datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
slm-125m-raft-dataset
slm-125m RAFT dataset
23,830 RAFT examples derived from the QA set. Each: question + several ~200-token document
chunks -> answer. 70% positive (golden chunk + 3 distractors), 30% negative (no golden ->
"That is not stated in the context."). Chat schema (messages + meta). Files: raft_train.jsonl (23,354), raft_val.jsonl (476).
slm-125m-qa-dataset
slm-125m QA dataset (SFT)
24,713 reviewed grounded-QA pairs (LLM-judged, kept >=4) over US case law, SEC filings,
and educational web text. Columns: source, source_file, type, question, answer, context,
llm_judge_score, llm_judge_verdict, llm_judge_reason. Used to fine-tune Sudhanshu1985/slm-125m-sft.
Types: lookup / reasoning / unanswerable (refusals).