All pull requests

Pull requests

• [Web] Submit pending WebGPU commands every maxDispatchesPerFlush launches

  #20430·JCNTH opened Sep 24, 2026·

  ·

• [Web] Skip WebGPU uniform uploads whose contents are unchanged

  #20429·JCNTH opened Sep 24, 2026·

  ·

• [Web] Cache WebGPU bind groups of kernel launches in CacheState

  #20428·JCNTH opened Sep 24, 2026·

  ·

• [Web] Share one compute pass across consecutive WebGPU dispatches

  #20427·JCNTH opened Sep 24, 2026·

  ·

• [Relax][WebGPU] Sequence-sharded paged decode attention for long contexts

  #20426·JCNTH opened Sep 24, 2026·

  ·

• [Dlight][WebGPU] Add a WebGPU config to the GEMV inner-reduction schedule

  #20425·JCNTH opened Sep 24, 2026·

  ·

• [Backend][WebGPU] Fix off-by-one in the grid-packing guard

  #20424·JCNTH opened Sep 24, 2026·

  ·

• [REFACTOR][TVMSCRIPT] Introduce Transpiler based parser

  #20423·tqchen opened Sep 24, 2026Member·

  ·
  1

• [Fix][Relax] Scope BasePyModule's Python function registry entries to the module

  #20417·akaashrp opened Sep 22, 2026Contributor·

  ·
  2

• [Fix][Relax][ONNX] Support Hardmax over a symbolic axis extent

  #20416·LiRunGuo opened Sep 22, 2026·

  ·
  1

• [Fix][Relax][ONNX] Fix Upsample and Resize-10 import

  #20415·LiRunGuo opened Sep 22, 2026·

  ·
  1

• [Fix][Sym] Use a typed zero for the sign test in interval Mul/Div

  #20414·LiRunGuo opened Sep 22, 2026·

  ·
  1

• [Fix][Relax][VM] Stop BasePyModule from wiping the call_py_func registry

  #20413·LiRunGuo opened Sep 22, 2026·

  ·
  1

• [TIRx] Reject boolean div and mod

  #20412·Nanmur opened Sep 22, 2026Contributor·

  ·
  1

• [Fix][Relax][ONNX] Import Resize with empty scales tensor and symbolic sizes

  #20406·LiRunGuo opened Sep 22, 2026·

  ·
  5

• [Fix][Relax][Torch] Respect zero-dimensional tensor dtype promotion

  #20405·jack-k737 opened Sep 22, 2026·

  ·
  1

• [Fix][Relax][ONNX] Support symbolic dims in SAME auto_pad pooling

  #20404·LiRunGuo opened Sep 21, 2026·

  ·
  1

  1

• [FIX][Vulkan] Avoid device API destruction during Windows DLL shutdown

  #20397·sepcnt opened Sep 21, 2026·

  ·
  1

• [Metal] Add TVM_METAL_STORAGE_MODE opt-in for correctness at scale (#20157)

  #20391·zacharywhitley opened Sep 19, 2026·

  ·

• [Relax] Skip 1-D inputs in CombineParallelMatmul

  #20389·Nanmur opened Sep 19, 2026Contributor·

  ·
  1

• [Fix][CUDA] Preserve NaNs in floating-point min and max

  #20387·LngelKyo opened Sep 18, 2026Contributor·

  ·

• [Fix][Relax][Frontend][Torch] Compare shapes without == on symbolic dims in the reshape shortcut

  #20377·hiyufan opened Sep 17, 2026Contributor·

  ·
  1

• [Fix][Relax][Frontend][Torch] Make any and prod follow torch: missing overloads and result dtypes

  #20376·hiyufan opened Sep 17, 2026Contributor·

  ·
  1

• [Relax][Frontend][Torch] Add amax, amin and min.dim converters

  #20375·hiyufan opened Sep 17, 2026Contributor·

  ·

• [Fix][Relax][Frontend][Torch] Accumulate integer cumsum and cumprod in int64

  #20374·hiyufan opened Sep 17, 2026Contributor·

  ·