Commit history

1. Commits on Oct 28, 2026

   • Restore MLPerf GPTOSS 20B configs after 26.08 perf recipe refactor (#6473)

     Show description for e301c81
     

     8 peopleauthored
     e301c81View commit details

     Copy full SHA for e301c81

     Browse repository at this point

   • chore(deps): AUT-2659 bump Megatron-Core main to a2965235 (#6707)

     Show description for da40966
     svcnemo-autobot

     authored
     da40966View commit details

     Copy full SHA for da40966

     Browse repository at this point

   • perf(recipes): enable NCCL EP for B200 and B300 Lightning BF16 and MXFP8 (#6567)

     Show description for cb493d3
     rhmukundan

     authored
     cb493d3View commit details

     Copy full SHA for cb493d3

     Browse repository at this point

   • fix(training): fix local CUDA graph deletion by explicitly clearing local CG runners (#6681)

     Show description for e4b9e16
     ghadiaravi13

     authored
     e4b9e16View commit details

     Copy full SHA for e4b9e16

     Browse repository at this point
2. Commits on Oct 27, 2026

   • fix(training): align Muon distributed setup and checkpointing (#6443)

     Show description for eb09bfc
     

     3 peopleauthored
     eb09bfcView commit details

     Copy full SHA for eb09bfc

     Browse repository at this point

   • chore: AUT-2643 MCore main bump to 3ddfe0fe (#6688)

     Show description for d97e24e
     svcnemo-autobot

     authored
     d97e24eView commit details

     Copy full SHA for d97e24e

     Browse repository at this point

   • perf(recipe): add Nemotron 3.5 Super VL 64-GPU GB300 MXFP8 pretrain recipe (#6614)

     Show description for 7171584
     tomlifu

     authored
     7171584View commit details

     Copy full SHA for 7171584

     Browse repository at this point

   • feat(quant): save ModelOpt-quantized models as quantized HF checkpoints (#6657)

     Show description for 0652ea7
     

     kevalmorabia97

     and

     claude

     authored
     0652ea7View commit details

     Copy full SHA for 0652ea7

     Browse repository at this point

   • feat(training): let distill() take training callbacks (#6686)

     Show description for 033ca6f
     

     kevalmorabia97

     and

     claude

     authored
     033ca6fView commit details

     Copy full SHA for 033ca6f

     Browse repository at this point

   • refactor(model): build Qwen3 MoE from a builder-backed model config (#6592)

     Show description for 97f10ed
     kamran-nvidia

     authored
     97f10edView commit details

     Copy full SHA for 97f10ed

     Browse repository at this point

   • docs(model): re-pin Super VL GPU conversion entries to include the launcher fix (#6674)

     Show description for 219dc7c
     kamran-nvidia

     authored
     219dc7cView commit details

     Copy full SHA for 219dc7c

     Browse repository at this point

   • docs(model): pin Super VL cards to the public HF revision (#6616)

     Show description for f46415d
     

     pruprakash

     and

     cuichenx

     authored
     f46415dView commit details

     Copy full SHA for f46415d

     Browse repository at this point
3. Commits on Oct 26, 2026

   • [sync][training] fix: honor checkpoint sharding validation policy (#6434)

     Show description for 794e7da
     

     yaoyu-33

     and

     cuichenx

     authored
     794e7daView commit details

     Copy full SHA for 794e7da

     Browse repository at this point

   • fix(ci): AUT-2636 use supported verl FP8 rollout quantization (#6668)

     Show description for 6da1402
     svcnemo-autobot

     authored
     6da1402View commit details

     Copy full SHA for 6da1402

     Browse repository at this point

   • docs(model): re-verify BAGEL card on NeMo 26.10 at fd8ab826 (#6613)

     Show description for 45b213d
     

     pruprakash

     and

     kamran-nvidia

     authored
     45b213dView commit details

     Copy full SHA for 45b213d

     Browse repository at this point

   • docs(model): verify Beam GB200 training leaves at public pins (#6615)

     Show description for ab41eee
     pruprakash

     authored
     ab41eeeView commit details

     Copy full SHA for ab41eee

     Browse repository at this point

   • fix(model): pass --bagel-repo through the conversion launcher for BAGEL import (#6623)

     Show description for 520bef4
     

     pruprakash

     and

     kamran-nvidia

     authored
     520bef4View commit details

     Copy full SHA for 520bef4

     Browse repository at this point

   • test: isolate config loader retry clock mocks (#6505)

     Show description for 11ac788
     cuichenx

     authored
     11ac788View commit details

     Copy full SHA for 11ac788

     Browse repository at this point

   • fix(conversion): keep gathered adapter weights on rank 0 only in save_hf_adapter (#6596)

     Show description for bbbc274
     

     mreso

     and

     gautham-kollu

     authored
     bbbc274View commit details

     Copy full SHA for bbbc274

     Browse repository at this point

   • docs(model): publish a two-node H100 geometry for Super VL conversion (#6606)

     Show description for 57664ce
     

     pruprakash

     and

     cuichenx

     authored
     57664ceView commit details

     Copy full SHA for 57664ce

     Browse repository at this point

   • fix(model): start text-only Super VL H100 SFT card leaf from the import (#6599)

     Show description for 4c8a14e
     

     pruprakash

     and

     cuichenx

     authored
     4c8a14eView commit details

     Copy full SHA for 4c8a14e

     Browse repository at this point

   • test: use test hostname in repository link fixture (#6664)

     Show description for 25cdf97
     balasaajay

     authored
     25cdf97View commit details

     Copy full SHA for 25cdf97

     Browse repository at this point

   • chore: AUT-2633 bump MCore main to ed9c2c8c (#6660)

     Show description for f73818d
     svcnemo-autobot

     authored
     f73818dView commit details

     Copy full SHA for f73818d

     Browse repository at this point

   • fix(build): pin leptonai for NeMo-Run API compatibility (#6576)

     Show description for c51adac
     

     balasaajay

     and

     malay-nagda

     authored
     c51adacView commit details

     Copy full SHA for c51adac

     Browse repository at this point
4. Commits on Oct 24, 2026

   • chore: AUT-2625 bump MCore main to 20527695 (#6654)

     Show description for afcc7ee
     svcnemo-autobot

     authored
     afcc7eeView commit details

     Copy full SHA for afcc7ee

     Browse repository at this point

   • chore: AUT-2615 bump MCore main to 3630e44a (#6631)

     Show description for f8e2933
     svcnemo-autobot

     authored
     f8e2933View commit details

     Copy full SHA for f8e2933

     Browse repository at this point

   • refactor(model): build Gemma 4 text models with ModelBuilder (#6612)

     Show description for 6bca0e5
     kamran-nvidia

     authored
     6bca0e5View commit details

     Copy full SHA for 6bca0e5

     Browse repository at this point
5. Commits on Oct 23, 2026

   • ci: stop norecursedirs from hiding 52 unit test files (#6500)

     Show description for f9a494e
     kamran-nvidia

     authored
     f9a494eView commit details

     Copy full SHA for f9a494e

     Browse repository at this point

   • refactor(model): build Qwen3-Next and Qwen3.5 from builder-backed model configs (#6598)

     Show description for 0d7eade
     kamran-nvidia

     authored
     0d7eadeView commit details

     Copy full SHA for 0d7eade

     Browse repository at this point

   • refactor(model): build Mistral from a builder-backed model config (#6593)

     Show description for 1009b42
     kamran-nvidia

     authored
     1009b42View commit details

     Copy full SHA for 1009b42

     Browse repository at this point

   • fix(ckpt): copy source artifacts from the local HF cache when the Hub is offline (#6643)

     Show description for ad774f0
     kamran-nvidia

     authored
     ad774f0View commit details

     Copy full SHA for ad774f0

     Browse repository at this point

   • refactor(model): build Nemotron-H from a builder-backed hybrid model config (#6609)

     Show description for c8fb26a
     kamran-nvidia

     authored
     c8fb26aView commit details

     Copy full SHA for c8fb26a

     Browse repository at this point

   • refactor(model): finish the Llama and OLMoE builder migrations (#6594)

     Show description for 7bffbd3
     kamran-nvidia

     authored
     7bffbd3View commit details

     Copy full SHA for 7bffbd3

     Browse repository at this point

   • docs(tutorials): replace removed run_recipe.py flags in data tutorials (#6618)

     Show description for eb8e7a3
     Threepointone4

     authored
     eb8e7a3View commit details

     Copy full SHA for eb8e7a3

     Browse repository at this point

   • fix(misc): run Pyxis conversion and inference tasks from the repo root (#6625)

     Show description for 50226bf
     pruprakash

     authored
     50226bfView commit details

     Copy full SHA for 50226bf

     Browse repository at this point