LoneStriker's picture
ExLLaMA V2 quant of Yarn-Mistral-7b-64k-6.0bpw-h6-exl2
7fb2dab
---
datasets:
- emozilla/yarn-train-tokenized-16k-mistral
metrics:
- perplexity
library_name: transformers
license: apache-2.0
language:
- en
---
# Model Card: Nous-Yarn-Mistral-7b-64k
[Preprint (arXiv)](https://arxiv.org/abs/2309.00071)
[GitHub](https://github.com/jquesnelle/yarn)
![yarn](https://raw.githubusercontent.com/jquesnelle/yarn/mistral/data/proofpile-long-small-mistral.csv.png)
## Model Description
Nous-Yarn-Mistral-7b-64k is a state-of-the-art language model for long context, further pretrained on long context data for 1000 steps using the YaRN extension method.
It is an extension of [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) and supports a 64k token context window.
To use, pass `trust_remote_code=True` when loading the model, for example
```python
model = AutoModelForCausalLM.from_pretrained("NousResearch/Yarn-Mistral-7b-64k",
use_flash_attention_2=True,
torch_dtype=torch.bfloat16,
device_map="auto",
trust_remote_code=True)
```
In addition you will need to use the latest version of `transformers` (until 4.35 comes out)
```sh
pip install git+https://github.com/huggingface/transformers
```
## Benchmarks
Long context benchmarks:
| Model | Context Window | 8k PPL | 16k PPL | 32k PPL | 64k PPL | 128k PPL |
|-------|---------------:|------:|----------:|-----:|-----:|------------:|
| [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) | 8k | 2.96 | - | - | - | - |
| [Yarn-Mistral-7b-64k](https://huggingface.co/NousResearch/Yarn-Mistral-7b-64k) | 64k | 3.04 | 2.65 | 2.44 | 2.20 | - |
| [Yarn-Mistral-7b-128k](https://huggingface.co/NousResearch/Yarn-Mistral-7b-128k) | 128k | 3.08 | 2.68 | 2.47 | 2.24 | 2.19 |
Short context benchmarks showing that quality degradation is minimal:
| Model | Context Window | ARC-c | Hellaswag | MMLU | Truthful QA |
|-------|---------------:|------:|----------:|-----:|------------:|
| [Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) | 8k | 59.98 | 83.31 | 64.16 | 42.15 |
| [Yarn-Mistral-7b-64k](https://huggingface.co/NousResearch/Yarn-Mistral-7b-64k) | 64k | 59.38 | 81.21 | 61.32 | 42.50 |
| [Yarn-Mistral-7b-128k](https://huggingface.co/NousResearch/Yarn-Mistral-7b-128k) | 128k | 58.87 | 80.58 | 60.64 | 42.46 |
## Collaborators
- [bloc97](https://github.com/bloc97): Methods, paper and evals
- [@theemozilla](https://twitter.com/theemozilla): Methods, paper, model training, and evals
- [@EnricoShippole](https://twitter.com/EnricoShippole): Model training
- [honglu2875](https://github.com/honglu2875): Paper and evals
The authors would like to thank LAION AI for their support of compute for this model.
It was trained on the [JUWELS](https://www.fz-juelich.de/en/ias/jsc/systems/supercomputers/juwels) supercomputer.