Commit
•
6f7d473
1
Parent(s):
3d7da74
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,35 @@
|
|
| 1 |
---
|
| 2 |
license: other
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: other
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
# Overview
|
| 6 |
+
Fine-tuned [Llama-2 7B](https://huggingface.co/TheBloke/Llama-2-7B-fp16) with an uncensored/unfiltered Wizard-Vicuna conversation dataset [ehartford/wizard_vicuna_70k_unfiltered](https://huggingface.co/datasets/ehartford/wizard_vicuna_70k_unfiltered).
|
| 7 |
+
Used QLoRA for fine-tuning. Trained for one epoch on a 24GB GPU (NVIDIA A10G) instance, took ~19 hours to train.
|
| 8 |
+
|
| 9 |
+
# Prompt style
|
| 10 |
+
The model was trained with the following prompt style:
|
| 11 |
+
```
|
| 12 |
+
### HUMAN:
|
| 13 |
+
Hello
|
| 14 |
+
|
| 15 |
+
### RESPONSE:
|
| 16 |
+
Hi, how are you?
|
| 17 |
+
|
| 18 |
+
### HUMAN:
|
| 19 |
+
I'm fine.
|
| 20 |
+
|
| 21 |
+
### RESPONSE:
|
| 22 |
+
How can I help you?
|
| 23 |
+
...
|
| 24 |
+
```
|
| 25 |
+
|
| 26 |
+
# Training code
|
| 27 |
+
Code used to train the model is available [here](https://github.com/georgesung/llm_qlora).
|
| 28 |
+
|
| 29 |
+
To reproduce the results:
|
| 30 |
+
```
|
| 31 |
+
git clone https://github.com/georgesung/llm_qlora
|
| 32 |
+
cd llm_qlora
|
| 33 |
+
pip install -r requirements.txt
|
| 34 |
+
python train.pyy configs/llama2_7b_chat_uncensored.yaml
|
| 35 |
+
```
|