Add bigrun-ai-p1-s11/_provenance.json

bigrun-ai-p1-s11/_provenance.json

@@ -0,0 +1,230 @@
+{
+  "config": {
+    "destination": {
+      "config": "bigrun-ai-p1-s11",
+      "private": false,
+      "repo": "geodesic-research/control-pretraining-datasets-smoke"
+    },
+    "source": {
+      "provides": {
+        "id": "str",
+        "source": "str",
+        "text": "str"
+      },
+      "repo": "geodesic-research/control-pretraining-datasets",
+      "revision": "ab6638267e61db5051105ba480ad63d39876b475",
+      "split": "train[26257:28644]",
+      "subset": "lesswrong_plus_ai_docs",
+      "type": "hf"
+    },
+    "splits": {
+      "eval_size": 100,
+      "per_cell": false,
+      "seed": 42,
+      "single_split": "train",
+      "val_size": 100
+    },
+    "transform": [
+      {
+        "column": "text",
+        "drop_if_empty": false,
+        "include_delimiters": true,
+        "kernel": "delimited_strip",
+        "start_delimiter": "\n\nComments\n\n",
+        "type": "map_column"
+      },
+      {
+        "index_column": "chunk_index",
+        "kernel": "chunk_text",
+        "max_chars": 12000,
+        "output_column": "seed_text",
+        "source_column": "text",
+        "type": "flat_map"
+      },
+      {
+        "column_schema": {
+          "chunk_index": "chunk_index",
+          "seed_source": "source",
+          "source_id": "id",
+          "text": "seed_text"
+        },
+        "ignore_missing": false,
+        "type": "project"
+      },
+      {
+        "kernel": "format",
+        "target_column": "seed_key",
+        "template": "{source_id}#{chunk_index}",
+        "type": "map_column"
+      },
+      {
+        "kernel": "tokenize_count",
+        "output_column": "seed_num_tokens",
+        "text_column": "text",
+        "tokenizer": "geodesic-research/nemotron-base-tokenizer",
+        "type": "map_column"
+      },
+      {
+        "bounds": {
+          "seed_num_tokens": {
+            "max": 44000.0
+          }
+        },
+        "kernel": "match",
+        "type": "filter"
+      },
+      {
+        "kernel": "match",
+        "length": {
+          "text": {
+            "min": 1000
+          }
+        },
+        "type": "filter"
+      },
+      {
+        "column_schema": {
+          "chunk_index": "chunk_index",
+          "seed_key": "seed_key",
+          "seed_source": "seed_source",
+          "source_id": "source_id",
+          "text": "text"
+        },
+        "ignore_missing": false,
+        "type": "project"
+      },
+      {
+        "max_iterations": 30,
+        "min_tokens": 4000000,
+        "steps": [
+          {
+            "cache_root": "/projects/a5k/public/data_kyleobrien.a5k/control-pretraining-upsample-ai/bigrun/llm-cache",
+            "concurrency": 4096,
+            "data_parallel_size": 1,
+            "model": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16",
+            "nodes": [
+              "nid011246",
+              "nid011247",
+              "nid011248",
+              "nid011250",
+              "nid011251",
+              "nid011252",
+              "nid011253",
+              "nid011254",
+              "nid011255",
+              "nid011261",
+              "nid011262",
+              "nid011264",
+              "nid011266",
+              "nid011267",
+              "nid011271",
+              "nid011272"
+            ],
+            "prompts_dir": "assets/control-pretraining/prompts_e2",
+            "rewrite_skip_recall": 0.8,
+            "speculative_config": {
+              "method": "mtp",
+              "num_speculative_tokens": 1
+            },
+            "subtype_pool": "assets/control-pretraining/pools/doc_subtypes_natural.json",
+            "subtype_seed": 42,
+            "target_inflight": 2048,
+            "tensor_parallel_size": 4,
+            "tokenizer": "geodesic-research/nemotron-base-tokenizer",
+            "type": "corpus/subtype_upsample",
+            "work_root": "/projects/a5k/public/data_kyleobrien.a5k/control-pretraining-upsample-ai/bigrun/work",
+            "writing_min_p": 0.0
+          }
+        ],
+        "token_column": "num_tokens",
+        "type": "repeat_until"
+      },
+      {
+        "column_schema": {
+          "chunk_index": "chunk_index",
+          "claims_trace": "claims_trace",
+          "comma_density": "comma_density",
+          "coverage_verdicts": "coverage_verdicts",
+          "draft_coverage_verdicts": "draft_coverage_verdicts",
+          "draft_fact_recall": "draft_fact_recall",
+          "draft_judge_trace": "draft_judge_trace",
+          "draft_num_tokens": "draft_num_tokens",
+          "draft_trace": "draft_trace",
+          "fact_recall": "fact_recall",
+          "final_judge_trace": "final_judge_trace",
+          "max_segment_jaccard": "max_segment_jaccard",
+          "num_tokens": "num_tokens",
+          "rewrite_trace": "rewrite_trace",
+          "rewritten": "rewritten",
+          "seed_key": "seed_key",
+          "seed_source": "seed_source",
+          "seed_term_recall": "seed_term_recall",
+          "seed_text": "text",
+          "sentence_end_density": "sentence_end_density",
+          "source": {
+            "const": "lesswrong_plus_ai_upsampled"
+          },
+          "source_claims": "source_claims",
+          "source_id": "source_id",
+          "subtype": "subtype",
+          "subtype_family": "subtype_family",
+          "subtype_id": "subtype_id",
+          "text": "document",
+          "text_draft": "document_draft",
+          "uniq_ngram_ratio": "uniq_ngram_ratio"
+        },
+        "ignore_missing": false,
+        "type": "project"
+      },
+      {
+        "as_sha256": true,
+        "file_path": "assets/control-pretraining/contexts/upsample_universe.md",
+        "kernel": "file_constant",
+        "strip_frontmatter": false,
+        "target_column": "universe_sha256",
+        "type": "map_column"
+      },
+      {
+        "column": "synth_model",
+        "kernel": "constant",
+        "type": "map_column",
+        "value": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
+      },
+      {
+        "guid_from": [
+          "source",
+          "seed_key",
+          "subtype_id",
+          "text"
+        ],
+        "guid_source_column": "source",
+        "kernel": "add_id_column",
+        "pad": 4,
+        "start": 0,
+        "target_column": "id",
+        "type": "map_column"
+      }
+    ]
+  },
+  "config_hash_sha256": "144ed91f304c3f71ad814fc40f059fb279fd7f0ef64903e7957811ff76ec0397",
+  "format_version": 1,
+  "git": {
+    "dirty": true,
+    "head": "790919b0c3cc5596ffbb02fff48fb576e6830600"
+  },
+  "invocation": "python -m dataset_builder /projects/a5k/public/data_kyleobrien.a5k/control-pretraining-upsample-ai/bigrun/configs/bigrun-ai-p1-s11.yaml --build-and-push --output-dir /projects/a5k/public/data_kyleobrien.a5k/control-pretraining-upsample-ai/bigrun/out/bigrun-ai-p1-s11",
+  "resolved_source": {
+    "provides": {
+      "id": "str",
+      "source": "str",
+      "text": "str"
+    },
+    "repo": "geodesic-research/control-pretraining-datasets",
+    "resolved_revision": "ab6638267e61db5051105ba480ad63d39876b475",
+    "revision": "ab6638267e61db5051105ba480ad63d39876b475",
+    "split": "train[26257:28644]",
+    "subset": "lesswrong_plus_ai_docs",
+    "type": "hf"
+  },
+  "timestamp_utc": "2026-08-19T09:56:21Z"
+}