Pythia-31M-deduped

Model Details

Developed by: EleutherAI
Model type: Transformer-based Language Model
Language: English
Learn more: Pythia's GitHub repository for training procedure, config files, and details on how to use. See paper for more evals and implementation details.
Library: GPT-NeoX
License: Apache 2.0
Contact: Join the EleutherAI Discord and post in #release-discussion. For general correspondence: contact@eleuther.ai

Note: Pythia-31M-deduped was trained after the original Pythia suite at the request of interpretability researchers who wanted a smaller model with the same training setup. It uses the same data, tokenizer, hyperparameter conventions, and checkpoint schedule as the rest of the Pythia suite. Unlike the original Pythia models (70M--12B), Pythia-31M-deduped was only trained on the deduplicated Pile and does not have a separate non-deduplicated variant.

Model Configuration:

Pythia model	Non-Embedding Params	Layers	Model Dim	Heads	Batch Size	Learning Rate	Equivalent Models
14M	1,189,888	6	128	4	2M	1.0 x 10^-3	—
31M	4,739,072	6	256	8	2M	1.0 x 10^-3	—
70M	18,915,328	6	512	8	2M	1.0 x 10^-3	—
160M	85,056,000	12	768	12	2M	6.0 x 10^-4	GPT-Neo 125M, OPT-125M
410M	302,311,424	24	1024	16	2M	3.0 x 10^-4	OPT-350M
1.0B	805,736,448	16	2048	8	2M	3.0 x 10^-4	—
1.4B	1,208,602,624	24	2048	16	2M	2.0 x 10^-4	GPT-Neo 1.3B, OPT-1.3B
2.8B	2,517,652,480	32	2560	32	2M	1.6 x 10^-4	GPT-Neo 2.7B, OPT-2.7B
6.9B	6,444,163,072	32	4096	32	2M	1.2 x 10^-4	OPT-6.7B
12B	11,327,027,200	36	5120	40	2M	1.2 x 10^-4	—

Uses and Limitations

Intended Use

The primary intended use of Pythia is research on the behavior, functionality, and limitations of large language models. This suite is intended to provide a controlled setting for performing scientific experiments. We provide 154 checkpoints per model:

Initial step0
10 log-spaced checkpoints step{1,2,4...512}
143 evenly-spaced checkpoints from step1000 to step143000

These checkpoints are hosted on Hugging Face as branches. Note that branch 143000 corresponds exactly to the model checkpoint on the main branch of each model.

You may further fine-tune and adapt Pythia-31M-deduped for deployment, as long as your use is in accordance with the Apache 2.0 license. Pythia models work with the Hugging Face Transformers Library. If you use pre-trained Pythia-31M-deduped as a basis for your fine-tuned model, please conduct your own risk and bias assessment.

Out-of-scope use

The Pythia Suite is not intended for deployment. It is not a product and cannot be used for human-facing interactions. For example, the model may generate harmful or offensive text. Please evaluate the risks associated with your particular use case.

Pythia models are English-language only and are not suitable for translation or generating text in other languages.

Pythia-31M-deduped has not been fine-tuned for downstream contexts in which language models are commonly deployed, such as writing genre prose or commercial chatbots. This means Pythia-31M-deduped will not respond to a given prompt the way a product like ChatGPT does. ChatGPT was fine-tuned using methods such as Reinforcement Learning from Human Feedback (RLHF) to better "follow" human instructions.

Limitations and biases

The core functionality of a large language model is to take a string of text and predict the next token. The token used by the model need not produce the most "accurate" text. Never rely on Pythia-31M-deduped to produce factually accurate output.

This model was trained on the Pile, a dataset known to contain profanity and texts that are lewd or otherwise offensive. See Section 6 of the Pile paper for a discussion of documented biases with regards to gender, religion, and race. Pythia-31M-deduped may produce socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive.

If you plan on using text generated through, for example, the Hosted Inference API, we recommend having a human curate the outputs of this language model before presenting it to other people. Please inform your audience that the text was generated by Pythia-31M-deduped.

Quickstart

Pythia models can be loaded and used via the following code, demonstrated here for the third pythia-31m-deduped checkpoint:

from transformers import GPTNeoXForCausalLM, AutoTokenizer

model = GPTNeoXForCausalLM.from_pretrained(
  "EleutherAI/pythia-31m-deduped",
  revision="step3000",
  cache_dir="./pythia-31m-deduped/step3000",
)

tokenizer = AutoTokenizer.from_pretrained(
  "EleutherAI/pythia-31m-deduped",
  revision="step3000",
  cache_dir="./pythia-31m-deduped/step3000",
)

inputs = tokenizer("Hello, I am", return_tensors="pt")
tokens = model.generate(**inputs)
tokenizer.decode(tokens[0])

Revision/branch step143000 corresponds exactly to the model checkpoint on the main branch of each model.

For more information on how to use all Pythia models, see documentation on GitHub.

Training

Training data

The Pile is a 825GiB general-purpose dataset in English. It was created by EleutherAI specifically for training large language models. It contains texts from 22 diverse sources, roughly broken down into five categories:

Academic writing (e.g. arXiv)
Internet (e.g. CommonCrawl)
Prose (e.g. Project Gutenberg)
Dialogue (e.g. YouTube subtitles)
Miscellaneous (e.g. GitHub, Enron Emails)

See the Pile paper for a breakdown of all data sources, methodology, and ethical implications. Consult the datasheet for more detailed documentation about the Pile and its component datasets.

The Pile can be downloaded from the official website or from a community mirror.

Pythia-31M-deduped was trained on the deduplicated version of the Pile.

Training procedure

All models were trained on the exact same data, in the exact same order. Each model saw 299,892,736,000 tokens during training, and 143 checkpoints for each model are saved every 2,097,152,000 tokens, spaced evenly throughout training, from step1000 to step143000 (which is the same as main). Early checkpoints are also provided: step0 and step{1,2,4...512}.

All Pythia models trained for 143,000 steps at a batch size of 2M (2,097,152 tokens).

See GitHub for more details on training procedure, including how to reproduce it.

Pythia uses the same tokenizer as GPT-NeoX-20B.

Evaluations

All 16 Pythia models were evaluated using the LM Evaluation Harness. You can access the results by model and step at results/json/* in the GitHub repository.

Changelog

This section compares differences between previously released Pythia v0 and the current models. See Appendix B of the Pythia paper for further discussion of these changes and the motivation behind them. We found that retraining Pythia had no impact on benchmark performance.

All model sizes are now trained with uniform batch size of 2M tokens. Previously, the models of size 160M, 410M, and 1.4B parameters were trained with batch sizes of 4M tokens.
We added checkpoints at initialization (step 0) and steps {1,2,4,8,16,32,64,128,256,512} in addition to every 1000 training steps.
Flash Attention was used in the new retrained suite.
We remedied a minor inconsistency that existed in the original suite: all models of size 2.8B parameters or smaller had a learning rate (LR) schedule which decayed to a minimum LR of 10% the starting LR rate, but the 6.9B and 12B models all used an LR schedule which decayed to a minimum LR of 0. In the redone training runs, we rectified this inconsistency: all models now were trained with LR decaying to a minimum of 0.1x their maximum LR.

Naming convention and parameter count

Pythia models were renamed in January 2023. It is possible that the old naming convention still persists in some documentation by accident. The current naming convention (70M, 160M, etc.) is based on total parameter count.

Current Pythia suffix	Old suffix	Total params	Non-embedding params
14M	—	14,067,712	1,189,888
31M	—	30,494,720	4,739,072
70M	19M	70,426,624	18,915,328
160M	125M	162,322,944	85,056,000
410M	350M	405,334,016	302,311,424
1B	800M	1,011,781,632	805,736,448
1.4B	1.3B	1,414,647,808	1,208,602,624
2.8B	2.7B	2,775,208,960	2,517,652,480
6.9B	6.7B	6,857,302,016	6,444,163,072
12B	13B	11,846,072,320	11,327,027,200

Citation

@article{biderman2023pythia,
  title={Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling},
  author={Biderman, Stella and Schoelkopf, Hailey and Anthony, Quentin Gregory and Bradley, Herbie and O'Brien, Kyle and Hallahan, Eric and Khan, Mohammad Aflah and Purohit, Shivanshu and Prashanth, USVSN Sai and Raff, Edward and others},
  journal={International Conference on Machine Learning},
  year={2023}
}