|
--- |
|
language: |
|
- en |
|
tags: |
|
- pytorch |
|
- causal-lm |
|
- pythia |
|
license: apache-2.0 |
|
datasets: |
|
- EleutherAI/pile |
|
--- |
|
|
|
The *Pythia Scaling Suite* is a collection of models developed to facilitate |
|
interpretability research [(see paper)](https://arxiv.org/pdf/2304.01373.pdf). |
|
It contains two sets of eight models of sizes |
|
70M, 160M, 410M, 1B, 1.4B, 2.8B, 6.9B, and 12B. For each size, there are two |
|
models: one trained on the Pile, and one trained on the Pile after the dataset |
|
has been globally deduplicated. All 8 model sizes are trained on the exact |
|
same data, in the exact same order. We also provide 154 intermediate |
|
checkpoints per model, hosted on Hugging Face as branches. |
|
|
|
The Pythia model suite was deliberately designed to promote scientific |
|
research on large language models, especially interpretability research. |
|
Despite not centering downstream performance as a design goal, we find the |
|
models <a href="#evaluations">match or exceed</a> the performance of |
|
similar and same-sized models, such as those in the OPT and GPT-Neo suites. |
|
|
|
<details> |
|
<summary style="font-weight:600">Details on previous early release and naming convention.</summary> |
|
|
|
Previously, we released an early version of the Pythia suite to the public. |
|
However, we decided to retrain the model suite to address a few hyperparameter |
|
discrepancies. This model card <a href="#changelog">lists the changes</a>; |
|
see appendix B in the Pythia paper for further discussion. We found no |
|
difference in benchmark performance between the two Pythia versions. |
|
The old models are |
|
[still available](https://huggingface.co/models?other=pythia_v0), but we |
|
suggest the retrained suite if you are just starting to use Pythia.<br> |
|
**This is the current release.** |
|
|
|
Please note that all models in the *Pythia* suite were renamed in January |
|
2023. For clarity, a <a href="#naming-convention-and-parameter-count">table |
|
comparing the old and new names</a> is provided in this model card, together |
|
with exact parameter counts. |
|
</details> |
|
<br> |
|
|
|
# Pythia-160M |
|
|
|
## Model Details |
|
|
|
- Developed by: [EleutherAI](http://eleuther.ai) |
|
- Model type: Transformer-based Language Model |
|
- Language: English |
|
- Learn more: [Pythia's GitHub repository](https://github.com/EleutherAI/pythia) |
|
for training procedure, config files, and details on how to use. |
|
[See paper](https://arxiv.org/pdf/2304.01373.pdf) for more evals and implementation |
|
details. |
|
- Library: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox) |
|
- License: Apache 2.0 |
|
- Contact: to ask questions about this model, join the [EleutherAI |
|
Discord](https://discord.gg/zBGx3azzUn), and post them in `#release-discussion`. |
|
Please read the existing *Pythia* documentation before asking about it in the |
|
EleutherAI Discord. For general correspondence: [contact@eleuther. |
|
ai](mailto:contact@eleuther.ai). |
|
|
|
<figure> |
|
|
|
| Pythia model | Non-Embedding Params | Layers | Model Dim | Heads | Batch Size | Learning Rate | Equivalent Models | |
|
| -----------: | -------------------: | :----: | :-------: | :---: | :--------: | :-------------------: | :--------------------: | |
|
| 70M | 18,915,328 | 6 | 512 | 8 | 2M | 1.0 x 10<sup>-3</sup> | — | |
|
| 160M | 85,056,000 | 12 | 768 | 12 | 2M | 6.0 x 10<sup>-4</sup> | GPT-Neo 125M, OPT-125M | |
|
| 410M | 302,311,424 | 24 | 1024 | 16 | 2M | 3.0 x 10<sup>-4</sup> | OPT-350M | |
|
| 1.0B | 805,736,448 | 16 | 2048 | 8 | 2M | 3.0 x 10<sup>-4</sup> | — | |
|
| 1.4B | 1,208,602,624 | 24 | 2048 | 16 | 2M | 2.0 x 10<sup>-4</sup> | GPT-Neo 1.3B, OPT-1.3B | |
|
| 2.8B | 2,517,652,480 | 32 | 2560 | 32 | 2M | 1.6 x 10<sup>-4</sup> | GPT-Neo 2.7B, OPT-2.7B | |
|
| 6.9B | 6,444,163,072 | 32 | 4096 | 32 | 2M | 1.2 x 10<sup>-4</sup> | OPT-6.7B | |
|
| 12B | 11,327,027,200 | 36 | 5120 | 40 | 2M | 1.2 x 10<sup>-4</sup> | — | |
|
<figcaption>Engineering details for the <i>Pythia Suite</i>. Deduped and |
|
non-deduped models of a given size have the same hyperparameters. “Equivalent” |
|
models have <b>exactly</b> the same architecture, and the same number of |
|
non-embedding parameters.</figcaption> |
|
</figure> |
|
|
|
## 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 also provide |
|
154 checkpoints per model: initial `step0`, 10 log-spaced checkpoints |
|
`step{1,2,4...512}`, and 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 also further fine-tune and adapt Pythia-160M 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](https://huggingface.co/docs/transformers/index). If you decide to use |
|
pre-trained Pythia-160M 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 in itself |
|
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-160M 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-160M will **not** |
|
respond to a given prompt the way a product like ChatGPT does. This is because, |
|
unlike this model, 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-160M to produce factually accurate |
|
output. |
|
|
|
This model was trained on [the Pile](https://pile.eleuther.ai/), a dataset |
|
known to contain profanity and texts that are lewd or otherwise offensive. |
|
See [Section 6 of the Pile paper](https://arxiv.org/abs/2101.00027) for a |
|
discussion of documented biases with regards to gender, religion, and race. |
|
Pythia-160M 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-160M. |
|
|
|
### Quickstart |
|
|
|
Pythia models can be loaded and used via the following code, demonstrated here |
|
for the third `pythia-70m-deduped` checkpoint: |
|
|
|
```python |
|
from transformers import GPTNeoXForCausalLM, AutoTokenizer |
|
|
|
model = GPTNeoXForCausalLM.from_pretrained( |
|
"EleutherAI/pythia-70m-deduped", |
|
revision="step3000", |
|
cache_dir="./pythia-70m-deduped/step3000", |
|
) |
|
|
|
tokenizer = AutoTokenizer.from_pretrained( |
|
"EleutherAI/pythia-70m-deduped", |
|
revision="step3000", |
|
cache_dir="./pythia-70m-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.<br> |
|
For more information on how to use all Pythia models, see [documentation on |
|
GitHub](https://github.com/EleutherAI/pythia). |
|
|
|
## Training |
|
|
|
### Training data |
|
|
|
[The Pile](https://pile.eleuther.ai/) 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), and |
|
miscellaneous (e.g. GitHub, Enron Emails). See [the Pile |
|
paper](https://arxiv.org/abs/2101.00027) for a breakdown of all data sources, |
|
methodology, and a discussion of ethical implications. Consult [the |
|
datasheet](https://arxiv.org/abs/2201.07311) for more detailed documentation |
|
about the Pile and its component datasets. The Pile can be downloaded from |
|
the [official website](https://pile.eleuther.ai/), or from a [community |
|
mirror](https://the-eye.eu/public/AI/pile/).<br> |
|
The Pile was **not** deduplicated before being used to train Pythia-160M. |
|
|
|
### 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`). In addition, we |
|
also provide frequent early checkpoints: `step0` and `step{1,2,4...512}`. |
|
This corresponds to training for just under 1 epoch on the Pile for |
|
non-deduplicated models, and about 1.5 epochs on the deduplicated Pile. |
|
|
|
All *Pythia* models trained for 143000 steps at a batch size |
|
of 2M (2,097,152 tokens).<br> |
|
See [GitHub](https://github.com/EleutherAI/pythia) for more details on training |
|
procedure, including [how to reproduce |
|
it](https://github.com/EleutherAI/pythia/blob/main/README.md#reproducing-training).<br> |
|
Pythia uses the same tokenizer as [GPT-NeoX- |
|
20B](https://huggingface.co/EleutherAI/gpt-neox-20b). |
|
|
|
## Evaluations |
|
|
|
All 16 *Pythia* models were evaluated using the [LM Evaluation |
|
Harness](https://github.com/EleutherAI/lm-evaluation-harness). You can access |
|
the results by model and step at `results/json/*` in the [GitHub |
|
repository](https://github.com/EleutherAI/pythia/tree/main/results/json/).<br> |
|
Expand the sections below to see plots of evaluation results for all |
|
Pythia and Pythia-deduped models compared with OPT and BLOOM. |
|
|
|
<details> |
|
<summary>LAMBADA – OpenAI</summary> |
|
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/lambada_openai_v1.png" style="width:auto"/> |
|
</details> |
|
|
|
<details> |
|
<summary>Physical Interaction: Question Answering (PIQA)</summary> |
|
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/piqa_v1.png" style="width:auto"/> |
|
</details> |
|
|
|
<details> |
|
<summary>WinoGrande</summary> |
|
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/winogrande_v1.png" style="width:auto"/> |
|
</details> |
|
|
|
<details> |
|
<summary>AI2 Reasoning Challenge—Easy Set</summary> |
|
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/arc_easy_v1.png" style="width:auto"/> |
|
</details> |
|
|
|
<details> |
|
<summary>SciQ</summary> |
|
<img src="/EleutherAI/pythia-12b/resolve/main/eval_plots/sciq_v1.png" style="width:auto"/> |
|
</details> |
|
|
|
## Changelog |
|
|
|
This section compares differences between previously released |
|
[Pythia v0](https://huggingface.co/models?other=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 0.1× 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. |
|
|
|
<figure style="width:32em"> |
|
|
|
| current Pythia suffix | old suffix | total params | non-embedding params | |
|
| --------------------: | ---------: | -------------: | -------------------: | |
|
| 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 | |
|
</figure> |
