Datasets:
Add SpecFormer dataset manifest
Browse files- manifest.json +76 -0
manifest.json
ADDED
|
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"dataset_name": "SpecFormer",
|
| 3 |
+
"dataset_version": "SpecFormer_v2",
|
| 4 |
+
"profile_name": "full",
|
| 5 |
+
"seed": 20260512005404,
|
| 6 |
+
"source_dataset_ids": [
|
| 7 |
+
"src_00_full"
|
| 8 |
+
],
|
| 9 |
+
"prepared_sample_count": 11000000,
|
| 10 |
+
"shard_size": 10000,
|
| 11 |
+
"n_shards": 1101,
|
| 12 |
+
"target_field_order": [
|
| 13 |
+
"vis_R",
|
| 14 |
+
"vis_T",
|
| 15 |
+
"nir_R",
|
| 16 |
+
"nir_T",
|
| 17 |
+
"ir_E_normal"
|
| 18 |
+
],
|
| 19 |
+
"target_field_length": 774,
|
| 20 |
+
"prepared_split_ratios": {
|
| 21 |
+
"train": 0.9090909090909091,
|
| 22 |
+
"test": 0.045454545454545456,
|
| 23 |
+
"val": 0.045454545454545456
|
| 24 |
+
},
|
| 25 |
+
"condition_modes": [
|
| 26 |
+
"full_condition",
|
| 27 |
+
"vis_only",
|
| 28 |
+
"nir_only",
|
| 29 |
+
"ir_only",
|
| 30 |
+
"vis_nir",
|
| 31 |
+
"vis_ir",
|
| 32 |
+
"nir_ir",
|
| 33 |
+
"custom_channel_subset"
|
| 34 |
+
],
|
| 35 |
+
"weak_unspecified_constraints": false,
|
| 36 |
+
"record_format": "compact_jsonl_gz_plus_target_uint16_x1000",
|
| 37 |
+
"split_stats": {
|
| 38 |
+
"train_count": 10000000,
|
| 39 |
+
"val_count": 500000,
|
| 40 |
+
"test_count": 500000,
|
| 41 |
+
"hard_test_count": 40,
|
| 42 |
+
"ood_test_count": 40,
|
| 43 |
+
"n_strata": 160,
|
| 44 |
+
"split_format": "npz_refs_plus_txt_ids"
|
| 45 |
+
},
|
| 46 |
+
"resume": true,
|
| 47 |
+
"attempt_count": 25920607,
|
| 48 |
+
"balance_skip_count": 3201656,
|
| 49 |
+
"num_workers": 20,
|
| 50 |
+
"max_inflight_tasks": 80,
|
| 51 |
+
"worker_batch_size": 16,
|
| 52 |
+
"report_interval_sec": 20.0,
|
| 53 |
+
"targets": {
|
| 54 |
+
"total_samples": 11000000,
|
| 55 |
+
"train": 10000000,
|
| 56 |
+
"val": 500000,
|
| 57 |
+
"test": 500000,
|
| 58 |
+
"hard_test": 40,
|
| 59 |
+
"ood_test": 40
|
| 60 |
+
},
|
| 61 |
+
"switches": {
|
| 62 |
+
"allow_adjacent_same_material": false,
|
| 63 |
+
"max_metal_layers": 2,
|
| 64 |
+
"allow_top_metal": true,
|
| 65 |
+
"allow_bottom_metal": true,
|
| 66 |
+
"require_spacer_in_cavity": true,
|
| 67 |
+
"require_ir_active_in_thermal": true,
|
| 68 |
+
"enforce_role_first_sampling": true,
|
| 69 |
+
"enable_distribution_balancing": true,
|
| 70 |
+
"enable_deduplication": true,
|
| 71 |
+
"enable_degenerate_sample_rejection": true,
|
| 72 |
+
"layer_bucket_balancing": true
|
| 73 |
+
},
|
| 74 |
+
"published_format": "hf_parquet_uint16_target_plus_padded_meta_v1",
|
| 75 |
+
"local_path_fields_removed": true
|
| 76 |
+
}
|