A implementação dos transformadores huggingface do Retriever multimodal de interação tardia de granulação fina.
A implementação oficial está aqui.
Os detalhes do modelo e pontos de verificação podem ser encontrados aqui.
Os detalhes para reproduzir os conjuntos de dados e avaliação no artigo podem ser encontrados aqui.
| Modelo | WIT Recall@10 | IGLUE Recall@1 | Rechamada KVQA@5 | Rechamada MSMARCO@5 | Rechamada do FORNO@5 | Rechamada LLaVA@1 | Rechamada de EVQA@5 | Pseudo Rechamada EVQA@5 | Rechamada OKVQA@5 | OKVQA Pseudo Rechamada@5 | Recuperação do Infoseek@5 | Pseudo-recall do Infoseek@5 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LinWeizheDragon/PreFLMR_ViT-G? | 0,619 | 0,718 | 0,419 | 0,783 | 0,643 | 0,726 | 0,625 | 0,721 | 0,302 | 0,674 | 0,392 | 0,577 |
| LinWeizheDragon/PreFLMR_ViT-L? | 0,605 | 0,699 | 0,440 | 0,779 | 0,608 | 0,729 | 0,609 | 0,708 | 0,314 | 0,690 | 0,374 | 0,578 |
| LinWeizheDragon/PreFLMR_ViT-B? | 0,427 | 0,574 | 0,294 | 0,786 | 0,468 | 0,673 | 0,550 | 0,663 | 0,272 | 0,658 | 0,260 | 0,496 |
Nota: Convertemos os pontos de verificação de PyTorch para transformadores Huggingface, cujos resultados de benchmark diferem ligeiramente dos números relatados no artigo original. Você pode reproduzir os resultados do artigo acima consultando as instruções deste documento.
Crie virtualenv:
conda create -n FLMR python=3.10 -y
conda activate FLMR
Instale o Pytorch:
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
Instale o faiss
conda install -c pytorch -c nvidia faiss-gpu=1.7.4 mkl=2021 blas=1.0=mkl
Teste se faiss gera erro
python -c "import faiss"
Instalar FLMR
git clone https://github.com/LinWeizheDragon/FLMR.git
cd FLMR
pip install -e .
Instale o motor ColBERT
cd third_party/ColBERT
pip install -e .
Instale outras dependências
pip install ujson gitpython easydict ninja datasets transformers
Carregar modelos pré-treinados
import os
import torch
import pandas as pd
import numpy as np
from torchvision . transforms import ToPILImage
from transformers import AutoImageProcessor
from flmr import index_custom_collection
from flmr import FLMRQueryEncoderTokenizer , FLMRContextEncoderTokenizer , FLMRModelForRetrieval
# load models
checkpoint_path = "LinWeizheDragon/PreFLMR_ViT-G"
image_processor_name = "laion/CLIP-ViT-bigG-14-laion2B-39B-b160k"
query_tokenizer = FLMRQueryEncoderTokenizer . from_pretrained ( checkpoint_path , subfolder = "query_tokenizer" )
context_tokenizer = FLMRContextEncoderTokenizer . from_pretrained (
checkpoint_path , subfolder = "context_tokenizer"
)
model = FLMRModelForRetrieval . from_pretrained (
checkpoint_path ,
query_tokenizer = query_tokenizer ,
context_tokenizer = context_tokenizer ,
)
image_processor = AutoImageProcessor . from_pretrained ( image_processor_name )Crie coleções de documentos
num_items = 100
feature_dim = 1664
passage_contents = [ f"This is test sentence { i } " for i in range ( num_items )]
# Option 1. text-only documents
custom_collection = passage_contents
# Option 2. multi-modal documents with pre-extracted image features
# passage_image_features = np.random.rand(num_items, feature_dim)
# custom_collection = [
# (passage_content, passage_image_feature, None) for passage_content, passage_image_feature in zip(passage_contents, passage_image_features)
# ]
# Option 3. multi-modal documents with images
# random_images = torch.randn(num_items, 3, 224, 224)
# to_img = ToPILImage()
# if not os.path.exists("./test_images"):
# os.makedirs("./test_images")
# for i, image in enumerate(random_images):
# image = to_img(image)
# image.save(os.path.join("./test_images", "{}.jpg".format(i)))
# image_paths = [os.path.join("./test_images", "{}.jpg".format(i)) for i in range(num_items)]
# custom_collection = [
# (passage_content, None, image_path)
# for passage_content, image_path in zip(passage_contents, image_paths)
# ]Execute a indexação na coleção personalizada
index_custom_collection (
custom_collection = custom_collection ,
model = model ,
index_root_path = "." ,
index_experiment_name = "test_experiment" ,
index_name = "test_index" ,
nbits = 8 , # number of bits in compression
doc_maxlen = 512 , # maximum allowed document length
overwrite = True , # whether to overwrite existing indices
use_gpu = False , # whether to enable GPU indexing
indexing_batch_size = 64 ,
model_temp_folder = "tmp" ,
nranks = 1 , # number of GPUs used in indexing
)Criar dados de consulta de brinquedos
num_queries = 2
query_instructions = [ f"instruction { i } " for i in range ( num_queries )]
query_texts = [ f" { query_instructions [ i ] } : query { i } " for i in range ( num_queries )]
query_images = torch . zeros ( num_queries , 3 , 224 , 224 )
query_encoding = query_tokenizer ( query_texts )
query_pixel_values = image_processor ( query_images , return_tensors = "pt" )[ 'pixel_values' ]Obtenha embeddings de consulta com modelo
inputs = dict (
input_ids = query_encoding [ 'input_ids' ],
attention_mask = query_encoding [ 'attention_mask' ],
pixel_values = query_pixel_values ,
)
# Run model query encoding
res = model . query ( ** inputs )
queries = { i : query_texts [ i ] for i in range ( num_queries )}
query_embeddings = res . late_interaction_outputPesquise a coleção
from flmr import search_custom_collection , create_searcher
# initiate a searcher
searcher = create_searcher (
index_root_path = "." ,
index_experiment_name = "test_experiment" ,
index_name = "test_index" ,
nbits = 8 , # number of bits in compression
use_gpu = True , # whether to enable GPU searching
)
# Search the custom collection
ranking = search_custom_collection (
searcher = searcher ,
queries = queries ,
query_embeddings = query_embeddings ,
num_document_to_retrieve = 5 , # how many documents to retrieve for each query
)
# Analyse retrieved documents
ranking_dict = ranking . todict ()
for i in range ( num_queries ):
print ( f"Query { i } retrieved documents:" )
retrieved_docs = ranking_dict [ i ]
retrieved_docs_indices = [ doc [ 0 ] for doc in retrieved_docs ]
retrieved_doc_scores = [ doc [ 2 ] for doc in retrieved_docs ]
retrieved_doc_texts = [ passage_contents [ doc_idx ] for doc_idx in retrieved_docs_indices ]
data = {
"Confidence" : retrieved_doc_scores ,
"Content" : retrieved_doc_texts ,
}
df = pd . DataFrame . from_dict ( data )
print ( df ) import torch
from flmr import FLMRQueryEncoderTokenizer , FLMRContextEncoderTokenizer , FLMRModelForRetrieval
checkpoint_path = "LinWeizheDragon/PreFLMR_ViT-L"
image_processor_name = "openai/clip-vit-large-patch14"
query_tokenizer = FLMRQueryEncoderTokenizer . from_pretrained ( checkpoint_path , subfolder = "query_tokenizer" )
context_tokenizer = FLMRContextEncoderTokenizer . from_pretrained ( checkpoint_path , subfolder = "context_tokenizer" )
model = FLMRModelForRetrieval . from_pretrained ( checkpoint_path ,
query_tokenizer = query_tokenizer ,
context_tokenizer = context_tokenizer ,
)
Q_encoding = query_tokenizer ([ "Using the provided image, obtain documents that address the subsequent question: What is the capital of France?" , "Extract documents linked to the question provided in conjunction with the image: What is the capital of China?" ])
D_encoding = context_tokenizer ([ "Paris is the capital of France." , "Beijing is the capital of China." ,
"Paris is the capital of France." , "Beijing is the capital of China." ])
Q_pixel_values = torch . zeros ( 2 , 3 , 224 , 224 )
inputs = dict (
query_input_ids = Q_encoding [ 'input_ids' ],
query_attention_mask = Q_encoding [ 'attention_mask' ],
query_pixel_values = Q_pixel_values ,
context_input_ids = D_encoding [ 'input_ids' ],
context_attention_mask = D_encoding [ 'attention_mask' ],
use_in_batch_negatives = True ,
)
res = model . forward ( ** inputs )
print ( res )Observe que os exemplos neste bloco de código são apenas para fins de demonstração. Eles mostram que o modelo pré-treinado dá pontuações mais altas para corrigir documentos. No treinamento real, você sempre precisa passar os documentos na ordem "doc positivo para consulta1, doc1 negativo para consulta1, doc2 negativo para consulta1, ..., documento positivo para consulta2, doc1 negativo para consulta2, doc2 negativo para consulta2, ...". Você pode querer ler a seção posterior que fornece um exemplo de script de ajuste fino.
pip install transformers
from transformers import AutoConfig , AutoModel , AutoImageProcessor , AutoTokenizer
import torch
checkpoint_path = "LinWeizheDragon/PreFLMR_ViT-L"
image_processor_name = "openai/clip-vit-large-patch14"
query_tokenizer = AutoTokenizer . from_pretrained ( checkpoint_path , subfolder = "query_tokenizer" , trust_remote_code = True )
context_tokenizer = AutoTokenizer . from_pretrained ( checkpoint_path , subfolder = "context_tokenizer" , trust_remote_code = True )
model = AutoModel . from_pretrained ( checkpoint_path ,
query_tokenizer = query_tokenizer ,
context_tokenizer = context_tokenizer ,
trust_remote_code = True ,
)
image_processor = AutoImageProcessor . from_pretrained ( image_processor_name )Fornecemos dois scripts para mostrar como os modelos pré-treinados podem ser usados na avaliação:
examples/example_use_flmr.py : um script de exemplo para avaliar FLMR (com 10 ROIs) em OK-VQA.examples/example_use_preflmr.py : um script de exemplo para avaliar PreFLMR no E-VQA. cd examples/ Baixe KBVQA_data aqui e descompacte as pastas de imagens. Os resultados de ROI/legenda/detecção de objetos foram incluídos.
Execute o seguinte comando (remova --run_indexing se você já executou a indexação uma vez):
python example_use_flmr.py
--use_gpu --run_indexing
--index_root_path " . "
--index_name OKVQA_GS
--experiment_name OKVQA_GS
--indexing_batch_size 64
--image_root_dir /path/to/KBVQA_data/ok-vqa/
--dataset_path BByrneLab/OKVQA_FLMR_preprocessed_data
--passage_dataset_path BByrneLab/OKVQA_FLMR_preprocessed_GoogleSearch_passages
--use_split test
--nbits 8
--Ks 1 5 10 20 50 100
--checkpoint_path LinWeizheDragon/FLMR
--image_processor_name openai/clip-vit-base-patch32
--query_batch_size 8
--num_ROIs 9 Você pode baixar as imagens E-VQA em https://github.com/google-research/google-research/tree/master/encyclopedic_vqa. Adicionaremos um link de conjunto de dados aqui em breve.
cd examples/ Execute o seguinte comando (remova --run_indexing se você já executou a indexação uma vez):
python example_use_preflmr.py
--use_gpu --run_indexing
--index_root_path " . "
--index_name EVQA_PreFLMR_ViT-G
--experiment_name EVQA
--indexing_batch_size 64
--image_root_dir /rds/project/rds-hirYTW1FQIw/shared_space/vqa_data/KBVQA_data/EVQA/eval_image/
--dataset_hf_path BByrneLab/multi_task_multi_modal_knowledge_retrieval_benchmark_M2KR
--dataset EVQA
--use_split test
--nbits 8
--Ks 1 5 10 20 50 100 500
--checkpoint_path LinWeizheDragon/PreFLMR_ViT-G
--image_processor_name laion/CLIP-ViT-bigG-14-laion2B-39B-b160k
--query_batch_size 8
--compute_pseudo_recall Aqui, carregamos todos os conjuntos de dados M2KR em um conjunto de dados HF BByrneLab/multi_task_multi_modal_knowledge_retrieval_benchmark_M2KR com diferentes conjuntos de dados como subconjunto. Para reproduzir os resultados dos outros conjuntos de dados no artigo, você pode alterar o --dataset para OKVQA , KVQA , LLaVA , OVEN , Infoseek , WIT , IGLUE e EVQA .
Atualizações :
--compute_pseudo_recall para calcular pseudo recall para conjuntos de dados como EVQA/OKVQA/Infoseek--Ks 1 5 10 20 50 100 500 : max(Ks) precisa ser 500 para corresponder ao desempenho relatado no documento PreFLMR. Altere os caminhos raiz da imagem em examples/evaluate_all.sh e execute:
cd examples
bash evaluate_all.shObtenha o relatório por:
python report.pyVocê precisará instalar o pytorch-lightning:
pip install pytorch-lightning==2.1.0
python example_finetune_preflmr.py
--image_root_dir /path/to/EVQA/images/
--dataset_hf_path BByrneLab/multi_task_multi_modal_knowledge_retrieval_benchmark_M2KR
--dataset EVQA
--freeze_vit
--log_with_wandb
--model_save_path saved_models
--checkpoint_path LinWeizheDragon/PreFLMR_ViT-G
--image_processor_name laion/CLIP-ViT-bigG-14-laion2B-39B-b160k
--batch_size 8
--accumulate_grad_batches 8
--valid_batch_size 16
--test_batch_size 64
--mode train
--max_epochs 99999999
--learning_rate 0.000005
--warmup_steps 100
--accelerator auto
--devices auto
--strategy ddp_find_unused_parameters_true
--num_sanity_val_steps 2
--precision bf16
--val_check_interval 2000
--save_top_k -1 python example_use_preflmr.py
--use_gpu --run_indexing
--index_root_path " . "
--index_name EVQA_PreFLMR_ViT-G_finetuned_model_step_10156
--experiment_name EVQA
--indexing_batch_size 64
--image_root_dir /path/to/EVQA/images/
--dataset_hf_path BByrneLab/multi_task_multi_modal_knowledge_retrieval_benchmark_M2KR
--dataset EVQA
--use_split test
--nbits 8
--num_gpus 1
--Ks 1 5 10 20 50 100 500
--checkpoint_path saved_models/model_step_10156
--image_processor_name laion/CLIP-ViT-bigG-14-laion2B-39B-b160k
--query_batch_size 8 Ao executar o script acima, podemos obter o seguinte desempenho de ajuste fino:
| Etapa | Pseudo Recall@5 no EVQA |
|---|---|
| 2500 | 73,6 |
| 10.000 | 73,55 |
| 12.000 | 74,21 |
| 14.000 | 73,73 |
(Pontos de verificação com baixas perdas de validação foram escolhidos e testados, executados em 2 GPUs A100)
O modelo FLMR é implementado seguindo o estilo de documentação dos transformers . Você pode encontrar documentação detalhada nos arquivos de modelagem.
Se nosso trabalho ajudou sua pesquisa, por favor, cite nosso artigo para FLMR e PreFLMR.
@inproceedings{
lin2023finegrained,
title={Fine-grained Late-interaction Multi-modal Retrieval for Retrieval Augmented Visual Question Answering},
author={Weizhe Lin and Jinghong Chen and Jingbiao Mei and Alexandru Coca and Bill Byrne},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=IWWWulAX7g}
}
@inproceedings{lin-etal-2024-preflmr,
title = "{P}re{FLMR}: Scaling Up Fine-Grained Late-Interaction Multi-modal Retrievers",
author = "Lin, Weizhe and
Mei, Jingbiao and
Chen, Jinghong and
Byrne, Bill",
editor = "Ku, Lun-Wei and
Martins, Andre and
Srikumar, Vivek",
booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = aug,
year = "2024",
address = "Bangkok, Thailand",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.acl-long.289",
pages = "5294--5316",
abstract = "Large Multimodal Models (LMMs) excel in natural language and visual understanding but are challenged by exacting tasks such as Knowledge-based Visual Question Answering (KB-VQA) which involve the retrieval of relevant information from document collections to use in shaping answers to questions. We present an extensive training and evaluation framework, M2KR, for KB-VQA. M2KR contains a collection of vision and language tasks which we have incorporated into a single suite of benchmark tasks for training and evaluating general-purpose multi-modal retrievers. We use M2KR to develop PreFLMR, a pre-trained version of the recently developed Fine-grained Late-interaction Multi-modal Retriever (FLMR) approach to KB-VQA, and we report new state-of-the-art results across a range of tasks. We also present investigations into the scaling behaviors of PreFLMR intended to be useful in future developments in general-purpose multi-modal retrievers.",
}
