File size: 2,890 Bytes
0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 a6316f5 0a81b20 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 | ---
pipeline_tag: text-generation
license: mit
base_model: openai-community/gpt2-medium
library_name: zeromodels
language:
- en
tags:
- keras
- zeromodels
- gpt2
- gpt-2
- text-generation
- pytorch
- jax
- tf
---
# Run GPT-2 with Keras 3: JAX, PyTorch, or TensorFlow
[](https://github.com/IMvision12/ZeroModels) [](https://imvision12.github.io/ZeroModels/gpt2/)
# zeromodels/gpt2_medium
Paper: [Language Models are Unsupervised Multitask Learners (Radford et al., 2019)](https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf)
GPT-2 is OpenAI's decoder-only transformer language model trained on WebText:
learned absolute position embeddings, pre-LayerNorm blocks, `gelu_new`
activations, a tied output head, and a byte-level BPE tokenizer. This is the
**355M** variant, a base completion model (no chat template).
For more details, see the upstream [model card](https://huggingface.co/openai-community/gpt2-medium).
Pure-**Keras 3** conversion of [`openai-community/gpt2-medium`](https://huggingface.co/openai-community/gpt2-medium) for
[zeromodels](https://github.com/IMvision12/ZeroModels). One implementation runs unmodified on
**TensorFlow / Torch / JAX**.
## Quick start
```python
import os
os.environ["KERAS_BACKEND"] = "torch" # or "jax" / "tensorflow"
from zeromodels.models.gpt2 import GPT2TextGenerate, GPT2Tokenizer
model = GPT2TextGenerate.from_weights("zeromodels/gpt2_medium")
tokenizer = GPT2Tokenizer.from_weights("zeromodels/gpt2_medium")
inputs = tokenizer("The meaning of life is")
outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0]))
```
All GPT-2 sizes load the same way with `from_weights("zeromodels/<variant>")`:
| Variant | Hub | Params |
|---|---|---|
| `gpt2` | [`zeromodels/gpt2`](https://huggingface.co/zeromodels/gpt2) | 124M |
| `gpt2_medium` | [`zeromodels/gpt2_medium`](https://huggingface.co/zeromodels/gpt2_medium) | 355M |
| `gpt2_large` | [`zeromodels/gpt2_large`](https://huggingface.co/zeromodels/gpt2_large) | 774M |
| `gpt2_xl` | [`zeromodels/gpt2_xl`](https://huggingface.co/zeromodels/gpt2_xl) | 1.5B |
## Tips
- Set `KERAS_BACKEND` **before** importing Keras / zeromodels.
- This is a base completion model: it continues a prompt and is not
instruction-tuned.
- See the [GPT-2 docs](https://imvision12.github.io/ZeroModels/gpt2/) and [Loading Weights](https://imvision12.github.io/ZeroModels/loading_weights/).
- Upstream safetensors still work via the `hf:` prefix, e.g.
`GPT2TextGenerate.from_weights("hf:openai-community/gpt2-medium")`.
## Special Thanks
A huge thank you to the OpenAI GPT-2 authors for creating and releasing these models.
License: MIT, inherited from the upstream OpenAI GPT-2 release.
|