Commits

Branch selector

 main

User selector

HaiShaw

Datepicker

All time

Commit History

Commits on Jun 6, 2025

• AITER backend extension and workload optimizations (#6838)

  Show description for b819381
  

  3 peopleauthored
  b819381

  Copy full SHA for b819381

Commits on May 27, 2025

• DeepSeek: enable none block-quant FP8 quantizations (#6638)

  HaiShaw

  authored
  183d9f9

  Copy full SHA for 183d9f9

Commits on May 21, 2025

• aiter attention-backend (default enabled on AMD/ROCm) (#6381)

  HaiShaw

  authored
  5c0b38f

  Copy full SHA for 5c0b38f

Commits on May 20, 2025

• Address performance regression: disable multiple streams on ROCm (#6412)

  HaiShaw

  authored
  6317c5c

  Copy full SHA for 6317c5c

Commits on Apr 28, 2025

• ROCm: update AITER (#5816)

  HaiShaw

  authored
  d364b9b

  Copy full SHA for d364b9b

Commits on Apr 22, 2025

• Revert "Support aiter RMSNorm in AMD" (#5646)

  HaiShaw

  authored
  b0feda0

  Copy full SHA for b0feda0

Commits on Apr 11, 2025

• ROCm/AITER CK_MoE: update 2-stage kernels & support both Activations (#5228)

  HaiShaw

  authored
  8879944

  Copy full SHA for 8879944

Commits on Apr 10, 2025

• ROCm sgl-kernel: compatible to later torch (#5167)

  HaiShaw

  authored
  d050df3

  Copy full SHA for d050df3

Commits on Apr 7, 2025

• Fix refactor error - fp8.py (#5106)

  Show description for 8199247
  

  HaiShaw

  and

  merrymercy

  authored
  8199247

  Copy full SHA for 8199247

Commits on Mar 7, 2025

• ROCm: Flex Attention Enablement with custom backends (#4178)

  Show description for 0beea45
  

  HaiShaw

  and

  linsun12

  authored
  0beea45

  Copy full SHA for 0beea45

Commits on Mar 6, 2025

• ROCm: enable trillion-parameter MoE models with INT4-FP8 single node (#4152)

  HaiShaw

  authored
  13bc39c

  Copy full SHA for 13bc39c

Commits on Mar 5, 2025

• Fix the moe padding conditional logic (#4081)

  HaiShaw

  authored
  71ab0da

  Copy full SHA for 71ab0da

Commits on Mar 4, 2025

• ROCm: update aiter and its usage to fused moe (bloat16, fp8, fp8 block-quant) (#4053)

  HaiShaw

  authored
  51d2540

  Copy full SHA for 51d2540

Commits on Feb 21, 2025

• AMD/ROCm: update AITER repo to ROCm/aiter (#3747)

  HaiShaw

  authored
  5c54ef0

  Copy full SHA for 5c54ef0

Commits on Feb 20, 2025

• revert BLOCK and num_warps on HIP (#3722)

  HaiShaw

  authored
  6252ade

  Copy full SHA for 6252ade

Commits on Feb 14, 2025

• ROCm docker: triton update (#3584)

  HaiShaw

  authored
  d973c78

  Copy full SHA for d973c78

Commits on Feb 11, 2025

• MI30x: More graph captures for larger batch sizes and concurrencies (#3420)

  HaiShaw

  authored
  d81ac44

  Copy full SHA for d81ac44

Commits on Feb 4, 2025

• ROCm: sgl-kernel enablement starting with sgl_moe_align_block (#3287)

  HaiShaw

  authored
  2c1a695

  Copy full SHA for 2c1a695

Commits on Feb 2, 2025

• ROCm: bump 6.3.0 (#3259)

  HaiShaw

  authored
  566d61d

  Copy full SHA for 566d61d

Commits on Feb 1, 2025

• update ENV to ROCm dockers (#3248)

  HaiShaw

  authored
  17dbf97

  Copy full SHA for 17dbf97

Commits on Jan 7, 2025

• Use Optional with None default (#2770)

  HaiShaw

  authored
  6d08ce2

  Copy full SHA for 6d08ce2

Commits on Dec 30, 2024

• AMD DeepSeek_V3 FP8 Numerical fix (#2667)

  HaiShaw

  authored
  c5210df

  Copy full SHA for c5210df

• fix typo in python/sglang/srt/layers/quantization/fp8.py (#2655)

  HaiShaw

  authored
  e6f523b

  Copy full SHA for e6f523b

Commits on Dec 29, 2024

• AMD: set weights and scaling numbers properly for block FP8 (#2637)

  HaiShaw

  authored
  30828e7

  Copy full SHA for 30828e7

Commits on Dec 27, 2024

• Regression fix to AMD/ROCm from recent change (#2606)

  HaiShaw

  authored
  7722c11

  Copy full SHA for 7722c11

Commits on Dec 7, 2024

• Fp8 MoE optimizations on AMD (#2388)

  HaiShaw

  authored
  95f93f4

  Copy full SHA for 95f93f4

Commits on Dec 4, 2024

• Adding SGLang FP8 Utils (#2348)

  HaiShaw

  authored
  b2986d7

  Copy full SHA for b2986d7

Commits on Dec 3, 2024

• ROCm Container: set SGLANG_SET_CPU_AFFINITY=1 (#2328)

  HaiShaw

  authored
  0639bf1

  Copy full SHA for 0639bf1

• Relax to include more AMD GPUs (#2319)

  HaiShaw

  authored
  69e2d4f

  Copy full SHA for 69e2d4f

Commits on Dec 2, 2024

• Use rocminfo instead of rocm-smi for more OS/WSL support (#2310)

  HaiShaw

  authored
  c54bda3

  Copy full SHA for c54bda3

Commits on Nov 28, 2024

• Update Install Method 2. From source (#2232)

  HaiShaw

  authored
  b79fffd

  Copy full SHA for b79fffd

• Rename tuned MI300X config files for fused_moe_triton (#2228)

  HaiShaw

  authored
  cd51758

  Copy full SHA for cd51758

Commits on Nov 25, 2024

• [Performance]: Process affinity to CPU cores with multiple sockets support (#2171)

  HaiShaw

  authored
  10189d0

  Copy full SHA for 10189d0

Commits on Nov 21, 2024

• ROCm: Fix MoE padding for none FP8 cases (#2111)

  HaiShaw

  authored
  f35cb46

  Copy full SHA for f35cb46

Commits on Nov 19, 2024

• Use native fp8 format on MI300X (#2094)

  HaiShaw

  authored
  e57c3e1

  Copy full SHA for e57c3e1

Pagination

Previous

Next