Overview
Fine-tuned Llama-2 7B with an uncensored/unfiltered Wizard-Vicuna conversation dataset (originally from ehartford/wizard_vicuna_70k_unfiltered). Used QLoRA for fine-tuning. Trained for one epoch on a 24GB GPU (NVIDIA A10G) instance, took ~19 hours to train.
The version here is the fp16 HuggingFace model.
GGML & GPTQ versions
Thanks to TheBloke, he has created the GGML and GPTQ versions:
- https://huggingface.co/TheBloke/llama2_7b_chat_uncensored-GGML
- https://huggingface.co/TheBloke/llama2_7b_chat_uncensored-GPTQ
Running in Ollama
https://ollama.com/library/llama2-uncensored
Prompt style
The model was trained with the following prompt style:
### HUMAN:
Hello
### RESPONSE:
Hi, how are you?
### HUMAN:
I'm fine.
### RESPONSE:
How can I help you?
...
Training code
Code used to train the model is available here.
To reproduce the results:
git clone https://github.com/georgesung/llm_qlora
cd llm_qlora
pip install -r requirements.txt
python train.py configs/llama2_7b_chat_uncensored.yaml
Fine-tuning guide
https://georgesung.github.io/ai/qlora-ift/
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 43.39 |
| ARC (25-shot) | 53.58 |
| HellaSwag (10-shot) | 78.66 |
| MMLU (5-shot) | 44.49 |
| TruthfulQA (0-shot) | 41.34 |
| Winogrande (5-shot) | 74.11 |
| GSM8K (5-shot) | 5.84 |
| DROP (3-shot) | 5.69 |
- Downloads last month
- 2,735
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.