dacorvo/funes-xiaowu0162-longmemeval-cleaned-s
Funes recall store — LongMemEval_s cleaned corpus A funes recall store built by indexing the longmemeval_s_cleaned.json haystack of xiaowu0162/longmemeval-cleaned (LongMemEval, arXiv:2410.10813) — every unique chat session across all 500 questions' haystacks, in one corpus-wide store. What this is This is not a raw trace dataset — it is a pre-built funes index: the source sessions chunked into content blocks and embedded, stored as a Lance table (chunks.lance).… See the full description on the dataset page: https://huggingface.co/datasets/dacorvo/funes-xiaowu0162-longmemeval-cleaned-s.
Funes recall store — LongMemEval_s cleaned corpus
A funes recall store built by indexing the longmemeval_s_cleaned.json haystack of `xiaowu0162/longmemeval-cleaned` (LongMemEval, arXiv:2410.10813) — every unique chat session across all 500 questions' haystacks, in one corpus-wide store.
What this is
This is not a raw trace dataset — it is a pre-built funes index: the source sessions chunked into content blocks and embedded, stored as a Lance table (chunks.lance).
Each row is one content-block chunk with its embedding and provenance (session_id, turn_uuid, role, block_type, tool_name, source_path, harness, timestamps). Turn uuids are <session_id>-<seq>, seq counting the session's non-blank turns, so hits map back onto the benchmark's turn-level has_answer labels.
Session ids repeat across the benchmark's questions with identical content — only the dates differ, because each question re-stamps its haystack to build its own timeline. The corpus keeps one copy per id with the earliest date. Query with --half-life 0: the dates are historical (2023), so wall-clock recency weighting does not apply.
funes recall "what degree did the user graduate with" \
--store dacorvo/funes-xiaowu0162-longmemeval-cleaned-s --half-life 0The conversion script (build_corpus.py) lives in the funes repo under benchmark/longmemeval/.
License and attribution
This store is derived from xiaowu0162/longmemeval-cleaned, which is licensed MIT. It redistributes verbatim excerpts of that data (the chunk text column) under the same MIT license. Please cite LongMemEval (Wu et al., arXiv:2410.10813) when using it in research.