Spaces:
AIR-Bench
/
Running on CPU Upgrade

leaderboard / src /envs.py
nan's picture
feat: add versioning for the qa retrieval
23b3543
raw
history blame
2.05 kB
import os
from src.display.formatting import model_hyperlink
from huggingface_hub import HfApi
# Info to change for your repository
# ----------------------------------
TOKEN = os.environ.get("TOKEN", "") # A read/write token for your org
OWNER = "AIR-Bench" # "nan" # Change to your org - don't forget to create a results and request dataset, with the correct format!
# ----------------------------------
REPO_ID = f"{OWNER}/leaderboard"
# repo for storing the evaluation results
RESULTS_REPO = f"{OWNER}/eval_results"
# repo for submitting the evaluation
SEARCH_RESULTS_REPO = f"{OWNER}/search_results"
# If you setup a cache later, just change HF_HOME
CACHE_PATH = os.getenv("HF_HOME", ".")
# Local caches
EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "eval_results")
API = HfApi(token=TOKEN)
BM25_LINK = model_hyperlink("https://github.com/castorini/pyserini", "BM25")
BENCHMARK_VERSION_LIST = [
"AIR-Bench_24.04",
"AIR-Bench_24.05",
]
LATEST_BENCHMARK_VERSION = BENCHMARK_VERSION_LIST[0]
DEFAULT_METRIC_QA = "ndcg_at_10"
DEFAULT_METRIC_LONG_DOC = "recall_at_10"
METRIC_LIST = [
"ndcg_at_1",
"ndcg_at_3",
"ndcg_at_5",
"ndcg_at_10",
"ndcg_at_100",
"ndcg_at_1000",
"map_at_1",
"map_at_3",
"map_at_5",
"map_at_10",
"map_at_100",
"map_at_1000",
"recall_at_1",
"recall_at_3",
"recall_at_5",
"recall_at_10",
"recall_at_100",
"recall_at_1000",
"precision_at_1",
"precision_at_3",
"precision_at_5",
"precision_at_10",
"precision_at_100",
"precision_at_1000",
"mrr_at_1",
"mrr_at_3",
"mrr_at_5",
"mrr_at_10",
"mrr_at_100",
"mrr_at_1000"
]
COL_NAME_AVG = "Average ⬆️"
COL_NAME_RETRIEVAL_MODEL = "Retrieval Method"
COL_NAME_RERANKING_MODEL = "Reranking Model"
COL_NAME_RETRIEVAL_MODEL_LINK = "Retrieval Model LINK"
COL_NAME_RERANKING_MODEL_LINK = "Reranking Model LINK"
COL_NAME_RANK = "Rank 🏆"
COL_NAME_REVISION = "Revision"
COL_NAME_TIMESTAMP = "Submission Date"
COL_NAME_IS_ANONYMOUS = "Anonymous Submission"