Code and samples from the paper ["Language Models are Unsupervised Multitask Learners"](https://d4mucfpksywv.cloudfront.net/better-language-models/language-models.pdf).
For now, we have only released a smaller (117M parameter) version of GPT-2.
See more details in our [blog post](https://blog.openai.com/better-language-models/).
This repository is meant to be a starting point for researchers and engineers to experiment with GPT-2-117M. While GPT-2-117M is less proficient than GPT-2-1.5B, it is useful for a wide range of research and applications which could also apply to larger models.
### Some caveats
- GPT-2-117M robustness and worst case behaviors are not well-understood. As with any machine-learned model, carefully evaluate GPT-2-117M for your use case, especially if used without fine-tuning or in safety-critical applications where reliability is important.
- The dataset our GPT-2-117M was trained on contains many texts with [biases](https://twitter.com/TomerUllman/status/1101485289720242177) and factual inaccuracies, and thus GPT-2-117M is likely to be biased and inaccurate as well.
- To avoid having samples mistaken as human-written, we recommend clearly labeling samples as synthetic before wide dissemination. Our models are often incoherent or inaccurate in subtle ways, which takes more than a quick read for a human to notice.
### Work with us
Please [let us know](mailto:languagequestions@openai.com) if you’re doing interesting research with or working on applications of GPT-2-117M! We’re especially interested in hearing from and potentially working with those who are studying
- Potential malicious use cases and defenses against them (e.g. the detectability of synthetic text)
- The extent of problematic content (e.g. bias) being baked into the models and effective mitigations
We show conditional samples, with contexts drawn from `WebText`'s test set, with default settings (temperature 1 and no truncation), with temperature 0.7, and with truncation with top_k 40.
