sync EmbedEd artifacts (phase01-v5)

.gitattributes

@@ -25,3 +25,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+artifacts/codexglue_data.jsonl filter=lfs diff=lfs merge=lfs -text
+artifacts/fragments.jsonl filter=lfs diff=lfs merge=lfs -text

artifacts/codexglue_data.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8bc51e62deddcc45bd26c5b57f5add2a2cf377f13b9f6c2fb656fbc8fca4dd2
+size 15174797

artifacts/fragments.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9cbcb4fce6b1a2dc8d01b6856f89186392347aa7c6eea5d02e5d2a2a489e6b2f
+size 14045244

artifacts/manifest.json

@@ -0,0 +1,10 @@
+{
+  "version": "phase01-v5",
+  "n_data_lines": 9126,
+  "n_fragments": 8063,
+  "n_pairs": {
+    "train": 901028,
+    "valid": 415416,
+    "test": 415416
+  }
+}

artifacts/overlap.json

@@ -0,0 +1,6 @@
+{
+  "train_frag": 6527,
+  "test_frag": 860,
+  "overlap": 76,
+  "overlap_frac_of_test": 0.0884
+}

artifacts/token_lengths.json

@@ -0,0 +1,10 @@
+{
+  "tokenizer": "microsoft/graphcodebert-base",
+  "n": 8063,
+  "p50": 474,
+  "p90": 1535,
+  "p95": 2233,
+  "p99": 5944,
+  "max": 36823,
+  "recommended_max_len": 512
+}

report/measurements.md

@@ -0,0 +1,471 @@
+# Measurements
+
+## Phase 0 — measured numbers
+
+generated by `python -m embeded.data.prepare_data` — version `phase01-v5`
+
+| quantity | spec value | measured | ok |
+|---|---|---|---|
+| data.jsonl lines | 9126 | 9126 | ✅ |
+| unique fragments | 8063 | 8063 | ✅ |
+| train pairs | 901028 | 901028 | ✅ |
+| valid pairs | 415416 | 415416 | ✅ |
+| test pairs | 415416 | 415416 | ✅ |
+
+**Train/test fragment overlap** (§4.2, text space): 76 of 860 test fragments (8.8%) also appear in a train pair. Standard-benchmark F1 is an optimistic upper bound to this extent; the generalisation test is there to expose it.
+
+**Token lengths** (microsoft/graphcodebert-base): p50=474 p90=1535 p95=2233 p99=5944 max=36823 → **max_len = 512**.
+
+
+## Phase 0.5 — throughput (measured, fixes BATCH / TRAIN_PAIRS_CAP / EPOCHS)
+
+generated by `python -m scripts.phase0_throughput --candidates 512:4 512:8 512:16 512:32 --minutes 0.5` — version `phase01-v5`, recipe `AMP_DTYPE=float16`, `GRAD_CHECKPOINT=False` (Colab Tesla T4, 14.5 GB, 2026-09-26)
+
+| max_len:batch (triples) | triples/s | peak GB | cap for one 40-min epoch |
+|---|---|---|---|
+| 512:4 | 11.9 | 4.37 | 28,648 |
+| 512:8 | 13.5 | 6.71 | 32,367 |
+| 512:16 | 14.1 | 11.39 | 33,951 |
+| 512:32 | OOM on Tesla T4 | — | — |
+
+Suggested: **BATCH = 8, TRAIN_PAIRS_CAP = 32,367, EPOCHS = 1** (rule: fastest-within-10% with the lowest peak memory; 13.5 vs top 14.1 triples/s). `settings.py` at run time: BATCH = 8, TRAIN_PAIRS_CAP = 32,000, EPOCHS = 1.
+
+Frozen in `settings.py` as `TRAIN_PAIRS_CAP = 32_000` (= 1,600 anchors × k=20 = 4,000 steps of 8; measured cap rounded down). Context: the same test in plain fp32 with the original candidates OOMed on its first candidate (256:32 = 96 sequences/step, 14.3 GB), which is why the training recipe is fp16 AMP.
+
+## s′ acquisition (Phase 0.7)
+
+**verdict: PASS** — Clones paper replication package release zip/datasets/bcb_v2_sampled_bf/data_bcb_v2_sampled_bf.pickle: 4600 pairs, labels {1: 2300, 0: 2300}, functionality cols {'functionality_id': 23} (+4 duplicate copy/copies: Clones paper replication package release zip/llms/dataset.pickle, Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/blocks.pickle, Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pair_ids.pickle, Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pairs.pickle)
+
+```json
+{
+  "n_files": 2129,
+  "n_git_files_skipped": 0,
+  "files": [
+    {
+      "path": "Detecting_clones_of_unseen_functionality_ASE25.zip",
+      "size": 138423097
+    },
+    {
+      "path": "Clones paper replication package release zip/datasets/bcb_v2_sampled/bcb_funcs_all.tsv",
+      "size": 81658277
+    },
+    {
+      "path": "Clones paper replication package release zip/llms/llm_data_oj.csv",
+      "size": 35637449
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/ast.pkl",
+      "size": 31429135
+    },
+    {
+      "path": "Clones paper replication package release zip/datasets/oj/programs.pkl",
+      "size": 30572619
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/ast.pkl",
+      "size": 28254515
+    },
+    {
+      "path": "Clones paper replication package release zip/llms/dataset.pickle",
+      "size": 13348027
+    },
+    {
+      "path": "Clones paper replication package release zip/datasets/bcb_v2_sampled_bf/data_bcb_v2_sampled_bf.pickle",
+      "size": 13347993
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/methods.pickle",
+      "size": 8286455
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/blocks.pickle",
+      "size": 7942669
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/methods.pickle",
+      "size": 7518647
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/blocks.pickle",
+      "size": 6950754
+    },
+    {
+      "path": "Clones paper replication package release zip/llms/llm_data.csv",
+      "size": 5699841
+    },
+    {
+      "path": "Clones paper replication package release zip/oj_dataset_experiments/astnn/data/c/oj_and_scb/programs.pkl",
+      "size": 4682591
+    },
+    {
+      "path": "Clones paper replication package release zip/codegrid/data/word2vec_bcb_v2_sampled_bf_and_scb.bin",
+      "size": 3767507
+    },
+    {
+      "path": "Clones paper replication package release zip/codegrid/data/word2vec_bcb_v2_sampled_bf.bin",
+      "size": 3536861
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/embedding/node_w2v_128",
+      "size": 3153555
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/embedding/node_w2v_128",
+      "size": 3086471
+    },
+    {
+      "path": "Clones paper replication package release zip/oj_dataset_experiments/astnn/data/c/oj_and_scb/pair_ids.pkl",
+      "size": 2139613
+    },
+    {
+      "path": "Clones paper replication package release zip/datasets/bcb_v2_sampled/bcb_pair_ids.pkl",
+      "size": 1167949
+    },
+    {
+      "path": "Clones paper replication package release zip/oj_dataset_experiments/codegrid/data/word2vec_oj_and_scb.bin",
+      "size": 565457
+    },
+    {
+      "path": "Clones paper replication package release zip/datasets/oj/oj_clone_ids.pkl",
+      "size": 401007
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/pair_ids.pickle",
+      "size": 264781
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/pairs.pickle",
+      "size": 264781
+    },
+    {
+      "path": "Clones paper replication package release zip/oj_dataset_experiments/codegrid/data/word2vec_oj.bin",
+      "size": 216582
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pair_ids.pickle",
+      "size": 184724
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pairs.pickle",
+      "size": 184724
+    },
+    {
+      "path": "Clones paper replication package release zip/llms/llm_experiments.ipynb",
+      "size": 50156
+    },
+    {
+      "path": "Clones paper replication package release zip/astnn/create_dataset_for_astnn_pipeline.ipynb",
+      "size": 43078
+    },
+    {
+      "path": "Clones paper replication package release zip/assets/contrastive_learning_architecture.drawio.pdf",
+      "size": 37880
+    }
+  ],
+  "probes": {
+    "Clones paper replication package release zip/datasets/bcb_v2_sampled/bcb_funcs_all.tsv": {
+      "kind": "text",
+      "rows": 1674982,
+      "cols": [
+        "74",
+        "\"    static void copy(String src, String dest) throws IOException {"
+      ]
+    },
+    "Clones paper replication package release zip/llms/llm_data_oj.csv": {
+      "kind": "text",
+      "rows": 2268479,
+      "cols": [
+        "id1",
+        "id2",
+        "label",
+        "code1",
+        "functionality_id1",
+        "code2",
+        "functionality_id2",
+        "functionality_id"
+      ]
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/ast.pkl": {
+      "kind": "pickle",
+      "error": "load failed: ModuleNotFoundError: No module named 'javalang'"
+    },
+    "Clones paper replication package release zip/datasets/oj/programs.pkl": {
+      "kind": "pickle",
+      "error": "load failed: ModuleNotFoundError: No module named 'pandas.core.indexes.numeric'"
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/ast.pkl": {
+      "kind": "pickle",
+      "error": "load failed: ModuleNotFoundError: No module named 'javalang'"
+    },
+    "Clones paper replication package release zip/llms/dataset.pickle": {
+      "kind": "dataframe",
+      "rows": 4600,
+      "columns": [
+        "code1",
+        "code2",
+        "label",
+        "functionality_id"
+      ],
+      "label_counts": {
+        "1": 2300,
+        "0": 2300
+      },
+      "unique_label": 2,
+      "unique_functionality_id": 23
+    },
+    "Clones paper replication package release zip/datasets/bcb_v2_sampled_bf/data_bcb_v2_sampled_bf.pickle": {
+      "kind": "dataframe",
+      "rows": 4600,
+      "columns": [
+        "code1",
+        "code2",
+        "label",
+        "functionality_id"
+      ],
+      "label_counts": {
+        "1": 2300,
+        "0": 2300
+      },
+      "unique_label": 2,
+      "unique_functionality_id": 23
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/methods.pickle": {
+      "kind": "dataframe",
+      "rows": 5371,
+      "columns": [
+        "id",
+        "code"
+      ],
+      "unique_id": 5371
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/blocks.pickle": {
+      "kind": "dataframe",
+      "rows": 6594,
+      "columns": [
+        "id1",
+        "id2",
+        "label",
+        "functionality_id",
+        "code_x",
+        "code_y"
+      ],
+      "label_counts": {
+        "1": 3297,
+        "0": 3297
+      },
+      "unique_id1": 4073,
+      "unique_id2": 2481,
+      "unique_label": 2
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/methods.pickle": {
+      "kind": "dataframe",
+      "rows": 4140,
+      "columns": [
+        "id",
+        "code"
+      ],
+      "unique_id": 4140
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/blocks.pickle": {
+      "kind": "dataframe",
+      "rows": 4600,
+      "columns": [
+        "id1",
+        "id2",
+        "label",
+        "functionality_id",
+        "code_x",
+        "code_y"
+      ],
+      "label_counts": {
+        "1": 2300,
+        "0": 2300
+      },
+      "unique_id1": 3072,
+      "unique_id2": 1515,
+      "unique_label": 2,
+      "unique_functionality_id": 23
+    },
+    "Clones paper replication package release zip/llms/llm_data.csv": {
+      "kind": "text",
+      "rows": 123799,
+      "cols": [
+        "code1",
+        "code2",
+        "label",
+        "functionality_id"
+      ]
+    },
+    "Clones paper replication package release zip/oj_dataset_experiments/astnn/data/c/oj_and_scb/programs.pkl": {
+      "kind": "dataframe",
+      "rows": 8290,
+      "columns": [
+        "id",
+        "code"
+      ],
+      "unique_id": 8290
+    },
+    "Clones paper replication package release zip/oj_dataset_experiments/astnn/data/c/oj_and_scb/pair_ids.pkl": {
+      "kind": "dataframe",
+      "rows": 50920,
+      "columns": [
+        "id1",
+        "id2",
+        "label",
+        "functionality_id",
+        "functionality_id2"
+      ],
+      "label_counts": {
+        "0": 47174,
+        "1": 3746
+      },
+      "unique_id1": 8099,
+      "unique_id2": 8095,
+      "unique_label": 2
+    },
+    "Clones paper replication package release zip/datasets/bcb_v2_sampled/bcb_pair_ids.pkl": {
+      "kind": "dataframe",
+      "rows": 97535,
+      "columns": [
+        "id1",
+        "id2",
+        "label"
+      ],
+      "label_counts": {
+        "4": 20000,
+        "5": 20000,
+        "0": 20000,
+        "3": 18317,
+        "1": 15555,
+        "2": 3663
+      },
+      "unique_label": 6
+    },
+    "Clones paper replication package release zip/datasets/oj/oj_clone_ids.pkl": {
+      "kind": "dataframe",
+      "rows": 50000,
+      "columns": [
+        "id1",
+        "id2",
+        "label"
+      ],
+      "label_counts": {
+        "0": 46714,
+        "1": 3286
+      },
+      "unique_label": 2
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/pair_ids.pickle": {
+      "kind": "dataframe",
+      "rows": 6600,
+      "columns": [
+        "id1",
+        "id2",
+        "label",
+        "functionality_id"
+      ],
+      "label_counts": {
+        "1": 3300,
+        "0": 3300
+      },
+      "unique_id1": 4076,
+      "unique_id2": 2484,
+      "unique_label": 2
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf_and_scb/pairs.pickle": {
+      "kind": "dataframe",
+      "rows": 6600,
+      "columns": [
+        "id1",
+        "id2",
+        "label",
+        "functionality_id"
+      ],
+      "label_counts": {
+        "1": 3300,
+        "0": 3300
+      },
+      "unique_id1": 4076,
+      "unique_id2": 2484,
+      "unique_label": 2
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pair_ids.pickle": {
+      "kind": "dataframe",
+      "rows": 4600,
+      "columns": [
+        "id1",
+        "id2",
+        "label",
+        "functionality_id"
+      ],
+      "label_counts": {
+        "1": 2300,
+        "0": 2300
+      },
+      "unique_id1": 3072,
+      "unique_id2": 1515,
+      "unique_label": 2,
+      "unique_functionality_id": 23
+    },
+    "Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pairs.pickle": {
+      "kind": "dataframe",
+      "rows": 4600,
+      "columns": [
+        "id1",
+        "id2",
+        "label",
+        "functionality_id"
+      ],
+      "label_counts": {
+        "1": 2300,
+        "0": 2300
+      },
+      "unique_id1": 3072,
+      "unique_id2": 1515,
+      "unique_label": 2,
+      "unique_functionality_id": 23
+    },
+    "Clones paper replication package release zip/llms/functionalities.pickle": {
+      "kind": "dataframe",
+      "rows": 43,
+      "columns": [
+        "name",
+        "description",
+        "id",
+        "search_heuristic"
+      ],
+      "unique_name": 43,
+      "unique_id": 43
+    }
+  },
+  "scb": {
+    "Clones paper replication package release zip/datasets/scb/C": {
+      "Stand Alone Clones": {
+        "files": 1000,
+        "clone_named": 1000
+      }
+    },
+    "Clones paper replication package release zip/datasets/scb/Java": {
+      "Stand Alone Clones": {
+        "files": 1000,
+        "clone_named": 1000
+      }
+    }
+  },
+  "verdict": [
+    "PASS",
+    "Clones paper replication package release zip/datasets/bcb_v2_sampled_bf/data_bcb_v2_sampled_bf.pickle: 4600 pairs, labels {1: 2300, 0: 2300}, functionality cols {'functionality_id': 23} (+4 duplicate copy/copies: Clones paper replication package release zip/llms/dataset.pickle, Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/blocks.pickle, Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pair_ids.pickle, Clones paper replication package release zip/astnn/data/java/bcb_v2_sampled_bf/pairs.pickle)"
+  ]
+}
+```
+
+## Gate G1 — hardness check
+
+### run 2026-09-26 (Colab T4, first mining run) — version `phase01-v5`, margin 0.02
+
+- C1: mean cos(anchor, negative) = 0.96068 (sd 0.026, n = 20000)
+- C2: mean cos(anchor, negative) = 0.97526 (sd 0.022, n = 20000)
+- C3: mean cos(anchor, negative) = 0.98845 (sd 0.007, n = 20000)
+- **verdict: FAIL** — C1→C2 gap +0.0146 < margin 0.02 (mining produced equally easy negatives)
+
+Context recorded at the time: ordering C1 < C2 ≤ C3 holds; the C1→C2 gap is ≈70 standard errors (n = 20,000) and d ≈ 0.6, but the full C1→C3 range of this untuned, anisotropic space is only 0.028, so the absolute 0.02 margin (set in `settings.py` before the spread was known) asks C2 to cover ~75% of it. Per SCOPE P6-3 training is blocked; next step is `python -m scripts.hardness_diagnostics` (percentile-rank view), then either a mining fix or an explicit, documented re-operationalisation of "visible gap" — decided before any training, with this run kept here either way.

sync_manifest.json

@@ -0,0 +1,15 @@
+{
+  "version": "phase01-v5",
+  "synced_at_utc": "2026-09-26T15:24:12Z",
+  "staged_files": [
+    "artifacts/codexglue_data.jsonl",
+    "artifacts/fragments.jsonl",
+    "artifacts/manifest.json",
+    "artifacts/overlap.json",
+    "artifacts/pairs_test.tsv",
+    "artifacts/pairs_train.tsv",
+    "artifacts/pairs_valid.tsv",
+    "artifacts/token_lengths.json",
+    "report/measurements.md"
+  ]
+}