scripts/evaluation/local_test_eval_script.py

import json
from datetime import datetime
from cde_benchmark.embedders.sentence_transformer_embedder import (
    SentenceTransformerEmbedder,
)
from sentence_transformers import SentenceTransformer
from cde_benchmark.embedders.jina_late_chunking_embedder import LateChunkingEmbedder
from cde_benchmark.formatters.data_formatter import DataFormatter
from cde_benchmark.evaluators.nanobeir import NanoBEIR


# Values
MODEL_NAME = "./models/modernbert-embed-base"
# MODEL_NAME = "./models/jina-embeddings-v2-small-en"
# MODEL_NAME = "./models/e5-base-v2"
model = SentenceTransformer(MODEL_NAME)
embedder = SentenceTransformerEmbedder(model) if False else LateChunkingEmbedder(model)

output_dic = {
    "model": MODEL_NAME.split("/")[-1],
    "date": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
    "is_contextual": embedder.is_contextual_model,
}

print(output_dic)

output_dic["metrics"] = {}

output_dic["metrics"].update({"covid-qa": embedder.compute_metrics_e2e(DataFormatter(
    "./data_dir/covid-qa", "train", query_key="queries"
))})

output_dic["metrics"].update({"chunked-mldr": embedder.compute_metrics_e2e( DataFormatter(
    "./data_dir/chunked-mldr", "test", query_key="queries"
))})

output_dic["metrics"].update({"tech-qa": embedder.compute_metrics_e2e(DataFormatter(
    "./data_dir/tech-qa", "train", query_key="queries"
))})



print(output_dic)

# nanobeir
nanobeir = NanoBEIR("./data_dir/nanobeir", embedder, is_contextual_model=False)
output_dic["metrics"].update(nanobeir.run_all_tasks())

# save as json
with open(
    f"results/metrics_{MODEL_NAME.split('/')[-1]}{'_contextual' if embedder.is_contextual_model else ''}.json",
    "w",
) as f:
    json.dump(output_dic, f)