-
Notifications
You must be signed in to change notification settings - Fork 301
All issues
Issue creation is restricted in this repository
- #3278 路 HuiyingLi opened
on Jul 28, 2026 - #2958 路 khazic opened
on Jul 7, 2026 7 - #3576 路 yuhezhang-ai opened
on Aug 18, 2026 1
Issues
is:issue state:open
is:issue state:open
Search results
- Status: Open.#3857聽In NVIDIA-NeMo/Automodel;
- Status: Open.#3846聽In NVIDIA-NeMo/Automodel;
- Status: Open.#3837聽In NVIDIA-NeMo/Automodel;
[TEST] Continuous day-0 release validation
enhancementNew feature or requestNew feature or requestStatus: Open.#3836聽In NVIDIA-NeMo/Automodel;Sequence-classification recipe scales gradients by grad_accumulation_steps
waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respondStatus: Open.#3829聽In NVIDIA-NeMo/Automodel;grad clipping computes a different "global" norm on every EP group with TE experts
bugSomething isn't workingSomething isn't workingwaiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respondStatus: Open.#3816聽In NVIDIA-NeMo/Automodel;- Status: Open.#3807聽In NVIDIA-NeMo/Automodel;
ChunkedCrossEntropy returns num_chunks x the loss for reduction='mean' and wrong-shaped output for 'none'
waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respondStatus: Open.#3803聽In NVIDIA-NeMo/Automodel;VLM LengthGroupedSampler logs a 1.2x pre-filter threshold but applies max_length + 512
waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respondStatus: Open.#3798聽In NVIDIA-NeMo/Automodel;Fully-masked batch yields NaN loss in 3 of 4 losses; only MaskedCrossEntropy guards it
waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respondStatus: Open.#3796聽In NVIDIA-NeMo/Automodel;Mistral3ForConditionalGeneration: forward does not declare logits_to_keep, so fused losses are silently replaced
waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respondStatus: Open.#3778聽In NVIDIA-NeMo/Automodel;--nproc-per-node is overridden by the local GPU count
waiting-on-maintainersWaiting on maintainers to respondWaiting on maintainers to respondStatus: Open.#3774聽In NVIDIA-NeMo/Automodel;