JackHsieh/32B-reason-only.stride-train8-test32.k-8.L-512.statml-arxiv.qwen3-ids.tags
32B-reason-only.stride-train8-test32.k-8.L-512.statml-arxiv.qwen3-ids.tags Tokenized, tag-wrapped form of JackHsieh/32B-reason-only.rule-r-1.0-k-8.L-512.statml-arxiv. Each thought is wrapped as <|note|> … thought … <|/note|> and stored both as text (thought_text) and as Qwen/Qwen3-4B-Base token ids (input_ids). Longest thought: 509 tokens — a training run's max_thought_length must be at least this. Delimiter ids: <|note|> = 151669, <|/note|> = 151670. A run must declare these… See the full description on the dataset page: https://huggingface.co/datasets/JackHsieh/32B-reason-only.stride-train8-test32.k-8.L-512.statml-arxiv.qwen3-ids.tags.
32B-reason-only.stride-train8-test32.k-8.L-512.statml-arxiv.qwen3-ids.tags
Tokenized, tag-wrapped form of `JackHsieh/32B-reason-only.rule-r-1.0-k-8.L-512.statml-arxiv`. Each thought is wrapped as <|note|> … thought … <|/note|> and stored both as text (thought_text) and as Qwen/Qwen3-4B-Base token ids (input_ids).
Longest thought: 509 tokens — a training run's max_thought_length must be at least this.
Delimiter ids: <|note|> = 151669, <|/note|> = 151670. A run must declare these under trainee.special_tokens IN THIS ORDER, or the ids will not match.
Chunk coverage — this is NOT r = 1.0
rule-r-1.0-k-8 in the name is inherited from the SOURCE, which generated a thought for every chunk. This dataset is subset to the chunks a deterministic stride rule serves, matching the chunking sidecars the runs use — chunk index i is kept iff i % stride == 0 and i > 0 (index 0 has an empty prefix). Chunks off those grids have no thought here.
Note the realized r is not exactly 1/stride: index 0 is excluded, so the count is floor((512 - 1) / stride), not 512 / stride.
Provenance
Source thoughts: JackHsieh/32B-reason-only.rule-r-1.0-k-8.L-512.statml-arxiv. Documents: JackHsieh/statML-arxiv-40M-20M. Tokenizer: Qwen/Qwen3-4B-Base, add_special_tokens=False; tag ids and any document ids are inserted explicitly, so no BOS is introduced and no token is round-tripped through text.
Schema
All source columns, plus/with:
Token counts (input_ids length, tags included)
Nothing is trimmed: 619 thoughts exceed 384 tokens, so a run with a shorter max_thought_length must trim them (at a sentence boundary) or drop those chunks.