From 4a190a98fd50e07fa011b17daf6a78ae1de15c4a Mon Sep 17 00:00:00 2001
From: oobabooga <112222186+oobabooga@users.noreply.github.com>
Date: Mon, 29 May 2023 14:56:05 -0300
Subject: [PATCH] Update GPTQ-models-(4-bit-mode).md

---
 docs/GPTQ-models-(4-bit-mode).md | 67 +++++++++++++++++++++++++++-----
 1 file changed, 57 insertions(+), 10 deletions(-)

diff --git a/docs/GPTQ-models-(4-bit-mode).md b/docs/GPTQ-models-(4-bit-mode).md
index 0ec28fa6..deb69555 100644
--- a/docs/GPTQ-models-(4-bit-mode).md
+++ b/docs/GPTQ-models-(4-bit-mode).md
@@ -1,12 +1,59 @@
 In 4-bit mode, models are loaded with just 25% of their regular VRAM usage. So LLaMA-7B fits into a 6GB GPU, and LLaMA-30B fits into a 24GB GPU.
 
-This is possible thanks to [@qwopqwop200](https://github.com/qwopqwop200/GPTQ-for-LLaMa)'s adaptation of the GPTQ algorithm for LLaMA: https://github.com/qwopqwop200/GPTQ-for-LLaMa
 
 GPTQ is a clever quantization algorithm that lightly reoptimizes the weights during quantization so that the accuracy loss is compensated relative to a round-to-nearest quantization. See the paper for more details: https://arxiv.org/abs/2210.17323
 
-## GPTQ-for-LLaMa branches
+## AutoGPTQ
 
-Different branches of GPTQ-for-LLaMa are available:
+AutoGPTQ is the recommended way to create new quantized models: https://github.com/PanQiWei/AutoGPTQ
+
+#### Installation
+
+To load a model quantized with AutoGPTQ in the web UI, manual installation is currently necessary:
+
+```
+conda activate textgen
+git clone https://github.com/PanQiWei/AutoGPTQ.git && cd AutoGPTQ
+pip install .
+```
+
+You are going to need to have `nvcc` installed (see the [instructions below](https://github.com/oobabooga/text-generation-webui/blob/main/docs/GPTQ-models-(4-bit-mode).md#step-0-install-nvcc)).
+
+#### Usage
+
+Place the output folder generated by AutoGPTQ in your `models/` folder and load it with the `--autogptq` flag:
+
+```
+python server.py --autogptq --model model_name
+```
+
+Alternatively, check the `autogptq` box in the "Model" tab of the UI before loading the model.
+
+#### Offloading
+
+In order to do CPU offloading or multi-cpu inference with AutoGPTQ, use the `--gpu-memory` flag. It is currently somewhat slower than offloading with the `--pre_layer` option in GPTQ-for-LLaMA.
+
+For CPU offloading:
+
+```
+python server.py --autogptq --gpu-memory 3000MiB --model model_name
+```
+
+For multi-GPU:
+
+```
+python server.py --autogptq --gpu-memory 3000MiB 6000MiB --model model_name
+```
+
+#### Applying LoRAs
+
+Not supported yet.
+
+## GPTQ-for-LLaMa
+
+GPTQ-for-LLaMa is the original adaptation of GPTQ for the LLaMA model. It was made by [@qwopqwop200](https://github.com/qwopqwop200/GPTQ-for-LLaMa) in this repository: https://github.com/qwopqwop200/GPTQ-for-LLaMa
+
+Different branches of GPTQ-for-LLaMa are currently available, including:
 
 | Branch | Comment |
 |----|----|
@@ -16,9 +63,9 @@ Different branches of GPTQ-for-LLaMa are available:
 
 Overall, I recommend using the old CUDA branch. It is included by default in the one-click-installer for this web UI.
 
-## Installation
+### Installation
 
-### Step 0: install nvcc
+#### Step 0: install nvcc
 
 ```
 conda activate textgen
@@ -29,7 +76,7 @@ The command above takes some 10 minutes to run and shows no progress bar or upda
 
 See this issue for more details: https://github.com/oobabooga/text-generation-webui/issues/416#issuecomment-1475078571
 
-### Step 1: install GPTQ-for-LLaMa
+#### Step 1: install GPTQ-for-LLaMa
 
 Clone the GPTQ-for-LLaMa repository into the `text-generation-webui/repositories` subfolder and install it:
 
@@ -64,14 +111,14 @@ git clone https://github.com/qwopqwop200/GPTQ-for-LLaMa.git -b triton
 
 https://github.com/qwopqwop200/GPTQ-for-LLaMa
 
-### Step 2: get the pre-converted weights
+#### Step 2: get the pre-converted weights
 
 * Converted without `group-size` (better for the 7b model): https://github.com/oobabooga/text-generation-webui/pull/530#issuecomment-1483891617
 * Converted with `group-size` (better from 13b upwards): https://github.com/oobabooga/text-generation-webui/pull/530#issuecomment-1483941105 
 
 ⚠️ The tokenizer files in the sources above may be outdated. Make sure to obtain the universal LLaMA tokenizer as described [here](https://github.com/oobabooga/text-generation-webui/blob/main/docs/LLaMA-model.md#option-1-pre-converted-weights).
 
-### Step 3: Start the web UI:
+#### Step 3: Start the web UI:
 
 For the models converted without `group-size`:
 
@@ -91,7 +138,7 @@ The command-line flags `--wbits` and `--groupsize` are automatically detected ba
 python server.py --model llama-13b-4bit-128g --wbits 4 --groupsize 128
 ```
 
-## CPU offloading
+### CPU offloading
 
 It is possible to offload part of the layers of the 4-bit model to the CPU with the `--pre_layer` flag. The higher the number after `--pre_layer`, the more layers will be allocated to the GPU.
 
@@ -109,7 +156,7 @@ Output generated in 123.79 seconds (1.61 tokens/s, 199 tokens)
 
 You can also use multiple GPUs with `pre_layer` if using the oobabooga fork of GPTQ, eg `--pre_layer 30 60` will load a LLaMA-30B model half onto your first GPU and half onto your second, or `--pre_layer 20 40` will load 20 layers onto GPU-0, 20 layers onto GPU-1, and 20 layers offloaded to CPU.
 
-## Using LoRAs in 4-bit mode
+### Using LoRAs with GPTQ-for-LLaMa
 
 At the moment, this feature is not officially supported by the relevant libraries, but a patch exists and is supported by this web UI: https://github.com/johnsmith0031/alpaca_lora_4bit