pytorch-training-debugger / validation /reports /fidelity_report.json
omkarrr88
minor changes
206438f
{
"methodology": "Real PyTorch 20-epoch mini-training with fault injection",
"torch_version": "2.5.1+cpu",
"models": [
"SimpleCNN (~50K params)",
"SimpleMLP (~20K params)"
],
"training_approach": "Real forward+backward passes on random CIFAR-10 style data, cached per (task_id, seed)",
"results": [
{
"task": "task_001",
"fault": "exploding_gradients",
"checks": {
"gradient_instability_detected": true,
"loss_shows_instability": true,
"max_gradient_norm": 111.8,
"max_loss": 43.27,
"real_pytorch_training": true
},
"pass": true
},
{
"task": "task_002",
"fault": "vanishing_gradients",
"checks": {
"vanishing_detected": true,
"min_gradient_norm": 0.0,
"real_pytorch_gradients": true
},
"pass": true
},
{
"task": "task_003",
"fault": "data_leakage",
"checks": {
"class_overlap_above_0.5": true,
"class_overlap_score": 0.83,
"real_training_runs": true,
"has_confusion_matrix": true
},
"pass": true
},
{
"task": "task_004",
"fault": "overfitting",
"checks": {
"real_training_runs": true,
"clean_data": true,
"final_train_loss": 0.1017,
"final_val_loss": 2.6519
},
"pass": true
},
{
"task": "task_005",
"fault": "batchnorm_eval_mode",
"checks": {
"all_layers_in_eval_mode": true,
"no_layer_is_exploding": true,
"real_training_runs": true,
"real_model_eval_mode": true,
"red_herring_spike_layer": "conv1"
},
"pass": true
},
{
"task": "task_006",
"fault": "code_bug",
"checks": {
"variants_tested": 4,
"variant_results": {
"eval_mode": {
"correct_fix_accepted": true,
"wrong_fix_rejected": true
},
"detach_loss": {
"correct_fix_accepted": true,
"wrong_fix_rejected": true
},
"zero_grad_missing": {
"correct_fix_accepted": true,
"wrong_fix_rejected": true
},
"inplace_relu": {
"correct_fix_accepted": true,
"wrong_fix_rejected": true
}
},
"fix_validation_pipeline": "normalize -> tokenize -> semantic -> AST"
},
"pass": true
},
{
"task": "task_007",
"fault": "scheduler_misconfigured",
"checks": {
"real_training_runs": true,
"scheduler_gamma": 0.0001,
"scheduler_step_size": 2,
"final_loss": 2.5911
},
"pass": true
},
{
"task": "architecture",
"fault": "dual_model_support",
"checks": {
"cnn_output_shape": [
4,
10
],
"mlp_output_shape": [
4,
10
],
"cnn_params": 66890,
"mlp_params": 411658,
"both_produce_10_classes": true
},
"pass": true
}
],
"summary": {
"total": 8,
"passed": 8,
"failed": 0
}
}