Skip to main content
sorry
Projects
trlx
Reports
Overview
Workspace
Runs
Automat.
Sweeps
Reports
Artifacts
Reports
Created by
Created On
Last edited
Difference due to the change in base_trainer.decode
0
sorry
2023-09-05
Difference due to changes in base_trainer.decode
0
sorry
2023-09-05
w/ evol-15.08-beluga-13b-threshold
0
sorry
2023-08-16
lora sentiments multigpu comparison
0
sorry
2023-06-22
Different position_ids calculations
0
sorry
2023-06-21
zero3 offload
0
sorry
2023-05-12
Single Q vs two Qs on randomwalks
0
sorry
2023-05-09
update-requirements v. main
0
sorry
2023-05-01
Pad prompts to the right in T5 examples and add EOS token to seq2seq prompts #422
0
sorry
2023-04-27
Sequence-wise v. token-wise mean KL
0
sorry
2023-04-19
[fix] Fix ILQL head sync under ZeRO3 #387
https://github.com/CarperAI/trlx/pull/387
0
sorry
2023-03-23
Cuda OOM with PPO on GPT2-medium #372
https://github.com/CarperAI/trlx/issues/372
0
sorry
2023-03-21
Timing difference
0
sorry
2023-03-16
Untitled Report
0
sorry
2023-03-06
...
0
sorry
2023-03-06
...
0
sorry
2023-03-06
...
0
sorry
2023-03-06
Convert the rest of configs from ymls #346
https://github.com/CarperAI/trlx/pull/346
0
sorry
2023-03-01
Add batch_size option for the reward model #322
https://github.com/CarperAI/trlx/pull/322
0
sorry
2023-02-21
Make gather_for_metrics usage more strict #315
seq2seq example, the difference in reward/mean reflects the difference in len(eval_samples), rest of the metrics are the same
0
sorry
2023-02-20
Make gather_for_metrics more strict #315
https://github.com/CarperAI/trlx/pull/315
0
sorry
2023-02-19
Make gather_for_metrics more strict #315
https://github.com/CarperAI/trlx/pull/315, no changes in behavior
0
sorry
2023-02-17
Untitled Report
0
sorry
2023-02-13
Gather experience samples #305
https://github.com/CarperAI/trlx/pull/305
0
sorry
2023-02-13
Gather experience samples #305
no changes with 1. determinstic reward_fn and 2. a single process runs with usual sentiment pipeline
0
sorry
2023-02-11
Gather experience samples #305
https://github.com/CarperAI/trlx/pull/305
0
sorry
2023-02-10
Add Accelerate SFT Trainer #280
0
sorry
2023-02-08
Add Accelerate SFT Trainer #280
https://github.com/CarperAI/trlx/pull/280 CUDA_VISIBLE_DEVICES=0 python examples/ppo_sentiments.py & CUDA_VISIBLE_DEVICES=1 python examples/sft_sentiments.py & CUDA_VISIBLE_DEVICES=2 python examples/ilql_sentiments.py &
0
sorry
2023-02-07
Set deepspeed's fp16 auto_cast to false #279
0
sorry
2023-02-06
[fix] Set deepspeed's fp16 auto_cast to false #279
https://github.com/CarperAI/trlx/pull/279
0
sorry
2023-02-06
Improve PPO readability #210
https://github.com/CarperAI/trlx/pull/210
0
sorry
2023-02-05
Fix distributed dataloaders & deduplicate eval #276
https://github.com/CarperAI/trlx/pull/276
0
sorry
2023-02-04
Toy example for PPO does learn as much as expected
0
sorry
2023-02-03
Fix empty elements
0
sorry
2023-02-03
Fix heads dtype
zero3 now works for ILQL, no changes for PPO
0
sorry
2023-01-23
random_walks_document v. main
0
sorry
2023-01-22
gptj-rm-static v. gptj-rm-hh ?
0
sorry
2023-01-19
Difference truncation_side left/right
0
sorry
2023-01-13
Update generation utilities #172
ppo_sentiments
0
sorry
2023-01-12
Update generation utilities #172
ilql_sentiments
0
sorry
2023-01-12
Untitled Report
0
sorry
2023-01-04
Fix ppo calculation with unequal generation lenghts
0
sorry
2022-12-14