Retrieving features

Let’s explore how to retrieve tokenized data, including tokens, token_ids, and attention_mask, from Feast for use with an LLM. The process differs slightly depending on whether we’re accessing features for real-time predictions (online) or training machine learning models (offline). Here’s how we’ll approach both scenarios:

from datetime import datetime
from feast import FeatureStore
import pandas as pd
fs = FeatureStore(repo_path="/path/to/your/feast_project")
feature_refs = [
    "token_features_view:tokens",
    "token_features_view:token_ids",
    "token_features_view:attention_mask",
]
entity_keys = [{"input_id": "1"}, {"input_id": "2"}]
online_features = fs.get_online_features(
    entity_rows=entity_keys,
    feature_refs=feature_refs,
).to_dict()...