All pull requests

Pull requests

• [Core] Serialize captured graph replay with session runs

  #32902·Jiajia Qin (qjia7) opened Sep 29, 2026Contributor·

  ·

• [CUDA] Add tinygemm2 as a GEMM auto-tune candidate for small-M MatMul on SM 9.0+

  #32901·Tianlei Wu (tianleiwu) opened Sep 29, 2026Contributor·

  ·

• Plugin EP shared allocator and data transfer documentation.

  #32900·Scott McKay (skottmckay) opened Sep 29, 2026Contributor·

  ·
  3

• Improve the app-EP weightless contract

  #32899·Javier Martinez (javier-intel) opened Sep 29, 2026Contributor·

  ·

• Honor deterministic compute for CUDA MatMulNBits tactic selection

  #32898·kunal-vaishnavi opened Sep 29, 2026Contributor·

  ·
  5

• [WebGPU] Use shader helpers for scalar and vector storage access

  #32897·Yang Gu (gyagp) opened Sep 29, 2026Contributor·

  ·
  2

• ORT Mobile infrastructure documentation

  #32895·Scott McKay (skottmckay) opened Sep 29, 2026Contributor·

  ·
  2

• Handle invalid partition config without exceptions

  #32894·Mustapha Jaber (mustjab) opened Sep 29, 2026Contributor·

  ·
  5

• Fix MatmulBNFusion dropping graph outputs produced by in-between nodes

  #32893·losewayy (losewayy) opened Sep 29, 2026·

  ·
  1

  2

• Fix LayerNorm fusions silently dropping a non-constant epsilon

  #32892·losewayy (losewayy) opened Sep 29, 2026·

  ·
  1

  4

• Speed up CPU Transpose for adjacent axis groups

  #32891· Stern (Hosi121) opened Sep 29, 2026Contributor·

  ·
  2

• Interop API documentation and bugfix

  #32890·Scott McKay (skottmckay) opened Sep 29, 2026Contributor·

  ·
  1

• Model Editor documentation and fixes

  #32889·Scott McKay (skottmckay) opened Sep 29, 2026Contributor·

  ·
  7

• Gate variable-length convolution state capture

  #32888·kunal-vaishnavi opened Sep 28, 2026Contributor·

  ·
  4

• Add CUDA QMoE block-scaled FP8 support with routed-expert compaction

  #32887·Akshay Sonawane (apsonawane) opened Sep 28, 2026Contributor·

  ·
  2

• [js/web] Add LoRA adapter support for the standard(non-proxy) run path

  #32886·Saravanan A R (saravanan-a-r) opened Sep 28, 2026·

  ·
  16

• Cpu a16 nbits large block

  #32885·mirounga opened Sep 28, 2026Contributor·

  ·
  2

• [CUDA] Remove memset nodes from decode CUDA graphs and prefer GEMV at small M

  #32884·Tianlei Wu (tianleiwu) opened Sep 28, 2026Contributor·

  ·
  1

• Parse model package metadata without exceptions

  #32883·Mustapha Jaber (mustjab) opened Sep 28, 2026Contributor·

  ·
  1

• Fix dead links in docs and CONTRIBUTING

  #32880·Pratik Gandhi (pratikgx) opened Sep 28, 2026·

  ·
  4

• Add WebGPU INT2 and mixed-width QMoE support

  #32878·Sayan Shaw (sayanshaw24) opened Sep 28, 2026Contributor·

  ·

• Document Phi-4-mini memory diagnostics and logits-allocation findings

  #32877·BBaditi opened Sep 28, 2026·

  ·

• [CUDA] Auto-tune small-M decode MatMul (cuBLAS vs small-N GEMV) and speed up MatMulNBits GEMV

  #32876·Tianlei Wu (tianleiwu) opened Sep 28, 2026Contributor·

  ·
  3

• Handle optimizer size overflow without exceptions

  #32875·Mustapha Jaber (mustjab) opened Sep 28, 2026Contributor·

  ·
  8

• [WebGPU] Separate command recording from buffer reuse tracking

  #32874·Jiajia Qin (qjia7) opened Sep 28, 2026Contributor·

  ·