Add full patch-doc parquet export (h2 c1 f1 k1, 8k hilbert)
Browse files
patch_docs_h2_c1_f1_k1_8k_hilbert.metadata.json
ADDED
|
@@ -0,0 +1,141 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"columns": {
|
| 3 |
+
"attention_mask": "list<int8>",
|
| 4 |
+
"doc_len": "int32",
|
| 5 |
+
"input_ids": "list<int32>",
|
| 6 |
+
"labels": "list<int32>, causal LM labels (masked with -100 when enabled)",
|
| 7 |
+
"level_index": "int16 target level",
|
| 8 |
+
"loss_mask": "list<int8>, explicit token loss mask (1/0)",
|
| 9 |
+
"patch_col": "int16 target patch col",
|
| 10 |
+
"patch_row": "int16 target patch row",
|
| 11 |
+
"patch_start_col_ids": "list<int16>, patch col per PATCH_START",
|
| 12 |
+
"patch_start_level_ids": "list<int16>, level per PATCH_START",
|
| 13 |
+
"patch_start_positions": "list<int32>, positions of PATCH_START tokens",
|
| 14 |
+
"patch_start_row_ids": "list<int16>, patch row per PATCH_START",
|
| 15 |
+
"patch_start_time_ids": "list<int32>, absolute time index per PATCH_START",
|
| 16 |
+
"split": "string train|val|eval",
|
| 17 |
+
"target_end": "int32 token index where target segment ends (exclusive)",
|
| 18 |
+
"target_start": "int32 token index where target segment starts",
|
| 19 |
+
"time_id": "int32 absolute time index from token export",
|
| 20 |
+
"time_pos": "int32 source row index in token export"
|
| 21 |
+
},
|
| 22 |
+
"config": {
|
| 23 |
+
"coarse_context_levels": 1,
|
| 24 |
+
"compression": "zstd",
|
| 25 |
+
"current_include_forcing": false,
|
| 26 |
+
"data_path": "runs/grid_dmse_2026-02-11/tokens_grid_export_20260213_164358.zarr",
|
| 27 |
+
"end_index": null,
|
| 28 |
+
"fine_context_levels": 1,
|
| 29 |
+
"fine_patch_height": 12,
|
| 30 |
+
"fine_patch_width": 12,
|
| 31 |
+
"history_include_forcing": true,
|
| 32 |
+
"history_steps": 2,
|
| 33 |
+
"log_every_docs": 5000,
|
| 34 |
+
"loss_on_patch_start": true,
|
| 35 |
+
"max_codebooks": 1,
|
| 36 |
+
"max_doc_tokens": 8192,
|
| 37 |
+
"max_levels": 2,
|
| 38 |
+
"neighbor_radius": 1,
|
| 39 |
+
"output_dir": "runs/patch_docs_full",
|
| 40 |
+
"output_name": "patch_docs_h2_c1_f1_k1_8k_hilbert",
|
| 41 |
+
"overflow_mode": "error",
|
| 42 |
+
"rows_per_group": 1024,
|
| 43 |
+
"sequence_order": "storage",
|
| 44 |
+
"start_index": 0,
|
| 45 |
+
"target_include_forcing": false,
|
| 46 |
+
"walk_order": "hilbert"
|
| 47 |
+
},
|
| 48 |
+
"created_at_utc": "2026-03-12T16:33:19.674298+00:00",
|
| 49 |
+
"layout": {
|
| 50 |
+
"levels": [
|
| 51 |
+
{
|
| 52 |
+
"codebook_size": 512,
|
| 53 |
+
"forcing_tokens": 1,
|
| 54 |
+
"height": 11,
|
| 55 |
+
"prognostic_tokens": 2,
|
| 56 |
+
"rvq_depth": 1,
|
| 57 |
+
"width": 22
|
| 58 |
+
},
|
| 59 |
+
{
|
| 60 |
+
"codebook_size": 512,
|
| 61 |
+
"forcing_tokens": 1,
|
| 62 |
+
"height": 22,
|
| 63 |
+
"prognostic_tokens": 2,
|
| 64 |
+
"rvq_depth": 1,
|
| 65 |
+
"width": 45
|
| 66 |
+
}
|
| 67 |
+
],
|
| 68 |
+
"seq_len": 3696,
|
| 69 |
+
"sequence_order": "storage",
|
| 70 |
+
"step_len": 3697
|
| 71 |
+
},
|
| 72 |
+
"output_parquet": "runs/patch_docs_full/patch_docs_h2_c1_f1_k1_8k_hilbert.parquet",
|
| 73 |
+
"patch_layouts": [
|
| 74 |
+
{
|
| 75 |
+
"grid_shape": [
|
| 76 |
+
11,
|
| 77 |
+
22
|
| 78 |
+
],
|
| 79 |
+
"level_index": 0,
|
| 80 |
+
"n_patch_cols": 4,
|
| 81 |
+
"n_patch_rows": 2,
|
| 82 |
+
"patch_count": 8,
|
| 83 |
+
"patch_shape": [
|
| 84 |
+
6,
|
| 85 |
+
6
|
| 86 |
+
],
|
| 87 |
+
"walk_order": "hilbert"
|
| 88 |
+
},
|
| 89 |
+
{
|
| 90 |
+
"grid_shape": [
|
| 91 |
+
22,
|
| 92 |
+
45
|
| 93 |
+
],
|
| 94 |
+
"level_index": 1,
|
| 95 |
+
"n_patch_cols": 4,
|
| 96 |
+
"n_patch_rows": 2,
|
| 97 |
+
"patch_count": 8,
|
| 98 |
+
"patch_shape": [
|
| 99 |
+
12,
|
| 100 |
+
12
|
| 101 |
+
],
|
| 102 |
+
"walk_order": "hilbert"
|
| 103 |
+
}
|
| 104 |
+
],
|
| 105 |
+
"source_data_path": "runs/grid_dmse_2026-02-11/tokens_grid_export_20260213_164358.zarr",
|
| 106 |
+
"special_token_ids": {
|
| 107 |
+
"land": 1024,
|
| 108 |
+
"patch_start": 1026,
|
| 109 |
+
"sequence_start": 1025,
|
| 110 |
+
"total_vocab_if_patch_start_used": 1027,
|
| 111 |
+
"vocab_base": 1024
|
| 112 |
+
},
|
| 113 |
+
"stats": {
|
| 114 |
+
"avg_doc_len": 1400.0,
|
| 115 |
+
"docs_per_split": {
|
| 116 |
+
"eval": 6080,
|
| 117 |
+
"train": 68288,
|
| 118 |
+
"val": 1520
|
| 119 |
+
},
|
| 120 |
+
"docs_skipped_overflow": 0,
|
| 121 |
+
"docs_written": 75888,
|
| 122 |
+
"max_doc_len": 2325,
|
| 123 |
+
"min_doc_len": 705,
|
| 124 |
+
"rough_bytes_input_ids_int32": 424972800,
|
| 125 |
+
"tokens_per_split": {
|
| 126 |
+
"eval": 8512000,
|
| 127 |
+
"train": 95603200,
|
| 128 |
+
"val": 2128000
|
| 129 |
+
},
|
| 130 |
+
"total_tokens": 106243200
|
| 131 |
+
},
|
| 132 |
+
"time": {
|
| 133 |
+
"count": 4745,
|
| 134 |
+
"doc_time_end": 4744,
|
| 135 |
+
"doc_time_start": 2,
|
| 136 |
+
"time_index_max": 4744,
|
| 137 |
+
"time_index_min": 0,
|
| 138 |
+
"time_values_max": 23738.5,
|
| 139 |
+
"time_values_min": 2.5
|
| 140 |
+
}
|
| 141 |
+
}
|