xgen-7b-8k-base / README.md
tomaarsen's picture
tomaarsen HF staff
Resolve broken URL in citation
09bcd3a
|
raw
history blame
1.76 kB
---
license: apache-2.0
---
# XGen-7B-8K-Base
Official research release for the family of **XGen** models (`7B`) by Salesforce AI Research:
*Title*: [Long Sequence Modeling with XGen: A 7B LLM Trained on 8K Input Sequence Length](https://blog.salesforceairesearch.com/xgen/)
## Models
### Base models
* [XGen-7B-4K-Base](https://huggingface.co/Salesforce/xgen-7b-4k-base): XGen-7B model pre-trained under 4K sequence length.
* License: Apache-2.0
* [XGen-7B-8K-Base](https://huggingface.co/Salesforce/xgen-7b-8k-base): XGen-7B model pre-trained under 8K sequence length.
* License: Apache-2.0
### Instruction-finetuned models
Supervised finetuned model on public domain instructional data. Released for ***research purpose*** only.
* [XGen-7B-8K-Inst](https://huggingface.co/Salesforce/xgen-7b-8k-inst)
## How to run
The training data for the models are tokenized with OpenAI Tiktoken library.
To use this model, install the package via `pip`:
```sh
pip install tiktoken
```
The models can be used as auto-regressive samplers as follows:
```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("Salesforce/xgen-7b-8k-base", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("Salesforce/xgen-7b-8k-base", torch_dtype=torch.bfloat16)
inputs = tokenizer("The world is", return_tensors="pt")
sample = model.generate(**inputs, max_length=128)
print(tokenizer.decode(sample[0]))
```
## Citation
```bibtex
@misc{XGen,
title={Long Sequence Modeling with XGen: A 7B LLM Trained on 8K Input Sequence Length},
author={Salesforce AI Research},
howpublished={Salesforce AI Research Blog},
year={2023},
url={https://blog.salesforceairesearch.com/xgen/}
}
```