Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on May 18, 2026

• Tier offload followup (#42529)

  Show description for 8474748
  ronensc

  authored
  8474748

  Copy full SHA for 8474748

• [ROCm] Guard AITER GDN decode fast path by layout (#42880)

  Show description for 8fc1c28
  tuukkjs

  authored
  8fc1c28

  Copy full SHA for 8fc1c28

• [Docs] update attribution to reflect EDEN foundation (#41666)

  Show description for ce88f01
  amitport

  authored
  ce88f01

  Copy full SHA for ce88f01

• [Refactor] Remove dead cuda kernels (#42767)

  Show description for 00e20e7
  yewentao256

  authored
  00e20e7

  Copy full SHA for 00e20e7

• [BugFix] support PP for Cohere vision model (#42819)

  Show description for 9758a6e
  czhu-cohere

  authored
  9758a6e

  Copy full SHA for 9758a6e

• [ROCm][Quantization][3/N] Refactor quark_moe w4a4 w/ oracle (#41436)

  Show description for a2c8fc6
  BowenBao

  authored
  a2c8fc6

  Copy full SHA for a2c8fc6

• [Bugfix] fix swiglu limit issue for humming backend + deepseek v4 (#42541)

  Show description for 6859ca7
  

  jinzhen-lin

  and

  mgoin

  authored
  6859ca7

  Copy full SHA for 6859ca7

• [Bugfix] Fix DSV4 MTP after ROCm mHC integration (#42930)

  Show description for 67f58ce
  mmangkad

  authored
  67f58ce

  Copy full SHA for 67f58ce

• [Perf] Re-enable flashinfer autotune by default and cleanup (#42857)

  Show description for 8c296de
  wzhao18

  authored
  8c296de

  Copy full SHA for 8c296de

• Fix `--convert` passed without `--runner` on causal models (#42935)

  Show description for b12745e
  hmellor

  authored
  b12745e

  Copy full SHA for b12745e

• [Model Runner V2] Fix prompt logprobs calculation `Sizes of tensors must match` error (#42778)

  Show description for e267369
  yewentao256

  authored
  e267369

  Copy full SHA for e267369

• [Bugfix] mamba: run single-token extends as decodes (#42430)

  Show description for 47829b1
  netanel-haber

  authored
  47829b1

  Copy full SHA for 47829b1

• [Model] Add Apertus Tool Parser (#41154)

  Show description for 4a39b4f
  blancsw

  authored
  4a39b4f

  Copy full SHA for 4a39b4f

• Refactor AWQ Marlin MoE onto modular WNA16 oracle (#42483)

  Show description for 78e7a7b
  

  5 peopleauthored
  78e7a7b

  Copy full SHA for 78e7a7b

• [Model Runner v2] Support update_config (#42783)

  Show description for f5d3dc7
  

  mgoin

  and

  claude

  authored
  f5d3dc7

  Copy full SHA for f5d3dc7

• Revert "[torch.compile] Add patch for fullgraph compilation" (#42686) (#42913)

  Show description for 1ac10f1
  

  3 peopleauthored
  1ac10f1

  Copy full SHA for 1ac10f1

• [KV Connector][Offloading] Flush all pending jobs on last step (#42611)

  Show description for e541765
  

  3 peopleauthored
  e541765

  Copy full SHA for e541765

• [XPU][CI] Temporarily skip test_moe_lora_align_block_size_mixed_base_and_lora[1] in Intel GPU CI (#42954)

  Show description for 2e40faf
  zxd1997066

  authored
  2e40faf

  Copy full SHA for 2e40faf

• [MRv2] Default to MRv1 when a connector is present (#42955)

  Show description for 69c91d0
  NickLucche

  authored
  69c91d0

  Copy full SHA for 69c91d0

• [Bugfix][Hybrid][NemotronH] Fix mamba_cache_mode=all + speculative decoding crash (#41233)

  Show description for 737bfa3
  roikoren755

  authored
  737bfa3

  Copy full SHA for 737bfa3

• [Bugfix][KV Offload] count appended GPU blocks in store group_sizes (#42945)

  Show description for e414e1f
  kfirtoledo

  authored
  e414e1f

  Copy full SHA for e414e1f

• fix: remove unused norm for dpskv4 (#41710)

  Show description for df852ed
  

  inisis

  and

  zyongye

  authored
  df852ed

  Copy full SHA for df852ed

• [CPU] Add MXFP4 W4A16 MoE support (#41922)

  Show description for 88a860d
  yuwenzho

  authored
  88a860d

  Copy full SHA for 88a860d

• [CPU Backend] Improve cpu thread utilization (#42666)

  Show description for cac81b6
  

  tianmu-li

  and

  mergify[bot]

  authored
  cac81b6

  Copy full SHA for cac81b6

• [CPU] Add fused GDN support for AMX CPU platform (#42707)

  Show description for b4601ad
  bigPYJ1151

  authored
  b4601ad

  Copy full SHA for b4601ad

• [Kernel] Pack topk id/weights triton kernel (#42527)

  Show description for 2267f70
  jeejeelee

  authored
  2267f70

  Copy full SHA for 2267f70

• [CPU] Specify required KV cache layout for CPU attention backend (#42740)

  Show description for 965d076
  

  hlin99

  and

  bigPYJ1151

  authored
  965d076

  Copy full SHA for 965d076

• delete xpu ci (#42582)

  Show description for c38bed4
  

  3 peopleauthored
  c38bed4

  Copy full SHA for c38bed4

• [Perf] Add do_not_specialize in fused FP8 RoPE kernel (#42849)

  Show description for 998714b
  xyang16

  authored
  998714b

  Copy full SHA for 998714b

• Revert checkpoint specific workaround in Transformers modelling backend (#42923)

  Show description for 9537542
  hmellor

  authored
  9537542

  Copy full SHA for 9537542

• [Model] [Perf] Use flatten for Qwen3.5's GDN output projection (#42311)

  Show description for 5ab6d1b
  rishaps

  authored
  5ab6d1b

  Copy full SHA for 5ab6d1b

• [LoRA] Support 2D and 3D MoE LoRA adapter at the same time (#42242)

  Show description for 7d5b033
  

  3 peopleauthored
  7d5b033

  Copy full SHA for 7d5b033

• [Bugfix] moe lora align kernel grid (#40131)

  Show description for e3aeee5
  

  TheDuyIT

  and

  jeejeelee

  authored
  e3aeee5

  Copy full SHA for e3aeee5

• Improve logging when docs build is skipped (#42929)

  Show description for c1f7854
  hmellor

  authored
  c1f7854

  Copy full SHA for c1f7854

• [BugFix] Kimi-K2.5: skip vision tower dtype conversion when using quantization (#42869)

  Show description for 23c15ac
  gaozihao-shy

  authored
  23c15ac

  Copy full SHA for 23c15ac