-
Notifications
You must be signed in to change notification settings - Fork 550
Pull requests: pytorch/torchtune
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
(draft/discussion) GRPO LoRA
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2467
opened Mar 9, 2025 by
ianbarber
Loading…
12 tasks
Only offload if activation is on CUDA
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2466
opened Mar 6, 2025 by
janeyx99
Loading…
3 of 13 tasks
Reference-free DPO losses in torchtune.
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2465
opened Mar 6, 2025 by
krammnic
Loading…
3 of 13 tasks
Add validation dataset loss to distributed SFT recipies
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2464
opened Mar 6, 2025 by
bzz
Loading…
5 of 13 tasks
[WIP][RFC] Refactored and Simplified Recipes
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2447
opened Feb 28, 2025 by
pbontrager
•
Draft
Enable PPO on Intel XPU using a tiny model
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2446
opened Feb 27, 2025 by
songhappy
Loading…
3 tasks done
[WIP] Add This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
StatefulDataLoader
to all recipes except knowledge_single
CLA Signed
#2441
opened Feb 26, 2025 by
krammnic
Loading…
2 of 13 tasks
Scale grads by dp size rather than world size
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2440
opened Feb 26, 2025 by
joecummings
•
Draft
13 tasks
[WIP] Add Llama3.3 tokenizer & replace 'ipython' role with 'tool'
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2438
opened Feb 26, 2025 by
Ankur-singh
•
Draft
6 of 13 tasks
Bugfixes: Grad norm scaling in TP
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
Rough draft for integration with DCP HF Storage Reader / Writer
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2435
opened Feb 25, 2025 by
joecummings
•
Draft
Attempt to make the reward function customizable in GRPO
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2433
opened Feb 25, 2025 by
imenelydiaker
Loading…
1 of 13 tasks
[WIP] Padding bug in GRPO
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2425
opened Feb 22, 2025 by
krammnic
Loading…
2 of 13 tasks
[RFC] Proposal to Update PPO Test to Add LR Scheduler
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2423
opened Feb 22, 2025 by
Seoley
Loading…
6 of 13 tasks
GRPO datasets
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2422
opened Feb 22, 2025 by
krammnic
Loading…
[WIP] Federated learning
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
(WIP/RFC) Hybrid Sharding
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2415
opened Feb 20, 2025 by
nathan-az
Loading…
1 of 11 tasks
(WIP/RFC) FP8 full finetune distributed
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
Added Scalable Softmax module
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2397
opened Feb 15, 2025 by
AbdullahCompE
Loading…
7 of 13 tasks
Kd recipe update
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2395
opened Feb 14, 2025 by
rajuptvs
Loading…
4 of 13 tasks
[WIP] max-autotune
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2393
opened Feb 13, 2025 by
krammnic
Loading…
1 of 13 tasks
Implemented FIRE positional encoding module
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2388
opened Feb 12, 2025 by
kaddu341
Loading…
6 of 13 tasks
[WIP] Flux finetuning
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2386
opened Feb 12, 2025 by
calvinpelletier
•
Draft
Implement step based checkpointing
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2384
opened Feb 11, 2025 by
joecummings
•
Draft
8 of 13 tasks
fix: Moved dev deps from optional-dependencies to dependency-groups
CLA Signed
This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#2376
opened Feb 10, 2025 by
bogdansalyp
•
Draft
1 of 13 tasks
Previous Next
ProTip!
Adding no:label will show everything without a label.