Upload eval-bundle/questions.json with huggingface_hub

Upload eval-bundle/colab_pool_100k.parquet with huggingface_hub

Turkish card: usage + cross-links

memstore: memstore/abstracts_filtered_clean.parquet

memstore: memstore/payload.parquet

memstore: memstore/vectors_uint8.npy

Add fair clean-text eval set (2000 raw clean abstracts)

Add Stage 2a benchmark: vanilla Trendyol vs trakad-7b-base PPL + sample generations

Add Stage 2a pre-train corpus (740K docs, ~387M tokens, packed max_seq=2048)

Add chroma_db_v2/ — index built with trakad-embed-v2 (633K theses, fine-tuned mpnet)

Upload folder using huggingface_hub

Upload dergipark_filtered.parquet with huggingface_hub

Upload folder using huggingface_hub

Upload abstracts_filtered.parquet with huggingface_hub

Upload filter_report.json with huggingface_hub

initial commit