-
Notifications
You must be signed in to change notification settings - Fork 301
Pull requests: NVIDIA-NeMo/Automodel
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
refactor(optim): split DTensor param groups and avoid TE master-weight duplication for FP32
#3860
opened Sep 10, 2026 by
akoumpa
Contributor
Loading…
fix(qwen3_5): keep vision tower rotary inv_freq in fp32 across the dt…
community-request
#3856
opened Sep 10, 2026 by
haowen-han
Loading…
feat: add DeepSeek-V4.1-Flash model support
community-request
waiting-on-customer
Waiting on the original author to respond
#3855
opened Sep 10, 2026 by
khazic
Contributor
Loading…
3 of 5 tasks
fix(loss): handle strided soft cross-entropy inputs
community-request
#3854
opened Sep 10, 2026 by
MrCapricornLiu
Loading…
3 tasks done
fix(loss): apply per-token gradients in parallel cross entropy
community-request
#3853
opened Sep 10, 2026 by
MrCapricornLiu
Loading…
3 tasks done
feat(vlm): Nemotron 3.5 Super VL text-to-SQL LoRA recipe and guide
#3852
opened Sep 10, 2026 by
HuiyingLi
Contributor
Loading…
fix(loss): scale InfoNCE logits before masking padded negatives
community-request
#3850
opened Sep 10, 2026 by
MrCapricornLiu
Loading…
3 tasks done
fix(data): support Mistral tokenizers in the distillation collator
community-request
#3849
opened Sep 9, 2026 by
MrCapricornLiu
Loading…
3 tasks done
fix(loss): preserve student gradients for fully ignored KD batches
community-request
#3848
opened Sep 9, 2026 by
MrCapricornLiu
Loading…
3 tasks done
fix: preserve Nemotron router state during HF load
#3845
opened Sep 9, 2026 by
fujial-code
Loading…
3 tasks done
fix(peft): export merged grouped-expert adapters
#3834
opened Sep 8, 2026 by
pstjohn
Contributor
Loading…
3 tasks done
fix(vlm): skip dense masks for compact packing backends
community-request
#3831
opened Sep 8, 2026 by
Butterfingrz
Contributor
Loading…
3 tasks done
fix(recipes): normalize seq-cls loss and gradients by a global label count
community-request
#3830
opened Sep 8, 2026 by
kabirvashisht4-glitch
Contributor
Loading…
3 tasks done
fix(distributed): synchronize tensor-parallel replicas
#3822
opened Sep 5, 2026 by
yuhezhang-ai
Contributor
Loading…
4 of 5 tasks
fix(training): reduce grad-clip norms across the expert-parallel axis
community-request
#3817
opened Sep 5, 2026 by
stanley1208
Contributor
Loading…
perf(training): accelerate grad clipping with fused Triton kernel
community-request
waiting-on-customer
Waiting on the original author to respond
#3815
opened Sep 5, 2026 by
piyushumate
Contributor
Loading…
3 tasks done
fix(checkpoint): restore serialized module extra state safely
community-request
waiting-on-customer
Waiting on the original author to respond
#3814
opened Sep 5, 2026 by
jquesnelle
Loading…
3 tasks done
fix(models): honor TE linear backend and native initialization
community-request
waiting-on-customer
Waiting on the original author to respond
#3813
opened Sep 5, 2026 by
jquesnelle
Loading…
3 tasks done
feat(quantization): support copy-safe TE module exclusions
community-request
waiting-on-maintainers
Waiting on maintainers to respond
#3812
opened Sep 5, 2026 by
jquesnelle
Loading…
3 tasks done
feat(minimax-m3): add SM100 MSA sparse attention with custom backward
community-request
#3809
opened Sep 4, 2026 by
Butterfingrz
Contributor
Loading…
2 of 3 tasks
fix(init): initialize DTensor-sharded models whose embedding has a padding_idx
#3808
opened Sep 4, 2026 by
yisongbetter
Contributor
Loading…
3 tasks done
fix(loss): make ChunkedCrossEntropy reductions independent of chunk_len
community-request
waiting-on-customer
Waiting on the original author to respond
#3804
opened Sep 3, 2026 by
kabirvashisht4-glitch
Contributor
Loading…
3 tasks done
perf(moe): fuse router weight multiply with Triton kernel
community-request
#3800
opened Sep 3, 2026 by
piyushumate
Contributor
Loading…
3 tasks done
Previous Next
ProTip!
What’s not been updated in a month: updated:<2026-08-10.