Skip to content

Actions: NVIDIA-NeMo/Megatron-Bridge

Actions

PR Review Guidance

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
claude-review.yml
2,500+ workflow runs
2,500+ workflow runs

Filter by Event

⌥ + click/return to exclude

Filter by Status

⌥ + click/return to exclude

Filter by Branch

⌥ + click/return to exclude

Filter by Actor

⌥ + click/return to exclude
fix(training): realign the distributed optimizer's parameter index with its group order PR Review Guidance #16815: Issue comment #6426 (comment) created by kamran-nvidia
9s
9s
fix(training): realign the distributed optimizer's parameter index with its group order PR Review Guidance #16814: Issue comment #6426 (comment) created by copy-pr-bot Bot
1s
1s
PR Review Guidance PR Review Guidance #16813: requested by copy-pr-bot Bot
9s
9s
fix(model): default MODEL_CONFIG_CLASS to None and opt in per family PR Review Guidance #16812: Issue comment #6305 (comment) created by aroshanghias-nvd
10s
10s
fix(training): exclude dropped anchor blocks from the DSpark loss and metrics PR Review Guidance #16811: Issue comment #5242 (comment) created by aroshanghias-nvd
11s
11s
PR Review Guidance PR Review Guidance #16810: requested by copy-pr-bot Bot
9s
9s
perf(recipe): add an 8-GPU GB300 MXFP8 pretrain recipe for Qwen3.5-VL 35B-A3B PR Review Guidance #16809: Issue comment #6413 (comment) created by tomlifu
1s
1s
[bug] Qwen3-30B-A3B GB200 MXFP8 perf preset (full_iteration CUDA graphs + HybridEP) OOMs in first optimizer step on PyTorch >=2.14 / TE >=2.18 stacks PR Review Guidance #16808: Issue comment #6307 (comment) created by janbernloehr
10s
10s
fix(recipe): use partial CUDA graphs for Qwen3 GB200 MXFP8 PR Review Guidance #16807: Issue comment #6364 (comment) created by janbernloehr
10s
10s
PR Review Guidance PR Review Guidance #16806: requested by copy-pr-bot Bot
10s
10s
feat(training): add DPO training with offline-scored reference model PR Review Guidance #16805: Issue comment #6092 (comment) created by Josephasafg
7s
7s
docs(model): re-verify Nemotron 3.5 Lightning training at a recipe-containing revision PR Review Guidance #16804: Issue comment #6425 (comment) created by copy-pr-bot Bot
1s
1s
fix(model): default MODEL_CONFIG_CLASS to None and opt in per family PR Review Guidance #16803: Issue comment #6305 (comment) created by jayzuccarelli
1s
1s
fix(checkpoint): use global FSDP v1 pipeline layer keys PR Review Guidance #16802: Issue comment #6424 (comment) created by copy-pr-bot Bot
9s
9s
PR Review Guidance PR Review Guidance #16801: requested by copy-pr-bot Bot
8s
8s
cp: fix(recipe)!: run Nemotron 3 Ultra H100 SFT on the documented 48-node layout (6376) into r0.7.0 PR Review Guidance #16800: Issue comment #6423 (comment) created by copy-pr-bot Bot
1s
1s
cp: fix(recipe)!: run Nemotron 3 Ultra H100 SFT on the documented 48-node layout (6376) into r0.7.0 PR Review Guidance #16799: Issue comment #6423 (comment) created by svcnvidia-nemo-ci
7s
7s
PR Review Guidance PR Review Guidance #16798: requested by copy-pr-bot Bot
9s
9s
feat(nemotron): Add GTP + partial CG + cuDNN LN support for NT3 Ultra MXFP8 GB200/GB300/VRNVL72 recipes PR Review Guidance #16797: Issue comment #6228 (comment) created by ghadiaravi13
1s
1s
docs(ckpt): clarify config-only Hugging Face export PR Review Guidance #16796: Issue comment #6351 (comment) created by aroshanghias-nvd
11s
11s
fix(model): default MODEL_CONFIG_CLASS to None and opt in per family PR Review Guidance #16795: Issue comment #6305 (comment) created by aroshanghias-nvd
10s
10s
PR Review Guidance PR Review Guidance #16794: requested by copy-pr-bot Bot
9s
9s
fix(training): reuse Megatron-Core determinism setup under --deterministic PR Review Guidance #16793: Issue comment #6271 (comment) created by ZhiyuLi-Nvidia
1s
1s
Support for NVCRE init containers PR Review Guidance #16792: Issue comment #6375 (comment) created by sudostock
1s
1s
PR Review Guidance PR Review Guidance #16791: requested by copy-pr-bot Bot
11s
11s