Add DoRA+ training report: llama3.2-1b_D_1k [20260826_070303]

reports/v2.0/doraplus/reports_training/llama3.2-1b_D_1k_20260826_070303.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "llama3.2-1b",
+  "model_id": "meta-llama/Llama-3.2-1B-Instruct",
+  "lora_config": "D",
+  "lora_config_desc": "Heavy — full attention + MLP, rank 32",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_070303",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 22921216,
+  "total_params": 1258735616,
+  "trainable_pct": 1.821,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 4603.5,
+  "total_training_time_s": 1870.8,
+  "final_train_loss": 0.109895,
+  "final_eval_loss": 0.391134,
+  "final_val_accuracy": 0.95,
+  "final_train_accuracy": 0.98,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/llama3.2-1b_DoRA+_D_1k_20260826-070251",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.3256666946411133,
+      "grad_norm": 0.5068637132644653,
+      "learning_rate": 5.352691903491303e-08,
+      "entropy": 0.2557024986273609,
+      "num_tokens": 318964.0,
+      "mean_token_accuracy": 0.921875,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.14365483820438385,
+      "eval_model_preparation_time": 0.0119,
+      "eval_runtime": 12.8011,
+      "eval_samples_per_second": 7.812,
+      "eval_steps_per_second": 1.016,
+      "eval_entropy": 0.06792628077360299,
+      "eval_num_tokens": 318964.0,
+      "eval_mean_token_accuracy": 0.9759615384615384,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.0622498083114624,
+      "grad_norm": 0.04171453043818474,
+      "learning_rate": 4.994647308096509e-05,
+      "entropy": 0.018900299305896623,
+      "num_tokens": 637928.0,
+      "mean_token_accuracy": 0.991875,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.09829603880643845,
+      "eval_model_preparation_time": 0.0119,
+      "eval_runtime": 12.8073,
+      "eval_samples_per_second": 7.808,
+      "eval_steps_per_second": 1.015,
+      "eval_entropy": 0.049963345392965354,
+      "eval_num_tokens": 637928.0,
+      "eval_mean_token_accuracy": 0.9711538461538461,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.94
+    },
+    {
+      "loss": 0.04880781650543213,
+      "grad_norm": 0.06020014360547066,
+      "learning_rate": 4.803679899192337e-07,
+      "entropy": 0.01415753465546004,
+      "num_tokens": 956892.0,
+      "mean_token_accuracy": 0.991875,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.206960529088974,
+      "eval_model_preparation_time": 0.0119,
+      "eval_runtime": 12.7941,
+      "eval_samples_per_second": 7.816,
+      "eval_steps_per_second": 1.016,
+      "eval_entropy": 0.013269053825822014,
+      "eval_num_tokens": 956892.0,
+      "eval_mean_token_accuracy": 0.9711538461538461,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.0028571030497550965,
+      "grad_norm": 3.4570674896240234,
+      "learning_rate": 4.867325323737765e-05,
+      "entropy": 0.0038452082678122677,
+      "num_tokens": 1275856.0,
+      "mean_token_accuracy": 0.998125,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.39113354682922363,
+      "eval_model_preparation_time": 0.0119,
+      "eval_runtime": 12.8061,
+      "eval_samples_per_second": 7.809,
+      "eval_steps_per_second": 1.015,
+      "eval_entropy": 0.0032777108908004057,
+      "eval_num_tokens": 1275856.0,
+      "eval_mean_token_accuracy": 0.9711538461538461,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.95
+    },
+    {
+      "train_runtime": 1870.3047,
+      "train_samples_per_second": 1.711,
+      "train_steps_per_second": 0.107,
+      "total_flos": 1.0943369400188928e+16,
+      "train_loss": 0.10989535562694072,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/llama3.2-1b_D_1k_20260826_070303.json"
+}