Update README.md

open_flamingo/eval/README.md

@@ -44,7 +44,7 @@ print(checkpoint_path)
 
 This should place the OpenFlamingo model at the expected location in the evaluation script.
 
-For TextVQA and VizWiz we expect annotations to be formatted differently than the original datasets. We provide the custom annotations in `open_flamingo/open_flamingo/eval/data/`.
+For TextVQA and VizWiz we expect annotations to be formatted differently than the original datasets. We provide the custom annotations in `open_flamingo/open_flamingo/eval/data/`. We have also uploaded all the annotation files in a [huggingface dataset](https://huggingface.co/datasets/openflamingo/eval_benchmark/tree/main) for easy access.
 
 # Evaluating using RICES (Retrieval-based In-Context Example Selection)
 
@@ -65,4 +65,4 @@ python cache_rices_features.py \
  --output_dir /path/to/coco/features 
 ```
 
-This will create a directory at `/path/to/coco/features` containing a file named `coco.pkl` with the extracted features. You can then use this directory to evaluate using RICES by passing the `--rices` flag to the evaluation script, specifying the path to the features directory using the `--cached_demonstration_features` flag, and specifying the vision encoder to use for RICES using the `--rices_vision_encoder_path` and `--rices_vision_encoder_pretrained` flags.
+This will create a directory at `/path/to/coco/features` containing a file named `coco.pkl` with the extracted features. You can then use this directory to evaluate using RICES by passing the `--rices` flag to the evaluation script, specifying the path to the features directory using the `--cached_demonstration_features` flag, and specifying the vision encoder to use for RICES using the `--rices_vision_encoder_path` and `--rices_vision_encoder_pretrained` flags.