Add DoRA+ training report: qwen3-0.6b_D_1k [20260826_052412]

reports/v2.0/doraplus/reports_training/qwen3-0.6b_D_1k_20260826_052412.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "qwen3-0.6b",
+  "model_id": "Qwen/Qwen3-0.6B",
+  "lora_config": "D",
+  "lora_config_desc": "Heavy — full attention + MLP, rank 32",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_052412",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": false,
+  "trainable_params": 20529152,
+  "total_params": 616579072,
+  "trainable_pct": 3.3295,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 15515.7,
+  "total_training_time_s": 889.6,
+  "final_train_loss": 0.086283,
+  "final_eval_loss": 0.038565,
+  "final_val_accuracy": 0.98,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen3-0.6b_DoRA+_D_1k_20260826-052401",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.23928045272827148,
+      "grad_norm": 0.11181318759918213,
+      "learning_rate": 5.352691903491303e-08,
+      "entropy": 0.09659767880453728,
+      "num_tokens": 304665.0,
+      "mean_token_accuracy": 0.9625000017881393,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.14118249714374542,
+      "eval_model_preparation_time": 0.0215,
+      "eval_runtime": 8.7855,
+      "eval_samples_per_second": 11.382,
+      "eval_steps_per_second": 1.48,
+      "eval_entropy": 0.032727361334344514,
+      "eval_num_tokens": 304665.0,
+      "eval_mean_token_accuracy": 0.9807692261842581,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.94
+    },
+    {
+      "loss": 0.04970855236053467,
+      "grad_norm": 2.6163506507873535,
+      "learning_rate": 4.994647308096509e-05,
+      "entropy": 0.02492827997135464,
+      "num_tokens": 609330.0,
+      "mean_token_accuracy": 0.9908333352208137,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.12043285369873047,
+      "eval_model_preparation_time": 0.0215,
+      "eval_runtime": 8.7947,
+      "eval_samples_per_second": 11.37,
+      "eval_steps_per_second": 1.478,
+      "eval_entropy": 0.05699912630594694,
+      "eval_num_tokens": 609330.0,
+      "eval_mean_token_accuracy": 0.9647435866869413,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.91
+    },
+    {
+      "loss": 0.04468375682830811,
+      "grad_norm": 0.40879589319229126,
+      "learning_rate": 4.803679899192337e-07,
+      "entropy": 0.040532562478911134,
+      "num_tokens": 913995.0,
+      "mean_token_accuracy": 0.9908333352208137,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.027890466153621674,
+      "eval_model_preparation_time": 0.0215,
+      "eval_runtime": 8.7836,
+      "eval_samples_per_second": 11.385,
+      "eval_steps_per_second": 1.48,
+      "eval_entropy": 0.030569107844852485,
+      "eval_num_tokens": 913995.0,
+      "eval_mean_token_accuracy": 0.9967948702665476,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.99
+    },
+    {
+      "loss": 0.011460347175598145,
+      "grad_norm": 0.004923271015286446,
+      "learning_rate": 4.867325323737765e-05,
+      "entropy": 0.006250606294124736,
+      "num_tokens": 1218660.0,
+      "mean_token_accuracy": 0.9983333337306977,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.03856474906206131,
+      "eval_model_preparation_time": 0.0215,
+      "eval_runtime": 8.7756,
+      "eval_samples_per_second": 11.395,
+      "eval_steps_per_second": 1.481,
+      "eval_entropy": 0.01171558703152606,
+      "eval_num_tokens": 1218660.0,
+      "eval_mean_token_accuracy": 0.9935897405330951,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.98
+    },
+    {
+      "train_runtime": 889.1128,
+      "train_samples_per_second": 3.599,
+      "train_steps_per_second": 0.225,
+      "total_flos": 4906693000691712.0,
+      "train_loss": 0.0862832772731781,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/qwen3-0.6b_D_1k_20260826_052412.json"
+}