
2026-08-11 18:23:08 +08:00 |
|
|
Mick
|
dd8c5849af
|
[diffusion] refactor: move dit execution capabilities to runtime models (#34249)
|
2026-08-11 18:20:43 +08:00 |
|
|
Xiaoyu ZhangandClaude Fable 5
|
071f0f1e9d
|
[diffusion] ERNIE-Image: fuse rotate-half RoPE + GELU-mul and hoist rope cos/sin (denoise -16.2% H100 / -12.7% H200, bit-exact) (#34306)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
|
2026-08-11 18:18:37 +08:00 |
|
|
Xiaoyu ZhangandClaude Fable 5
|
ba3dc16401
|
[diffusion] weight-only FP8: dequantize linear weights once at first use (Ideogram-4 denoise -18.8% H200 / -7.8% H100, bit-exact) (#34305)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
|
2026-08-11 18:15:26 +08:00 |
|
|
Zhiqiang XieandTingwei Huang
|
5469faec45
|
HiSparse: shared-index (IndexShare) plan-then-IO swap-in prefetch (#34329)
Co-authored-by: Tingwei Huang <huangtingwei9988@gmail.com>
|
2026-08-11 01:58:28 -07:00 |
|
|
|
e74ea5b1d7
|
[ROCm/gfx95] Fix fp8 per-channel attention for Kimi-K2.7-code-mxfp4 o… (#31105)
Co-authored-by: Hung <Emmanuel0612@users.noreply.github.com>
Co-authored-by: HaiShaw <hixiao@gmail.com>
|
2026-08-11 00:51:57 -07:00 |
|
|
ybyang
|
9d4be40124
|
Fix DSpark + DeepSeek V4 prefill CP compatibility (#33865)
|
2026-08-10 23:26:29 -07:00 |
|
|
Tanmay patil
|
afa2d5570b
|
[Diffusion] Add online FP8 support for Krea-2 (#34136)
|
2026-08-10 23:12:05 -07:00 |
|
|
YAMYandShangming Cai
|
667e18d99d
|
[PD] Support pipeline-parallel prefill with Mooncake staging buffer (#33807)
Co-authored-by: Shangming Cai <csmthu@gmail.com>
|
2026-08-11 13:53:17 +08:00 |
|
|
Xinyuan Tong
|
13aeb91b6e
|
[Fix] Update multimodal CUDA VMM helper import (#34358)
|
2026-08-10 22:41:21 -07:00 |
|
|
Liangsheng YinandAlex Nails
|
b3c02cbce7
|
[perf] Collapse the DP attention scheduler sync to a single D2H copy (#34338)
Co-authored-by: Alex Nails <alex.nails@radixark.ai>
|
2026-08-10 21:52:29 -07:00 |
|
|
Liangsheng Yin
|
585c3c6816
|
[Refactor] Split the FlashInfer autotune dummy-run flag from the LM-head policy (#34336)
|
2026-08-10 21:20:05 -07:00 |
|
|
EchO
|
a58fa0388e
|
[Fix] Correct W4AFP8 DeepEP scaling and mode-specific dtypes (#33669)
|
2026-08-11 11:21:42 +08:00 |
|
|
forrestlandClaude Opus 5
|
704808ed27
|
[MiniMax-H3] SubBlock: training-free block-sparse attention for the DiT (#34148)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
2026-08-11 11:11:11 +08:00 |
|
|
Zhiyao Jiang
|
dd20826e0a
|
[AMD] Preserve the AITER expert mask across torch_memory_saver pause/resume (#34220)
|
2026-08-10 19:59:15 -07:00 |
|
|
Baizhou Zhang
|
b498f46271
|
[CI] Add MegaMoE runner compatibility alias (#34326)
|
2026-08-10 19:47:38 -07:00 |
|
|
Mick
|
aeab1de1de
|
[diffusion] optimization: support cuda graph for Pi-0.5 prefix encoding (#34256)
|
2026-08-11 09:14:53 +08:00 |
|
|
cctry
|
df986c4d5e
|
Consolidate CUDA VMM allocation helpers (#34199)
|
2026-08-10 18:11:11 -07:00 |
|
|
Mick
|
ba5183fe10
|
[diffusion] UX: fix CI server warmup progress logging (#34301)
|
2026-08-11 09:09:32 +08:00 |
|
|
Mick
|
418975ba64
|
[EPD] feat: pipeline owner-only multimodal preprocessing (#34206)
|
2026-08-11 09:08:05 +08:00 |
|
|
elvischenvandClaude Opus 5
|
03c942dfab
|
[Fix] Make wait_port_available actually wait timeout_s seconds (#33086)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
2026-08-10 17:35:50 -07:00 |
|
|
milesial
|
d59c1ddf70
|
fix(dflash): account for DCP in draft KV pool sizing (#33912)
Signed-off-by: Alexandre Milesi <milesial@users.noreply.github.com>
|
2026-08-10 17:15:01 -07:00 |
|
|
Khoa PhamandClaude Opus 5
|
0967885121
|
[DCP] Drop two per-layer launches from the MLA target-verify path (#34240)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
2026-08-10 17:11:03 -07:00 |
|
|
weireweireandweireweire
|
2d5009d130
|
[DSV4] Avoid host syncs in EAGLE prefill (#33662)
Co-authored-by: weireweire <20922698+weireweire@users.noreply.github.com>
|
2026-08-10 17:09:55 -07:00 |
|
|
Michael Gschwind
|
7c7326ccb3
|
Fix DeepSeek-V4/DeepSeek-V4-Pro DP-attention gather semantics (#31700)
|
2026-08-10 17:03:19 -07:00 |
|
|
Baizhou Zhang
|
56e8bb49b8
|
fix(ci): constrain NeMo Skills evaluator dependencies (#34322)
|
2026-08-10 16:56:46 -07:00 |
|
|
Liangsheng Yin
|
c30872fa00
|
[Fix] Read trace level via get_global_trace_level() in trace_async (#34325)
|
2026-08-10 16:46:54 -07:00 |
|
|
Mohammad Miadh Angkad
|
8c5d5f75bf
|
Fix DSV4 DSpark shared expert loading (#33312)
|
2026-08-11 07:36:50 +08:00 |
|
|
Cheng WanandClaude Fable 5
|
ceeaec2078
|
[PD] Support --enable-unified-memory with PD disaggregation (kimi-linear MLA hybrid-Mamba) (#33362)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
|
2026-08-10 16:07:58 -07:00 |
|
|
Cheng WanandClaude Fable 5
|
8a7c8a72d6
|
Fix NaN logits from deterministic Triton extend on the unified memory pool (#33517)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
|
2026-08-10 16:05:04 -07:00 |
|
|
|
0661eb1c50
|
plugins: don't directly set quant class (#34252)
Signed-off-by: Devashish Lal <devcode@fb.com>
Co-authored-by: Devashish Lal <devcode@fb.com>
Co-authored-by: Alex Nails <alex.nails@radixark.ai>
|
2026-08-10 16:01:31 -07:00 |
|
|
Zhiyao Jiang
|
2c72323d90
|
[AMD] Fix AITER custom reduce-scatter CUDA-graph capture crash under torch_memory_saver (#34203)
|
2026-08-10 15:36:15 -07:00 |
|
|
Liangsheng Yin
|
c80a38edcd
|
[Fix] Pin cuda-tile to 1.6.0rc5 to unblock Python 3.10 x86_64 installs (#34321)
|
2026-08-10 15:09:47 -07:00 |
|
|
AMD-yanfeiwang
|
ca0f8a0f4c
|
perf(hisparse): fuse the DSv4 value and scale swap-in copy on ROCm (#33484)
|
2026-08-10 14:30:15 -07:00 |
|
|
QIN2DIM
|
166c6f7181
|
[Kernel] cutedsl_bf16_gemm: trailing cluster barrier for 2-CTA TGV kernel exit (#32907) (#32954)
|
2026-08-10 20:31:18 +00:00 |
|
|
AMD-yanfeiwang
|
1a8e4876b6
|
perf(hisparse): 128-bit non-temporal swap-in copy on ROCm (#33085)
|
2026-08-10 13:27:15 -07:00 |
|
|
Qiaolin Yu
|
733c05c887
|
[spec decoding] support inkling dspark (#31847)
|
2026-08-10 13:21:35 -07:00 |
|
|
Lifan ShenandXinyuan Tong
|
a2161ce682
|
Support thinking budget for Inkling (#33146)
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
|
2026-08-11 02:01:08 +08:00 |
|