All pull requests

Pull requests

• [OMNIML-6273] Run /claude review on the shared review prompt

  #2977·shengliangxu opened Oct 23, 2026Collaborator·

  ·
  3

• Add offline AR validation for DFlash drafts on dumped target hidden states

  #2976·h-guo18 opened Oct 23, 2026Contributor·

  ·
  3

• [NVBug 6901733] fix(hf_ptq): use bundled code for dense Nemotron-H on Transformers 5.5

  cherry-pick-0.48.0
  #2975·juhi10071998 opened Oct 23, 2026Contributor·

  ·
  5

  

• Add examples/specdec_aiperf: speculative-decoding AL and throughput with AIPerf

  #2974·h-guo18 opened Oct 23, 2026Contributor·

  ·
  3

• [NVBug 6980441] Fix Diffusers subfolder ModelOpt state restoration

  cherry-pick-0.48.0
  #2971·yingguo-trt opened Oct 23, 2026Contributor·

  ·
  11

• [17/17] Opt-in FP16 strong typing for DLA

  #2969·mgohil-png opened Oct 23, 2026Contributor·16/16·

  ·
  5

• [16/17] DLA public integration and reshape cleanup

  #2968·mgohil-png opened Oct 23, 2026Contributor·15/16·

  ·
  5

• [15/17] DLA TopK and dtype compatibility

  #2967·mgohil-png opened Oct 23, 2026Contributor·14/16·

  ·
  5

• [14/17] DLA boolean and comparison rewrites

  #2966·mgohil-png opened Oct 23, 2026Contributor·13/16·

  ·
  5

• [13/17] Standalone operation rank-conversion compatibility

  #2965·mgohil-png opened Oct 23, 2026Contributor·12/16·

  ·
  11

• [12/17] DLA rank conversion

  #2964·mgohil-png opened Oct 23, 2026Contributor·11/16·

  ·
  9

• [11/17] DLA indexing, reductions and attention

  #2963·mgohil-png opened Oct 23, 2026Contributor·10/16·

  ·
  12

• [10/17] DLA model boundary promotion

  #2962·mgohil-png opened Oct 23, 2026Contributor·9/16·

  ·
  10

• [9/17] DLA LSTM decomposition

  #2961·mgohil-png opened Oct 23, 2026Contributor·8/16·

  ·
  9

• [8/17] DLA MatMul and Gemm convolution conversion

  #2960·mgohil-png opened Oct 23, 2026Contributor·7/16·

  ·
  9

• [7/17] DLA dequantization and reshape stages

  #2959·mgohil-png opened Oct 23, 2026Contributor·6/16·

  ·
  12

• [6/17] Standalone five-dimensional reshape conversion

  #2958·mgohil-png opened Oct 23, 2026Contributor·5/16·

  ·
  13

• [5/17] DLA QDQ and constant preprocessing

  #2957·mgohil-png opened Oct 23, 2026Contributor·4/16·

  ·
  50

• [4/17] Private ordered DLA scaffold and node ordering

  #2956·mgohil-png opened Oct 23, 2026Contributor·3/16·

  ·
  39

• [3/17] Standalone DLA graph cleanup

  #2955·mgohil-png opened Oct 23, 2026Contributor·2/16·

  ·
  85

• Bump the github-actions group across 1 directory with 4 updates

  dependenciesgithub_actions
  #2954·dependabot[bot] opened Oct 23, 2026Bot·

  ·
  3

• fix(speculative): EAGLE3 on heterogeneous per-layer configs, and a first-batch hidden-state sanity check

  #2952·aniskumar-nv opened Oct 23, 2026·

  ·
  19

• [OMNIML-6257] Update launcher Megatron-LM submodule for checkpoint rewrite

  #2949·ChenhanYu opened Oct 22, 2026Collaborator·

  ·
  3

• [3/3] Place GPTQ Hessians by measured headroom instead of a 65% usage cap

  #2945·cjluo-nv opened Oct 22, 2026Collaborator·2/2·

  ·
  4

• [2/17] Add shared ONNX model saving for DLA surgeries

  #2939·mgohil-png opened Oct 22, 2026Contributor·1/16·

  ·
  50