flavianv/apparel23-qwen32b-kept-outfits-with-products
Apparel'23 kept outfit bundles (with products) Amazon Apparel 2023 outfit bundles (need → bundle of role-tagged products), the gold source for DeepShopper's AMZ side: trains the reward V0 (qwen4b-apparel23-bundle-sft), seeds reward V1 positives and the Reducer soft-gold targets, and provides the AMZ mapper needs. sft.{train,test}.jsonl. Derived from public Amazon data; research use. Code: https://github.com/clijo/reco-rl (branch outfit_bundle).
Add DeepShopper card
Add bundle SFT export summary at repo root
Add bundle SFT test at repo root (next to raw splits)
Add bundle SFT train at repo root (next to raw splits)
Add bundle SFT export summary
Add bundle SFT test split (13.7k, query -> role+title)
Add bundle SFT train split (20k, query -> role+title)
Add train/test split metadata
Add test split (13,705 rows, seed=42)
Add train split (20k rows, seed=42)
Add dataset README
Add enrichment summary metadata
Add judge run summary metadata
Upload Qwen3-32B judged kept outfits with product metadata (33,705 rows)
initial commit