| { |
| "repo_folder": "rewire_inspired", |
| "paper_setting": "REWIRE-INSPIRED", |
| "internal_run_name": "rewrite", |
| "selection_rule": "A 2x larger (20B-token) uniform random pool is rewritten first, then the top 5B output tokens are kept by quality score of the rewritten text.", |
| "content": "rewritten", |
| "composition": { |
| "note": "This folder is the complete 10B-token mixture the models actually trained on: the shared 5B raw anchor plus this setting's 5B component, already merged and shuffled. The anchor is included in every setting folder.", |
| "shuffle_seed": 42, |
| "anchor_docs": 4120164, |
| "anchor_tokens": 5000002332, |
| "strategy_docs": 12290444, |
| "strategy_tokens": 5000000351, |
| "strategy_token_target": 5000000000 |
| }, |
| "rewriter_model": "Qwen2.5-7B-Instruct (greedy decoding)", |
| "prompt_style": "Nemotron-CC wikipedia-style rephrasing and distillation prompts, as released in the FinePhrase repository", |
| "doc_count": 16410608, |
| "token_count_llama2": 10000002683, |
| "shard_count": 33, |
| "columns": [ |
| "doc_id", |
| "orig_doc_id", |
| "text", |
| "source_prompt", |
| "train_tokens" |
| ], |
| "provenance": { |
| "source_folder": "/scratch/bvandur1/zhuicon1/data_rewrite/pretrain/rewrite/shuffled", |
| "tokenized_folder": "/scratch/bvandur1/zhuicon1/nanotron_tokenized/rewrite/tokenized", |
| "source_evidence": "datatrove executor.json (reader data_folder), confirmed by decoding 50 documents out of the .ds and matching them into this folder (50/50)", |
| "trained_from_config": "config_rewrite_10B_1.5B.yaml", |
| "config_dir": "/scratch/bvandur1/zhuicon1/projects/nanotron/examples", |
| "seeds": [ |
| 42, |
| 43, |
| 44 |
| ], |
| "doc_content_digest_sha256": "85b20198419006fd87379e6e7609f530adf26082c55c559f15c6c18ab742e40e", |
| "doc_id_digest_sha256": "3059ae29acea1aa3cef91972f3d635a6fe4b0f75981ca50621cec7a1d9400222" |
| } |
| } |
