File size: 3,410 Bytes
c6e456a
b926638
023a3de
c6e456a
85afc56
 
 
 
 
 
 
 
 
 
ccf4998
 
85afc56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b926638
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
---
license: gpl
inference: false
---

# gpt4-x-vicuna-13B-GPTQ

This repo contains 4bit GPTQ format quantised models of [NousResearch's gpt4-x-vicuna-13b](https://huggingface.co/NousResearch/gpt4-x-vicuna-13b).

It is the result of quantising to 4bit using [GPTQ-for-LLaMa](https://github.com/qwopqwop200/GPTQ-for-LLaMa).

## Repositories available

* [4bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/gpt4-x-vicuna-13B-GPTQ).
* [4bit and 5bit GGML models for CPU inference](https://huggingface.co/TheBloke/gpt4-x-vicuna-13B-GGML).
* [float16 model in HF format for GPU inference](https://huggingface.co/TheBloke/gpt4-x-vicuna-13B-HF).
 
## How to easily download and use this model in text-generation-webui

Open the text-generation-webui UI as normal.

1. Click the **Model tab**.
2. Under **Download custom model or LoRA**, enter `TheBloke/gpt4-x-vicuna-13B-GPTQ`.
3. Click **Download**.
4. Wait until it says it's finished downloading.
5. Click the **Refresh** icon next to **Model** in the top left.
6. In the **Model drop-down**: choose the model you just downloaded, `gpt4-x-vicuna-13B-GPTQ`.
7. If you see an error in the bottom right, ignore it - it's temporary.
8. Fill out the `GPTQ parameters` on the right: `Bits = 4`, `Groupsize = 128`, `model_type = Llama`
9. Click **Save settings for this model** in the top right.
10. Click **Reload the Model** in the top right.
11. Once it says it's loaded, click the **Text Generation tab** and enter a prompt!

## Provided files

**Compatible file - GPT4-x-Vicuna-13B-GPTQ-4bit-128g.compat.act-order.safetensors**

In the `main` branch - the default one - you will find `GPT4-x-Vicuna-13B-GPTQ-4bit-128g.compat.act-order.safetensors`

This will work with all versions of GPTQ-for-LLaMa. It has maximum compatibility

It was created without the `--act-order` parameter. It may have slightly lower inference quality compared to the other file, but is guaranteed to work on all versions of GPTQ-for-LLaMa and text-generation-webui.

* `GPT4-x-Vicuna-13B-GPTQ-4bit-128g.compat.act-order.safetensors`
  * Works with all versions of GPTQ-for-LLaMa code, both Triton and CUDA branches
  * Works with text-generation-webui one-click-installers
  * Parameters: Groupsize = 128g. No act-order.
  * Command used to create the GPTQ:
    ```
    CUDA_VISIBLE_DEVICES=0 python3 llama.py GPT4All-13B-snoozy c4 --wbits 4 --true-sequential --groupsize 128 --save_safetensors GPT4-x-Vicuna-13B-GPTQ-4bit-128g.compat.act-order.safetensors
    ```

# Original model card

As a base model used https://huggingface.co/eachadea/vicuna-13b-1.1

Finetuned on Teknium's GPTeacher dataset, unreleased Roleplay v2 dataset, GPT-4-LLM dataset, and Nous Research Instruct Dataset

Approx 180k instructions, all from GPT-4, all cleaned of any OpenAI censorship/"As an AI Language Model" etc.

Base model still has OpenAI censorship. Soon, a new version will be released with cleaned vicuna from https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltere

Trained on 8 A100-80GB GPUs for 5 epochs following Alpaca deepspeed training code.

Nous Research Instruct Dataset will be released soon.

GPTeacher, Roleplay v2 by https://huggingface.co/teknium

Wizard LM by https://github.com/nlpxucan

Nous Research Instruct Dataset by https://huggingface.co/karan4d and https://huggingface.co/huemin

Compute provided by our project sponsor https://redmond.ai/