mmq

mmq project

This repo contains results, notebooks, and code related to quantizing blip2 with various configs. To get an idea of the main logic, look at the below diagram:

Installing Dependencies

Install dependencies for BLIP-2 tasks:

pip3 install -r requirements_blip2.txt

Install dependencies for LLAVA tasks:

pip3 install r requirements_llava.txt

IMPORTANT: The scoring part of this pipeline relies on the pycocoevalcap python submodule. To also clone this into the repo run git clone --recurse-submodules https://github.com/gautomdas/blip2-coco or if you already downloaded the repo and the pycocoevalcap folder is still empty, run git submodule init && git submodule update.

Installing Datasets

COCO:

python3 download_coco.py

Flickr30k (1K test set):

python3 download_flickr.py

VQAv2:

python3 download_vqav2.py

GQA:

python3 download_gqa.py

Running Evaluations

The run scripts generally follow this structure:

python3 run_[quantization_method].py --task <task_name> --config <path_to_config.json>

| Argument        | Type         | Default     | Description |
|-----------------|--------------|-------------|-------------|
| `--distributed` | flag         | False       | Whether to use distributed inference in a single node (default: False); only supported for image captioning and VQA tasks. 
| `--batch_size`  | int          | 64          | Batch size used during inference. 
| `--num_workers` | int          | 1           | Number of worker threads for dataset loading. 
| `--task`        | string       | —           | Task to run.
| `--config`      | string       | —           | Path to the quantization config JSON file. 
| `--max_samples` | int or None  | None        | If set, restricts evaluation to the first n samples of the dataset. 
| `--dataset_dir` | string       | `./data`    | Path to the dataset directory. 
| `--output_dir`  | string       | `./output`  | Directory where results will be saved.

Available Tasks

Uniform Quantization:

blip2-image_captioning
blip2-image_retrieval

GPTQ:

blip2-image_captioning
blip2-image_text_retrieval
blip2-vqav2
blip2-gqa
llava-vqav2
llava-gqa

AWQ:

blip2-image_captioning
blip2-image_text_retrieval
blip2-vqav2
blip2-gqa
llava-vqav2
llava-gqa

To Recreate the Demo File

Download the coco data set to the data folder using the following script (assumes you have the environment loaded): python download_coco.py
From there you should be able to run all of demo.ipynb
demo.ipynb goes over the 3 main steps in the diagram above

The following files are as follows:

blip_quantizer.py: The quantization class that quantizes a the blip2 model.
inference_pipeline.py: The inference class that takes a model and tasks to produce results/<#>.json.
scoring_pipeline.py: The scoring class used to convert results to scores based on task. This is separate from the inferencer/quantizer because it only requires the CPU to run.
quant_functions.py: Functions that are Tensor->Tensor and perform quantization.
utils.py: Additional utils used for config loading and model printing.
multi_sbatch.py: Runs the main.py script over many GPUs and different configs.

Notebooks

demo.ipynb: The above figure demonstrated in a ipynb
blip2_analysis.ipynb: Counting linear layers and params for the BLIP2 model
blip2_dropoff_coco.ipynb: A look at drop off between different quantizations over the whole model
dataset_usage.ipynb: A simple file showing how the COCO dataset (and others) are loaded

config_creator.ipynb: Create all combinations of configs based on:

for each bit width:
for each model part (ViT, LLM, QFormer):
  for each of the 8 combinations of front/middle/end:
    try with 2 other models quantized, not quantized, 1 of each, and 1 of each the other way

mmq