Skip to content

Pull requests: allenai/open-instruct

New pull request New
51 Open 941 Closed
51 Open 941 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
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
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

CPU offload toggles feature
#1228 opened Nov 25, 2025 by ShotaKaji5207 • Draft
Switches CodeVerifier to use aiohttp instead of requests, enabling native async, and avoiding using asyncio.to_thread.
#1227 opened Nov 24, 2025 by finbarrtimbers Loading…
9
Switches generation to use vllm's OpenAI API, rather than going through vLLM directly.
#1226 opened Nov 24, 2025 by finbarrtimbers • Draft
@finbarrtimbers
Now, reward_fn lives in LLMRayActor.
#1225 opened Nov 24, 2025 by finbarrtimbers • Draft
1
fix metrics in grpo_fast.py
#1223 opened Nov 24, 2025 by mnoukhov Loading…
5
increase prompt_Q size to account for eval prompts and small fixes
#1222 opened Nov 23, 2025 by mnoukhov Loading…
8
Adds retries to checkpointing.
#1213 opened Nov 19, 2025 by finbarrtimbers Loading…
7
Implements an OLMo-core compatible data loader to load HF datasets from get_cached_dataset_tulu.
#1208 opened Nov 18, 2025 by finbarrtimbers • Draft
Simplify pending query map increments codex
#1207 opened Nov 18, 2025 by finbarrtimbers Loading…
3
added staleness metrics
#1204 opened Nov 18, 2025 by mnoukhov Loading…
14
Adds StreamingDataLoader class following the OLMo-core definition, and changes grpo_fast.py to use it.
#1202 opened Nov 17, 2025 by finbarrtimbers • Draft
add filestore fix for multi-learn node augsta checkpointing
#1198 opened Nov 16, 2025 by saurabh111233212 Loading…
8
Makes checkpointing and evaluation bookkeeping run asynchronously
#1196 opened Nov 15, 2025 by finbarrtimbers • Draft
1
pass in model_name_or_path that is only augusta and it works
#1183 opened Nov 12, 2025 by mnoukhov Loading…
11
Now, we cache logprobs to disk in DPO.
#1182 opened Nov 12, 2025 by finbarrtimbers • Draft
Pad out 32b
#1177 opened Nov 12, 2025 by hamishivi • Draft
Changes to make DPO run faster
#1175 opened Nov 11, 2025 by finbarrtimbers • Draft
Added NCCL flags from DPO
#1165 opened Nov 10, 2025 by finbarrtimbers Loading…
3
RL Zero Scripts
#1162 opened Nov 10, 2025 by mnoukhov • Draft
1
Correct loss accumulation for grpo_fast
#1161 opened Nov 10, 2025 by hamishivi Loading…
13
Fix comment clarity in maybe_evaluate function in grpo_fast.py
#1159 opened Nov 10, 2025 by MohdElgaar Loading…
4
Adds DPO and SFT to CI and combines push-image with beaker-experiment
#1148 opened Nov 5, 2025 by finbarrtimbers Loading…
1
Adds a DPO implementation that uses OLMo-core
#1144 opened Nov 4, 2025 by finbarrtimbers • Draft
Simplifies data_preparation_thread
#1142 opened Nov 4, 2025 by finbarrtimbers • Draft
Cleaned up make_internal_command function from mason.py.
#1141 opened Nov 4, 2025 by finbarrtimbers • Draft
ProTip! Filter pull requests by the default branch with base:main.