thealper2/graphcodebert-code-clone-detection

1{2  "model_name": "microsoft/graphcodebert-base",3  "model_dir": "/mnt/d/work2/graphcodebert-code-clone-detection/models/graphcodebert-clone-detection",4  "dataset_name": "PoolC/1-fold-clone-detection-600k-5fold",5  "dataset_split_strategy": {6    "train_split": "train",7    "heldout_split": "val",8    "note": "The repository provides one of 5 predefined folds as `train` + `val`; those groups are disjoint and are kept as-is. `val` is partitioned further into validation/test along problem-group boundaries.",9    "heldout_groups": 59,10    "validation_groups": 29,11    "test_groups": 30,12    "dropped_cross_boundary_pairs": 337398,13    "validation": {14      "num_examples": 483738,15      "negatives_label_0": 157558,16      "positives_label_1": 326180,17      "positive_ratio": 0.674291,18      "num_groups": 2919    },20    "test": {21      "num_examples": 503224,22      "negatives_label_0": 167224,23      "positives_label_1": 336000,24      "positive_ratio": 0.667695,25      "num_groups": 3026    }27  },28  "num_train_examples": 50000,29  "num_validation_examples": 20000,30  "num_test_examples": 20000,31  "class_distributions": {32    "train": {33      "num_examples": 50000,34      "negatives_label_0": 25000,35      "positives_label_1": 25000,36      "positive_ratio": 0.5,37      "num_groups": 24038    },39    "validation": {40      "num_examples": 20000,41      "negatives_label_0": 10000,42      "positives_label_1": 10000,43      "positive_ratio": 0.5,44      "num_groups": 2945    },46    "test": {47      "num_examples": 20000,48      "negatives_label_0": 10000,49      "positives_label_1": 10000,50      "positive_ratio": 0.5,51      "num_groups": 3052    }53  },54  "class_weighting": {55    "mode": "auto",56    "threshold": 0.6,57    "majority_class_share": 0.5,58    "applied": false,59    "weights": null,60    "reason": "Measured majority-class share 0.5000 is within the 0.6 threshold, so weighted cross entropy is NOT used."61  },62  "sequence_length": 512,63  "data_flow_length": 128,64  "total_sequence_length": 640,65  "per_device_train_batch_size": 16,66  "gradient_accumulation_steps": 1,67  "effective_batch_size": 16,68  "learning_rate": 2e-05,69  "num_train_epochs": 3.0,70  "optimizer": "adamw_torch",71  "scheduler": "linear",72  "warmup_ratio": 0.1,73  "warmup_steps": 938,74  "weight_decay": 0.01,75  "max_grad_norm": 1.0,76  "mixed_precision": "fp16",77  "gradient_checkpointing": false,78  "seed": 42,79  "training_time_seconds": 6921.9,80  "training_time_hours": 1.923,81  "train_runtime_metrics": {82    "train_runtime": 6918.8074,83    "train_samples_per_second": 21.68,84    "train_steps_per_second": 1.355,85    "total_flos": 0.0,86    "train_loss": 0.43145505716959637,87    "epoch": 3.088  },89  "gpu": {90    "cuda_available": true,91    "torch_version": "2.11.0+cu128",92    "transformers_version": "5.17.0",93    "python_version": "3.12.3",94    "platform": "Linux-6.18.33.2-microsoft-standard-WSL2-x86_64-with-glibc2.39",95    "gpu_name": "NVIDIA GeForce RTX 5060 Ti",96    "gpu_count": 1,97    "gpu_total_memory_gb": 15.9,98    "gpu_capability": "12.0",99    "cuda_version": "12.8"100  },101  "parameters": {102    "trainable_parameters": 125236994,103    "total_parameters": 125236994104  },105  "best_validation_f1": 0.8671882190520018,106  "best_checkpoint": "./outputs/checkpoint-9000",107  "validation_metrics": {108    "loss": 0.34459105134010315,109    "accuracy": 0.8557,110    "precision": 0.8032395566922421,111    "recall": 0.9422,112    "f1": 0.8671882190520018,113    "macro_f1": 0.8546121719233737,114    "tp": 9422,115    "tn": 7692,116    "fp": 2308,117    "fn": 578,118    "confusion_matrix": [119      [120        7692,121        2308122      ],123      [124        578,125        9422126      ]127    ],128    "confusion_matrix_layout": "[[TN, FP], [FN, TP]]",129    "support": {130      "num_examples": 20000,131      "label_0": 10000,132      "label_1": 10000133    },134    "runtime": 173.9737,135    "samples_per_second": 114.96,136    "steps_per_second": 3.592,137    "num_examples": 20000,138    "eval_seconds": 174.0139  },140  "test_metrics": {141    "loss": 0.3165612816810608,142    "model_preparation_time": 0.0022,143    "accuracy": 0.87465,144    "precision": 0.8409939018840448,145    "recall": 0.924,146    "f1": 0.8805450993472149,147    "macro_f1": 0.874343974488208,148    "tp": 9240,149    "tn": 8253,150    "fp": 1747,151    "fn": 760,152    "confusion_matrix": [153      [154        8253,155        1747156      ],157      [158        760,159        9240160      ]161    ],162    "confusion_matrix_layout": "[[TN, FP], [FN, TP]]",163    "support": {164      "num_examples": 20000,165      "label_0": 10000,166      "label_1": 10000167    },168    "runtime": 181.6455,169    "samples_per_second": 110.105,170    "steps_per_second": 3.441,171    "num_examples": 20000,172    "eval_seconds": 181.7173  },174  "preprocessing": {175    "snippet_pool": {176      "num_unique_snippets": 44950,177      "scan_seconds": 46.3,178      "snippets_with_multiple_groups": 1,179      "cross_split_snippet_overlap": {180        "train|val:snippets": 0,181        "train|val:groups": 0182      },183      "per_split": {184        "train": {185          "num_examples": 5388622,186          "negatives_label_0": 2694311,187          "positives_label_1": 2694311,188          "positive_ratio": 0.5,189          "num_groups": 240190        },191        "val": {192          "num_examples": 1324360,193          "negatives_label_0": 662180,194          "positives_label_1": 662180,195          "positive_ratio": 0.5,196          "num_groups": 59197        }198      }199    },200    "feature_extraction": {201      "num_snippets": 44950,202      "extraction_seconds": 64.1,203      "status_counts": {204        "ok": 44930,205        "comment_strip_failed": 13,206        "dfg_failed": 7207      },208      "snippets_with_empty_dataflow": 263,209      "dataflow_nodes_mean": 44.22,210      "dataflow_nodes_p50": 33,211      "dataflow_nodes_p95": 127,212      "dataflow_nodes_max": 193,213      "code_tokens_mean": 137.9,214      "code_tokens_truncated": 885,215      "total_dataflow_edges": 2481388,216      "sequence_length": 640217    },218    "subsampling": {219      "train": {220        "subsampled": true,221        "kept": 50000,222        "balanced": true223      },224      "validation": {225        "subsampled": true,226        "kept": 20000,227        "balanced": true228      },229      "test": {230        "subsampled": true,231        "kept": 20000,232        "balanced": true233      }234    }235  },236  "sanity_check": {237    "passed": true,238    "device": "cuda",239    "checks": [240      {241        "check": "1_dataset_loading",242        "passed": true,243        "detail": "train=50000 val=20000 test=20000"244      },245      {246        "check": "2_column_detection",247        "passed": true,248        "detail": "code columns=code1/code2, label=similar; excluded from features: ['code1_group', 'code2_group', 'pair_id', 'question_pair_id']"249      },250      {251        "check": "3_label_correctness",252        "passed": true,253        "detail": "labels in {0,1}; validation positive ratio=0.5"254      },255      {256        "check": "4_tokenisation",257        "passed": true,258        "detail": "<s>...</s> wrapping OK, 93 code tokens, decodes to 'import math a , b , c = map ( int , input ( ) . split ( ) ) '"259      },260      {261        "check": "5_dataflow_extraction",262        "passed": true,263        "detail": "44687/44950 snippets have a non-empty data-flow graph, 2481388 edges, mean nodes=44.22"264      },265      {266        "check": "6_attention_mask",267        "passed": true,268        "detail": "shape=(640, 640) (code_length 512 + data_flow_length 128), 93 code tokens, 30 nodes, density=0.0215, padding rows empty"269      },270      {271        "check": "7_forward_pass",272        "passed": true,273        "detail": "logits shape=(16, 2) for batch of 16 pairs; mask tensor=(16, 640, 640)"274      },275      {276        "check": "8_finite_loss",277        "passed": true,278        "detail": "loss=0.6978 (chance level ~0.6931)"279      },280      {281        "check": "9_training_step",282        "passed": true,283        "detail": "loss=0.6798, grad_norm=2.4929, weights updated (amp=on, dtype=torch.float16)"284      }285    ],286    "parameters": {287      "trainable_parameters": 125236994,288      "total_parameters": 125236994289    },290    "sequence_length": 640291  },292  "test_metrics_provenance": {293    "note": "Re-scored after fixing a transformers v5 dataloader-caching bug: Trainer.get_eval_dataloader caches under the key 'eval' when a Dataset object is passed and dataloader_persistent_workers=True, so the original in-training test evaluation silently re-scored the validation split. evaluate_split now uses Trainer.predict and verifies the returned labels.",294    "superseded_test_f1": 0.8671882190520018,295    "corrected_test_f1": 0.8805450993472149,296    "validation_f1_unchanged": 0.8671882190520018297  }298}