Models Trained with DoReMi Data Mixture
This is a collection of the language models trained using RegMix data mxiture, each with approximately 1B parameters, trained on different seeds. This models aims to server as the strong baseline for our RegMix approach (https://huggingface.co/papers/2407.01492).
- Model Size: 5 separate models trained with different seeds, each with ~1B parameters
- Training Data: our method drived automatic data mixture on the RegMix-Data dataset
- Purpose: To verify the effectiveness of our proposed method
Dataset
The models were trained using the RegMix-Data dataset, which is split into different domains from The Pile dataset.
Training Hyperparameters
Hyperparameter | Value |
---|---|
Batch Size | 1M tokens |
Learning Rate | 4e-4 |
Minimum Learning Rate | 1e-5 |
Learning Rate Schedule | Cosine |
Warmup Ratio | 4% |
Total Tokens | 25B |
How to Load a Model
You can load any model using the corresponding branch with the Hugging Face Transformers library:
from transformers import AutoModel, AutoTokenizer
model = AutoModel.from_pretrained("sail/data-mixture-regmix-1b", revision="seed-1")
tokenizer = AutoTokenizer.from_pretrained("sail/data-mixture-regmix-1b", revision="seed-1")
Data Mixture
The specific data mixture used for training this 1B model is as follows, which can be also found in our code:
train:
train_the_pile_arxiv: 0.0012046169821426883
train_the_pile_freelaw: 0.001454510048554701
train_the_pile_nih_exporter: 0.001231640306882902
train_the_pile_pubmed_central: 0.003108561825532002
train_the_pile_wikipedia_en: 0.01593264140324679
train_the_pile_dm_mathematics: 0.00031106907908634156
train_the_pile_github: 0.00022861228152440253
train_the_pile_philpapers: 1.329107360676338e-05
train_the_pile_stackexchange: 0.00029547405933203174
train_the_pile_enron_emails: 0.0016691646199353991
train_the_pile_gutenberg_pg_19: 0.001612531300038395
train_the_pile_pile_cc: 0.8701291419934237
train_the_pile_ubuntu_irc: 0.06417728505869834
train_the_pile_europarl: 2.9166170357771267e-06
train_the_pile_hackernews: 0.011925517591888925
train_the_pile_pubmed_abstracts: 0.02424425081714838
train_the_pile_uspto_backgrounds: 0.0024587749419225434
valid:
valid_the_pile_pile_cc: 1.0
model_name: tinyllama_1_1b
Model Variants
To access different model variants, simply change the revision
parameter in the from_pretrained
method to the desired seed (e.g., "seed-2", "seed-3"), and the maxium seed is 5.
Model Performance
We evaluated each model using lm-evaluation-harness. The performance metric for each task is the average of 0-shot to 5-shot accnorm
(accuracy normalized, if available) or acc
(accuracy) scores.
Seed | PIQA | LAMBADA | MultiRC | LogiQA | SocialIQA | Winogrande | RACE | OpenBookQA | COPA | HellaSwag | SciQ | ARC Easy | QQP | Average |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | 69.33 | 34.20 | 51.70 | 25.76 | 33.77 | 53.08 | 31.34 | 30.30 | 70.17 | 44.19 | 82.75 | 51.68 | 58.34 | 48.97 |
2 | 69.47 | 34.02 | 50.71 | 26.97 | 33.45 | 52.06 | 30.99 | 29.64 | 70.40 | 44.17 | 82.90 | 51.50 | 54.94 | 48.56 |
3 | 69.24 | 31.99 | 54.07 | 23.66 | 33.38 | 51.16 | 30.70 | 30.32 | 69.40 | 43.74 | 82.60 | 52.95 | 53.43 | 48.20 |
4 | 69.18 | 33.29 | 54.21 | 25.35 | 33.34 | 52.27 | 31.67 | 29.28 | 69.20 | 44.00 | 82.34 | 53.32 | 55.07 | 48.65 |
5 | 68.39 | 31.01 | 53.43 | 25.38 | 33.57 | 51.87 | 31.44 | 29.40 | 70.40 | 43.74 | 83.46 | 51.28 | 56.49 | 48.45 |
Usage Notes
- These models are primarily intended for research purposes.
- Performance may vary depending on the specific task and domain.
Citation
If you use these models in your research, please cite the RegMix paper:
@article{liu2024regmix,
title={RegMix: Data Mixture as Regression for Language Model Pre-training},
author={Liu, Qian and Zheng, Xiaosen and Muennighoff, Niklas and Zeng, Guangtao and Dou, Longxu and Pang, Tianyu and Jiang, Jing and Lin, Min},
journal={arXiv preprint arXiv:2407.01492},
year={2024}
}
For more information about the RegMix methodology and its applications, please refer to the original paper.
- Downloads last month
- 17