Skip to content

Files

Latest commit

 

History

History
146 lines (106 loc) · 11.8 KB

Development.md

File metadata and controls

146 lines (106 loc) · 11.8 KB

Develop new Benchmark / MLLM

🛠️ How to implement a new Benchmark / VLM in VLMEvalKit?

Implement a new benchmark

Example PR: Math-Vision Benchmark (#292)

In VLMEvalKit, benchmarks are organized as dataset classes. When you try to implement a new benchmark, you can either reuse existing dataset classes (e.g., You can reuse ImageMCQDataset when implementing a new multi-choice benchmark), or support a new dataset class. Each dataset must have the following two member functions (either reuse the one of the parent class or implement your own):

  • build_prompt(self, line): The function input line is an integer (the sample index) or a pd.Series object (the raw record of the sample). The function outputs a multi-modal message, serving as the input of an MLLM. The multi-modal message is an interleaved list of multi-modal messages adopting the following format (the example includes an image and a text message): [dict(type='image', value=IMAGE_PTH), dict(type='text', value=prompt)].
  • evaluate(self, eval_file, **judge_kwargs): The function input eval_file is the MLLM prediction (typically in .xlsx format). If the benchmark requires an external LLM (typically GPT) for evaluation, then judge_kwargs can pass the arguments for the LLM. The function outputs the benchmark evaluation results (metrics) in the form of dict or pd.DataFrame.

We then brief the typical steps to implement a new benchmark under VLMEvalKit:

1. Prepare your benchmark tsv file

Currently, we organize a benchmark as one single TSV file. During inference, the data file will be automatically downloaded from the definited DATASET_URL link to $LMUData file (default path is $HOME/LMUData, if not set explicitly). You can upload the prepared TSV file to a downloadable address (e.g., Huggingface) or send it to us at opencompass@pjlab.org.cn. We will assist in uploading the dataset to the server. You can also customize LMUData path in the environment variable LMUData=/path/to/your/data.

The contents of the TSV file consist of:

Dataset Name \ Fields index image image_path question hint multi-choice
options
answer category l2-category split
MMBench_DEV_[CN/EN]
MMBench_TEST_[CN/EN]
CCBench
SEEDBench_IMG
MME
CORE_MM
MMVet
MMMU_DEV_VAL
COCO_VAL
OCRVQA_[TEST/TESTCORE]
TextVQA_VAL
VCR_[EN/ZH]_[EASY/HARD]_[ALL/500/100]
MMMB_[en/cn/pt/ar/tr/ru]
MMBench_dev_[en/cn/pt/ar/tr/ru]
Table 1. TSV fields of supported datasets.

Intro to mandatory fields in the TSV file:

  • index: Integer, Unique for each line in tsv
  • image: The base64 of the image, you can use APIs implemented in vlmeval/smp/vlm.py for encoding and decoding:
    • Encoding: encode_image_to_base64 (for PIL Image) / encode_image_file_to_base64 (for image file path)
    • Decoding: decode_base64_to_image(for PIL Image) / decode_base64_to_image_file (for image file path)
  • question: The question corresponding to the image, a string
  • answer: The answer to the question, a string. The test split does not need this field

2. Cutomize your benchmark prompt

ImageBaseDataset defines the default prompt format. If you need to add prompts specific to the dataset or input data in the Interleave format to the model, you can implement this through the build_prompt(line) function. This function takes a line from a TSV file as input, containing fields such as index, image, question, etc. The function returns a dictionary list of multimodal messages msg in the format [dict(type='image', value=IMAGE_PTH), dict(type='text', value=prompt)], including the image path and the text prompt to be input into VLMs. For interleave type inputs, you can directly place the dictionary of the image path at the image token position.

3. Cutomize your benchmark metrics

To add evaluation for a new benchmark, you need to customize a class object to implement the dataset’s metrics calculation. Multimodal datasets inherit from the ImageBaseDataset object in vlmeval/dataset/image_base.py. The TYPE defines the type of dataset, DATASET_URL is the download address of the dataset, and DATASET_MD5 is the MD5 checksum for consistency checking of the dataset file.

In this class, you need to implement the evaluate(eval_file, **judge_kwargs) class function to calculate metrics and output results for the custom dataset. The function input eval_file is the path to the model prediction results file {model_name}_{dataset}.xlsx. This file can be read as a pandas.DataFrame using the load(eval_file) method, containing fields such as index, question, answer, category, prediction, etc. The judge_kwargs will pass a dictionary related to evaluation, such as the name of the judge model, the number of API request threads, etc. The return value of the function is the calculated accuracy and other metrics, formatted as a dictionary composed of lists, organized into a pandas.DataFrame.

Implement a new model

Example PR: Support LLaVA-Next-Interleave (#294)

1. Support generate_inner API (mandatory).

All existing models are implemented in vlmeval/vlm. For a minimal model, your model class must implement the method generate_inner(msgs, dataset=None). In this function, you feed a multi-modal message to your VLM and return the VLM prediction (which is a string). The optional argument dataset can be used as the flag for the model to switch among various inference strategies.

The multi-modal messages msgs is a list of dictionaries, each dictionary has two keys: type and value:

  • type: We currently support two types, choices are ["image", "text"].
  • value: When type=='text' , the value is the text message (a single string); when type=='image', the value can be the local path of an image file, or the image URL.

Currently a multi-modal message may contain arbitrarily interleaved images and texts. If your model do not support that, a practice can be taking the 1st image and concatenated text messages as the input. You can set the INTERLEAVE = False in your model class and use self.message_to_promptimg(message, dataset=dataset) to build your prompt and the first image's path.

Here are some examples of multi-modal messages:

IMAGE_PTH = 'assets/apple.jpg'
IMAGE_URL = 'https://raw.githubusercontent.com/open-compass/VLMEvalKit/main/assets/apple.jpg'
msg1 = [
    dict(type='image', value=IMAGE_PTH),
    dict(type='text', value='What is in this image?')
]
msg2 = [
    dict(type='image', value=IMAGE_URL),
    dict(type='image', value=IMAGE_URL),
    dict(type='text', value='How many apples are there in these images?')
]
response = model.generate(msg1)

For convenience sake, we also support to take a list of string as inputs. In that case, we will check if a string is an image path or image URL and automatically convert it to the list[dict] format:

IMAGE_PTH = 'assets/apple.jpg'
IMAGE_URL = 'https://raw.githubusercontent.com/open-compass/VLMEvalKit/main/assets/apple.jpg'
msg1 = [IMAGE_PTH, 'What is in this image?']
msg2 = [IMAGE_URL, IMAGE_URL,  'How many apples are there in these images?']
response = model.generate(msg1)

Support Custom Prompt (optional).

Besides, your model can support custom prompt building by implementing two optional methods: use_custom_prompt(dataset) and build_prompt(line, dataset=None).

Both functions take the dataset name as the input:

  • use_custom_prompt(dataset) returns a boolean flag, indicating whether the model should use the custom prompt building strategy.
  • If use_custom_prompt(dataset) returns True, build_prompt(line, dataset) should return a customly bulit multimodal message for the corresponding dataset, given line, which is a dictionary that includes the necessary information of a data sample. If use_custom_prompt(dataset) returns False, the default prompt building strategy will be used.

Support multi-turn chatting (optional).

You can also support the multi-turn chatting and evaluation with your VLM by supporting the chat_inner(message, dataset) function. The function outputs a single string response, and the message is a list of chat history, following the below format.

# Assume msg1, msg2, msg3, ... are multi-modal messages following the previously described format
# `chat_inner` take the following chat history list as input:
message = [
    dict(role='user', content=msg1),
    dict(role='assistant', content=msg2),
    dict(role='user', content=msg3),
    dict(role='assistant', content=msg4),
	......
    dict(role='user', content=msgn),
]
# `message` should contain an odd number of chat utterances, the role of utterances should be interleaved "user" and "assistant", with the role of the last utterance to be "user".
# The chat function will call `chat_inner`
response = model.chat(message)

Example PRs:

Contribute to VLMEvalKit

If you want to contribute codes to VLMEvalKit, please do the pre-commit check before you submit a PR. That helps to keep the code tidy.

# Under the directory of VLMEvalKit, install the pre-commit hook:
pip install pre-commit
pre-commit install
pre-commit run --all-files
# Then you can commit your code.