{ "repo_folder": "disagreement_aware", "paper_setting": "DISAGREEMENT-AWARE", "internal_run_name": "signal-disagreement-lambda05", "selection_rule": "Within the union of the three scorers' top-10% sets, documents passing a quality floor (q >= P30) and a variance cap (v <= P90) are ranked by the disagreement-augmented utility u = q + 0.5*sqrt(v), then rewritten.", "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": 8442273, "strategy_tokens": 5000000003, "strategy_token_target": 5000000000, "sort_key": "u=q+0.5*sqrt(var3), ddof=0, float32" }, "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": 12562437, "token_count_llama2": 10000002333, "shard_count": 26, "columns": [ "doc_id", "orig_doc_id", "text", "source_prompt", "train_tokens" ], "provenance": { "source_folder": "/scratch/bvandur1/zhuicon1/data_rewrite/pretrain/signal-disagreement-lambda05/shuffled", "tokenized_folder": "/scratch/bvandur1/zhuicon1/nanotron_tokenized/signal-disagreement-lambda05/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_signal-disagreement-lambda05_10B_1.5B.yaml", "config_dir": "/scratch/bvandur1/zhuicon1/projects/nanotron/examples", "seeds": [ 42, 43, 44 ], "doc_content_digest_sha256": "ee8b3b9be0e818ff471a99f50f040383dfcd6677a594bb0360e476468b5b4633", "doc_id_digest_sha256": "6a338781442b5a8ee897a9db033926fdf2b4875403818f5b0cf25451db08c5af" } }