Datasets:

Modalities:
Image
Text
Formats:
parquet
Languages:
Chinese
ArXiv:
Libraries:
Datasets
Dask
License:
tianyu-z commited on
Commit
326c95d
1 Parent(s): 460b3e8
Files changed (1) hide show
  1. README.md +11 -7
README.md CHANGED
@@ -74,9 +74,10 @@ We support open-source model_id:
74
  "THUDM/cogvlm2-llama3-chat-19B",
75
  "echo840/Monkey-Chat",]
76
  ```
77
- For the models not on list, they are not intergated with huggingface, please refer to their github repo to create the evaluation pipeline.
78
 
79
  ```bash
 
80
  # We use HuggingFaceM4/idefics2-8b and vcr_wiki_en_easy as an example
81
  # Inference from the VLMs and save the results to {model_id}_{difficulty}_{language}.json
82
  cd src/evaluation
@@ -90,19 +91,23 @@ python3 gather_results.py --jsons_path .
90
  ```
91
 
92
  ### Close-source evaluation
93
- We provide the evaluation script for the close-source model: `GPT-4o`, `GPT-4-Turbo`, `Claude-3-Opus` in the `evaluation` folder.
94
 
95
  You need an API Key, a pre-saved testing dataset and specify the path of the data saving the paper
96
  ```bash
 
97
  cd src/evaluation
98
- # save the testing dataset to the path
99
  python3 save_image_from_dataset.py --output_path .
 
 
 
100
 
101
- # Inference Put your API key and Image Path in the evaluation script (e.g. gpt-4o.py)
102
- python3 gpt-4o.py
103
 
104
  # Evaluate the results and save the evaluation metrics to {model_id}_{difficulty}_{language}_evaluation_result.json
105
- python3 evaluation_metrics.py --model_id gpt4o --output_path . --json_filename "gpt4o_en_easy.json" --dataset_handler "vcr-org/VCR-wiki-en-easy-test"
106
 
107
  # To get the mean score of all the `{model_id}_{difficulty}_{language}_evaluation_result.json` in `jsons_path` (and the std, confidence interval if `--bootstrap`) of the evaluation metrics
108
  python3 gather_results.py --jsons_path .
@@ -115,7 +120,6 @@ pip install git+https://github.com/EvolvingLMMs-Lab/lmms-eval.git
115
  # We use HuggingFaceM4/idefics2-8b and vcr_wiki_en_easy as an example
116
  python3 -m accelerate.commands.launch --num_processes=8 -m lmms_eval --model idefics2 --model_args pretrained="HuggingFaceM4/idefics2-8b" --tasks vcr_wiki_en_easy --batch_size 1 --log_samples --log_samples_suffix HuggingFaceM4_idefics2-8b_vcr_wiki_en_easy --output_path ./logs/
117
  ```
118
-
119
  `lmms-eval` supports the following VCR `--tasks` settings:
120
 
121
  * English
 
74
  "THUDM/cogvlm2-llama3-chat-19B",
75
  "echo840/Monkey-Chat",]
76
  ```
77
+ For the models not on list, they are not intergated with huggingface, please refer to their github repo to create the evaluation pipeline. Examples of the inference logic are in `src/evaluation/inference.py`
78
 
79
  ```bash
80
+ pip install -r requirements.txt
81
  # We use HuggingFaceM4/idefics2-8b and vcr_wiki_en_easy as an example
82
  # Inference from the VLMs and save the results to {model_id}_{difficulty}_{language}.json
83
  cd src/evaluation
 
91
  ```
92
 
93
  ### Close-source evaluation
94
+ We provide the evaluation script for the close-source models in `src/evaluation/closed_source_eval.py`.
95
 
96
  You need an API Key, a pre-saved testing dataset and specify the path of the data saving the paper
97
  ```bash
98
+ pip install -r requirements.txt
99
  cd src/evaluation
100
+ # [download images to inference locally option 1] save the testing dataset to the path using script from huggingface
101
  python3 save_image_from_dataset.py --output_path .
102
+ # [download images to inference locally option 2] save the testing dataset to the path using github repo
103
+ # use en-easy-test-500 as an example
104
+ git clone https://github.com/tianyu-z/VCR-wiki-en-easy-test-500.git
105
 
106
+ # specify your image path if you would like to inference using the image stored locally by --image_path "path_to_image", otherwise, the script will streaming the images from github repo
107
+ python3 closed_source_eval.py --model_id gpt4o --dataset_handler "VCR-wiki-en-easy-test-500" --api_key "Your_API_Key"
108
 
109
  # Evaluate the results and save the evaluation metrics to {model_id}_{difficulty}_{language}_evaluation_result.json
110
+ python3 evaluation_metrics.py --model_id gpt4o --output_path . --json_filename "gpt4o_en_easy.json" --dataset_handler "vcr-org/VCR-wiki-en-easy-test-500"
111
 
112
  # To get the mean score of all the `{model_id}_{difficulty}_{language}_evaluation_result.json` in `jsons_path` (and the std, confidence interval if `--bootstrap`) of the evaluation metrics
113
  python3 gather_results.py --jsons_path .
 
120
  # We use HuggingFaceM4/idefics2-8b and vcr_wiki_en_easy as an example
121
  python3 -m accelerate.commands.launch --num_processes=8 -m lmms_eval --model idefics2 --model_args pretrained="HuggingFaceM4/idefics2-8b" --tasks vcr_wiki_en_easy --batch_size 1 --log_samples --log_samples_suffix HuggingFaceM4_idefics2-8b_vcr_wiki_en_easy --output_path ./logs/
122
  ```
 
123
  `lmms-eval` supports the following VCR `--tasks` settings:
124
 
125
  * English