All pull requests

Pull requests

• [BugFix] online attention tiling type error

  #25122·ppppqp opened Oct 24, 2026·

  ·
  1

  2

• [Codegen][CPU] Add RISC-V V bf16 vfwmaccbf16 inner_tiled MMA.

  #25120·TMahlatini opened Oct 23, 2026Contributor·

  ·

• [LLVMCPU] Write embedded executable object data before linking

  #25117·rpqt opened Oct 23, 2026·

  ·
  4

• [GlobalOpt] Fold broadcasts into batch_matmul indexing maps

  #25114·Manewing opened Oct 23, 2026Contributor·2/2·

  ·
  3

• [DispatchCreation] Collapse named contractions with user indexing maps

  #25113·Manewing opened Oct 23, 2026Contributor·1/2·

  ·
  1

  1

• [GlobalOptimization] Do not demote contraction inputs with a tiny reduction extent

  #25105·tonykuttai opened Oct 22, 2026·

  ·
  1

  2

• [Torch] Convert torchao affine quantization to LinalgExt

  #25104·roberto-laudani opened Oct 21, 2026Contributor·

  ·
  3

  

• [CodeGen] Fix linkExecutablesInto creating incorrect executable references

  #25102·devtbi opened Oct 21, 2026Contributor·

  ·
  1

• [Stream] Fix scf.while region signature conversion

  #25100·anutosh491 opened Oct 21, 2026Contributor·

  ·
  5

• fix(compiler/GlobalOptimization): change pass ordering so img2col convolutions in quantized conv models will result in integer math

  #25099·Manewing opened Oct 21, 2026Contributor·

  ·
  5

• [Stream] Avoid repeated transfers for concrete await results

  #25098·zhangpan2001 opened Oct 21, 2026Contributor·

  ·
  1

• [Runtime] reject bytecode modules with more than 65535 function descriptors

  #25097·yl120064227 opened Oct 20, 2026·

  ·
  6

• [Runtime] validate gguf alignment and guard the tensor data offset against underflow

  #25096·yl120064227 opened Oct 20, 2026·

  ·
  2

• [Stream] Honor parameter subrange offsets in split encoders

  #25091·crystarm opened Oct 20, 2026·

  ·
  1

  2

• [DispatchCreation] Skip split reduction candidates with unsupported combiners

  #25090·losewayy opened Oct 20, 2026·

  ·
  1

  2

• [Encoding] Replace packed_storage special cases with interface queries

  #25089·Prajit-Rahul opened Oct 19, 2026Contributor·

  ·
  1

  1

• [WIP][CI] Run WebGPU HAL tests on Linux

  #25079·anutosh491 opened Oct 19, 2026Contributor·

  ·

• [TorchAX] Add isolated stateless export and composite registration

  #25074·jschuhmacher opened Oct 19, 2026Contributor·10/10·

  ·

• [LLVMCPU] Bound attention tiles while preserving whole-head contractions

  #25073·jschuhmacher opened Oct 19, 2026Contributor·8/10·

  ·

• [StableHLO] Lower Torch SDPA composites with dynamic grouped-query attention

  #25072·jschuhmacher opened Oct 19, 2026Contributor·9/10·

  ·

• [LinalgExt] Preserve attention score coordinates and masked-tile state

  #25071·jschuhmacher opened Oct 19, 2026Contributor·7/10·

  ·

• [WebGPU] Fix staging buffer mapping lifetimes

  #25069·anutosh491 opened Oct 19, 2026Contributor·

  ·
  6

• [StableHLO] Add plugin registration for composite and custom-call rewrites

  #25067·jschuhmacher opened Oct 18, 2026Contributor·1/10·

  ·

• [StableHLO] Preserve shape assertions through optimization and CPU dispatch

  #25066·jschuhmacher opened Oct 18, 2026Contributor·6/10·

  ·

• [StableHLO] Support dynamic slices, updates and concatenation

  #25065·jschuhmacher opened Oct 18, 2026Contributor·5/10·

  ·