Add LoRA+ training report: qwen3-0.6b_C_1k [20260826_181953]

reports/v2.0/loraplus/reports_training/qwen3-0.6b_C_1k_20260826_181953.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": "LoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_181953",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 10092544,
+  "total_params": 606142464,
+  "trainable_pct": 1.665,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 2707.4,
+  "total_training_time_s": 738.5,
+  "final_train_loss": 0.108563,
+  "final_eval_loss": 0.123684,
+  "final_val_accuracy": 0.92,
+  "final_train_accuracy": 0.96,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen3-0.6b_LoRA+_C_1k_20260826-181949",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.30177635192871094,
+      "grad_norm": 0.6231783628463745,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.14148966075852512,
+      "num_tokens": 304665.0,
+      "mean_token_accuracy": 0.9454166662693023,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.12591677904129028,
+      "eval_model_preparation_time": 0.0181,
+      "eval_runtime": 4.7154,
+      "eval_samples_per_second": 21.207,
+      "eval_steps_per_second": 2.757,
+      "eval_entropy": 0.12716996039335543,
+      "eval_num_tokens": 304665.0,
+      "eval_mean_token_accuracy": 0.974358971302326,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.9,
+      "eval_accuracy": 0.91
+    },
+    {
+      "loss": 0.06612724781036378,
+      "grad_norm": 12.33492660522461,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.04182387651177123,
+      "num_tokens": 609330.0,
+      "mean_token_accuracy": 0.987083330154419,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.06929491460323334,
+      "eval_model_preparation_time": 0.0181,
+      "eval_runtime": 4.7357,
+      "eval_samples_per_second": 21.116,
+      "eval_steps_per_second": 2.745,
+      "eval_entropy": 0.029510275066758577,
+      "eval_num_tokens": 609330.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.01614093542098999,
+      "grad_norm": 0.016738321632146835,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.0147365781548433,
+      "num_tokens": 913995.0,
+      "mean_token_accuracy": 0.9958333313465119,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.09634676575660706,
+      "eval_model_preparation_time": 0.0181,
+      "eval_runtime": 4.7424,
+      "eval_samples_per_second": 21.086,
+      "eval_steps_per_second": 2.741,
+      "eval_entropy": 0.01893678502752804,
+      "eval_num_tokens": 913995.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.96
+    },
+    {
+      "loss": 0.050209183692932126,
+      "grad_norm": 5.338968753814697,
+      "learning_rate": 9.956320346634876e-05,
+      "entropy": 0.010882307557185413,
+      "num_tokens": 1218660.0,
+      "mean_token_accuracy": 0.9904166662693024,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.12368405610322952,
+      "eval_model_preparation_time": 0.0181,
+      "eval_runtime": 4.7425,
+      "eval_samples_per_second": 21.086,
+      "eval_steps_per_second": 2.741,
+      "eval_entropy": 0.07795339765457007,
+      "eval_num_tokens": 1218660.0,
+      "eval_mean_token_accuracy": 0.974358971302326,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.92
+    },
+    {
+      "train_runtime": 738.099,
+      "train_samples_per_second": 4.335,
+      "train_steps_per_second": 0.271,
+      "total_flos": 5429276835840000.0,
+      "train_loss": 0.10856342971324921,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/loraplus/reports_training/qwen3-0.6b_C_1k_20260826_181953.json"
+}