README.md

---
pipeline_tag: text-generation
inference: true
widget:
- text: 'def factorial(n):'
  example_title: Factorial
  group: Python
- text: 'def recur_fibo(n):'
  example_title: Recursive Fibonacci
  group: Python
license: llama2
library_name: transformers
tags:
- text-generation
- code
language:
- en
---

# lemur-70b-v1

<p align="center">
  <img src="https://huggingface.co/datasets/OpenLemur/assets/resolve/main/lemur_icon.png" width="300" height="300" alt="Lemur">
</p>


<div align="center">
  <img src="https://huggingface.co/datasets/OpenLemur/assets/resolve/main/lemur_base_radar.png">
</div>

📄Paper: https://arxiv.org/abs/2310.06830

👩‍💻Code: https://github.com/OpenLemur/Lemur

## Use

### Setup

First, we have to install all the libraries listed in `requirements.txt` in [GitHub](https://github.com/OpenLemur/lemur-v1):

```bash
pip install -r requirements.txt
```

### Intended use

Since it is not trained on instruction following corpus, it won't respond well to questions like "What is the Python code to do quick sort?". 

### Generation

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OpenLemur/lemur-70b-v1")
model = AutoModelForCausalLM.from_pretrained("OpenLemur/lemur-70b-v1", device_map="auto", load_in_8bit=True)

# Text Generation Example
prompt = "The world is "
input = tokenizer(prompt, return_tensors="pt")
output = model.generate(**input, max_length=50, num_return_sequences=1)
generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
print(generated_text)

# Code Generation Example
prompt = """
def factorial(n):
    if n == 0:
        return 1
"""
input = tokenizer(prompt, return_tensors="pt")
output = model.generate(**input, max_length=200, num_return_sequences=1)
generated_code = tokenizer.decode(output[0], skip_special_tokens=True)
print(generated_code)
```

# License
The model is licensed under the Llama-2 community license agreement.

# Acknowledgements
The Lemur project is an open collaborative research effort between [XLang Lab](https://www.xlang.ai/) and Salesforce Research. We thank Salesforce, Google Research and Amazon AWS for their gift support.
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_OpenLemur__lemur-70b-v1)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 54.03   |
| ARC (25-shot)         | 64.33          |
| HellaSwag (10-shot)   | 85.72    |
| MMLU (5-shot)         | 65.85         |
| TruthfulQA (0-shot)   | 44.78   |
| Winogrande (5-shot)   | 83.03   |
| GSM8K (5-shot)        | 28.73        |
| DROP (3-shot)         | 5.74         |