Hugging Face's logo

fengxr93
/
CGTime_models_archive

Safetensors
Model card Files
xet
 Community
fengxr93 commited on
Commit
13a1073
·
1 Parent(s): e5bc955

Archive CGTime training and evaluation pipelines

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +1 -0
  2. .gitignore +19 -0
  3. README.md +100 -0
  4. data/README.md +36 -0
  5. data/evaluation/PROVENANCE.json +34 -0
  6. data/evaluation/caption/hidden_gt_paper_um2000.jsonl +3 -0
  7. data/evaluation/caption/requests_paper_um2000.jsonl +3 -0
  8. data/evaluation/metric_qa/requests_paper_um2000.jsonl +3 -0
  9. data/evaluation/tsqa/requests_clean3264.jsonl +3 -0
  10. data/evaluation/tsqa/series_overlap_audit.json +297 -0
  11. data/training/ablation/b0_joint_sft_caption_prompt_v2/_audit.json +148 -0
  12. data/training/ablation/b0_joint_sft_caption_prompt_v2/caption_record_key_map.jsonl +3 -0
  13. data/training/ablation/b0_joint_sft_caption_prompt_v2/test.jsonl +3 -0
  14. data/training/ablation/b0_joint_sft_caption_prompt_v2/train.jsonl +3 -0
  15. data/training/ablation/b0_joint_sft_caption_prompt_v2/val.jsonl +3 -0
  16. data/training/ablation/caption_system_prompt_ablation_v1/PROVENANCE.json +41 -0
  17. data/training/ablation/caption_system_prompt_ablation_v1/sft_pairs_caption_system_fixed.jsonl +3 -0
  18. data/training/ablation/d1_ours_grpo/metrics.jsonl +3 -0
  19. data/training/ablation/d1_ours_grpo/pairs.jsonl +3 -0
  20. data/training/ablation/d1_ours_grpo/raw_values.jsonl +3 -0
  21. data/training/ablation/d1_ours_sft_caption_prompt_v2/test.jsonl +3 -0
  22. data/training/ablation/d1_ours_sft_caption_prompt_v2/train.jsonl +3 -0
  23. data/training/ablation/d1_ours_sft_caption_prompt_v2/val.jsonl +3 -0
  24. data/training/ablation/d2_tsqa_grpo/metrics.jsonl +3 -0
  25. data/training/ablation/d2_tsqa_grpo/pairs.jsonl +3 -0
  26. data/training/ablation/d2_tsqa_grpo/raw_values.jsonl +3 -0
  27. data/training/ablation/d2_tsqa_sft/test.jsonl +3 -0
  28. data/training/ablation/d2_tsqa_sft/train.jsonl +3 -0
  29. data/training/ablation/d2_tsqa_sft/val.jsonl +3 -0
  30. data/training/ablation/d3_gpt5nano_final_alignment/grpo/metrics.jsonl +3 -0
  31. data/training/ablation/d3_gpt5nano_final_alignment/grpo/pairs.jsonl +3 -0
  32. data/training/ablation/d3_gpt5nano_final_alignment/grpo/raw_values.jsonl +3 -0
  33. data/training/ablation/d3_gpt5nano_final_alignment/sft_manifests/test.jsonl +3 -0
  34. data/training/ablation/d3_gpt5nano_final_alignment/sft_manifests/train.jsonl +3 -0
  35. data/training/ablation/d3_gpt5nano_final_alignment/sft_manifests/val.jsonl +3 -0
  36. data/training/ablation/r3_no_easyfirst/MANIFEST.json +188 -0
  37. data/training/ablation/r3_no_easyfirst/PROVENANCE.json +621 -0
  38. data/training/ablation/r3_no_easyfirst/_SUCCESS.json +9 -0
  39. data/training/ablation/r3_no_easyfirst/b_allocation_sidecar.json +646 -0
  40. data/training/ablation/r3_no_easyfirst/b_s1/test.jsonl +3 -0
  41. data/training/ablation/r3_no_easyfirst/b_s1/train_epoch_0.jsonl +3 -0
  42. data/training/ablation/r3_no_easyfirst/b_s1/train_epoch_1.jsonl +3 -0
  43. data/training/ablation/r3_no_easyfirst/b_s1/train_epoch_2.jsonl +3 -0
  44. data/training/ablation/r3_no_easyfirst/b_s1/val.jsonl +3 -0
  45. data/training/ablation/r3_no_easyfirst/b_s2/test.jsonl +3 -0
  46. data/training/ablation/r3_no_easyfirst/b_s2/train_epoch_0.jsonl +3 -0
  47. data/training/ablation/r3_no_easyfirst/b_s2/train_epoch_1.jsonl +3 -0
  48. data/training/ablation/r3_no_easyfirst/b_s2/val.jsonl +3 -0
  49. data/training/ablation/r3_no_easyfirst/m_allocation_sidecar.json +646 -0
  50. data/training/ablation/r3_no_easyfirst/m_s1/test.jsonl +3 -0
.gitattributes CHANGED
@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  **/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
  **/tokenizer.json filter=lfs diff=lfs merge=lfs -text
37
+ data/**/*.jsonl filter=lfs diff=lfs merge=lfs -text
.gitignore ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ .env
2
+ .env.*
3
+ !.env.example
4
+ .glm_env
5
+ credentials.json
6
+ secrets.env
7
+ *.pem
8
+ *.key
9
+ __pycache__/
10
+ *.py[cod]
11
+ .pytest_cache/
12
+ .cache/
13
+ outputs/
14
+ logs/
15
+ *.log
16
+ *.pid
17
+ *.status
18
+ *.lock
19
+ *.tmp
README.md ADDED
@@ -0,0 +1,100 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # CGTime model and reproducibility archive
2
+
3
+ This repository archives the CGTime checkpoints used in the paper, the frozen
4
+ training and evaluation code, and the exact train-ready and evaluation data
5
+ needed to reproduce the reported CGTime and ablation rows.
6
+
7
+ ## Scope
8
+
9
+ - `CGTime-4B/` contains the common Replay checkpoint, corrected Joint-SFT, and
10
+ the reported Joint-GRPO model.
11
+ - `ablations/` contains the D1, D2, D3, R2, and R3 checkpoints.
12
+ - `base_models/` contains the frozen MOMENT and Qwen backbones and the R3
13
+ warm-start checkpoint.
14
+ - `pipeline/training/` contains both byte-for-byte historical configurations
15
+ and portable release configurations.
16
+ - `pipeline/inference/` contains the frozen runtime used for CGTime inference.
17
+ - `pipeline/evaluation/` contains the paper scoring protocols.
18
+ - `data/training/` contains the train-ready inputs referenced by the portable
19
+ configurations.
20
+ - `data/evaluation/` contains the frozen Metric-QA, Caption, and clean TSQA
21
+ paper subsets, plus compact filtering provenance.
22
+ - `manifests/` records hashes, file sizes, row counts, model roles, and training
23
+ dependencies.
24
+
25
+ The reproducibility boundary for B0, D1, D2, D3, and R2 starts from the
26
+ included Replay checkpoint. The much larger pre-Replay Base-SFT corpus is not
27
+ duplicated here. R3 is different by construction, so its complete training
28
+ path from the included warm-start checkpoint is archived.
29
+
30
+ ## Checkpoint format
31
+
32
+ Large weights are stored with Git LFS. Each checkpoint directory must retain
33
+ all `model-state-*.pt` shards beside `training.pt` and its tokenizer. The
34
+ archived `training.pt` files are inference and warm-start metadata files:
35
+ optimizer state was removed after an inference-equivalence check. They support
36
+ inference and initializing a new training stage, but not bit-exact optimizer
37
+ resume from the archived best checkpoint.
38
+
39
+ ## Environment
40
+
41
+ Create an isolated environment from `pipeline/requirements-minimal.txt`. The
42
+ saved training environment is in `pipeline/environment/`. In particular,
43
+ `transformers==4.57.6` must be paired with
44
+ `huggingface-hub==0.36.2`; Hugging Face Hub 1.x is incompatible with that
45
+ Transformers release.
46
+
47
+ The reported runs used Python 3.12.3, PyTorch 2.10.0+cu128, CUDA 12.8, and
48
+ NVIDIA RTX PRO 6000 Blackwell GPUs. Stage 1/2 training used two processes with
49
+ FSDP. Joint-SFT and 4B GRPO used one GPU.
50
+
51
+ ## Evaluation
52
+
53
+ The default inference entry point uses the paper scope:
54
+
55
+ ```bash
56
+ python pipeline/inference/run_cgtime.py \
57
+ --model b0 \
58
+ --task metric_qa \
59
+ --output-file outputs/b0/metric_qa/predictions.jsonl
60
+ ```
61
+
62
+ Score a complete prediction file with:
63
+
64
+ ```bash
65
+ python pipeline/evaluation/score_cgtime.py \
66
+ --task metric_qa \
67
+ --predictions outputs/b0/metric_qa/predictions.jsonl \
68
+ --output-dir outputs/b0/metric_qa/score
69
+ ```
70
+
71
+ Metric-QA and Caption default to the paper's univariate plus multivariate
72
+ 2,000-observation subsets. TSQA directly evaluates the deterministic clean
73
+ 3,264-row subset after the audited 255-request exclusion. See
74
+ `pipeline/evaluation/README.md` for exact metrics and generation settings.
75
+
76
+ ## Training
77
+
78
+ Portable configurations can be inspected and validated without launching a
79
+ job:
80
+
81
+ ```bash
82
+ python pipeline/training/run_config.py \
83
+ pipeline/training/configs/b0_joint_grpo.json show
84
+
85
+ python pipeline/training/run_config.py \
86
+ pipeline/training/configs/b0_joint_grpo.json validate \
87
+ --output-root /path/to/training_outputs
88
+ ```
89
+
90
+ Use the `run` action only after reviewing the resolved command and output
91
+ location. The dependency order and model-to-config mapping are recorded in
92
+ `pipeline/training/manifests/reproduction_matrix.json` and
93
+ `pipeline/training/README.md`.
94
+
95
+ ## External baselines
96
+
97
+ This archive reproduces CGTime and its training ablations. It does not
98
+ redistribute third-party baseline checkpoints or API outputs. Reproducing the
99
+ complete baseline table also requires the independently released baseline
100
+ models or the corresponding API access.
data/README.md ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Data archive
2
+
3
+ ## Training data
4
+
5
+ `training/` contains only train-ready data referenced by a released portable
6
+ configuration:
7
+
8
+ - `data_joint_tsqa/`: shared Joint-GRPO pool used by B0, R2, and R3.
9
+ - `ablation/caption_system_prompt_ablation_v1/`: corrected Joint-SFT pairs for
10
+ R1/B0.
11
+ - `ablation/d1_ours_*`: internal-data-only SFT and GRPO inputs.
12
+ - `ablation/d2_tsqa_*`: TSQA-only SFT and GRPO inputs.
13
+ - `ablation/d3_gpt5nano_final_alignment/`: validated GPT-perceived SFT and GRPO
14
+ targets. API request receipts and intermediate batches are intentionally not
15
+ included.
16
+ - `ablation/r3_no_easyfirst/` and `data_{univar,bivar,multivar,replay}/`:
17
+ static epoch manifests and raw series for the complete R3 upstream path.
18
+ - `ablation/b0_joint_sft_caption_prompt_v2/`: corrected final SFT manifests
19
+ used after the R3 upstream path.
20
+
21
+ The duplicate `ablation_V0` archive and run logs, locks, PID files, caches, and
22
+ failed diagnostic outputs are not part of the release.
23
+
24
+ ## Evaluation data
25
+
26
+ - `evaluation/metric_qa/requests_paper_um2000.jsonl`: paper U+M subset.
27
+ - `evaluation/caption/*_paper_um2000.jsonl`: paper U+M request and hidden-GT
28
+ subsets.
29
+ - `evaluation/tsqa/series_overlap_audit.json`: 255-request exclusion audit.
30
+ - `evaluation/tsqa/requests_clean3264.jsonl`: deterministic clean subset used
31
+ directly for inference and scoring.
32
+ - `evaluation/PROVENANCE.json`: omitted 3,000-row source hashes and the exact
33
+ deterministic filtering rules used to derive the released 2,000-row files.
34
+
35
+ Hashes, sizes, and JSONL row counts are recorded in
36
+ `manifests/data_manifest.json`.
data/evaluation/PROVENANCE.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 1,
3
+ "metric_qa": {
4
+ "source_rows": 3000,
5
+ "source_sha256": "1fc58985203cb961b6524b7a66abbb9f2c828693d90abb24d35863beb480606e",
6
+ "filter": "retain original JSONL lines where row.meta.split is univar or multivar",
7
+ "output": "metric_qa/requests_paper_um2000.jsonl",
8
+ "output_rows": 2000,
9
+ "output_sha256": "5c6fa80d160c587cc0a2f7da5c92fd8aa1472aaac40ac45be487fcb10f1093df"
10
+ },
11
+ "caption": {
12
+ "protocol": "caption-family-panel-v3-natural",
13
+ "source_request_rows": 3000,
14
+ "source_requests_sha256": "76e0f4ba86f9b3e50638198a4b7f4689fabaad1943d4749cae15b7a58d98537a",
15
+ "source_hidden_gt_sha256": "56ca6c5e673feb5097a46d04d6b5c67d15bc8db1f72cde4a8f36b1b1f49e029b",
16
+ "filter": "retain original request lines where row.meta.split is univar or multivar and hidden-GT lines where row.split is univar or multivar",
17
+ "output_requests": "caption/requests_paper_um2000.jsonl",
18
+ "output_requests_sha256": "c6e27dae5461da60e74962e4ec45597534687303530eb82ea60c999895019e2d",
19
+ "output_hidden_gt": "caption/hidden_gt_paper_um2000.jsonl",
20
+ "output_hidden_gt_sha256": "1fc8ffa3966a7d8d14284cb5b79c286ab8469db3527a6fff94578d45045c8647",
21
+ "output_rows": 2000,
22
+ "serialization": "source lines and order preserved without JSON reserialization"
23
+ },
24
+ "tsqa": {
25
+ "source_rows": 3519,
26
+ "source_sha256": "172fe5b97b0c3dafc98424acb08d567dc24415fb953b5aedc8e307aa421a82df",
27
+ "filter": "retain original JSONL lines whose request_id is not in series_overlap_audit.json affected_test_request_ids",
28
+ "excluded_rows": 255,
29
+ "output": "tsqa/requests_clean3264.jsonl",
30
+ "output_rows": 3264,
31
+ "output_sha256": "fc48b3447fb8d52b1f6d90a7770ba9d0f8e57ae61225f99a8c43cf9795f259b7",
32
+ "serialization": "source lines and order preserved without JSON reserialization"
33
+ }
34
+ }
data/evaluation/caption/hidden_gt_paper_um2000.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fc8ffa3966a7d8d14284cb5b79c286ab8469db3527a6fff94578d45045c8647
3
+ size 2140744
data/evaluation/caption/requests_paper_um2000.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c6e27dae5461da60e74962e4ec45597534687303530eb82ea60c999895019e2d
3
+ size 37860229
data/evaluation/metric_qa/requests_paper_um2000.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5c6fa80d160c587cc0a2f7da5c92fd8aa1472aaac40ac45be487fcb10f1093df
3
+ size 33807874
data/evaluation/tsqa/requests_clean3264.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc48b3447fb8d52b1f6d90a7770ba9d0f8e57ae61225f99a8c43cf9795f259b7
3
+ size 3055430
data/evaluation/tsqa/series_overlap_audit.json ADDED
@@ -0,0 +1,297 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "method": "SHA-256 of the complete decoded numeric array after canonicalizing all numbers to float and normalizing signed zero",
3
+ "historical_signature_method": "MD5(len, round(first 10 values, 3), round(last 10 values, 3), round(sum, 2)); reproduced from changelog/2026-06-25/commands.md",
4
+ "train_file": "/home/__ts-align_project_home/oos_eval_results/Benchmark_eval/TSQA/train/raw_values.jsonl",
5
+ "train_file_sha256": "03356fbed51efdea3a450c8965ae4818b838234c25da7bfeed636418b01dcf8c",
6
+ "train_sft_file": "/home/__ts-align_project_home/oos_eval_results/Benchmark_eval/TSQA/train/sft_pairs.jsonl",
7
+ "train_sft_file_sha256": "ec0d94ac0ada0511be687f29f3f16f829fb2325374ca5285815db530a6cf0f57",
8
+ "test_file": "/home/__ts-align_project_home/oos_eval_results/Benchmark_eval/TSQA/test_requests.jsonl",
9
+ "test_file_sha256": "172fe5b97b0c3dafc98424acb08d567dc24415fb953b5aedc8e307aa421a82df",
10
+ "train_rows": 31556,
11
+ "train_unique_series": 30293,
12
+ "test_rows": 3519,
13
+ "test_unique_series": 3498,
14
+ "overlapping_unique_series": 247,
15
+ "affected_test_rows": 255,
16
+ "affected_test_fraction": 0.07246376811594203,
17
+ "affected_test_rows_by_task": {
18
+ "multiple_choice": 64,
19
+ "open_ended_question": 145,
20
+ "true_false": 46
21
+ },
22
+ "remaining_test_rows": 3264,
23
+ "remaining_test_rows_by_task": {
24
+ "multiple_choice": 848,
25
+ "open_ended_question": 1784,
26
+ "true_false": 632
27
+ },
28
+ "historical_signature_affected_test_rows": 255,
29
+ "historical_signature_matches_full_array_ids": true,
30
+ "byte_exact_json_diagnostic": {
31
+ "overlapping_unique_series": 246,
32
+ "affected_test_rows": 254,
33
+ "additional_rows_found_after_numeric_normalization": [
34
+ "timemqa:multiple_choice:35307"
35
+ ],
36
+ "note": "Byte-exact JSON distinguishes 0.0 from -0.0."
37
+ },
38
+ "exact_series_prompt_answer_task_overlap_rows": 0,
39
+ "interpretation": "The old cross-version duplicate-QA contamination is absent, but the current row-level split is not series-disjoint.",
40
+ "affected_test_request_ids": [
41
+ "timemqa:true_false:37384",
42
+ "timemqa:open_ended_question:15125",
43
+ "timemqa:true_false:30785",
44
+ "timemqa:open_ended_question:32967",
45
+ "timemqa:open_ended_question:31322",
46
+ "timemqa:open_ended_question:30897",
47
+ "timemqa:open_ended_question:22934",
48
+ "timemqa:true_false:19461",
49
+ "timemqa:true_false:25125",
50
+ "timemqa:multiple_choice:5780",
51
+ "timemqa:open_ended_question:34399",
52
+ "timemqa:open_ended_question:32424",
53
+ "timemqa:open_ended_question:39",
54
+ "timemqa:multiple_choice:21698",
55
+ "timemqa:open_ended_question:6649",
56
+ "timemqa:open_ended_question:8820",
57
+ "timemqa:multiple_choice:14497",
58
+ "timemqa:open_ended_question:26957",
59
+ "timemqa:multiple_choice:33228",
60
+ "timemqa:open_ended_question:26249",
61
+ "timemqa:open_ended_question:4921",
62
+ "timemqa:multiple_choice:13096",
63
+ "timemqa:multiple_choice:751",
64
+ "timemqa:multiple_choice:6836",
65
+ "timemqa:open_ended_question:24626",
66
+ "timemqa:true_false:10243",
67
+ "timemqa:multiple_choice:35470",
68
+ "timemqa:open_ended_question:20282",
69
+ "timemqa:multiple_choice:10954",
70
+ "timemqa:open_ended_question:26823",
71
+ "timemqa:multiple_choice:35669",
72
+ "timemqa:open_ended_question:11674",
73
+ "timemqa:multiple_choice:30361",
74
+ "timemqa:open_ended_question:1045",
75
+ "timemqa:open_ended_question:24250",
76
+ "timemqa:true_false:35742",
77
+ "timemqa:multiple_choice:5307",
78
+ "timemqa:true_false:2925",
79
+ "timemqa:open_ended_question:25480",
80
+ "timemqa:open_ended_question:29854",
81
+ "timemqa:multiple_choice:23250",
82
+ "timemqa:open_ended_question:8743",
83
+ "timemqa:multiple_choice:37111",
84
+ "timemqa:multiple_choice:33901",
85
+ "timemqa:open_ended_question:8909",
86
+ "timemqa:open_ended_question:2383",
87
+ "timemqa:open_ended_question:25633",
88
+ "timemqa:multiple_choice:12186",
89
+ "timemqa:true_false:29450",
90
+ "timemqa:open_ended_question:7592",
91
+ "timemqa:true_false:10230",
92
+ "timemqa:open_ended_question:11090",
93
+ "timemqa:multiple_choice:10102",
94
+ "timemqa:open_ended_question:31828",
95
+ "timemqa:true_false:11280",
96
+ "timemqa:true_false:24788",
97
+ "timemqa:open_ended_question:925",
98
+ "timemqa:multiple_choice:7424",
99
+ "timemqa:multiple_choice:15588",
100
+ "timemqa:open_ended_question:10910",
101
+ "timemqa:true_false:35831",
102
+ "timemqa:open_ended_question:2879",
103
+ "timemqa:multiple_choice:13982",
104
+ "timemqa:open_ended_question:27452",
105
+ "timemqa:true_false:29734",
106
+ "timemqa:open_ended_question:23775",
107
+ "timemqa:open_ended_question:32370",
108
+ "timemqa:multiple_choice:941",
109
+ "timemqa:open_ended_question:23896",
110
+ "timemqa:open_ended_question:35982",
111
+ "timemqa:multiple_choice:6612",
112
+ "timemqa:multiple_choice:10883",
113
+ "timemqa:open_ended_question:20782",
114
+ "timemqa:true_false:25990",
115
+ "timemqa:open_ended_question:17625",
116
+ "timemqa:open_ended_question:3242",
117
+ "timemqa:open_ended_question:17053",
118
+ "timemqa:open_ended_question:25684",
119
+ "timemqa:true_false:15812",
120
+ "timemqa:open_ended_question:11606",
121
+ "timemqa:open_ended_question:33866",
122
+ "timemqa:multiple_choice:36359",
123
+ "timemqa:open_ended_question:27225",
124
+ "timemqa:open_ended_question:29619",
125
+ "timemqa:open_ended_question:12280",
126
+ "timemqa:open_ended_question:20895",
127
+ "timemqa:open_ended_question:13285",
128
+ "timemqa:open_ended_question:5331",
129
+ "timemqa:open_ended_question:34267",
130
+ "timemqa:open_ended_question:21076",
131
+ "timemqa:true_false:1659",
132
+ "timemqa:open_ended_question:24727",
133
+ "timemqa:open_ended_question:2554",
134
+ "timemqa:open_ended_question:14577",
135
+ "timemqa:multiple_choice:26516",
136
+ "timemqa:open_ended_question:2153",
137
+ "timemqa:true_false:10977",
138
+ "timemqa:open_ended_question:18670",
139
+ "timemqa:true_false:13221",
140
+ "timemqa:multiple_choice:10609",
141
+ "timemqa:multiple_choice:9979",
142
+ "timemqa:open_ended_question:3327",
143
+ "timemqa:open_ended_question:6624",
144
+ "timemqa:multiple_choice:3269",
145
+ "timemqa:open_ended_question:14322",
146
+ "timemqa:open_ended_question:21720",
147
+ "timemqa:multiple_choice:6638",
148
+ "timemqa:true_false:13541",
149
+ "timemqa:multiple_choice:29228",
150
+ "timemqa:open_ended_question:31486",
151
+ "timemqa:open_ended_question:12772",
152
+ "timemqa:multiple_choice:35307",
153
+ "timemqa:multiple_choice:2239",
154
+ "timemqa:open_ended_question:25380",
155
+ "timemqa:open_ended_question:2209",
156
+ "timemqa:true_false:15236",
157
+ "timemqa:open_ended_question:23653",
158
+ "timemqa:true_false:26808",
159
+ "timemqa:open_ended_question:10736",
160
+ "timemqa:open_ended_question:3421",
161
+ "timemqa:open_ended_question:28978",
162
+ "timemqa:open_ended_question:23491",
163
+ "timemqa:open_ended_question:16134",
164
+ "timemqa:open_ended_question:26227",
165
+ "timemqa:multiple_choice:17225",
166
+ "timemqa:open_ended_question:32231",
167
+ "timemqa:open_ended_question:16686",
168
+ "timemqa:open_ended_question:17746",
169
+ "timemqa:open_ended_question:17332",
170
+ "timemqa:open_ended_question:12841",
171
+ "timemqa:multiple_choice:31600",
172
+ "timemqa:multiple_choice:23013",
173
+ "timemqa:multiple_choice:36567",
174
+ "timemqa:open_ended_question:22198",
175
+ "timemqa:multiple_choice:29646",
176
+ "timemqa:multiple_choice:127",
177
+ "timemqa:open_ended_question:11747",
178
+ "timemqa:true_false:29253",
179
+ "timemqa:open_ended_question:19035",
180
+ "timemqa:open_ended_question:14065",
181
+ "timemqa:true_false:12829",
182
+ "timemqa:open_ended_question:12444",
183
+ "timemqa:true_false:29247",
184
+ "timemqa:true_false:37148",
185
+ "timemqa:multiple_choice:7676",
186
+ "timemqa:open_ended_question:8032",
187
+ "timemqa:multiple_choice:12996",
188
+ "timemqa:open_ended_question:22210",
189
+ "timemqa:open_ended_question:2211",
190
+ "timemqa:multiple_choice:16161",
191
+ "timemqa:open_ended_question:9752",
192
+ "timemqa:true_false:12724",
193
+ "timemqa:true_false:25699",
194
+ "timemqa:open_ended_question:21280",
195
+ "timemqa:true_false:28534",
196
+ "timemqa:true_false:12532",
197
+ "timemqa:multiple_choice:20633",
198
+ "timemqa:open_ended_question:19677",
199
+ "timemqa:multiple_choice:26078",
200
+ "timemqa:open_ended_question:132",
201
+ "timemqa:open_ended_question:12382",
202
+ "timemqa:true_false:18968",
203
+ "timemqa:open_ended_question:27875",
204
+ "timemqa:open_ended_question:710",
205
+ "timemqa:open_ended_question:20819",
206
+ "timemqa:open_ended_question:12815",
207
+ "timemqa:open_ended_question:852",
208
+ "timemqa:open_ended_question:10078",
209
+ "timemqa:open_ended_question:9288",
210
+ "timemqa:true_false:3104",
211
+ "timemqa:open_ended_question:27772",
212
+ "timemqa:open_ended_question:9641",
213
+ "timemqa:true_false:20525",
214
+ "timemqa:true_false:33304",
215
+ "timemqa:multiple_choice:6123",
216
+ "timemqa:open_ended_question:16030",
217
+ "timemqa:true_false:37361",
218
+ "timemqa:open_ended_question:11979",
219
+ "timemqa:open_ended_question:12670",
220
+ "timemqa:open_ended_question:30455",
221
+ "timemqa:open_ended_question:21938",
222
+ "timemqa:open_ended_question:31547",
223
+ "timemqa:multiple_choice:22065",
224
+ "timemqa:open_ended_question:33814",
225
+ "timemqa:true_false:33066",
226
+ "timemqa:multiple_choice:18222",
227
+ "timemqa:multiple_choice:37311",
228
+ "timemqa:multiple_choice:36095",
229
+ "timemqa:open_ended_question:26757",
230
+ "timemqa:multiple_choice:8357",
231
+ "timemqa:open_ended_question:29715",
232
+ "timemqa:multiple_choice:18565",
233
+ "timemqa:open_ended_question:36903",
234
+ "timemqa:open_ended_question:5491",
235
+ "timemqa:multiple_choice:31392",
236
+ "timemqa:open_ended_question:1840",
237
+ "timemqa:open_ended_question:2079",
238
+ "timemqa:open_ended_question:17171",
239
+ "timemqa:open_ended_question:15211",
240
+ "timemqa:open_ended_question:5032",
241
+ "timemqa:open_ended_question:27824",
242
+ "timemqa:open_ended_question:28150",
243
+ "timemqa:multiple_choice:28237",
244
+ "timemqa:open_ended_question:18384",
245
+ "timemqa:true_false:29530",
246
+ "timemqa:multiple_choice:31041",
247
+ "timemqa:multiple_choice:4794",
248
+ "timemqa:open_ended_question:32120",
249
+ "timemqa:open_ended_question:1745",
250
+ "timemqa:open_ended_question:12783",
251
+ "timemqa:open_ended_question:1424",
252
+ "timemqa:open_ended_question:29169",
253
+ "timemqa:true_false:22197",
254
+ "timemqa:true_false:12219",
255
+ "timemqa:open_ended_question:13424",
256
+ "timemqa:open_ended_question:22515",
257
+ "timemqa:open_ended_question:12802",
258
+ "timemqa:multiple_choice:23322",
259
+ "timemqa:true_false:14191",
260
+ "timemqa:multiple_choice:26493",
261
+ "timemqa:open_ended_question:10904",
262
+ "timemqa:multiple_choice:16059",
263
+ "timemqa:true_false:36112",
264
+ "timemqa:multiple_choice:20091",
265
+ "timemqa:true_false:28449",
266
+ "timemqa:true_false:21153",
267
+ "timemqa:true_false:34408",
268
+ "timemqa:multiple_choice:17569",
269
+ "timemqa:open_ended_question:17179",
270
+ "timemqa:open_ended_question:18623",
271
+ "timemqa:open_ended_question:18089",
272
+ "timemqa:open_ended_question:14049",
273
+ "timemqa:true_false:29064",
274
+ "timemqa:open_ended_question:20526",
275
+ "timemqa:open_ended_question:10272",
276
+ "timemqa:open_ended_question:3477",
277
+ "timemqa:multiple_choice:23311",
278
+ "timemqa:open_ended_question:23875",
279
+ "timemqa:open_ended_question:18468",
280
+ "timemqa:multiple_choice:36747",
281
+ "timemqa:open_ended_question:449",
282
+ "timemqa:open_ended_question:3954",
283
+ "timemqa:multiple_choice:7748",
284
+ "timemqa:true_false:37171",
285
+ "timemqa:open_ended_question:28080",
286
+ "timemqa:true_false:26314",
287
+ "timemqa:open_ended_question:32828",
288
+ "timemqa:open_ended_question:16754",
289
+ "timemqa:open_ended_question:34418",
290
+ "timemqa:open_ended_question:26397",
291
+ "timemqa:open_ended_question:12418",
292
+ "timemqa:multiple_choice:18144",
293
+ "timemqa:multiple_choice:26533",
294
+ "timemqa:open_ended_question:21038",
295
+ "timemqa:open_ended_question:63"
296
+ ]
297
+ }
data/training/ablation/b0_joint_sft_caption_prompt_v2/_audit.json ADDED
@@ -0,0 +1,148 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "builder": {
3
+ "path": "/root/autodl-tmp/_REPLICATION/caption_prompt_retrain_scheduler/materialize_corrected_manifests.py",
4
+ "sha256": "0f45f6b2acad734270d577f129bf39808977db1a01cff403d6aed288610f66ad"
5
+ },
6
+ "byte_identical_non_caption_rows": 59089,
7
+ "caption_system_prompt": {
8
+ "sha256": "369193c6795360c010a1975d88dc15ae0e00460d07895bc1e48b01201972489c",
9
+ "value": "你是专业的时间序列分析助手，请仅根据给定时间序列完成分析。"
10
+ },
11
+ "changed_caption_rows": 27492,
12
+ "created_at_utc": "2026-07-25T03:17:32.501477Z",
13
+ "d1_is_ordered_filtered_joint_subset": true,
14
+ "d1_source_id_overlap": {
15
+ "train:test": 297,
16
+ "train:val": 328,
17
+ "val:test": 5
18
+ },
19
+ "d1_unique_source_ids": 46612,
20
+ "families": {
21
+ "d1": {
22
+ "splits": {
23
+ "test": {
24
+ "rows": 1085,
25
+ "sha256": "ea096a38a87984b82a7f64e08ffed86117fad2626584e9dd91f7a891e5f6db6e",
26
+ "size_bytes": 1661100,
27
+ "task_counts": {
28
+ "caption": 527,
29
+ "metric_qa": 558
30
+ }
31
+ },
32
+ "train": {
33
+ "rows": 52832,
34
+ "sha256": "51b13eae32bf7b952202914c60c2a66c6848c380a660b52658f29ccfdcacca7d",
35
+ "size_bytes": 81166810,
36
+ "task_counts": {
37
+ "caption": 26416,
38
+ "metric_qa": 26416
39
+ }
40
+ },
41
+ "val": {
42
+ "rows": 1108,
43
+ "sha256": "3562fa50b0822674c4f07bde8d66e51959822155040105ab52a7b8f1a9a4c069",
44
+ "size_bytes": 1734079,
45
+ "task_counts": {
46
+ "caption": 549,
47
+ "metric_qa": 559
48
+ }
49
+ }
50
+ },
51
+ "unique_record_keys": 55025
52
+ },
53
+ "joint": {
54
+ "splits": {
55
+ "test": {
56
+ "rows": 1731,
57
+ "sha256": "ca12249e286c40e22d2ebcf9ecf58d779829c40f03c8a0a248fb997e1d0cac21",
58
+ "size_bytes": 2075385,
59
+ "task_counts": {
60
+ "caption": 527,
61
+ "choice": 311,
62
+ "metric_qa": 558,
63
+ "text": 335
64
+ }
65
+ },
66
+ "train": {
67
+ "rows": 83118,
68
+ "sha256": "451aade90885c464e2130b15ff3ca7bea4b1b1757e082afbc33ae882de8832be",
69
+ "size_bytes": 100698993,
70
+ "task_counts": {
71
+ "caption": 26416,
72
+ "choice": 13615,
73
+ "metric_qa": 26416,
74
+ "text": 16671
75
+ }
76
+ },
77
+ "val": {
78
+ "rows": 1732,
79
+ "sha256": "cd6bee7a6d4e7b29e114f1d47dc330154eedb61e5704d6004f0899a337ff9766",
80
+ "size_bytes": 2135521,
81
+ "task_counts": {
82
+ "caption": 549,
83
+ "choice": 284,
84
+ "metric_qa": 559,
85
+ "text": 340
86
+ }
87
+ }
88
+ },
89
+ "unique_record_keys": 86581
90
+ }
91
+ },
92
+ "host": "autodl-container-60n7t62jzr-d22d8646",
93
+ "key_map": {
94
+ "path": "caption_record_key_map.jsonl",
95
+ "rows": 27492,
96
+ "sha256": "6c95e34598117dc86fa26f893b446b96d0448deb83d99e601e1e7e2078c24f07",
97
+ "size_bytes": 7825338
98
+ },
99
+ "only_changed_fields": [
100
+ "system_prompt",
101
+ "record_key"
102
+ ],
103
+ "output_d1": "/root/autodl-tmp/_TRAIN_DATA/ablation/d1_ours_sft_caption_prompt_v2",
104
+ "output_joint": "/root/autodl-tmp/_TRAIN_DATA/ablation/b0_joint_sft_caption_prompt_v2",
105
+ "record_key_reason": "record_key identity includes system_prompt",
106
+ "schema_version": 1,
107
+ "source_d1": "/root/autodl-tmp/_TRAIN_DATA/ablation/d1_ours_sft",
108
+ "source_joint": "/root/autodl-tmp/_TRAIN_DATA/ablation/b0_joint_sft",
109
+ "split_builds": {
110
+ "test": {
111
+ "byte_identical_non_caption_rows": 1204,
112
+ "changed_caption_rows": 527,
113
+ "d1_rows": 1085,
114
+ "joint_rows": 1731,
115
+ "task_counts": {
116
+ "caption": 527,
117
+ "choice": 311,
118
+ "metric_qa": 558,
119
+ "text": 335
120
+ }
121
+ },
122
+ "train": {
123
+ "byte_identical_non_caption_rows": 56702,
124
+ "changed_caption_rows": 26416,
125
+ "d1_rows": 52832,
126
+ "joint_rows": 83118,
127
+ "task_counts": {
128
+ "caption": 26416,
129
+ "choice": 13615,
130
+ "metric_qa": 26416,
131
+ "text": 16671
132
+ }
133
+ },
134
+ "val": {
135
+ "byte_identical_non_caption_rows": 1183,
136
+ "changed_caption_rows": 549,
137
+ "d1_rows": 1108,
138
+ "joint_rows": 1732,
139
+ "task_counts": {
140
+ "caption": 549,
141
+ "choice": 284,
142
+ "metric_qa": 559,
143
+ "text": 340
144
+ }
145
+ }
146
+ },
147
+ "status": "verified"
148
+ }
data/training/ablation/b0_joint_sft_caption_prompt_v2/caption_record_key_map.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c95e34598117dc86fa26f893b446b96d0448deb83d99e601e1e7e2078c24f07
3
+ size 7825338
data/training/ablation/b0_joint_sft_caption_prompt_v2/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca12249e286c40e22d2ebcf9ecf58d779829c40f03c8a0a248fb997e1d0cac21
3
+ size 2075385
data/training/ablation/b0_joint_sft_caption_prompt_v2/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:451aade90885c464e2130b15ff3ca7bea4b1b1757e082afbc33ae882de8832be
3
+ size 100698993
data/training/ablation/b0_joint_sft_caption_prompt_v2/val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd6bee7a6d4e7b29e114f1d47dc330154eedb61e5704d6004f0899a337ff9766
3
+ size 2135521
data/training/ablation/caption_system_prompt_ablation_v1/PROVENANCE.json ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema_version": 1,
3
+ "created_at_utc": "2026-07-24T09:09:48.688171Z",
4
+ "host": "autodl-container-993f498a74-03c7321a",
5
+ "source": {
6
+ "requested_path": "/root/autodl-tmp/_TRAIN_DATA/data_joint_tsqa/sft_pairs.jsonl",
7
+ "resolved_path": "/root/autodl-tmp/_TRAIN_DATA/data_joint_tsqa/sft_pairs_joint.jsonl",
8
+ "size_bytes": 87147033,
9
+ "sha256": "d48d56deec88a86e489abcfe678426821b17ef8ea6755c3b5743400c6d999b62"
10
+ },
11
+ "output": {
12
+ "path": "/root/autodl-tmp/_TRAIN_DATA/ablation/caption_system_prompt_ablation_v1/sft_pairs_caption_system_fixed.jsonl",
13
+ "size_bytes": 89704626,
14
+ "sha256": "9a13caa7abf15967d4f44778c4c094fb8d3999d360b092e32bdcc7225ab35f37"
15
+ },
16
+ "treatment": {
17
+ "task_type": "caption",
18
+ "field": "system_prompt",
19
+ "value": "你是专业的时间序列分析助手，请仅根据给定时间序列完成分析。"
20
+ },
21
+ "build": {
22
+ "rows": 86593,
23
+ "changed_caption_rows": 27501,
24
+ "task_counts": {
25
+ "caption": 27501,
26
+ "choice": 14210,
27
+ "metric_qa": 27536,
28
+ "text": 17346
29
+ }
30
+ },
31
+ "verification": {
32
+ "rows": 86593,
33
+ "changed_caption_rows": 27501,
34
+ "byte_identical_non_caption_rows": 59092,
35
+ "only_treatment_field_changed": true
36
+ },
37
+ "builder": {
38
+ "path": "/root/autodl-tmp/_REPLICATION/caption_system_prompt_ablation/build_caption_system_fixed_sft.py",
39
+ "sha256": "a98e6d74cc3701542096c9071f3aee779da3955d702d3808b55b6dde96171332"
40
+ }
41
+ }
data/training/ablation/caption_system_prompt_ablation_v1/sft_pairs_caption_system_fixed.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a13caa7abf15967d4f44778c4c094fb8d3999d360b092e32bdcc7225ab35f37
3
+ size 89704626
data/training/ablation/d1_ours_grpo/metrics.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb5de0be361e8687577fcbdde04e023f0dd382af0812ffc403eb24eb99ddb126
3
+ size 71048310
data/training/ablation/d1_ours_grpo/pairs.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:71a8279ffe8d8fdf29c8e510399b73f0b7b1ec8e83842b088c04abf3f758156d
3
+ size 15872632
data/training/ablation/d1_ours_grpo/raw_values.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b8ad32068890ec8069620c739720464ca927e7805582884b3a956081b456e10
3
+ size 366858227
data/training/ablation/d1_ours_sft_caption_prompt_v2/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea096a38a87984b82a7f64e08ffed86117fad2626584e9dd91f7a891e5f6db6e
3
+ size 1661100
data/training/ablation/d1_ours_sft_caption_prompt_v2/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:51b13eae32bf7b952202914c60c2a66c6848c380a660b52658f29ccfdcacca7d
3
+ size 81166810
data/training/ablation/d1_ours_sft_caption_prompt_v2/val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3562fa50b0822674c4f07bde8d66e51959822155040105ab52a7b8f1a9a4c069
3
+ size 1734079
data/training/ablation/d2_tsqa_grpo/metrics.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1bb3ba3cf0af415ce60446ca7887c81a0f05d7a878b869c19835d40eae768fe9
3
+ size 4814370
data/training/ablation/d2_tsqa_grpo/pairs.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0a0a6896e97a7da407bff7aa1e887a603a1691d203e6863d656d424c11a7f0a6
3
+ size 5297538
data/training/ablation/d2_tsqa_grpo/raw_values.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec46aa4b8d5e7ecedc7fca4f49174ff4c5f780b230818d61850b50d07e35be3d
3
+ size 5751460
data/training/ablation/d2_tsqa_sft/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbd7c4ac094a20cc84e5f675fc5c587c8b59d6b1c7da6751cac4473c3b1d7349
3
+ size 414285
data/training/ablation/d2_tsqa_sft/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:43a10199cfb1d551c9ef49fa07e0683801b5eba32ec853993b2ce3e5992ded9e
3
+ size 19532183
data/training/ablation/d2_tsqa_sft/val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f14c068949f577c1543dd247b3dacb363f0ff6c3842941923d35f4ca2361d5d
3
+ size 401442
data/training/ablation/d3_gpt5nano_final_alignment/grpo/metrics.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc01d4a01bd53a3d66f90a168c2f5461be130e19e502481e61e90e1790c782f7
3
+ size 3404019
data/training/ablation/d3_gpt5nano_final_alignment/grpo/pairs.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29aad424c110af3a2d4bc2913323aa3f7f8f9d0e3afdaace1def0e0c2eb5765f
3
+ size 28246646
data/training/ablation/d3_gpt5nano_final_alignment/grpo/raw_values.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c18a3dc1c9be5d6b6d50fc92ca9d666e4bd59204e6b57e3a2f34a50b681943e6
3
+ size 636771616
data/training/ablation/d3_gpt5nano_final_alignment/sft_manifests/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9ddc94e3245e7e164b42692d8faacc3a70ce4ee56aafeb06c057969b79d3fb1c
3
+ size 1581156
data/training/ablation/d3_gpt5nano_final_alignment/sft_manifests/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:619236503adacf558e1f01bc1fd2a336ce3f7f21262f936d52798c3ef56760b1
3
+ size 76930494
data/training/ablation/d3_gpt5nano_final_alignment/sft_manifests/val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8502c116d5551075b0fb3395e61563d9aacc23c0aeb85d117e5b0d419be65169
3
+ size 1621746
data/training/ablation/r3_no_easyfirst/MANIFEST.json ADDED
@@ -0,0 +1,188 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "file_count": 36,
3
+ "files": [
4
+ {
5
+ "path": "PROVENANCE.json",
6
+ "sha256": "16f5ded471a60a62f787d26e7206d1e5bcd30a347b7ac8e0871d40bee3f15711",
7
+ "size_bytes": 22402
8
+ },
9
+ {
10
+ "path": "b_allocation_sidecar.json",
11
+ "sha256": "a9bb3203ad77b9fd63ad125f8fba2d40eaf7de20454b10ca3ec3a60734ddd8d4",
12
+ "size_bytes": 18886
13
+ },
14
+ {
15
+ "path": "b_s1/test.jsonl",
16
+ "sha256": "6f1b6961c2065fcde39fe9aafde3cde7c1dd7b7789a172485156d9f70962a4fc",
17
+ "size_bytes": 826581
18
+ },
19
+ {
20
+ "path": "b_s1/train_epoch_0.jsonl",
21
+ "sha256": "465c69b698ed3686080fa5aafdc8f1fcaeea06ba32b80e3fc315a045b840a3ca",
22
+ "size_bytes": 61933906
23
+ },
24
+ {
25
+ "path": "b_s1/train_epoch_1.jsonl",
26
+ "sha256": "94439b81436df2292cf09dd4bbf16256f34e7660c2a4e688f1ba16d8c335143d",
27
+ "size_bytes": 62018729
28
+ },
29
+ {
30
+ "path": "b_s1/train_epoch_2.jsonl",
31
+ "sha256": "c1ca406ee44dbad90184273dcaa61aaac4531549a63f4fa776fe83f1b2e5af0c",
32
+ "size_bytes": 61825773
33
+ },
34
+ {
35
+ "path": "b_s1/val.jsonl",
36
+ "sha256": "c906af13d05c6f2944e5d53bf5c144be518644fccdfa43ce10cccac3e6e8abac",
37
+ "size_bytes": 821997
38
+ },
39
+ {
40
+ "path": "b_s2/test.jsonl",
41
+ "sha256": "94a3126718b27386d06c248638d143e2de6c2560959b5329d61683048d648187",
42
+ "size_bytes": 1177107
43
+ },
44
+ {
45
+ "path": "b_s2/train_epoch_0.jsonl",
46
+ "sha256": "25f5db8257948a79a0db53169760da6df050aa25f3aaa04dcae595210e7cf670",
47
+ "size_bytes": 35098810
48
+ },
49
+ {
50
+ "path": "b_s2/train_epoch_1.jsonl",
51
+ "sha256": "6f55d2babe7855cc06fbb31aeca6d3c48c19f67cf1dec62aa053e5003a517cdf",
52
+ "size_bytes": 34988718
53
+ },
54
+ {
55
+ "path": "b_s2/val.jsonl",
56
+ "sha256": "b0c5360e96ca41997cdde098d0135925dac4807606f692b5be7d9501e23a7f8f",
57
+ "size_bytes": 1190097
58
+ },
59
+ {
60
+ "path": "m_allocation_sidecar.json",
61
+ "sha256": "1a6cecc4eb02e383d3f40a174a4c8b62a4253403e0c5d0f24e3e3d3c27c39254",
62
+ "size_bytes": 18887
63
+ },
64
+ {
65
+ "path": "m_s1/test.jsonl",
66
+ "sha256": "c2a30603cdb56613a931ef1362d842a4b9924aee1dcb7ac4590b79972ae8b132",
67
+ "size_bytes": 734345
68
+ },
69
+ {
70
+ "path": "m_s1/train_epoch_0.jsonl",
71
+ "sha256": "61b7d5fad7b28cc6169831cd112a709d2d48602f408701b491b4efd76830e40e",
72
+ "size_bytes": 61891787
73
+ },
74
+ {
75
+ "path": "m_s1/train_epoch_1.jsonl",
76
+ "sha256": "fc13648f9cd277dd63c0f49fc7143f016e1d57b2454310f1631e7178ec708a03",
77
+ "size_bytes": 62068142
78
+ },
79
+ {
80
+ "path": "m_s1/train_epoch_2.jsonl",
81
+ "sha256": "82bbd8f3e25230695255d3920c87fefa54545b3175b574652b4090677e7b30d6",
82
+ "size_bytes": 62110992
83
+ },
84
+ {
85
+ "path": "m_s1/val.jsonl",
86
+ "sha256": "9a97913b365119da108dd96ac3e09cc797ed3f6e41e670505c53d213f52c80fc",
87
+ "size_bytes": 733359
88
+ },
89
+ {
90
+ "path": "m_s2/test.jsonl",
91
+ "sha256": "4d2d5425a08fdddaea01cef98369fdfc193d752057a1aa8b11a55d805727c867",
92
+ "size_bytes": 1289680
93
+ },
94
+ {
95
+ "path": "m_s2/train_epoch_0.jsonl",
96
+ "sha256": "a2ac2ff0bfafbecb2b8c9628eb60d60168bb77e2f61a4b425134be4dcb08fd65",
97
+ "size_bytes": 35165030
98
+ },
99
+ {
100
+ "path": "m_s2/train_epoch_1.jsonl",
101
+ "sha256": "1f6cef3dacf382766db5ad26656a79696cb2cbcea9de96facd331b7065458546",
102
+ "size_bytes": 35060589
103
+ },
104
+ {
105
+ "path": "m_s2/val.jsonl",
106
+ "sha256": "a98c43da64518cf924740293c2706829f0a5caab2696a643b70ba545fb6021b8",
107
+ "size_bytes": 1302370
108
+ },
109
+ {
110
+ "path": "replay_s2/sidecar.json",
111
+ "sha256": "659d8f57e3eeaacf06c6aa583343fb27fc9f9e7795257001b06964c3585b249f",
112
+ "size_bytes": 7304
113
+ },
114
+ {
115
+ "path": "replay_s2/test.jsonl",
116
+ "sha256": "75009c6214215d064370a28a05685298b0584737cccb8b8273386e8a29b69253",
117
+ "size_bytes": 3603077
118
+ },
119
+ {
120
+ "path": "replay_s2/train_epoch_0.jsonl",
121
+ "sha256": "f4fb364bb9bf4fce4f488761f19546f6ab47078477336116ddc540a1e1caf25a",
122
+ "size_bytes": 181225280
123
+ },
124
+ {
125
+ "path": "replay_s2/train_epoch_1.jsonl",
126
+ "sha256": "36b2c037353ebdf96c5964a7f6507a504479f8f8f9c58917cfa36ec1f1f666fc",
127
+ "size_bytes": 181233054
128
+ },
129
+ {
130
+ "path": "replay_s2/val.jsonl",
131
+ "sha256": "a155850d45765012f41ecfec074e2c68cba8e27ecf7f454962c900cfce7515b8",
132
+ "size_bytes": 3557633
133
+ },
134
+ {
135
+ "path": "u_allocation_sidecar.json",
136
+ "sha256": "539f008f955621613c5d94c4df666de81b3083a76439992a01ccd47f4e5def2b",
137
+ "size_bytes": 21442
138
+ },
139
+ {
140
+ "path": "u_s1/test.jsonl",
141
+ "sha256": "f05c4c793fa88fc836a618c7c4f1ae047a5b4cb6434d6c4d78746a12c4394c4c",
142
+ "size_bytes": 658165
143
+ },
144
+ {
145
+ "path": "u_s1/train_epoch_0.jsonl",
146
+ "sha256": "6c82af5bd54a872668b557073f1f6b220c15296d213406e4a964e3bd038723e5",
147
+ "size_bytes": 51184761
148
+ },
149
+ {
150
+ "path": "u_s1/train_epoch_1.jsonl",
151
+ "sha256": "49e490e9f51d68b93c655233f42a210a2b12696c613e7baac01b76e329f1c8fe",
152
+ "size_bytes": 51251747
153
+ },
154
+ {
155
+ "path": "u_s1/train_epoch_2.jsonl",
156
+ "sha256": "cd658e5e523f0fc2cc9edf770bb0ddf03e28c175dc7e9a9be9f5faecf68b3d08",
157
+ "size_bytes": 51203535
158
+ },
159
+ {
160
+ "path": "u_s1/val.jsonl",
161
+ "sha256": "c5f77cd8c087549103d98a84492b7c08d22d048c2292c3b0d2f4cf4f8bdf228f",
162
+ "size_bytes": 655994
163
+ },
164
+ {
165
+ "path": "u_s2/test.jsonl",
166
+ "sha256": "4c9298ebaf26eaf67b4b9846fca3aa2908b8af91251cd51aea8f77c95ef2bfa7",
167
+ "size_bytes": 987438
168
+ },
169
+ {
170
+ "path": "u_s2/train_epoch_0.jsonl",
171
+ "sha256": "2c01758ba6ded6dcb139e30dd675a52db873d424b36bc03918aea6c9d00f8f91",
172
+ "size_bytes": 29634849
173
+ },
174
+ {
175
+ "path": "u_s2/train_epoch_1.jsonl",
176
+ "sha256": "53418e22ba10dacab3240a8cceaf982830bb4573ba93290fa057ae5f6ae9768d",
177
+ "size_bytes": 29714896
178
+ },
179
+ {
180
+ "path": "u_s2/val.jsonl",
181
+ "sha256": "4d05ac659f17bbec5a0d19a517fed196ac1997362472655f86447b759d758cd8",
182
+ "size_bytes": 968431
183
+ }
184
+ ],
185
+ "schema_version": 1,
186
+ "status": "complete",
187
+ "total_size_bytes": 1106205793
188
+ }
data/training/ablation/r3_no_easyfirst/PROVENANCE.json ADDED
@@ -0,0 +1,621 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "argv": [
3
+ "build_r3_data_once.py"
4
+ ],
5
+ "assertions": {
6
+ "all_exact_multiset_checks_pass": true,
7
+ "all_record_keys_globally_unique": true,
8
+ "all_sampler_padding_checks_pass": true,
9
+ "all_target_ownership_checks_pass": true,
10
+ "frozen_loader_hashes_match": true,
11
+ "input_line_counts_match": true
12
+ },
13
+ "code": {
14
+ "expected_hashes": {
15
+ "data.py": "b6a0de2c342e59262953a3ca91ccc65f643f363293690800515186f612596d14",
16
+ "qa_templates.py": "491da1c8d0b4be057dd82efbe6e591e0db79013129a72390d0d6e596da000b36",
17
+ "train_stage1.py": "74ce9e231cfc6ba8c4f0251aa29d7e813e0594bd03f6aa407712ce8353eb4bd9"
18
+ },
19
+ "files": {
20
+ "data.py": {
21
+ "path": "/root/autodl-tmp/_REPLICATION/multivar/src/ts_align/data.py",
22
+ "sha256": "b6a0de2c342e59262953a3ca91ccc65f643f363293690800515186f612596d14",
23
+ "size_bytes": 71827
24
+ },
25
+ "qa_templates.py": {
26
+ "path": "/root/autodl-tmp/_REPLICATION/multivar/src/ts_align/qa_templates.py",
27
+ "sha256": "491da1c8d0b4be057dd82efbe6e591e0db79013129a72390d0d6e596da000b36",
28
+ "size_bytes": 125687
29
+ },
30
+ "train_stage1.py": {
31
+ "path": "/root/autodl-tmp/_REPLICATION/multivar/src/ts_align/train_stage1.py",
32
+ "sha256": "74ce9e231cfc6ba8c4f0251aa29d7e813e0594bd03f6aa407712ce8353eb4bd9",
33
+ "size_bytes": 81413
34
+ }
35
+ },
36
+ "git": {
37
+ "commit": null,
38
+ "dirty": null,
39
+ "repository": null
40
+ },
41
+ "materializer": {
42
+ "path": "/root/autodl-tmp/_REPLICATION/ablation_data_once/build_r3_data_once.py",
43
+ "sha256": "6f13e3a95e0279be839c70484939fd5675a72bce490e04ea8b17143dacb8a268",
44
+ "size_bytes": 69590
45
+ }
46
+ },
47
+ "completed_at_utc": "2026-07-21T16:36:25.739411+00:00",
48
+ "host": "autodl-container-60n7t62jzr-d22d8646",
49
+ "output_summary": {
50
+ "global_unique_record_keys": 1463746,
51
+ "jsonl": [
52
+ {
53
+ "ordered_semantic_sha256": "82c1fcde99eef76b5b401cf47fb7c439f3e184ee04fb87cd3d66822afd8dd171",
54
+ "relative_path": "u_s1/train_epoch_0.jsonl",
55
+ "rows": 95950,
56
+ "sha256": "6c82af5bd54a872668b557073f1f6b220c15296d213406e4a964e3bd038723e5",
57
+ "size_bytes": 51184761,
58
+ "target_level_counts": {
59
+ "level_1": 39411,
60
+ "level_2": 39386,
61
+ "level_3": 10344,
62
+ "level_4": 6809
63
+ },
64
+ "unique_source_ids": 43204
65
+ },
66
+ {
67
+ "ordered_semantic_sha256": "eb7f93a34e5e233c1b5e17bc1fb85de863d71564d30f4c8a04b4bc40f11c8718",
68
+ "relative_path": "u_s1/train_epoch_1.jsonl",
69
+ "rows": 95950,
70
+ "sha256": "49e490e9f51d68b93c655233f42a210a2b12696c613e7baac01b76e329f1c8fe",
71
+ "size_bytes": 51251747,
72
+ "target_level_counts": {
73
+ "level_1": 39412,
74
+ "level_2": 39385,
75
+ "level_3": 10344,
76
+ "level_4": 6809
77
+ },
78
+ "unique_source_ids": 43260
79
+ },
80
+ {
81
+ "ordered_semantic_sha256": "65f69cc918a143397c53b76df41bd46b424d24f0a35a3e24207626b654e17fd3",
82
+ "relative_path": "u_s1/train_epoch_2.jsonl",
83
+ "rows": 95950,
84
+ "sha256": "cd658e5e523f0fc2cc9edf770bb0ddf03e28c175dc7e9a9be9f5faecf68b3d08",
85
+ "size_bytes": 51203535,
86
+ "target_level_counts": {
87
+ "level_1": 39412,
88
+ "level_2": 39386,
89
+ "level_3": 10344,
90
+ "level_4": 6808
91
+ },
92
+ "unique_source_ids": 43159
93
+ },
94
+ {
95
+ "ordered_semantic_sha256": "2fb189dd0c7f0ded2e4d82c409cface6e35d1fe650e6bd3cbc1eb5f3be5b5ff3",
96
+ "relative_path": "u_s2/train_epoch_0.jsonl",
97
+ "rows": 47975,
98
+ "sha256": "2c01758ba6ded6dcb139e30dd675a52db873d424b36bc03918aea6c9d00f8f91",
99
+ "size_bytes": 29634849,
100
+ "target_level_counts": {
101
+ "level_1": 19706,
102
+ "level_2": 19693,
103
+ "level_3": 5172,
104
+ "level_4": 3404
105
+ },
106
+ "unique_source_ids": 31734
107
+ },
108
+ {
109
+ "ordered_semantic_sha256": "e3c403c6a1bf67dc22abfb7e9ba59d1c11ff36fc5224a66a42ff303005e60435",
110
+ "relative_path": "u_s2/train_epoch_1.jsonl",
111
+ "rows": 47975,
112
+ "sha256": "53418e22ba10dacab3240a8cceaf982830bb4573ba93290fa057ae5f6ae9768d",
113
+ "size_bytes": 29714896,
114
+ "target_level_counts": {
115
+ "level_1": 19705,
116
+ "level_2": 19693,
117
+ "level_3": 5172,
118
+ "level_4": 3405
119
+ },
120
+ "unique_source_ids": 31662
121
+ },
122
+ {
123
+ "ordered_semantic_sha256": "90b64e117d6f8efcc2ee6057fb5a1d1cf16eeaf8799040582db42522d4f39fde",
124
+ "relative_path": "u_s1/val.jsonl",
125
+ "rows": 2000,
126
+ "sha256": "c5f77cd8c087549103d98a84492b7c08d22d048c2292c3b0d2f4cf4f8bdf228f",
127
+ "size_bytes": 655994,
128
+ "target_level_counts": {
129
+ "level_1": 1000,
130
+ "level_2": 1000,
131
+ "level_3": 0,
132
+ "level_4": 0
133
+ },
134
+ "unique_source_ids": 1000
135
+ },
136
+ {
137
+ "ordered_semantic_sha256": "68838bdd2e6fa91a0051f1a657d5d35f33f495221f0af8fa7544eca6d9d6cc40",
138
+ "relative_path": "u_s1/test.jsonl",
139
+ "rows": 1998,
140
+ "sha256": "f05c4c793fa88fc836a618c7c4f1ae047a5b4cb6434d6c4d78746a12c4394c4c",
141
+ "size_bytes": 658165,
142
+ "target_level_counts": {
143
+ "level_1": 999,
144
+ "level_2": 999,
145
+ "level_3": 0,
146
+ "level_4": 0
147
+ },
148
+ "unique_source_ids": 999
149
+ },
150
+ {
151
+ "ordered_semantic_sha256": "278b4ab99ecc0fe839027eaa8011e529f2ca80f970f395841d762cdff741ae74",
152
+ "relative_path": "u_s2/val.jsonl",
153
+ "rows": 1000,
154
+ "sha256": "4d05ac659f17bbec5a0d19a517fed196ac1997362472655f86447b759d758cd8",
155
+ "size_bytes": 968431,
156
+ "target_level_counts": {
157
+ "level_1": 158,
158
+ "level_2": 143,
159
+ "level_3": 434,
160
+ "level_4": 265
161
+ },
162
+ "unique_source_ids": 1000
163
+ },
164
+ {
165
+ "ordered_semantic_sha256": "14da21355cd5cbdaf3854524fce1d8369433566da1b0c3469bc25cb54d4c7f0d",
166
+ "relative_path": "u_s2/test.jsonl",
167
+ "rows": 999,
168
+ "sha256": "4c9298ebaf26eaf67b4b9846fca3aa2908b8af91251cd51aea8f77c95ef2bfa7",
169
+ "size_bytes": 987438,
170
+ "target_level_counts": {
171
+ "level_1": 157,
172
+ "level_2": 143,
173
+ "level_3": 434,
174
+ "level_4": 265
175
+ },
176
+ "unique_source_ids": 999
177
+ },
178
+ {
179
+ "ordered_semantic_sha256": "4ff45f94bea469c11f082ae9a3dc72da774cad7276873e8e2c9da5eb964c9719",
180
+ "relative_path": "b_s1/train_epoch_0.jsonl",
181
+ "rows": 96000,
182
+ "sha256": "465c69b698ed3686080fa5aafdc8f1fcaeea06ba32b80e3fc315a045b840a3ca",
183
+ "size_bytes": 61933906,
184
+ "target_level_counts": {
185
+ "level_1": 39433,
186
+ "level_2": 39406,
187
+ "level_3": 10349,
188
+ "level_4": 6812
189
+ },
190
+ "unique_source_ids": 43229
191
+ },
192
+ {
193
+ "ordered_semantic_sha256": "60f9b90fc0fd5554729286c373c8820274c9cc11600b851e4363d33ef1f276c3",
194
+ "relative_path": "b_s1/train_epoch_1.jsonl",
195
+ "rows": 96000,
196
+ "sha256": "94439b81436df2292cf09dd4bbf16256f34e7660c2a4e688f1ba16d8c335143d",
197
+ "size_bytes": 62018729,
198
+ "target_level_counts": {
199
+ "level_1": 39432,
200
+ "level_2": 39406,
201
+ "level_3": 10350,
202
+ "level_4": 6812
203
+ },
204
+ "unique_source_ids": 43246
205
+ },
206
+ {
207
+ "ordered_semantic_sha256": "c6c8070c205ea7134f07dd92c0ea7d54203c5e52ce43be40ad1477f292ec90db",
208
+ "relative_path": "b_s1/train_epoch_2.jsonl",
209
+ "rows": 96000,
210
+ "sha256": "c1ca406ee44dbad90184273dcaa61aaac4531549a63f4fa776fe83f1b2e5af0c",
211
+ "size_bytes": 61825773,
212
+ "target_level_counts": {
213
+ "level_1": 39433,
214
+ "level_2": 39406,
215
+ "level_3": 10350,
216
+ "level_4": 6811
217
+ },
218
+ "unique_source_ids": 43206
219
+ },
220
+ {
221
+ "ordered_semantic_sha256": "252f9b7d45a46e1a454677a936facaa97bca019cf7a9d4fd20b974720d67f7db",
222
+ "relative_path": "b_s2/train_epoch_0.jsonl",
223
+ "rows": 48000,
224
+ "sha256": "25f5db8257948a79a0db53169760da6df050aa25f3aaa04dcae595210e7cf670",
225
+ "size_bytes": 35098810,
226
+ "target_level_counts": {
227
+ "level_1": 19717,
228
+ "level_2": 19703,
229
+ "level_3": 5174,
230
+ "level_4": 3406
231
+ },
232
+ "unique_source_ids": 31668
233
+ },
234
+ {
235
+ "ordered_semantic_sha256": "9eeb5b28d742eccee5d57598d2c74bdfcb8eb78151403c542035206b9d65f68a",
236
+ "relative_path": "b_s2/train_epoch_1.jsonl",
237
+ "rows": 48000,
238
+ "sha256": "6f55d2babe7855cc06fbb31aeca6d3c48c19f67cf1dec62aa053e5003a517cdf",
239
+ "size_bytes": 34988718,
240
+ "target_level_counts": {
241
+ "level_1": 19716,
242
+ "level_2": 19703,
243
+ "level_3": 5175,
244
+ "level_4": 3406
245
+ },
246
+ "unique_source_ids": 31722
247
+ },
248
+ {
249
+ "ordered_semantic_sha256": "5aeb97061a0bfbd039732730bc63d725f4bcdd056daa8defff6b1708a8805b78",
250
+ "relative_path": "b_s1/val.jsonl",
251
+ "rows": 2000,
252
+ "sha256": "c906af13d05c6f2944e5d53bf5c144be518644fccdfa43ce10cccac3e6e8abac",
253
+ "size_bytes": 821997,
254
+ "target_level_counts": {
255
+ "level_1": 1000,
256
+ "level_2": 1000,
257
+ "level_3": 0,
258
+ "level_4": 0
259
+ },
260
+ "unique_source_ids": 1000
261
+ },
262
+ {
263
+ "ordered_semantic_sha256": "5859d048aa7ea27368695ff70cfc425a248cf17536484026c22d57c125de5a91",
264
+ "relative_path": "b_s1/test.jsonl",
265
+ "rows": 2000,
266
+ "sha256": "6f1b6961c2065fcde39fe9aafde3cde7c1dd7b7789a172485156d9f70962a4fc",
267
+ "size_bytes": 826581,
268
+ "target_level_counts": {
269
+ "level_1": 1000,
270
+ "level_2": 1000,
271
+ "level_3": 0,
272
+ "level_4": 0
273
+ },
274
+ "unique_source_ids": 1000
275
+ },
276
+ {
277
+ "ordered_semantic_sha256": "b345a78ea5e84c27ae179054553b076620768ac2ac8f466ff547e2c801d0230e",
278
+ "relative_path": "b_s2/val.jsonl",
279
+ "rows": 1000,
280
+ "sha256": "b0c5360e96ca41997cdde098d0135925dac4807606f692b5be7d9501e23a7f8f",
281
+ "size_bytes": 1190097,
282
+ "target_level_counts": {
283
+ "level_1": 158,
284
+ "level_2": 143,
285
+ "level_3": 434,
286
+ "level_4": 265
287
+ },
288
+ "unique_source_ids": 1000
289
+ },
290
+ {
291
+ "ordered_semantic_sha256": "c2de27e9f58cc4f4be4a75cb8404619e850392c59f0577394aaa62ffbe632e06",
292
+ "relative_path": "b_s2/test.jsonl",
293
+ "rows": 1000,
294
+ "sha256": "94a3126718b27386d06c248638d143e2de6c2560959b5329d61683048d648187",
295
+ "size_bytes": 1177107,
296
+ "target_level_counts": {
297
+ "level_1": 157,
298
+ "level_2": 143,
299
+ "level_3": 434,
300
+ "level_4": 266
301
+ },
302
+ "unique_source_ids": 1000
303
+ },
304
+ {
305
+ "ordered_semantic_sha256": "78fd620aae0e55ff62bced6805bfd7770422cc1434b83905cd9748eaf20162a2",
306
+ "relative_path": "m_s1/train_epoch_0.jsonl",
307
+ "rows": 96000,
308
+ "sha256": "61b7d5fad7b28cc6169831cd112a709d2d48602f408701b491b4efd76830e40e",
309
+ "size_bytes": 61891787,
310
+ "target_level_counts": {
311
+ "level_1": 39433,
312
+ "level_2": 39406,
313
+ "level_3": 10349,
314
+ "level_4": 6812
315
+ },
316
+ "unique_source_ids": 43141
317
+ },
318
+ {
319
+ "ordered_semantic_sha256": "82e85bf148ba83df46ffa207d42a3ed01e9339451f8a4b93d495374b29d6873b",
320
+ "relative_path": "m_s1/train_epoch_1.jsonl",
321
+ "rows": 96000,
322
+ "sha256": "fc13648f9cd277dd63c0f49fc7143f016e1d57b2454310f1631e7178ec708a03",
323
+ "size_bytes": 62068142,
324
+ "target_level_counts": {
325
+ "level_1": 39432,
326
+ "level_2": 39406,
327
+ "level_3": 10350,
328
+ "level_4": 6812
329
+ },
330
+ "unique_source_ids": 43229
331
+ },
332
+ {
333
+ "ordered_semantic_sha256": "61749cbff3d1919c25940b9aef967e17b674d37e7818b461e810aa2a09a7bc15",
334
+ "relative_path": "m_s1/train_epoch_2.jsonl",
335
+ "rows": 96000,
336
+ "sha256": "82bbd8f3e25230695255d3920c87fefa54545b3175b574652b4090677e7b30d6",
337
+ "size_bytes": 62110992,
338
+ "target_level_counts": {
339
+ "level_1": 39433,
340
+ "level_2": 39406,
341
+ "level_3": 10350,
342
+ "level_4": 6811
343
+ },
344
+ "unique_source_ids": 43165
345
+ },
346
+ {
347
+ "ordered_semantic_sha256": "5d47c8a487dac53f2102708bf74e866665b62387cd70251f1656d54784402c51",
348
+ "relative_path": "m_s2/train_epoch_0.jsonl",
349
+ "rows": 48000,
350
+ "sha256": "a2ac2ff0bfafbecb2b8c9628eb60d60168bb77e2f61a4b425134be4dcb08fd65",
351
+ "size_bytes": 35165030,
352
+ "target_level_counts": {
353
+ "level_1": 19717,
354
+ "level_2": 19703,
355
+ "level_3": 5174,
356
+ "level_4": 3406
357
+ },
358
+ "unique_source_ids": 31686
359
+ },
360
+ {
361
+ "ordered_semantic_sha256": "493efb22a0cf6c370cd48343622fe27d6ad1b649997e6f69ab5475502b1e831d",
362
+ "relative_path": "m_s2/train_epoch_1.jsonl",
363
+ "rows": 48000,
364
+ "sha256": "1f6cef3dacf382766db5ad26656a79696cb2cbcea9de96facd331b7065458546",
365
+ "size_bytes": 35060589,
366
+ "target_level_counts": {
367
+ "level_1": 19716,
368
+ "level_2": 19703,
369
+ "level_3": 5175,
370
+ "level_4": 3406
371
+ },
372
+ "unique_source_ids": 31621
373
+ },
374
+ {
375
+ "ordered_semantic_sha256": "8454d70177d057b630c72bd284593bc8f819c4d37695eb452f14fa99d5143bfc",
376
+ "relative_path": "m_s1/val.jsonl",
377
+ "rows": 2000,
378
+ "sha256": "9a97913b365119da108dd96ac3e09cc797ed3f6e41e670505c53d213f52c80fc",
379
+ "size_bytes": 733359,
380
+ "target_level_counts": {
381
+ "level_1": 1000,
382
+ "level_2": 1000,
383
+ "level_3": 0,
384
+ "level_4": 0
385
+ },
386
+ "unique_source_ids": 1000
387
+ },
388
+ {
389
+ "ordered_semantic_sha256": "18d97278af66ba06159e692cbe481be8028b57567c6636719e70df692edb58b0",
390
+ "relative_path": "m_s1/test.jsonl",
391
+ "rows": 2000,
392
+ "sha256": "c2a30603cdb56613a931ef1362d842a4b9924aee1dcb7ac4590b79972ae8b132",
393
+ "size_bytes": 734345,
394
+ "target_level_counts": {
395
+ "level_1": 1000,
396
+ "level_2": 1000,
397
+ "level_3": 0,
398
+ "level_4": 0
399
+ },
400
+ "unique_source_ids": 1000
401
+ },
402
+ {
403
+ "ordered_semantic_sha256": "3756ecd3de9cae0fbd2de757b5bab762be31c0f79506024925572ecf617b1a0c",
404
+ "relative_path": "m_s2/val.jsonl",
405
+ "rows": 1000,
406
+ "sha256": "a98c43da64518cf924740293c2706829f0a5caab2696a643b70ba545fb6021b8",
407
+ "size_bytes": 1302370,
408
+ "target_level_counts": {
409
+ "level_1": 158,
410
+ "level_2": 143,
411
+ "level_3": 434,
412
+ "level_4": 265
413
+ },
414
+ "unique_source_ids": 1000
415
+ },
416
+ {
417
+ "ordered_semantic_sha256": "451bb07883812e5a2bea12c082c7bebd3312aa650c35889f0990e6d54cc1cac1",
418
+ "relative_path": "m_s2/test.jsonl",
419
+ "rows": 1000,
420
+ "sha256": "4d2d5425a08fdddaea01cef98369fdfc193d752057a1aa8b11a55d805727c867",
421
+ "size_bytes": 1289680,
422
+ "target_level_counts": {
423
+ "level_1": 157,
424
+ "level_2": 143,
425
+ "level_3": 434,
426
+ "level_4": 266
427
+ },
428
+ "unique_source_ids": 1000
429
+ },
430
+ {
431
+ "ordered_semantic_sha256": "c25bac4dbd6859c5b26f7331c4df018da14b7c78902b6b9263b0ebafb927f782",
432
+ "relative_path": "replay_s2/train_epoch_0.jsonl",
433
+ "rows": 143975,
434
+ "sha256": "f4fb364bb9bf4fce4f488761f19546f6ab47078477336116ddc540a1e1caf25a",
435
+ "size_bytes": 181225280,
436
+ "target_level_counts": {
437
+ "level_1": 20572,
438
+ "level_2": 20551,
439
+ "level_3": 61920,
440
+ "level_4": 40932
441
+ },
442
+ "unique_source_ids": 143975
443
+ },
444
+ {
445
+ "ordered_semantic_sha256": "d9f0ff7e52baaad7d6d5bf222acd4903617bf5d43da8c2f36d7ab78c6322b626",
446
+ "relative_path": "replay_s2/train_epoch_1.jsonl",
447
+ "rows": 143975,
448
+ "sha256": "36b2c037353ebdf96c5964a7f6507a504479f8f8f9c58917cfa36ec1f1f666fc",
449
+ "size_bytes": 181233054,
450
+ "target_level_counts": {
451
+ "level_1": 20675,
452
+ "level_2": 20520,
453
+ "level_3": 61393,
454
+ "level_4": 41387
455
+ },
456
+ "unique_source_ids": 143975
457
+ },
458
+ {
459
+ "ordered_semantic_sha256": "a6fc13b6a9e933133459110d7adaabaeea4d3763952be3c1a62cf09f3a126612",
460
+ "relative_path": "replay_s2/val.jsonl",
461
+ "rows": 3000,
462
+ "sha256": "a155850d45765012f41ecfec074e2c68cba8e27ecf7f454962c900cfce7515b8",
463
+ "size_bytes": 3557633,
464
+ "target_level_counts": {
465
+ "level_1": 456,
466
+ "level_2": 425,
467
+ "level_3": 1306,
468
+ "level_4": 813
469
+ },
470
+ "unique_source_ids": 3000
471
+ },
472
+ {
473
+ "ordered_semantic_sha256": "05ab00c5c5fdfdcea59a4f083a6837be295eaf55283f7c746bec8e676f520c30",
474
+ "relative_path": "replay_s2/test.jsonl",
475
+ "rows": 2999,
476
+ "sha256": "75009c6214215d064370a28a05685298b0584737cccb8b8273386e8a29b69253",
477
+ "size_bytes": 3603077,
478
+ "target_level_counts": {
479
+ "level_1": 455,
480
+ "level_2": 425,
481
+ "level_3": 1306,
482
+ "level_4": 813
483
+ },
484
+ "unique_source_ids": 2999
485
+ }
486
+ ],
487
+ "jsonl_files": 31,
488
+ "jsonl_rows": 1463746
489
+ },
490
+ "parameters": {
491
+ "code_root": "/root/autodl-tmp/_REPLICATION/multivar/src",
492
+ "output_root": "/root/autodl-tmp/_TRAIN_DATA/ablation/r3_no_easyfirst",
493
+ "sampler_seed": 0,
494
+ "seed": 42,
495
+ "split_ratios": [
496
+ 0.96,
497
+ 0.02,
498
+ 0.02
499
+ ],
500
+ "stage2_level_weights": {
501
+ "level_1": 1.0,
502
+ "level_2": 1.0,
503
+ "level_3": 3.0,
504
+ "level_4": 2.0
505
+ },
506
+ "task_type_environment": "TS_ALIGN_PROMPT_TASK_TYPE=0",
507
+ "train_data_root": "/root/autodl-tmp/_TRAIN_DATA",
508
+ "world_size": 2
509
+ },
510
+ "platform": "Linux-5.15.0-78-generic-x86_64-with-glibc2.35",
511
+ "purpose": "R3 no-L1/L2-first one-shot epoch-aware data materialization",
512
+ "python": "3.12.3 | packaged by Anaconda, Inc. | (main, May 6 2024, 19:46:43) [GCC 11.2.0]",
513
+ "schema_version": 1,
514
+ "source_files": {
515
+ "b": {
516
+ "level12": {
517
+ "line_count": 50000,
518
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_bivar/multivar/level12_revised.jsonl",
519
+ "sha256": "ed265ef97839603a9033f51599bba33edf9caed98f2265f672437802fbcaf9f2",
520
+ "size_bytes": 214404527
521
+ },
522
+ "level3": {
523
+ "line_count": 50000,
524
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_bivar/multivar/level3_desc.jsonl",
525
+ "sha256": "2f6a66e9003327162ccfa2969d0c0211dc3375a75940caa46893b82e754eb751",
526
+ "size_bytes": 153036342
527
+ },
528
+ "level4": {
529
+ "line_count": 50000,
530
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_bivar/multivar/level4_desc.jsonl",
531
+ "sha256": "413ec60487b00f52faa43daa05bc327c125db5905682a0799692c879ef1011da",
532
+ "size_bytes": 203079998
533
+ },
534
+ "raw": {
535
+ "line_count": 50000,
536
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_bivar/multivar/raw_values_multivar.jsonl",
537
+ "sha256": "6a81f6dab77fb715d0ffc794b8055bea4ceeb7bc928b74dbf6d4fa68da145b95",
538
+ "size_bytes": 493080307
539
+ }
540
+ },
541
+ "m": {
542
+ "level12": {
543
+ "line_count": 50000,
544
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_multivar/multivar/level12_revised.jsonl",
545
+ "sha256": "fbfc2fe31a5dd303081f12e3c49eab9ce9e118ea913f21dc7a77b2134bad22eb",
546
+ "size_bytes": 146021503
547
+ },
548
+ "level3": {
549
+ "line_count": 50000,
550
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_multivar/multivar/level3_desc.jsonl",
551
+ "sha256": "cad4bbdb7ae6491d02f81ef86241696e745a4784a85778eed5d36b4573741d34",
552
+ "size_bytes": 239465227
553
+ },
554
+ "level4": {
555
+ "line_count": 50000,
556
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_multivar/multivar/level4_desc.jsonl",
557
+ "sha256": "b7bf30bd9842aaa728a2c2288bdbf97a6917c1daf609e9dbb43d680c84240b5d",
558
+ "size_bytes": 289491979
559
+ },
560
+ "raw": {
561
+ "line_count": 50000,
562
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_multivar/multivar/raw_values_multivar.jsonl",
563
+ "sha256": "52642aa00a84b0bf0aa925e4633f26998a248e59adbd6c40b2f07bdb14865390",
564
+ "size_bytes": 1535905350
565
+ }
566
+ },
567
+ "replay": {
568
+ "level12": {
569
+ "line_count": 150000,
570
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_replay/multivar/level12_merged.jsonl",
571
+ "sha256": "6b0f3bb8d90174048e8b5ad0a2f49ae81f14cba6fb7fc915e7f406cb47a3728b",
572
+ "size_bytes": 513341504
573
+ },
574
+ "level3": {
575
+ "line_count": 150000,
576
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_replay/multivar/level3_desc.jsonl",
577
+ "sha256": "f1acec5754c7a66307a837f41cd2fe13a2bff5f67d0d28177aa7771999e7ac84",
578
+ "size_bytes": 530194484
579
+ },
580
+ "level4": {
581
+ "line_count": 150000,
582
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_replay/multivar/level4_desc.jsonl",
583
+ "sha256": "4d98c32a5f4bc20bd2f606491c42eaa9df2533016f7f2386dafd19c52e0573d2",
584
+ "size_bytes": 636656457
585
+ },
586
+ "raw": {
587
+ "line_count": 150000,
588
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_replay/multivar/raw_values_multivar.jsonl",
589
+ "sha256": "c1f2b273795d8be386588b802bdeb95684ce29fadade9f32843c49ef52b24d51",
590
+ "size_bytes": 2125304159
591
+ }
592
+ },
593
+ "u": {
594
+ "level12": {
595
+ "line_count": 50000,
596
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_univar/multivar/level12_revised.jsonl",
597
+ "sha256": "db295b4719659d3f82806950662799dac0c8b1baacc84da9d96159da37ac9316",
598
+ "size_bytes": 152804258
599
+ },
600
+ "level3": {
601
+ "line_count": 50000,
602
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_univar/multivar/level3_desc.jsonl",
603
+ "sha256": "c1eab76f7c4321c69edcb276ebf5a32b6a07305bc76e0538554d6937440a86e2",
604
+ "size_bytes": 137581699
605
+ },
606
+ "level4": {
607
+ "line_count": 50000,
608
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_univar/multivar/level4_desc.jsonl",
609
+ "sha256": "095a534a13fd87ae59ee11414ede539c7d0cc0fc7a46bd1fa20a8c3f3334a97c",
610
+ "size_bytes": 143973264
611
+ },
612
+ "raw": {
613
+ "line_count": 50000,
614
+ "path": "/root/autodl-tmp/_TRAIN_DATA/data_univar/multivar/raw_values_multivar.jsonl",
615
+ "sha256": "0a8ebb8846dc037e54e76aa197e80a064c046e2fef1e2a8d08884bb3b009b56c",
616
+ "size_bytes": 96207286
617
+ }
618
+ }
619
+ },
620
+ "started_at_utc": "2026-07-21T16:25:09.481334+00:00"
621
+ }
data/training/ablation/r3_no_easyfirst/_SUCCESS.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "completed_at_utc": "2026-07-21T16:36:26.604281+00:00",
3
+ "jsonl_rows": 1463746,
4
+ "manifest_file_count": 36,
5
+ "manifest_sha256": "24c64dcb5be4180c073769eafa07e0e407c363b8d93768e91fe7471f0b349d4d",
6
+ "manifest_total_size_bytes": 1106205793,
7
+ "schema_version": 1,
8
+ "status": "complete"
9
+ }
data/training/ablation/r3_no_easyfirst/b_allocation_sidecar.json ADDED
@@ -0,0 +1,646 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "assertions": {
3
+ "all_five_train_epochs_contain_all_four_levels": true,
4
+ "exact_post_sampler_target_multiset_preserved": true,
5
+ "exact_pre_sampler_occurrence_multiset_preserved": true,
6
+ "exact_pre_sampler_target_multiset_preserved": true,
7
+ "fixed_historical_val_test_materialized": true,
8
+ "no_target_stage_train_val_test_source_overlap": true,
9
+ "sampler_padding_pinned": true,
10
+ "target_text_owned_by_row_source_id": true
11
+ },
12
+ "baseline": {
13
+ "aggregate_post_sampler": {
14
+ "rows": 384000,
15
+ "target_level_counts": {
16
+ "level_1": 157731,
17
+ "level_2": 157624,
18
+ "level_3": 41398,
19
+ "level_4": 27247
20
+ },
21
+ "target_multiset": {
22
+ "rows": 384000,
23
+ "sha256": "8af30d1a1032c28148afbc7eb291e502ab64835da62019d669f37568461e300c",
24
+ "unique_signatures": 152702
25
+ }
26
+ },
27
+ "aggregate_pre_sampler": {
28
+ "exposure_key_set_sha256": "f3ef44ae72232f59de3132cb562e8913d4dc850217cb5acad073b5522b3af95b",
29
+ "rows": 384000,
30
+ "target_level_counts": {
31
+ "level_1": 157731,
32
+ "level_2": 157624,
33
+ "level_3": 41398,
34
+ "level_4": 27247
35
+ },
36
+ "target_multiset": {
37
+ "rows": 384000,
38
+ "sha256": "8af30d1a1032c28148afbc7eb291e502ab64835da62019d669f37568461e300c",
39
+ "unique_signatures": 152702
40
+ }
41
+ },
42
+ "blocks": {
43
+ "s1_e0": {
44
+ "prompt_sequence_sha256": "a81a954205b3f629e10a6e8ebe6783a963e6ca28f8ac78920bb60dc06ddae33e",
45
+ "rows": 96000,
46
+ "target_level_counts": {
47
+ "level_1": 48000,
48
+ "level_2": 48000,
49
+ "level_3": 0,
50
+ "level_4": 0
51
+ }
52
+ },
53
+ "s1_e1": {
54
+ "prompt_sequence_sha256": "f83a1e5279cf319369fb7521145d4852f2415ff97fc00a2a575f926d412805e2",
55
+ "rows": 96000,
56
+ "target_level_counts": {
57
+ "level_1": 48000,
58
+ "level_2": 48000,
59
+ "level_3": 0,
60
+ "level_4": 0
61
+ }
62
+ },
63
+ "s1_e2": {
64
+ "prompt_sequence_sha256": "3703c789ccc36415fdfd3c98620beb0abeef51a31323814eca118bdb4f3a6458",
65
+ "rows": 96000,
66
+ "target_level_counts": {
67
+ "level_1": 48000,
68
+ "level_2": 48000,
69
+ "level_3": 0,
70
+ "level_4": 0
71
+ }
72
+ },
73
+ "s2_e0": {
74
+ "prompt_sequence_sha256": "4847abc97ac193e2037f3b04749a65fea7f27016a33b9aac6198764210e938f8",
75
+ "rows": 48000,
76
+ "target_level_counts": {
77
+ "level_1": 6882,
78
+ "level_2": 6846,
79
+ "level_3": 20781,
80
+ "level_4": 13491
81
+ }
82
+ },
83
+ "s2_e1": {
84
+ "prompt_sequence_sha256": "8adf1023a7f907c8985ff06d5cb06ec755b7f714da01b20b481977d5d5a3da3e",
85
+ "rows": 48000,
86
+ "target_level_counts": {
87
+ "level_1": 6849,
88
+ "level_2": 6778,
89
+ "level_3": 20617,
90
+ "level_4": 13756
91
+ }
92
+ }
93
+ },
94
+ "sampler_by_block": {
95
+ "s1_e0": {
96
+ "dataset_rows": 96000,
97
+ "padded_records": [],
98
+ "padding_rows": 0,
99
+ "post_sampler_rows": 96000,
100
+ "post_sampler_target_level_counts": {
101
+ "level_1": 48000,
102
+ "level_2": 48000,
103
+ "level_3": 0,
104
+ "level_4": 0
105
+ },
106
+ "post_sampler_target_multiset": {
107
+ "rows": 96000,
108
+ "sha256": "c791f8f2feb97a68bef59ed506f8baeca6f845698bda501cda9d9d2fa014cf33",
109
+ "unique_signatures": 96000
110
+ },
111
+ "rows_per_rank": [
112
+ 48000,
113
+ 48000
114
+ ],
115
+ "sampler_epoch": 0,
116
+ "sampler_seed": 0,
117
+ "world_size": 2
118
+ },
119
+ "s1_e1": {
120
+ "dataset_rows": 96000,
121
+ "padded_records": [],
122
+ "padding_rows": 0,
123
+ "post_sampler_rows": 96000,
124
+ "post_sampler_target_level_counts": {
125
+ "level_1": 48000,
126
+ "level_2": 48000,
127
+ "level_3": 0,
128
+ "level_4": 0
129
+ },
130
+ "post_sampler_target_multiset": {
131
+ "rows": 96000,
132
+ "sha256": "c791f8f2feb97a68bef59ed506f8baeca6f845698bda501cda9d9d2fa014cf33",
133
+ "unique_signatures": 96000
134
+ },
135
+ "rows_per_rank": [
136
+ 48000,
137
+ 48000
138
+ ],
139
+ "sampler_epoch": 1,
140
+ "sampler_seed": 0,
141
+ "world_size": 2
142
+ },
143
+ "s1_e2": {
144
+ "dataset_rows": 96000,
145
+ "padded_records": [],
146
+ "padding_rows": 0,
147
+ "post_sampler_rows": 96000,
148
+ "post_sampler_target_level_counts": {
149
+ "level_1": 48000,
150
+ "level_2": 48000,
151
+ "level_3": 0,
152
+ "level_4": 0
153
+ },
154
+ "post_sampler_target_multiset": {
155
+ "rows": 96000,
156
+ "sha256": "c791f8f2feb97a68bef59ed506f8baeca6f845698bda501cda9d9d2fa014cf33",
157
+ "unique_signatures": 96000
158
+ },
159
+ "rows_per_rank": [
160
+ 48000,
161
+ 48000
162
+ ],
163
+ "sampler_epoch": 2,
164
+ "sampler_seed": 0,
165
+ "world_size": 2
166
+ },
167
+ "s2_e0": {
168
+ "dataset_rows": 48000,
169
+ "padded_records": [],
170
+ "padding_rows": 0,
171
+ "post_sampler_rows": 48000,
172
+ "post_sampler_target_level_counts": {
173
+ "level_1": 6882,
174
+ "level_2": 6846,
175
+ "level_3": 20781,
176
+ "level_4": 13491
177
+ },
178
+ "post_sampler_target_multiset": {
179
+ "rows": 48000,
180
+ "sha256": "8247d76ee2991aa7d0d084ad1ecff1f191fb34bedf0a53c6a67821276bc8cf56",
181
+ "unique_signatures": 48000
182
+ },
183
+ "rows_per_rank": [
184
+ 24000,
185
+ 24000
186
+ ],
187
+ "sampler_epoch": 0,
188
+ "sampler_seed": 0,
189
+ "world_size": 2
190
+ },
191
+ "s2_e1": {
192
+ "dataset_rows": 48000,
193
+ "padded_records": [],
194
+ "padding_rows": 0,
195
+ "post_sampler_rows": 48000,
196
+ "post_sampler_target_level_counts": {
197
+ "level_1": 6849,
198
+ "level_2": 6778,
199
+ "level_3": 20617,
200
+ "level_4": 13756
201
+ },
202
+ "post_sampler_target_multiset": {
203
+ "rows": 48000,
204
+ "sha256": "26433a771ff373afe72e0ea7c0930520de9d4a2342dab1d2edf3422dfbfd6cbd",
205
+ "unique_signatures": 48000
206
+ },
207
+ "rows_per_rank": [
208
+ 24000,
209
+ 24000
210
+ ],
211
+ "sampler_epoch": 1,
212
+ "sampler_seed": 0,
213
+ "world_size": 2
214
+ }
215
+ }
216
+ },
217
+ "contract": {
218
+ "block_apportionment": "sequential Hamilton largest-remainder with exact row/column sums",
219
+ "description": "Simulate historical S1x3 plus S2x2 exposures, then redistribute the exact (source_id, target_level, target_text) occurrence multiset proportionally across 2N/2N/2N/N/N blocks.",
220
+ "level_weights_for_baseline_stage2": {
221
+ "level_1": 1.0,
222
+ "level_2": 1.0,
223
+ "level_3": 3.0,
224
+ "level_4": 2.0
225
+ },
226
+ "prompt_rng": "seed + target_epoch * target_dataset_size + target_index",
227
+ "stage1_level_draw_consumes_rng": false,
228
+ "stage1_system_prompt": null,
229
+ "stage2_level_draw_consumes_rng": true,
230
+ "stage2_system_prompt": "你是专业的时间序列分析助手，请仅根据给定时间序列完成分析。",
231
+ "task_type_environment": "TS_ALIGN_PROMPT_TASK_TYPE=0"
232
+ },
233
+ "modality": "b",
234
+ "r3": {
235
+ "aggregate_post_sampler": {
236
+ "rows": 384000,
237
+ "target_level_counts": {
238
+ "level_1": 157731,
239
+ "level_2": 157624,
240
+ "level_3": 41398,
241
+ "level_4": 27247
242
+ },
243
+ "target_multiset": {
244
+ "rows": 384000,
245
+ "sha256": "8af30d1a1032c28148afbc7eb291e502ab64835da62019d669f37568461e300c",
246
+ "unique_signatures": 152702
247
+ }
248
+ },
249
+ "aggregate_pre_sampler": {
250
+ "rows": 384000,
251
+ "target_level_counts": {
252
+ "level_1": 157731,
253
+ "level_2": 157624,
254
+ "level_3": 41398,
255
+ "level_4": 27247
256
+ },
257
+ "target_multiset": {
258
+ "rows": 384000,
259
+ "sha256": "8af30d1a1032c28148afbc7eb291e502ab64835da62019d669f37568461e300c",
260
+ "unique_signatures": 152702
261
+ }
262
+ },
263
+ "allocated_target_level_counts": {
264
+ "s1_e0": {
265
+ "level_1": 39433,
266
+ "level_2": 39406,
267
+ "level_3": 10349,
268
+ "level_4": 6812
269
+ },
270
+ "s1_e1": {
271
+ "level_1": 39432,
272
+ "level_2": 39406,
273
+ "level_3": 10350,
274
+ "level_4": 6812
275
+ },
276
+ "s1_e2": {
277
+ "level_1": 39433,
278
+ "level_2": 39406,
279
+ "level_3": 10350,
280
+ "level_4": 6811
281
+ },
282
+ "s2_e0": {
283
+ "level_1": 19717,
284
+ "level_2": 19703,
285
+ "level_3": 5174,
286
+ "level_4": 3406
287
+ },
288
+ "s2_e1": {
289
+ "level_1": 19716,
290
+ "level_2": 19703,
291
+ "level_3": 5175,
292
+ "level_4": 3406
293
+ }
294
+ },
295
+ "files": [
296
+ {
297
+ "ordered_semantic_sha256": "4ff45f94bea469c11f082ae9a3dc72da774cad7276873e8e2c9da5eb964c9719",
298
+ "relative_path": "b_s1/train_epoch_0.jsonl",
299
+ "rows": 96000,
300
+ "sha256": "465c69b698ed3686080fa5aafdc8f1fcaeea06ba32b80e3fc315a045b840a3ca",
301
+ "size_bytes": 61933906,
302
+ "target_level_counts": {
303
+ "level_1": 39433,
304
+ "level_2": 39406,
305
+ "level_3": 10349,
306
+ "level_4": 6812
307
+ },
308
+ "unique_source_ids": 43229
309
+ },
310
+ {
311
+ "ordered_semantic_sha256": "60f9b90fc0fd5554729286c373c8820274c9cc11600b851e4363d33ef1f276c3",
312
+ "relative_path": "b_s1/train_epoch_1.jsonl",
313
+ "rows": 96000,
314
+ "sha256": "94439b81436df2292cf09dd4bbf16256f34e7660c2a4e688f1ba16d8c335143d",
315
+ "size_bytes": 62018729,
316
+ "target_level_counts": {
317
+ "level_1": 39432,
318
+ "level_2": 39406,
319
+ "level_3": 10350,
320
+ "level_4": 6812
321
+ },
322
+ "unique_source_ids": 43246
323
+ },
324
+ {
325
+ "ordered_semantic_sha256": "c6c8070c205ea7134f07dd92c0ea7d54203c5e52ce43be40ad1477f292ec90db",
326
+ "relative_path": "b_s1/train_epoch_2.jsonl",
327
+ "rows": 96000,
328
+ "sha256": "c1ca406ee44dbad90184273dcaa61aaac4531549a63f4fa776fe83f1b2e5af0c",
329
+ "size_bytes": 61825773,
330
+ "target_level_counts": {
331
+ "level_1": 39433,
332
+ "level_2": 39406,
333
+ "level_3": 10350,
334
+ "level_4": 6811
335
+ },
336
+ "unique_source_ids": 43206
337
+ },
338
+ {
339
+ "ordered_semantic_sha256": "252f9b7d45a46e1a454677a936facaa97bca019cf7a9d4fd20b974720d67f7db",
340
+ "relative_path": "b_s2/train_epoch_0.jsonl",
341
+ "rows": 48000,
342
+ "sha256": "25f5db8257948a79a0db53169760da6df050aa25f3aaa04dcae595210e7cf670",
343
+ "size_bytes": 35098810,
344
+ "target_level_counts": {
345
+ "level_1": 19717,
346
+ "level_2": 19703,
347
+ "level_3": 5174,
348
+ "level_4": 3406
349
+ },
350
+ "unique_source_ids": 31668
351
+ },
352
+ {
353
+ "ordered_semantic_sha256": "9eeb5b28d742eccee5d57598d2c74bdfcb8eb78151403c542035206b9d65f68a",
354
+ "relative_path": "b_s2/train_epoch_1.jsonl",
355
+ "rows": 48000,
356
+ "sha256": "6f55d2babe7855cc06fbb31aeca6d3c48c19f67cf1dec62aa053e5003a517cdf",
357
+ "size_bytes": 34988718,
358
+ "target_level_counts": {
359
+ "level_1": 19716,
360
+ "level_2": 19703,
361
+ "level_3": 5175,
362
+ "level_4": 3406
363
+ },
364
+ "unique_source_ids": 31722
365
+ },
366
+ {
367
+ "ordered_semantic_sha256": "5aeb97061a0bfbd039732730bc63d725f4bcdd056daa8defff6b1708a8805b78",
368
+ "relative_path": "b_s1/val.jsonl",
369
+ "rows": 2000,
370
+ "sha256": "c906af13d05c6f2944e5d53bf5c144be518644fccdfa43ce10cccac3e6e8abac",
371
+ "size_bytes": 821997,
372
+ "target_level_counts": {
373
+ "level_1": 1000,
374
+ "level_2": 1000,
375
+ "level_3": 0,
376
+ "level_4": 0
377
+ },
378
+ "unique_source_ids": 1000
379
+ },
380
+ {
381
+ "ordered_semantic_sha256": "5859d048aa7ea27368695ff70cfc425a248cf17536484026c22d57c125de5a91",
382
+ "relative_path": "b_s1/test.jsonl",
383
+ "rows": 2000,
384
+ "sha256": "6f1b6961c2065fcde39fe9aafde3cde7c1dd7b7789a172485156d9f70962a4fc",
385
+ "size_bytes": 826581,
386
+ "target_level_counts": {
387
+ "level_1": 1000,
388
+ "level_2": 1000,
389
+ "level_3": 0,
390
+ "level_4": 0
391
+ },
392
+ "unique_source_ids": 1000
393
+ },
394
+ {
395
+ "ordered_semantic_sha256": "b345a78ea5e84c27ae179054553b076620768ac2ac8f466ff547e2c801d0230e",
396
+ "relative_path": "b_s2/val.jsonl",
397
+ "rows": 1000,
398
+ "sha256": "b0c5360e96ca41997cdde098d0135925dac4807606f692b5be7d9501e23a7f8f",
399
+ "size_bytes": 1190097,
400
+ "target_level_counts": {
401
+ "level_1": 158,
402
+ "level_2": 143,
403
+ "level_3": 434,
404
+ "level_4": 265
405
+ },
406
+ "unique_source_ids": 1000
407
+ },
408
+ {
409
+ "ordered_semantic_sha256": "c2de27e9f58cc4f4be4a75cb8404619e850392c59f0577394aaa62ffbe632e06",
410
+ "relative_path": "b_s2/test.jsonl",
411
+ "rows": 1000,
412
+ "sha256": "94a3126718b27386d06c248638d143e2de6c2560959b5329d61683048d648187",
413
+ "size_bytes": 1177107,
414
+ "target_level_counts": {
415
+ "level_1": 157,
416
+ "level_2": 143,
417
+ "level_3": 434,
418
+ "level_4": 266
419
+ },
420
+ "unique_source_ids": 1000
421
+ }
422
+ ],
423
+ "sampler_by_block": {
424
+ "s1_e0": {
425
+ "dataset_rows": 96000,
426
+ "padded_records": [],
427
+ "padding_rows": 0,
428
+ "post_sampler_rows": 96000,
429
+ "post_sampler_target_level_counts": {
430
+ "level_1": 39433,
431
+ "level_2": 39406,
432
+ "level_3": 10349,
433
+ "level_4": 6812
434
+ },
435
+ "post_sampler_target_multiset": {
436
+ "rows": 96000,
437
+ "sha256": "375f4c94ca84e47248d8cb193cdc79b505a8d3c4ffc8c67125b3400dc9fb01ab",
438
+ "unique_signatures": 74704
439
+ },
440
+ "rows_per_rank": [
441
+ 48000,
442
+ 48000
443
+ ],
444
+ "sampler_epoch": 0,
445
+ "sampler_seed": 0,
446
+ "world_size": 2
447
+ },
448
+ "s1_e1": {
449
+ "dataset_rows": 96000,
450
+ "padded_records": [],
451
+ "padding_rows": 0,
452
+ "post_sampler_rows": 96000,
453
+ "post_sampler_target_level_counts": {
454
+ "level_1": 39432,
455
+ "level_2": 39406,
456
+ "level_3": 10350,
457
+ "level_4": 6812
458
+ },
459
+ "post_sampler_target_multiset": {
460
+ "rows": 96000,
461
+ "sha256": "20d852d19aaf9daa1053b37b9c1f8b4c1d1d52925c6ec07acc9d2a33cd2f3d28",
462
+ "unique_signatures": 74556
463
+ },
464
+ "rows_per_rank": [
465
+ 48000,
466
+ 48000
467
+ ],
468
+ "sampler_epoch": 1,
469
+ "sampler_seed": 0,
470
+ "world_size": 2
471
+ },
472
+ "s1_e2": {
473
+ "dataset_rows": 96000,
474
+ "padded_records": [],
475
+ "padding_rows": 0,
476
+ "post_sampler_rows": 96000,
477
+ "post_sampler_target_level_counts": {
478
+ "level_1": 39433,
479
+ "level_2": 39406,
480
+ "level_3": 10350,
481
+ "level_4": 6811
482
+ },
483
+ "post_sampler_target_multiset": {
484
+ "rows": 96000,
485
+ "sha256": "9239dbaf6d1a6b521a5bbed570b1cd1b519eba04282fcc7d085fc17a8fcfab6c",
486
+ "unique_signatures": 74618
487
+ },
488
+ "rows_per_rank": [
489
+ 48000,
490
+ 48000
491
+ ],
492
+ "sampler_epoch": 2,
493
+ "sampler_seed": 0,
494
+ "world_size": 2
495
+ },
496
+ "s2_e0": {
497
+ "dataset_rows": 48000,
498
+ "padded_records": [],
499
+ "padding_rows": 0,
500
+ "post_sampler_rows": 48000,
501
+ "post_sampler_target_level_counts": {
502
+ "level_1": 19717,
503
+ "level_2": 19703,
504
+ "level_3": 5174,
505
+ "level_4": 3406
506
+ },
507
+ "post_sampler_target_multiset": {
508
+ "rows": 48000,
509
+ "sha256": "5aa74ab09212ef5f60dce22a4e0c4eca23219b1ed848de43e9b92ad07095e226",
510
+ "unique_signatures": 42237
511
+ },
512
+ "rows_per_rank": [
513
+ 24000,
514
+ 24000
515
+ ],
516
+ "sampler_epoch": 0,
517
+ "sampler_seed": 0,
518
+ "world_size": 2
519
+ },
520
+ "s2_e1": {
521
+ "dataset_rows": 48000,
522
+ "padded_records": [],
523
+ "padding_rows": 0,
524
+ "post_sampler_rows": 48000,
525
+ "post_sampler_target_level_counts": {
526
+ "level_1": 19716,
527
+ "level_2": 19703,
528
+ "level_3": 5175,
529
+ "level_4": 3406
530
+ },
531
+ "post_sampler_target_multiset": {
532
+ "rows": 48000,
533
+ "sha256": "27377eaa4ee751c4d7537ed14077502e788f349f6d17cf20636b64f0f10b2c59",
534
+ "unique_signatures": 42147
535
+ },
536
+ "rows_per_rank": [
537
+ 24000,
538
+ 24000
539
+ ],
540
+ "sampler_epoch": 1,
541
+ "sampler_seed": 0,
542
+ "world_size": 2
543
+ }
544
+ }
545
+ },
546
+ "record_counts": {
547
+ "stage1_splits": {
548
+ "test": {
549
+ "ordered_source_ids_sha256": "c3f06fd1c3e3641b4b944089b5a648801a2b8f8fc3c86fb83e3c5808bc441d1e",
550
+ "records": 1000
551
+ },
552
+ "train": {
553
+ "ordered_source_ids_sha256": "16a24a4c8a2f5bf865ec6aabf92ed3660fa2bf12832b0c94a594924ab40a3b19",
554
+ "records": 48000
555
+ },
556
+ "val": {
557
+ "ordered_source_ids_sha256": "3b2eac18833da66b3628624c9bfc3dd89a5bbc83f4a2342b2d5b8813d5a8a53f",
558
+ "records": 1000
559
+ }
560
+ },
561
+ "stage2_splits": {
562
+ "test": {
563
+ "ordered_source_ids_sha256": "963b1ae71037b0fa789f82cf866a1ba616caabd1edf07506cd2630cd42708056",
564
+ "records": 1000
565
+ },
566
+ "train": {
567
+ "ordered_source_ids_sha256": "02f9ddfc0bdda29d496565cdaa045414c6bb3d00a373f562f7568fcaadd20c79",
568
+ "records": 48000
569
+ },
570
+ "val": {
571
+ "ordered_source_ids_sha256": "3475da0bbae4a17a380fdea296796859dd6954bb85ac6b6636a0a81daeae0e60",
572
+ "records": 1000
573
+ }
574
+ },
575
+ "usable": 50000
576
+ },
577
+ "schema_version": 1,
578
+ "split_overlap_audit": {
579
+ "historical_cross_stage": {
580
+ "stage1_train_vs_stage2_val_test": {
581
+ "count": 1914,
582
+ "first_20": [
583
+ 208,
584
+ 336,
585
+ 350,
586
+ 949,
587
+ 1873,
588
+ 1959,
589
+ 4032,
590
+ 4493,
591
+ 4658,
592
+ 5248,
593
+ 5647,
594
+ 6332,
595
+ 6946,
596
+ 7043,
597
+ 7197,
598
+ 7224,
599
+ 7990,
600
+ 8433,
601
+ 8566,
602
+ 10403
603
+ ],
604
+ "sha256": "917c20b56b3608d93f9da9faf8d9bcd50fb1ccc07f41fc1e37b3a875a8bc7421"
605
+ },
606
+ "stage2_train_vs_stage1_val_test": {
607
+ "count": 1914,
608
+ "first_20": [
609
+ 179,
610
+ 729,
611
+ 1380,
612
+ 2196,
613
+ 2496,
614
+ 3839,
615
+ 4388,
616
+ 5065,
617
+ 5929,
618
+ 6376,
619
+ 6407,
620
+ 6479,
621
+ 6990,
622
+ 7007,
623
+ 7602,
624
+ 7765,
625
+ 8061,
626
+ 8345,
627
+ 8643,
628
+ 8870
629
+ ],
630
+ "sha256": "ef438d82907e68408396518e3948037900a9c5ce5078af69967b21b8e05fc5e3"
631
+ }
632
+ },
633
+ "r3_with_target_stage_val_test": {
634
+ "stage1": {
635
+ "count": 0,
636
+ "first_20": [],
637
+ "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
638
+ },
639
+ "stage2": {
640
+ "count": 0,
641
+ "first_20": [],
642
+ "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
643
+ }
644
+ }
645
+ }
646
+ }
data/training/ablation/r3_no_easyfirst/b_s1/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f1b6961c2065fcde39fe9aafde3cde7c1dd7b7789a172485156d9f70962a4fc
3
+ size 826581
data/training/ablation/r3_no_easyfirst/b_s1/train_epoch_0.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:465c69b698ed3686080fa5aafdc8f1fcaeea06ba32b80e3fc315a045b840a3ca
3
+ size 61933906
data/training/ablation/r3_no_easyfirst/b_s1/train_epoch_1.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94439b81436df2292cf09dd4bbf16256f34e7660c2a4e688f1ba16d8c335143d
3
+ size 62018729
data/training/ablation/r3_no_easyfirst/b_s1/train_epoch_2.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c1ca406ee44dbad90184273dcaa61aaac4531549a63f4fa776fe83f1b2e5af0c
3
+ size 61825773
data/training/ablation/r3_no_easyfirst/b_s1/val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c906af13d05c6f2944e5d53bf5c144be518644fccdfa43ce10cccac3e6e8abac
3
+ size 821997
data/training/ablation/r3_no_easyfirst/b_s2/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94a3126718b27386d06c248638d143e2de6c2560959b5329d61683048d648187
3
+ size 1177107
data/training/ablation/r3_no_easyfirst/b_s2/train_epoch_0.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25f5db8257948a79a0db53169760da6df050aa25f3aaa04dcae595210e7cf670
3
+ size 35098810
data/training/ablation/r3_no_easyfirst/b_s2/train_epoch_1.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f55d2babe7855cc06fbb31aeca6d3c48c19f67cf1dec62aa053e5003a517cdf
3
+ size 34988718
data/training/ablation/r3_no_easyfirst/b_s2/val.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b0c5360e96ca41997cdde098d0135925dac4807606f692b5be7d9501e23a7f8f
3
+ size 1190097
data/training/ablation/r3_no_easyfirst/m_allocation_sidecar.json ADDED
@@ -0,0 +1,646 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "assertions": {
3
+ "all_five_train_epochs_contain_all_four_levels": true,
4
+ "exact_post_sampler_target_multiset_preserved": true,
5
+ "exact_pre_sampler_occurrence_multiset_preserved": true,
6
+ "exact_pre_sampler_target_multiset_preserved": true,
7
+ "fixed_historical_val_test_materialized": true,
8
+ "no_target_stage_train_val_test_source_overlap": true,
9
+ "sampler_padding_pinned": true,
10
+ "target_text_owned_by_row_source_id": true
11
+ },
12
+ "baseline": {
13
+ "aggregate_post_sampler": {
14
+ "rows": 384000,
15
+ "target_level_counts": {
16
+ "level_1": 157731,
17
+ "level_2": 157624,
18
+ "level_3": 41398,
19
+ "level_4": 27247
20
+ },
21
+ "target_multiset": {
22
+ "rows": 384000,
23
+ "sha256": "ed9cd939e4658a8e67bb40496027e3f7c40805b547400e909d792e9482defe78",
24
+ "unique_signatures": 152728
25
+ }
26
+ },
27
+ "aggregate_pre_sampler": {
28
+ "exposure_key_set_sha256": "c66d7ea9087cf87c440dc1fb838198f93eec69478ea54fdfc62be16e7ddb3811",
29
+ "rows": 384000,
30
+ "target_level_counts": {
31
+ "level_1": 157731,
32
+ "level_2": 157624,
33
+ "level_3": 41398,
34
+ "level_4": 27247
35
+ },
36
+ "target_multiset": {
37
+ "rows": 384000,
38
+ "sha256": "ed9cd939e4658a8e67bb40496027e3f7c40805b547400e909d792e9482defe78",
39
+ "unique_signatures": 152728
40
+ }
41
+ },
42
+ "blocks": {
43
+ "s1_e0": {
44
+ "prompt_sequence_sha256": "991d51323b3e631795a99129629258f6bebbccf0cb2f3ed3da3a8eebff5a18c6",
45
+ "rows": 96000,
46
+ "target_level_counts": {
47
+ "level_1": 48000,
48
+ "level_2": 48000,
49
+ "level_3": 0,
50
+ "level_4": 0
51
+ }
52
+ },
53
+ "s1_e1": {
54
+ "prompt_sequence_sha256": "d8b3767c862cd1d37d573615d15b587f1cd8ac8730fbcdda615368e3fa3c064b",
55
+ "rows": 96000,
56
+ "target_level_counts": {
57
+ "level_1": 48000,
58
+ "level_2": 48000,
59
+ "level_3": 0,
60
+ "level_4": 0
61
+ }
62
+ },
63
+ "s1_e2": {
64
+ "prompt_sequence_sha256": "ead7c7c4217049e0664ccf919a7609a443298a5d069e4e01ea4b410c23cf8f05",
65
+ "rows": 96000,
66
+ "target_level_counts": {
67
+ "level_1": 48000,
68
+ "level_2": 48000,
69
+ "level_3": 0,
70
+ "level_4": 0
71
+ }
72
+ },
73
+ "s2_e0": {
74
+ "prompt_sequence_sha256": "ad8f6544ec1ff5949a35e324541c4cae87320462785e41516ae20320d491674a",
75
+ "rows": 48000,
76
+ "target_level_counts": {
77
+ "level_1": 6882,
78
+ "level_2": 6846,
79
+ "level_3": 20781,
80
+ "level_4": 13491
81
+ }
82
+ },
83
+ "s2_e1": {
84
+ "prompt_sequence_sha256": "906d9d7f7c9cbf76c03cdf4ff645fd46db4e816dc735a345db7f3f7dfeed24f6",
85
+ "rows": 48000,
86
+ "target_level_counts": {
87
+ "level_1": 6849,
88
+ "level_2": 6778,
89
+ "level_3": 20617,
90
+ "level_4": 13756
91
+ }
92
+ }
93
+ },
94
+ "sampler_by_block": {
95
+ "s1_e0": {
96
+ "dataset_rows": 96000,
97
+ "padded_records": [],
98
+ "padding_rows": 0,
99
+ "post_sampler_rows": 96000,
100
+ "post_sampler_target_level_counts": {
101
+ "level_1": 48000,
102
+ "level_2": 48000,
103
+ "level_3": 0,
104
+ "level_4": 0
105
+ },
106
+ "post_sampler_target_multiset": {
107
+ "rows": 96000,
108
+ "sha256": "046ba6a278a5ab9aa7690f09650037aae17a68afcf75a500e61e2223f6e962bd",
109
+ "unique_signatures": 96000
110
+ },
111
+ "rows_per_rank": [
112
+ 48000,
113
+ 48000
114
+ ],
115
+ "sampler_epoch": 0,
116
+ "sampler_seed": 0,
117
+ "world_size": 2
118
+ },
119
+ "s1_e1": {
120
+ "dataset_rows": 96000,
121
+ "padded_records": [],
122
+ "padding_rows": 0,
123
+ "post_sampler_rows": 96000,
124
+ "post_sampler_target_level_counts": {
125
+ "level_1": 48000,
126
+ "level_2": 48000,
127
+ "level_3": 0,
128
+ "level_4": 0
129
+ },
130
+ "post_sampler_target_multiset": {
131
+ "rows": 96000,
132
+ "sha256": "046ba6a278a5ab9aa7690f09650037aae17a68afcf75a500e61e2223f6e962bd",
133
+ "unique_signatures": 96000
134
+ },
135
+ "rows_per_rank": [
136
+ 48000,
137
+ 48000
138
+ ],
139
+ "sampler_epoch": 1,
140
+ "sampler_seed": 0,
141
+ "world_size": 2
142
+ },
143
+ "s1_e2": {
144
+ "dataset_rows": 96000,
145
+ "padded_records": [],
146
+ "padding_rows": 0,
147
+ "post_sampler_rows": 96000,
148
+ "post_sampler_target_level_counts": {
149
+ "level_1": 48000,
150
+ "level_2": 48000,
151
+ "level_3": 0,
152
+ "level_4": 0
153
+ },
154
+ "post_sampler_target_multiset": {
155
+ "rows": 96000,
156
+ "sha256": "046ba6a278a5ab9aa7690f09650037aae17a68afcf75a500e61e2223f6e962bd",
157
+ "unique_signatures": 96000
158
+ },
159
+ "rows_per_rank": [
160
+ 48000,
161
+ 48000
162
+ ],
163
+ "sampler_epoch": 2,
164
+ "sampler_seed": 0,
165
+ "world_size": 2
166
+ },
167
+ "s2_e0": {
168
+ "dataset_rows": 48000,
169
+ "padded_records": [],
170
+ "padding_rows": 0,
171
+ "post_sampler_rows": 48000,
172
+ "post_sampler_target_level_counts": {
173
+ "level_1": 6882,
174
+ "level_2": 6846,
175
+ "level_3": 20781,
176
+ "level_4": 13491
177
+ },
178
+ "post_sampler_target_multiset": {
179
+ "rows": 48000,
180
+ "sha256": "32fbb6341af674c539ce4b8b201ee178763f6d209886d5e5f5f1095d59c67e33",
181
+ "unique_signatures": 48000
182
+ },
183
+ "rows_per_rank": [
184
+ 24000,
185
+ 24000
186
+ ],
187
+ "sampler_epoch": 0,
188
+ "sampler_seed": 0,
189
+ "world_size": 2
190
+ },
191
+ "s2_e1": {
192
+ "dataset_rows": 48000,
193
+ "padded_records": [],
194
+ "padding_rows": 0,
195
+ "post_sampler_rows": 48000,
196
+ "post_sampler_target_level_counts": {
197
+ "level_1": 6849,
198
+ "level_2": 6778,
199
+ "level_3": 20617,
200
+ "level_4": 13756
201
+ },
202
+ "post_sampler_target_multiset": {
203
+ "rows": 48000,
204
+ "sha256": "119361444f3cd1d91629a3e626452eb4cf379c21bb05acd6745cc47d3c463a4b",
205
+ "unique_signatures": 48000
206
+ },
207
+ "rows_per_rank": [
208
+ 24000,
209
+ 24000
210
+ ],
211
+ "sampler_epoch": 1,
212
+ "sampler_seed": 0,
213
+ "world_size": 2
214
+ }
215
+ }
216
+ },
217
+ "contract": {
218
+ "block_apportionment": "sequential Hamilton largest-remainder with exact row/column sums",
219
+ "description": "Simulate historical S1x3 plus S2x2 exposures, then redistribute the exact (source_id, target_level, target_text) occurrence multiset proportionally across 2N/2N/2N/N/N blocks.",
220
+ "level_weights_for_baseline_stage2": {
221
+ "level_1": 1.0,
222
+ "level_2": 1.0,
223
+ "level_3": 3.0,
224
+ "level_4": 2.0
225
+ },
226
+ "prompt_rng": "seed + target_epoch * target_dataset_size + target_index",
227
+ "stage1_level_draw_consumes_rng": false,
228
+ "stage1_system_prompt": null,
229
+ "stage2_level_draw_consumes_rng": true,
230
+ "stage2_system_prompt": "你是专业的时间序列分析助手，请仅根据给定时间序列完成分析。",
231
+ "task_type_environment": "TS_ALIGN_PROMPT_TASK_TYPE=0"
232
+ },
233
+ "modality": "m",
234
+ "r3": {
235
+ "aggregate_post_sampler": {
236
+ "rows": 384000,
237
+ "target_level_counts": {
238
+ "level_1": 157731,
239
+ "level_2": 157624,
240
+ "level_3": 41398,
241
+ "level_4": 27247
242
+ },
243
+ "target_multiset": {
244
+ "rows": 384000,
245
+ "sha256": "ed9cd939e4658a8e67bb40496027e3f7c40805b547400e909d792e9482defe78",
246
+ "unique_signatures": 152728
247
+ }
248
+ },
249
+ "aggregate_pre_sampler": {
250
+ "rows": 384000,
251
+ "target_level_counts": {
252
+ "level_1": 157731,
253
+ "level_2": 157624,
254
+ "level_3": 41398,
255
+ "level_4": 27247
256
+ },
257
+ "target_multiset": {
258
+ "rows": 384000,
259
+ "sha256": "ed9cd939e4658a8e67bb40496027e3f7c40805b547400e909d792e9482defe78",
260
+ "unique_signatures": 152728
261
+ }
262
+ },
263
+ "allocated_target_level_counts": {
264
+ "s1_e0": {
265
+ "level_1": 39433,
266
+ "level_2": 39406,
267
+ "level_3": 10349,
268
+ "level_4": 6812
269
+ },
270
+ "s1_e1": {
271
+ "level_1": 39432,
272
+ "level_2": 39406,
273
+ "level_3": 10350,
274
+ "level_4": 6812
275
+ },
276
+ "s1_e2": {
277
+ "level_1": 39433,
278
+ "level_2": 39406,
279
+ "level_3": 10350,
280
+ "level_4": 6811
281
+ },
282
+ "s2_e0": {
283
+ "level_1": 19717,
284
+ "level_2": 19703,
285
+ "level_3": 5174,
286
+ "level_4": 3406
287
+ },
288
+ "s2_e1": {
289
+ "level_1": 19716,
290
+ "level_2": 19703,
291
+ "level_3": 5175,
292
+ "level_4": 3406
293
+ }
294
+ },
295
+ "files": [
296
+ {
297
+ "ordered_semantic_sha256": "78fd620aae0e55ff62bced6805bfd7770422cc1434b83905cd9748eaf20162a2",
298
+ "relative_path": "m_s1/train_epoch_0.jsonl",
299
+ "rows": 96000,
300
+ "sha256": "61b7d5fad7b28cc6169831cd112a709d2d48602f408701b491b4efd76830e40e",
301
+ "size_bytes": 61891787,
302
+ "target_level_counts": {
303
+ "level_1": 39433,
304
+ "level_2": 39406,
305
+ "level_3": 10349,
306
+ "level_4": 6812
307
+ },
308
+ "unique_source_ids": 43141
309
+ },
310
+ {
311
+ "ordered_semantic_sha256": "82e85bf148ba83df46ffa207d42a3ed01e9339451f8a4b93d495374b29d6873b",
312
+ "relative_path": "m_s1/train_epoch_1.jsonl",
313
+ "rows": 96000,
314
+ "sha256": "fc13648f9cd277dd63c0f49fc7143f016e1d57b2454310f1631e7178ec708a03",
315
+ "size_bytes": 62068142,
316
+ "target_level_counts": {
317
+ "level_1": 39432,
318
+ "level_2": 39406,
319
+ "level_3": 10350,
320
+ "level_4": 6812
321
+ },
322
+ "unique_source_ids": 43229
323
+ },
324
+ {
325
+ "ordered_semantic_sha256": "61749cbff3d1919c25940b9aef967e17b674d37e7818b461e810aa2a09a7bc15",
326
+ "relative_path": "m_s1/train_epoch_2.jsonl",
327
+ "rows": 96000,
328
+ "sha256": "82bbd8f3e25230695255d3920c87fefa54545b3175b574652b4090677e7b30d6",
329
+ "size_bytes": 62110992,
330
+ "target_level_counts": {
331
+ "level_1": 39433,
332
+ "level_2": 39406,
333
+ "level_3": 10350,
334
+ "level_4": 6811
335
+ },
336
+ "unique_source_ids": 43165
337
+ },
338
+ {
339
+ "ordered_semantic_sha256": "5d47c8a487dac53f2102708bf74e866665b62387cd70251f1656d54784402c51",
340
+ "relative_path": "m_s2/train_epoch_0.jsonl",
341
+ "rows": 48000,
342
+ "sha256": "a2ac2ff0bfafbecb2b8c9628eb60d60168bb77e2f61a4b425134be4dcb08fd65",
343
+ "size_bytes": 35165030,
344
+ "target_level_counts": {
345
+ "level_1": 19717,
346
+ "level_2": 19703,
347
+ "level_3": 5174,
348
+ "level_4": 3406
349
+ },
350
+ "unique_source_ids": 31686
351
+ },
352
+ {
353
+ "ordered_semantic_sha256": "493efb22a0cf6c370cd48343622fe27d6ad1b649997e6f69ab5475502b1e831d",
354
+ "relative_path": "m_s2/train_epoch_1.jsonl",
355
+ "rows": 48000,
356
+ "sha256": "1f6cef3dacf382766db5ad26656a79696cb2cbcea9de96facd331b7065458546",
357
+ "size_bytes": 35060589,
358
+ "target_level_counts": {
359
+ "level_1": 19716,
360
+ "level_2": 19703,
361
+ "level_3": 5175,
362
+ "level_4": 3406
363
+ },
364
+ "unique_source_ids": 31621
365
+ },
366
+ {
367
+ "ordered_semantic_sha256": "8454d70177d057b630c72bd284593bc8f819c4d37695eb452f14fa99d5143bfc",
368
+ "relative_path": "m_s1/val.jsonl",
369
+ "rows": 2000,
370
+ "sha256": "9a97913b365119da108dd96ac3e09cc797ed3f6e41e670505c53d213f52c80fc",
371
+ "size_bytes": 733359,
372
+ "target_level_counts": {
373
+ "level_1": 1000,
374
+ "level_2": 1000,
375
+ "level_3": 0,
376
+ "level_4": 0
377
+ },
378
+ "unique_source_ids": 1000
379
+ },
380
+ {
381
+ "ordered_semantic_sha256": "18d97278af66ba06159e692cbe481be8028b57567c6636719e70df692edb58b0",
382
+ "relative_path": "m_s1/test.jsonl",
383
+ "rows": 2000,
384
+ "sha256": "c2a30603cdb56613a931ef1362d842a4b9924aee1dcb7ac4590b79972ae8b132",
385
+ "size_bytes": 734345,
386
+ "target_level_counts": {
387
+ "level_1": 1000,
388
+ "level_2": 1000,
389
+ "level_3": 0,
390
+ "level_4": 0
391
+ },
392
+ "unique_source_ids": 1000
393
+ },
394
+ {
395
+ "ordered_semantic_sha256": "3756ecd3de9cae0fbd2de757b5bab762be31c0f79506024925572ecf617b1a0c",
396
+ "relative_path": "m_s2/val.jsonl",
397
+ "rows": 1000,
398
+ "sha256": "a98c43da64518cf924740293c2706829f0a5caab2696a643b70ba545fb6021b8",
399
+ "size_bytes": 1302370,
400
+ "target_level_counts": {
401
+ "level_1": 158,
402
+ "level_2": 143,
403
+ "level_3": 434,
404
+ "level_4": 265
405
+ },
406
+ "unique_source_ids": 1000
407
+ },
408
+ {
409
+ "ordered_semantic_sha256": "451bb07883812e5a2bea12c082c7bebd3312aa650c35889f0990e6d54cc1cac1",
410
+ "relative_path": "m_s2/test.jsonl",
411
+ "rows": 1000,
412
+ "sha256": "4d2d5425a08fdddaea01cef98369fdfc193d752057a1aa8b11a55d805727c867",
413
+ "size_bytes": 1289680,
414
+ "target_level_counts": {
415
+ "level_1": 157,
416
+ "level_2": 143,
417
+ "level_3": 434,
418
+ "level_4": 266
419
+ },
420
+ "unique_source_ids": 1000
421
+ }
422
+ ],
423
+ "sampler_by_block": {
424
+ "s1_e0": {
425
+ "dataset_rows": 96000,
426
+ "padded_records": [],
427
+ "padding_rows": 0,
428
+ "post_sampler_rows": 96000,
429
+ "post_sampler_target_level_counts": {
430
+ "level_1": 39433,
431
+ "level_2": 39406,
432
+ "level_3": 10349,
433
+ "level_4": 6812
434
+ },
435
+ "post_sampler_target_multiset": {
436
+ "rows": 96000,
437
+ "sha256": "754d7d17d1cd703698facae70fb7e06273e0448f6e8ffd09a1c24b4f1e7f9650",
438
+ "unique_signatures": 74742
439
+ },
440
+ "rows_per_rank": [
441
+ 48000,
442
+ 48000
443
+ ],
444
+ "sampler_epoch": 0,
445
+ "sampler_seed": 0,
446
+ "world_size": 2
447
+ },
448
+ "s1_e1": {
449
+ "dataset_rows": 96000,
450
+ "padded_records": [],
451
+ "padding_rows": 0,
452
+ "post_sampler_rows": 96000,
453
+ "post_sampler_target_level_counts": {
454
+ "level_1": 39432,
455
+ "level_2": 39406,
456
+ "level_3": 10350,
457
+ "level_4": 6812
458
+ },
459
+ "post_sampler_target_multiset": {
460
+ "rows": 96000,
461
+ "sha256": "52e7e67b73a53be614d8893fede02ce87dd7796e162e64dce005e4801043752e",
462
+ "unique_signatures": 74637
463
+ },
464
+ "rows_per_rank": [
465
+ 48000,
466
+ 48000
467
+ ],
468
+ "sampler_epoch": 1,
469
+ "sampler_seed": 0,
470
+ "world_size": 2
471
+ },
472
+ "s1_e2": {
473
+ "dataset_rows": 96000,
474
+ "padded_records": [],
475
+ "padding_rows": 0,
476
+ "post_sampler_rows": 96000,
477
+ "post_sampler_target_level_counts": {
478
+ "level_1": 39433,
479
+ "level_2": 39406,
480
+ "level_3": 10350,
481
+ "level_4": 6811
482
+ },
483
+ "post_sampler_target_multiset": {
484
+ "rows": 96000,
485
+ "sha256": "5e15704c9d77bfa5a93d03913ee68ef9ddcb0e7873e3fd10468aece277bb925c",
486
+ "unique_signatures": 74722
487
+ },
488
+ "rows_per_rank": [
489
+ 48000,
490
+ 48000
491
+ ],
492
+ "sampler_epoch": 2,
493
+ "sampler_seed": 0,
494
+ "world_size": 2
495
+ },
496
+ "s2_e0": {
497
+ "dataset_rows": 48000,
498
+ "padded_records": [],
499
+ "padding_rows": 0,
500
+ "post_sampler_rows": 48000,
501
+ "post_sampler_target_level_counts": {
502
+ "level_1": 19717,
503
+ "level_2": 19703,
504
+ "level_3": 5174,
505
+ "level_4": 3406
506
+ },
507
+ "post_sampler_target_multiset": {
508
+ "rows": 48000,
509
+ "sha256": "1d726e86b656a6ade37297099ac2c74df04bb1ee499e8c908db8c6f0d622ec74",
510
+ "unique_signatures": 42127
511
+ },
512
+ "rows_per_rank": [
513
+ 24000,
514
+ 24000
515
+ ],
516
+ "sampler_epoch": 0,
517
+ "sampler_seed": 0,
518
+ "world_size": 2
519
+ },
520
+ "s2_e1": {
521
+ "dataset_rows": 48000,
522
+ "padded_records": [],
523
+ "padding_rows": 0,
524
+ "post_sampler_rows": 48000,
525
+ "post_sampler_target_level_counts": {
526
+ "level_1": 19716,
527
+ "level_2": 19703,
528
+ "level_3": 5175,
529
+ "level_4": 3406
530
+ },
531
+ "post_sampler_target_multiset": {
532
+ "rows": 48000,
533
+ "sha256": "3e41fa7d73ce832d26ed8fb74f58ba71b56c341a66f1a02f634d64759e3c91aa",
534
+ "unique_signatures": 42149
535
+ },
536
+ "rows_per_rank": [
537
+ 24000,
538
+ 24000
539
+ ],
540
+ "sampler_epoch": 1,
541
+ "sampler_seed": 0,
542
+ "world_size": 2
543
+ }
544
+ }
545
+ },
546
+ "record_counts": {
547
+ "stage1_splits": {
548
+ "test": {
549
+ "ordered_source_ids_sha256": "1eee1eb35f9991fec6445b81d84db9fd93b47c20069375fd5cb3498ad5912bb0",
550
+ "records": 1000
551
+ },
552
+ "train": {
553
+ "ordered_source_ids_sha256": "49bf9b3d1c6b86e449366a04e76f02799f6c1e8123fff7690fe3436a38140500",
554
+ "records": 48000
555
+ },
556
+ "val": {
557
+ "ordered_source_ids_sha256": "563569dd771d397d4be6c2827bab971c178b0612a46ac032991b0fd8d59b7794",
558
+ "records": 1000
559
+ }
560
+ },
561
+ "stage2_splits": {
562
+ "test": {
563
+ "ordered_source_ids_sha256": "5b768da6b1c9e483bacde94b3b37c770c6c119391dcb6e4665d99dbd9e4f167d",
564
+ "records": 1000
565
+ },
566
+ "train": {
567
+ "ordered_source_ids_sha256": "5cb1a3b86162ac8e7fd9ded24a727cffbce1a023b8d9cdf34155d2cb864421d7",
568
+ "records": 48000
569
+ },
570
+ "val": {
571
+ "ordered_source_ids_sha256": "210112965f2a86d54bcd2c0f5198c2622aafa4c588fab0f3cd4c4ff27316cd7f",
572
+ "records": 1000
573
+ }
574
+ },
575
+ "usable": 50000
576
+ },
577
+ "schema_version": 1,
578
+ "split_overlap_audit": {
579
+ "historical_cross_stage": {
580
+ "stage1_train_vs_stage2_val_test": {
581
+ "count": 1929,
582
+ "first_20": [
583
+ 332,
584
+ 652,
585
+ 1236,
586
+ 2077,
587
+ 2204,
588
+ 3566,
589
+ 4410,
590
+ 4680,
591
+ 4806,
592
+ 5415,
593
+ 5831,
594
+ 5958,
595
+ 6277,
596
+ 6874,
597
+ 7187,
598
+ 7819,
599
+ 8019,
600
+ 8676,
601
+ 8944,
602
+ 8945
603
+ ],
604
+ "sha256": "a0a1428397089c4a3fad910ad4c8e16477a643956f002c1f30b06e67fa3174b4"
605
+ },
606
+ "stage2_train_vs_stage1_val_test": {
607
+ "count": 1929,
608
+ "first_20": [
609
+ 177,
610
+ 512,
611
+ 1239,
612
+ 2124,
613
+ 2393,
614
+ 3462,
615
+ 4276,
616
+ 4564,
617
+ 4878,
618
+ 5706,
619
+ 6074,
620
+ 6162,
621
+ 6202,
622
+ 7197,
623
+ 7198,
624
+ 7962,
625
+ 8094,
626
+ 8839,
627
+ 9026,
628
+ 9131
629
+ ],
630
+ "sha256": "8c836b94530f4f0b94250203b88509371e5f06eee6a9369d9dfaffbb594f0639"
631
+ }
632
+ },
633
+ "r3_with_target_stage_val_test": {
634
+ "stage1": {
635
+ "count": 0,
636
+ "first_20": [],
637
+ "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
638
+ },
639
+ "stage2": {
640
+ "count": 0,
641
+ "first_20": [],
642
+ "sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
643
+ }
644
+ }
645
+ }
646
+ }
data/training/ablation/r3_no_easyfirst/m_s1/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c2a30603cdb56613a931ef1362d842a4b9924aee1dcb7ac4590b79972ae8b132
3
+ size 734345