Add DoRA+ training report: qwen3-0.6b_C_1k [20260826_153242]

reports/v2.0/doraplus/reports_training/qwen3-0.6b_C_1k_20260826_153242.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "qwen3-0.6b",
+  "model_id": "Qwen/Qwen3-0.6B",
+  "lora_config": "C",
+  "lora_config_desc": "Wide — full attention + MLP, rank 16",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_153242",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 10436608,
+  "total_params": 606486528,
+  "trainable_pct": 1.7208,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 2949.4,
+  "total_training_time_s": 1428.5,
+  "final_train_loss": 0.105573,
+  "final_eval_loss": 0.225153,
+  "final_val_accuracy": 0.89,
+  "final_train_accuracy": 0.96,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen3-0.6b_DoRA+_C_1k_20260826-153232",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.2916320991516113,
+      "grad_norm": 2.743527412414551,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.1707103259768337,
+      "num_tokens": 304665.0,
+      "mean_token_accuracy": 0.9470833283662796,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.09938085824251175,
+      "eval_model_preparation_time": 0.0207,
+      "eval_runtime": 8.7257,
+      "eval_samples_per_second": 11.46,
+      "eval_steps_per_second": 1.49,
+      "eval_entropy": 0.08493176102638245,
+      "eval_num_tokens": 304665.0,
+      "eval_mean_token_accuracy": 0.9711538461538461,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.94,
+      "eval_accuracy": 0.92
+    },
+    {
+      "loss": 0.03447272777557373,
+      "grad_norm": 0.003604077035561204,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.01031205168867018,
+      "num_tokens": 609330.0,
+      "mean_token_accuracy": 0.9924999976158142,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.06900880485773087,
+      "eval_model_preparation_time": 0.0207,
+      "eval_runtime": 8.7269,
+      "eval_samples_per_second": 11.459,
+      "eval_steps_per_second": 1.49,
+      "eval_entropy": 0.004303412762685464,
+      "eval_num_tokens": 609330.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.004043597877025604,
+      "grad_norm": 2.7254981994628906,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.0042462384887039665,
+      "num_tokens": 913995.0,
+      "mean_token_accuracy": 0.9987499994039536,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.13582678139209747,
+      "eval_model_preparation_time": 0.0207,
+      "eval_runtime": 8.7306,
+      "eval_samples_per_second": 11.454,
+      "eval_steps_per_second": 1.489,
+      "eval_entropy": 0.006184923578985035,
+      "eval_num_tokens": 913995.0,
+      "eval_mean_token_accuracy": 0.9807692261842581,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.0921437931060791,
+      "grad_norm": 3.1604411602020264,
+      "learning_rate": 9.956320346634876e-05,
+      "entropy": 0.03684742745448602,
+      "num_tokens": 1218660.0,
+      "mean_token_accuracy": 0.9858333307504654,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.22515279054641724,
+      "eval_model_preparation_time": 0.0207,
+      "eval_runtime": 8.7218,
+      "eval_samples_per_second": 11.465,
+      "eval_steps_per_second": 1.491,
+      "eval_entropy": 0.02783593273936556,
+      "eval_num_tokens": 1218660.0,
+      "eval_mean_token_accuracy": 0.9615384615384616,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.89
+    },
+    {
+      "train_runtime": 1428.0695,
+      "train_samples_per_second": 2.241,
+      "train_steps_per_second": 0.14,
+      "total_flos": 5433422829060096.0,
+      "train_loss": 0.10557305447757244,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/qwen3-0.6b_C_1k_20260826_153242.json"
+}