---
license: other
inference: false
---

# OpenAssistant LLaMA 30B SFT 7 GPTQ

This in a repo of GPTQ format 4bit quantised models for [OpenAssistant's LLaMA 30B SFT 7](https://huggingface.co/OpenAssistant/oasst-sft-7-llama-30b-xor).

It is the result of merging the XORs from the above repo with the original Llama 30B weights, and then quantising to 4bit GPU inference using [GPTQ-for-LLaMa](https://github.com/qwopqwop200/GPTQ-for-LLaMa).

This is epoch 7 of OpenAssistant's training of their Llama 30B model.

**Please note that these models will need 24GB VRAM or greater to use effectively**

## Repositories available

* [4bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-GPTQ).
* [4bit and 5bit GGML models for CPU inference](https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-GGML).
* [Unquantised 16bit model in HF format](https://huggingface.co/TheBloke/OpenAssistant-SFT-7-Llama-30B-HF).

## PROMPT TEMPLATE

This model requires the following prompt template:

```
<|prompter|> prompt goes here
<|assistant|>:
```

## CHOICE OF MODELS

Two sets of models are provided:

* Groupsize = 1024
  * Should work reliably in 24GB VRAM
* Groupsize = 128
  * Optimal setting for highest inference quality
  * But may require more than 24GB VRAM, depending on response length
  * In my testing it ran out of VRAM on a 24GB card around 1500 tokens returned.

For each model, two versions are available:
* `compat.no-act-order.safetensor`
  * Works with all versions of GPTQ-for-LLaMa, including the version in text-generation-webui one-click-installers
* `latest.act-order.safetensors`
  * uses `--act-order` for higher inference quality
  * requires more recent GPTQ-for-LLaMa code, therefore will not currently work with one-click-installers
 
## HOW TO CHOOSE YOUR MODEL

I have used branches to separate the models. This means you can clone the branch you want and not got model files you don't need.

* Branch: **main** = groupsize 1024, `compat.no-act-order.safetensor` file
* Branch: **1024-latest** = groupsize 1024, `latest.no-act-order.safetensor` file
* Branch: **128-compat** = groupsize 128, `compat.no-act-order.safetensor` file
* Branch: **128-latest** = groupsize 128, `latest.no-act-order.safetensor` file

![branches](https://i.imgur.com/PdiHnLxm.png)
  
## How to easily download and run the 1024g compat 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/OpenAssistant-SFT-7-Llama-30B-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, `OpenAssistant-SFT-7-Llama-30B-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 = 1024`, `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!
 
## Manual instructions for `text-generation-webui`

The `compat.no-act-order.safetensors` files can be loaded the same as any other GPTQ file, without requiring any updates to [oobaboogas text-generation-webui](https://github.com/oobabooga/text-generation-webui).

[Instructions on using GPTQ 4bit files in text-generation-webui are here](https://github.com/oobabooga/text-generation-webui/wiki/GPTQ-models-\(4-bit-mode\)).

The `latest.act-order.safetensors` files were created using `--act-order` to give the maximum possible quantisation quality, but this means it requires that the latest GPTQ-for-LLaMa is used inside the UI.

If you want to use the act-order `safetensors` files and need to update the Triton branch of GPTQ-for-LLaMa, here are the commands I used to clone the Triton branch of GPTQ-for-LLaMa, clone text-generation-webui, and install GPTQ into the UI:
```
# Clone text-generation-webui, if you don't already have it
git clone https://github.com/oobabooga/text-generation-webui
# Make a repositories directory
mkdir text-generation-webui/repositories
cd text-generation-webui/repositories
# Clone the latest GPTQ-for-LLaMa code inside text-generation-webui
git clone https://github.com/qwopqwop200/GPTQ-for-LLaMa
```

Then install this model into `text-generation-webui/models` and launch the UI as follows:
```
cd text-generation-webui
python server.py --model OpenAssistant-SFT-7-Llama-30B-GPTQ --wbits 4 --groupsize 128 --model_type Llama # add any other command line args you want
```

To update the CUDA branch of GPTQ-for-LLaMa, you can do the following. **This requires a C/C++ compiler and the CUDA toolkit installed!**
```
# Clone text-generation-webui, if you don't already have it
git clone https://github.com/oobabooga/text-generation-webui
# Make a repositories directory
mkdir text-generation-webui/repositories
cd text-generation-webui/repositories
# Clone the latest GPTQ-for-LLaMa code inside text-generation-webui
git clone -b cuda https://github.com/qwopqwop200/GPTQ-for-LLaMa
cd GPTQ-for-LLaMa
pip uninstall quant-cuda # uninstall existing CUDA version
python setup_cuda.py install # install latest version
```

The above commands assume you have installed all dependencies for GPTQ-for-LLaMa and text-generation-webui. Please see their respective repositories for further information.

If you can't update GPTQ-for-LLaMa or don't want to, please use a `compat.no-act-order.safetensor` file.

# Original model card

```
llama-30b-sft-7:
  dtype: fp16
  log_dir: "llama_log_30b"
  learning_rate: 1e-5
  model_name: /home/ubuntu/Open-Assistant/model/model_training/.saved/llama-30b-super-pretrain/checkpoint-3500
  #model_name: OpenAssistant/llama-30b-super-pretrain
  output_dir: llama_model_30b
  deepspeed_config: configs/zero3_config_sft.json
  weight_decay: 0.0
  residual_dropout: 0.0
  max_length: 2048
  use_flash_attention: true
  warmup_steps: 20
  gradient_checkpointing: true
  gradient_accumulation_steps: 12
  per_device_train_batch_size: 2
  per_device_eval_batch_size: 3
  eval_steps: 101
  save_steps: 485
  num_train_epochs: 4
  save_total_limit: 3
  use_custom_sampler: true
  sort_by_length: false
  #save_strategy: steps
  save_strategy: epoch
  datasets:
    - oasst_export:
        lang: "bg,ca,cs,da,de,en,es,fr,hr,hu,it,nl,pl,pt,ro,ru,sl,sr,sv,uk"
        input_file_path: 2023-04-12_oasst_release_ready_synth.jsonl.gz
        val_split: 0.05
    - vicuna:
        val_split: 0.05
        max_val_set: 800
        fraction: 1.0
    - dolly15k:
        val_split: 0.05
        max_val_set: 300
    - grade_school_math_instructions:
        val_split: 0.05
    - code_alpaca:
        val_split: 0.05
        max_val_set: 250
```

- **OASST dataset paper:** https://arxiv.org/abs/2304.07327