---
license: apache-2.0
datasets:
- timdettmers/openassistant-guanaco
pipeline_tag: text-generation
---
Model that is fine-tuned in 4-bit precision using QLoRA on [timdettmers/openassistant-guanaco](https://huggingface.co/datasets/timdettmers/openassistant-guanaco) and sharded to be used on a free Google Colab instance that can be loaded with 4bits. 

It can be easily imported using the `AutoModelForCausalLM` class from `transformers`:

```
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
          "guardrail/llama-2-7b-guanaco-instruct-sharded",
          load_in_4bit=True)
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)

```
# [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_guardrail__llama-2-7b-guanaco-instruct-sharded)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 44.87   |
| ARC (25-shot)         | 53.75          |
| HellaSwag (10-shot)   | 78.69    |
| MMLU (5-shot)         | 46.65         |
| TruthfulQA (0-shot)   | 43.93   |
| Winogrande (5-shot)   | 72.61   |
| GSM8K (5-shot)        | 7.81        |
| DROP (3-shot)         | 10.65         |