mirror of https://github.com/wassname/Open-Assistant.git synced 2026-06-27 16:10:30 +08:00

Files

T

Kian-Meng Ang 1e321a6fca Fix typos (#1143 )

Found via `codespell -S .mypy_cache,yarn.lock,*.json,*.ipynb -L
rouge,nam,vie`

2023-02-05 20:18:03 +01:00

configs

[fix] Add working A100 config for deberta-xxlarge (deepspeed stuck during evaluation, deadlock?)

2023-01-29 03:17:11 +00:00

tests

[feature] add initial version of anthropic dataset

2023-01-25 05:03:43 +00:00

cls_dataset.py

apply pre-commit rules

2023-01-02 00:01:45 +00:00

experimental_dataset.py

[fix] rename old summary human feedback dataset to new one

2023-01-04 01:00:37 +00:00

models.py

removed old precommit pragma requirement

2023-01-04 20:51:19 -05:00

rank_datasets.py

Fix typos (#1143 )

2023-02-05 20:18:03 +01:00

README.md

run prettier with new params

2023-01-01 20:57:35 +00:00

requirements.txt

fixed linting

2023-01-03 20:47:33 -05:00

summary_quality_trainer.py

apply pre-commit rules

2023-01-02 00:01:45 +00:00

TODO.md

Fix typos (#1143 )

2023-02-05 20:18:03 +01:00

trainer.py

[fix] Added support for deepspeed

2023-01-28 00:55:40 +00:00

utils.py

[fix] Add working A100 config for deberta-xxlarge (deepspeed stuck during evaluation, deadlock?)

2023-01-29 03:17:11 +00:00

README.md

Sections to train Reward Model (RM)

Trainer code based on huggingface. Compatible with deepspeed or accelerate

Requirements

wandb
evaluate
datasets
transformers
torch==1.12

Start training reward model

python trainer.py configs/electra-base-dis-webgpt.yml

Additional axis labeling, this outputs a 4 summary quality evaluation metrics (score are normalized to 0-1 )

python summary_quality_trainer.py configs/test-bloomz-560m-quality.yml

The four summary are :

overall
accuracy
coverage
coherence

Dataset

For now we only supports webgpt and summary dataset from OpenAI. Once open-asisstant dataset are available it will be added here.

Model

Check out configs

Open-Assistant/model/reward/instructor/configs/
    bloomz-560m.yml
    electra-base-dis-webgpt.yml
    galactica-125m.yml
    galactica-1b.yml

You can add new huggingface model as you want.