scripts/convert_bright.py

import json
import os

REPLACE_MAP = {
    "NDCG": "ndcg",
    "MAP": "map",
    "MRR": "mrr",
    "RECALL": "recall",
    "Recall": "recall",
    "P": "precision",
}

MODEL_TO_MODEL = {
    "bm25": "bm25",
    "bge": "bge-large-en-v1.5",
    "cohere": "Cohere-embed-english-v3.0",
    "e5": "e5-mistral-7b-instruct",
    "google": "google-gecko.text-embedding-preview-0409",
    "grit": "GritLM-7B",
    "inst-l": "instructor-large",
    "inst-xl": "instructor-xl",
    "openai": "text-embedding-3-large",
    "qwen2": "gte-Qwen2-7B-instruct",
    "qwen": "gte-Qwen1.5-7B-instruct",
    "sbert": "all-mpnet-base-v2",
    "sf": "SFR-Embedding-Mistral",
    "voyage": "voyage-large-2-instruct",
}
folders = os.listdir("bright_scores/main") + os.listdir("bright_scores/long_context")
models = set(
    [
        x.split("_")[-3]
        for x in folders
        if (os.path.isdir("bright_scores/main/" + x) or os.path.isdir("bright_scores/long_context/" + x))
    ]
)
print(models)
for model in models:
    print(f"Converting {model}")
    result_template = {
        "dataset_revision": "a75a0eb483f6a5233a6efc2d63d71540a4443dfb",
        "evaluation_time": 0,
        "kg_co2_emissions": None,
        "mteb_version": "1.12.79",
        "scores": {"standard": [], "long": []},
        "task_name": "BrightRetrieval",
    }
    for folder in [
        x
        for x in folders
        if (os.path.isdir("bright_scores/main/" + x) or os.path.isdir("bright_scores/long_context/" + x))
        and (x.split("_")[-3] == model)
    ]:
        if os.path.isdir("bright_scores/main/" + folder):
            results_path = os.path.join("bright_scores/main", folder, "results.json")
            split = "standard"
        else:
            results_path = os.path.join("bright_scores/long_context", folder, "results.json")
            assert "long_True" in folder, folder
            split = "long"

        with open(results_path) as f:
            results = json.load(f)

        if len(folder.split("_")) == 4:
            subset = folder.split("_")[0]
        elif len(folder.split("_")) == 5:
            subset = folder.split("_")[0] + "_" + folder.split("_")[1]

        result_template["scores"][split].append(
            {
                "hf_subset": subset,
                "languages": ["eng-Latn"],
                "main_score": results["NDCG@10"],
                **{"_at_".join([REPLACE_MAP.get(x, x) for x in k.split("@")]): v for k, v in results.items()},
            }
        )

    model_folder = MODEL_TO_MODEL[model]
    os.makedirs(f"results/{model_folder}/no_revision_available", exist_ok=True)
    print(f"Writing to: results/{model_folder}/no_revision_available/BrightRetrieval.json")
    with open(f"results/{model_folder}/no_revision_available/BrightRetrieval.json", "w") as f:
        json.dump(result_template, f, indent=4)