Skip to content

拉取请求: NVIDIA-NeMo/Automodel

新建拉取请求 新建
223 Open 2,868 Closed
223 Open 2,868 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
项目
Filter by project
Loading
里程碑
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
新建est Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

拉取请求 list

fix(checkpoint): preserve class-based mappings through model wrappers
#3749 opened Aug 29, 2026 by jQizhang Contributor • Draft
1
ci: Update transformers to latest version 5.16.1
#3748 opened Aug 29, 2026 by svcnvidia-nemo-ci Contributor Loading…
1
feat(glm): add GLM-5.3 cuDNN training support
#3747 opened Aug 29, 2026 by HuiyingLi Contributor Loading…
3 tasks done
7
fix: preserve selected routing weights in FP32 in deepseek community-request
#3746 opened Aug 29, 2026 by edjson Contributor • Draft
2 of 10 tasks
1
fix(moe): replay HybridEP layouts under activation checkpointing
#3743 opened Aug 28, 2026 by HuiyingLi Contributor • Draft
3
feat(gemma4): support D512 FFPA attn on SM100 community-request
#3742 opened Aug 28, 2026 by Butterfingrz Contributor Loading…
8 tasks done
3
fix(fsdp): preserve layer ownership for mixed compute dtypes
#3740 opened Aug 28, 2026 by akoumpa Contributor • Draft
13
feat(dataset): add mask_generation_prompt to keep prompt tokens out of loss community-request
#3739 opened Aug 28, 2026 by khazic Contributor Loading…
1
12
chore: CI proxy for #3732
#3737 opened Aug 28, 2026 by HuiyingLi Contributor Loading…
2
chore: CI proxy for #3731
#3736 opened Aug 28, 2026 by HuiyingLi Contributor Loading…
2
feat(attention): add FA4 backend, enable it by default, fix VLM flash packing
#3734 opened Aug 28, 2026 by akoumpa Contributor Loading…
9
build: bump FA3/FA4 pin to pick up CuTe compile-key fixes
#3733 opened Aug 28, 2026 by akoumpa Contributor Loading…
3
perf(moe): Kimi-K3 memory and throughput optimizations community-request
#3732 opened Aug 28, 2026 by yisongbetter Loading…
4
fix(moe): Kimi-K3 correctness fixes (routing, NCCL warmup, PP, per-GPU tps) community-request
#3731 opened Aug 28, 2026 by yisongbetter Loading…
14
debug: test ephemeral v2
#3711 opened Aug 27, 2026 by chtruong814 Contributor Loading…
3 tasks
2
test(pp): add mistral3p5 PP and TP parity tests [4/4]
#3709 opened Aug 27, 2026 by athitten Contributor • Draft
1
fix(fsdp): shard untied embeddings as separate units
#3708 opened Aug 27, 2026 by yuhezhang-ai Contributor Loading…
5 tasks done
7
fix(datasets): correct attention mask for left-padded and pad-id-0 tokenizers
#3704 opened Aug 27, 2026 by akoumpa Contributor • Draft
2
fix(qwen2_5_omni): place thinker prefix inside base_model.model on PEFT saves community-request waiting-on-maintainers Waiting on maintainers to respond
#3700 opened Aug 27, 2026 by taran-dev4u Contributor Loading…
1
1
perf(training): fused multi-tensor gradient-norm reduction
#3685 opened Aug 26, 2026 by akoumpa Contributor • Draft
1
perf(moe): reuse the DeepEP dispatch layout on activation-checkpoint recompute
#3684 opened Aug 26, 2026 by akoumpa Contributor • Draft
5
feat(kimi-k3): add expert LoRA and tune full SFT
#3683 opened Aug 26, 2026 by HuiyingLi Contributor Loading…
3 tasks done
1
feat: support generic freeze-config selectors
#3681 opened Aug 26, 2026 by pstjohn Loading…
13
fix(moe): preserve DeepEP autograd for empty expert ranks
#3680 opened Aug 25, 2026 by Slyne Contributor • Draft
2
fix(diffusion): rank-symmetric DMD2 all-reduce and deferred FSDP2 grad sync
#3679 opened Aug 25, 2026 by akoumpa Contributor Loading…
1
ProTip! Exclude everything labeled bug with -label:bug.