-
-
Notifications
You must be signed in to change notification settings - Fork 1.4k
Pull requests: axolotl-ai-cloud/axolotl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
feat: Nemotron-3 latentmoe support for ScatterMoE/SonicMoE (non-gated experts + NVFP4 checkpoint loading)
run-gpu-tests
#3866
opened Jul 21, 2026 by
winglian
Collaborator
Loading…
fix: drop samples with no trainable tokens (all labels -100)
#3864
opened Jul 21, 2026 by
vineethsaivs
Contributor
Loading…
fix: multipack drop_last total_slots stale reference
#3863
opened Jul 21, 2026 by
Solaris-star
Loading…
1 of 3 tasks
refactor: replace magic number -100 with IGNORE_INDEX constant
#3862
opened Jul 21, 2026 by
lxcxjxhx
Contributor
Loading…
4 tasks done
fix: correct total_token_slots accounting when dropping incomplete multipack batch
#3861
opened Jul 20, 2026 by
SAY-5
Loading…
1 task done
fix(samplers): correct packing stats when drop_last discards a batch (#3848)
hold
don't merge this yet
waiting for reporter
#3849
opened Jul 17, 2026 by
Anai-Guo
Contributor
Loading…
fix: FA4 quack0.6 with latest transformers support
scheduled_release
This PR is slated for the upcoming release
#3847
opened Jul 17, 2026 by
NanoCode012
Collaborator
Loading…
Context Parallel / Sequence Parallel support for Flash Attention via kernels
under review
#3844
opened Jul 17, 2026 by
winglian
Collaborator
Loading…
fix(loaders): pin device_map per-rank for 4-bit/8-bit DDP (multi-GPU QLoRA collapses to world_size 1)
#3839
opened Jul 15, 2026 by
rm-winpra
Loading…
fix(schemas): honor micro_batch_size/gradient_accumulation_steps defaults in check_batch_size_fields
#3837
opened Jul 12, 2026 by
Anai-Guo
Contributor
Loading…
feat(kernels): add quack fused gated-MLP integration (quack_mlp_kernel)
#3829
opened Jul 11, 2026 by
NanoCode012
Collaborator
•
Draft
feat(dpo): support tools and tool_calls in chat_template strategy
#3825
opened Jul 10, 2026 by
inoutro
Loading…
2 tasks done
feat: external benchmark API plugin
#3818
opened Jul 10, 2026 by
thad0ctor
Contributor
Loading…
1 task done
feat(multimodal): sample packing for VLMs
#3817
opened Jul 10, 2026 by
thad0ctor
Contributor
Loading…
1 task done
fix(pretrain): use non-overlapping chunking so BOS/EOS appear only at true document boundaries
#3809
opened Jul 9, 2026 by
namantiwari2002
Loading…
feat(integrations): expert_offload — single-GPU CPU offload of frozen 4-bit MoE experts
hold
don't merge this yet
review requested
#3797
opened Jul 4, 2026 by
pjordanandrsn
Loading…
docs: document S3-compatible endpoint override
waiting for reporter
#3795
opened Jul 3, 2026 by
davidmyriel
Loading…
fix: bound chat template nesting depth in the jinja analyzer
review requested
#3768
opened Jun 30, 2026 by
eharris128
Loading…
1 of 3 tasks
feat(nvfp4): native NVFP4-GEMM training refactor as plugin
#3767
opened Jun 30, 2026 by
thad0ctor
Contributor
Loading…
4 tasks done
fix: auto-detect EOT token from chat template when eos_token is absent (Gemma)
invalid
This doesn't seem right
waiting for reporter
#3762
opened Jun 28, 2026 by
JSap0914
Contributor
Loading…
feat: agent skills framework + liger-upstream-sync agent skill
under review
#3752
opened Jun 24, 2026 by
thad0ctor
Contributor
Loading…
2 tasks done
Previous Next
ProTip!
Mix and match filters to narrow down what you’re looking for.