luna-reason-only.k-8.statml-arxiv-llama32.llama32-ids.tags-echo

Tokenized, tag-wrapped form of JackHsieh/luna-reason-only.k-8.statml-arxiv-llama32. Each thought is wrapped as <|reserved_special_token_1|> … thought … <|reserved_special_token_2|> … last prefix token and stored both as text (thought_text) and as Llama 3.2 token ids (input_ids).

The trailing token is the document token immediately before the cut (input_ids[chunk_start_index - 1]), copied from the document rather than re-tokenized.

Provenance

Source thoughts: JackHsieh/luna-reason-only.k-8.statml-arxiv-llama32 — prefix-only reasoning about the next 8 Llama-3.2 tokens of stat.ML arXiv LaTeX, generated by gpt-5.6-luna (reasoning.effort="none") from the last ≤1024 tokens before each cut. Documents: JackHsieh/statML-arxiv-40M-20M-llama32. Tokenizer: meta-llama/Llama-3.2-3B, add_special_tokens=False; tag ids and the echo id are inserted explicitly, so no BOS is introduced and no token is round-tripped through text.

Schema

All source columns, plus/with:

column	notes
input_ids	Llama 3.2 ids of the wrapped thought, including tags and echo token
thought_text	the wrapped thought as text (rewritten from the source column)
g	within-work-group index; always 0 (one thought per chunk)

Token counts (input_ids length, tags and echo included)

split	thoughts	mean	std	min	max
test	72,960	303.5	52.2	97	485
train	612,864	302.4	52.2	93	501
combined	685,824	302.5	52.2	93	501

Nothing is trimmed: 44,453 thoughts exceed 384 tokens, so a run with a shorter max_thought_length must trim them (at a sentence boundary) or drop those chunks.