allenai/RL4LMs

View on GitHub

A modular RL library to fine-tune language models to human preferences

Stars: 2.4k
Forks: 201
Open beginner issues: 0
Indexed issues: 47

Dominant language: Python
License: Apache-2.0
Last GitHub push: Mar 30, 2023
Latest indexed: Sep 13, 2026
Contributing guide: No contributing guide
Code of conduct: No code of conduct
Beginner labels: good first issue help wanted
PR merge metrics: No merged PRs in 30d

Open All

47 issues indexed so far

Any plans for Deepspeed/Accelerate integration? Open

allenai/RL4LMs#4 · 9 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Numbeams Open

beam_search bug

allenai/RL4LMs#5 · 5 comments · 0 reactions · 1 assignee · Indexed Sep 13, 2026
Top-K and Top-p sampling Open

allenai/RL4LMs#7 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
BART supervised Open

allenai/RL4LMs#10 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
OOM on summarization example Open

allenai/RL4LMs#12 · 15 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
100% likely that two function parameters have been merged by accident Open

code enhancement good first issue

allenai/RL4LMs#16 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Implementing self-play Open

allenai/RL4LMs#18 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Larger models like GPT-J and GPT-NeoX-20B Open

allenai/RL4LMs#21 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Just a warning that the package doesn't work with Transformers 4.25.1 Open

allenai/RL4LMs#22 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Off-policy RL algorithms support Open

enhancement help wanted

allenai/RL4LMs#23 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Problems with models that don't have the parallelize() function Open

allenai/RL4LMs#25 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
passing extra variable to the forward function Open

allenai/RL4LMs#26 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Reproducing IMDB results Open

allenai/RL4LMs#28 · 4 comments · 1 reaction · 1 assignee · Indexed Sep 13, 2026
Mix-Precision training Open

allenai/RL4LMs#29 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Evaluating a specific checkpoint Open

allenai/RL4LMs#32 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Is it possible to release the code based on Jax Open

allenai/RL4LMs#33 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Problem with BLEURT reward function Open

allenai/RL4LMs#34 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Persistent Variance in IMDB Open

allenai/RL4LMs#37 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Value is not broadcastable with batch_shape+event_shape Open

allenai/RL4LMs#38 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
_pickle.UnpicklingError: pickle data was truncated Open

allenai/RL4LMs#41 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Metric version incompatible Open

allenai/RL4LMs#42 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Bloom Supporting Open

allenai/RL4LMs#44 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
A question bother me a long time: What is the difference between RL-for-text-generation and delete-0-reward-model-predictions? Open

allenai/RL4LMs#46 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
In the paper, what is the detail setting of supervised learning? Is SL has additional supervised data? Open

allenai/RL4LMs#49 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
[Question] End-to-end example Open

allenai/RL4LMs#51 · 0 comments · 3 reactions · 0 assignees · Indexed Sep 13, 2026
Error with Accelerate integration + NLPO Open

allenai/RL4LMs#52 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Bug while loading t5 base model Open

allenai/RL4LMs#53 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
How can I inference data with the model after PPO training? Open

allenai/RL4LMs#54 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Using GPT-2 Open

allenai/RL4LMs#55 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
'GPT2Model' object has no attribute 'first_device' Open

allenai/RL4LMs#56 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 13, 2026
model.generate.scores returning two scores Open

allenai/RL4LMs#57 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 13, 2026
CPU Support Minor Bug Open

allenai/RL4LMs#59 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
is multi-dimensional reward supported? Open

allenai/RL4LMs#60 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Memory issue in metric evals? Open

allenai/RL4LMs#61 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Reproducing existing results on NarrativeQA Open

allenai/RL4LMs#62 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
NLPO Code Error and Query About gymnasium vs gym Usage Open

allenai/RL4LMs#63 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Pip install error with gym and torch Open

allenai/RL4LMs#64 · 3 comments · 5 reactions · 0 assignees · Indexed Sep 13, 2026
Do you have any plans to apply the recently published Reinforced Self-Training (ReST)? Open

allenai/RL4LMs#65 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Is PPO really better than SFT (in general)? under the condition of same amount of data Open

allenai/RL4LMs#66 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
how to stop env parallel multi-process to debug env.step()? Open

allenai/RL4LMs#68 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Upgrade to torch 2.0 Open

allenai/RL4LMs#69 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Question about the classifier used for IntentAccuracyDailyDialog. Open

allenai/RL4LMs#71 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
An issue happened when I set up Open

allenai/RL4LMs#72 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
Which version of pip should we use when we try to set up the environment? Open

allenai/RL4LMs#73 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
assert self.name.count("/") == 1 Open

allenai/RL4LMs#75 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
An Issues about RuntimeError Open

allenai/RL4LMs#76 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026
query regarding the reference model Open

allenai/RL4LMs#77 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 13, 2026

allenai/RL4LMs

Get new issues in your inbox