All pull requests

Pull requests

• [Testing][JIT] Remove unsupported BF16 scalar regression

  #3443·by LeiWang1999 was merged Oct 7, 2026Member·

  ·
  2

• [JIT][Runtime] Fix bfloat16 scalar inputs across CUDA backends

  #3442·by LeiWang1999 was closed Oct 7, 2026Member·

  ·
  1

  2

• [Fix] Preserve shared region offsets in LDSM copies

  #3441·by Ethereal49 was merged Oct 7, 2026Contributor·

  ·
  1

  3

• [BugFix][CUDA] Gate fp8 T.gemm on SM89 instead of trapping at launch

  #3439·by shuyilinn was merged Oct 7, 2026Contributor·

  ·
  1

  3

• [BugFix][CUDA] Guard the shared-side region offset when lowering T.copy to ldmatrix

  #3436·by ZeroKernel798 was closed Oct 7, 2026·

  ·
  1

  2

• [BugFix][JIT] Preserve variadic arguments when building lazy kernels

  #3435·by luicarus was merged Oct 7, 2026Contributor·

  ·
  3

• [CI] [pre-commit.ci] autoupdate

  #3431·by pre-commit-ci[bot] was merged Oct 7, 2026Bot·

  ·
  2

• [BugFix][Transform] Preserve explicit casts during index promotion

  #3429·by kobecai was merged Oct 5, 2026Contributor·

  ·
  1

  2

• [Docs] Explain the two Ascend 950 backends in the Ascend guide

  #3423·by 173787247 was closed Oct 4, 2026Contributor·

  ·
  7

• [BugFix] Accept bfloat16 in the remaining packed-API dtype guards

  #3421·by 173787247 was closed Oct 3, 2026Contributor·

  ·
  1

  3

• [Chore] merge two identical branches

  #3418·by y1lan was merged Oct 3, 2026Contributor·

  ·
  3

• [BugFix] Complete bfloat16 scalar input binding and Cython marshalling

  #3416·by dajiaohuang was closed Oct 3, 2026Contributor·

  ·
  9

• Support bfloat16 scalar parameters

  #3361·by dajiaohuang was merged Oct 2, 2026Contributor·

  ·
  7

• Report non-boolean if and assert conditions at the source

  #3360·by dajiaohuang was closed Oct 2, 2026Contributor·

  ·
  1

  6

• [Carver] Remove unreachable tensor-core shared-scope branches

  #3358·by dajiaohuang was closed Oct 2, 2026Contributor·

  ·
  3

• [BugFix] Refuse a cluster kernel on the tvm_ffi backend

  #3349·by 173787247 was closed Oct 1, 2026Contributor·

  ·
  5

• [BugFix] Decode the reserved NVFP4 UE4M3 NaN scale code

  #3334·by 173787247 was merged Oct 3, 2026Contributor·

  ·
  3

• pto-main

  #3314·by erhsh was closed Sep 30, 2026Contributor·Draft·

  ·
  2

• tmp

  #3312·by erhsh was closed Sep 30, 2026Contributor·Draft·

  ·
  2

• [Ascend] Add PTO backend for Ascend950

  #3310·by LLMZhangYC was merged Sep 30, 2026Contributor·

  ·
  23

• [Release] Bump version into 0.1.15

  #3309·by LeiWang1999 was merged Sep 29, 2026Member·

  ·
  2

• [Public Release 9/30] Introduce Ascend 950 backend

  #3308·by SiriusNEO was merged Sep 29, 2026Collaborator·

  ·
  2

• [BugFix][CUDA] Preserve destination types in atomic references

  #3306·by KareemMusleh was closed Sep 29, 2026·

  ·
  1

  2

• [Build][Windows] Restore ccache hits for clang-cl wheel builds

  #3305·by LeiWang1999 was merged Sep 29, 2026Member·

  ·
  2

• [Fix][TVM] Update TVM to preserve dynamic reduction tail guards

  #3294·by LeiWang1999 was merged Sep 28, 2026Member·

  ·
  2