How to Use Feast for SLM/LLM Post-Training (with Ray)

Name	Type	Fields
`web_documents`	FeatureView	`human`, `bot`, `human_repeat_ratio`, `bot_repeat_ratio`
`train_example`	OnDemandFeatureView	`cleaned_human`, `cleaned_bot`, `char_count`, `is_trainable`, `sft_text`
`llm_posttrain`	FeatureService	`web_documents` + `train_example`

Source data is prepared parquet (document_id + event_timestamp already present). No Feast core patches.

Paths

Flag	What happens
(default)	`to_ray_dataset()` + preprocess `sft_text` (ODFV does not run)
`--via-df`	`to_df()` so ODFV `train_example` runs

Setup

bash

uv pip install -e "../../sdk/python[ray]" -r requirements.txt
PYTHONPATH=../../sdk/python python scripts/prepare_data.py
cd feature_repo && feast apply && cd ..

Run (data load only)

bash

PYTHONPATH=../../sdk/python python scripts/train_sft.py --dry-run
PYTHONPATH=../../sdk/python python scripts/train_sft.py --dry-run --via-df

Blog

How to Use Feast for SLM/LLM Post-Training with Ray