Add DoRA+ training report: qwen2.5-0.5b_D_1k [20260826_044059]

reports/v2.0/doraplus/reports_training/qwen2.5-0.5b_D_1k_20260826_044059.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "qwen2.5-0.5b",
+  "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+  "lora_config": "D",
+  "lora_config_desc": "Heavy — full attention + MLP, rank 32",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_044059",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": false,
+  "trainable_params": 17900544,
+  "total_params": 511933312,
+  "trainable_pct": 3.4967,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 12313.8,
+  "total_training_time_s": 768.1,
+  "final_train_loss": 0.5071,
+  "final_eval_loss": 0.251979,
+  "final_val_accuracy": 0.81,
+  "final_train_accuracy": 0.94,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen2.5-0.5b_DoRA+_D_1k_20260826-044049",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.9479441833496094,
+      "grad_norm": 3.083339214324951,
+      "learning_rate": 5.352691903491303e-08,
+      "entropy": 0.869655083566904,
+      "num_tokens": 301465.0,
+      "mean_token_accuracy": 0.7337500038743019,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.833139181137085,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 7.2061,
+      "eval_samples_per_second": 13.877,
+      "eval_steps_per_second": 1.804,
+      "eval_entropy": 0.7924602169256943,
+      "eval_num_tokens": 301465.0,
+      "eval_mean_token_accuracy": 0.727564105620751,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.26,
+      "eval_accuracy": 0.18
+    },
+    {
+      "loss": 0.6631587982177735,
+      "grad_norm": 6.756750583648682,
+      "learning_rate": 4.994647308096509e-05,
+      "entropy": 0.6486798901110887,
+      "num_tokens": 602930.0,
+      "mean_token_accuracy": 0.7916666665673255,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.5930512547492981,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 7.219,
+      "eval_samples_per_second": 13.852,
+      "eval_steps_per_second": 1.801,
+      "eval_entropy": 0.6510261709873493,
+      "eval_num_tokens": 602930.0,
+      "eval_mean_token_accuracy": 0.8237179517745972,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.56,
+      "eval_accuracy": 0.45
+    },
+    {
+      "loss": 0.28018630981445314,
+      "grad_norm": 6.3215460777282715,
+      "learning_rate": 4.803679899192337e-07,
+      "entropy": 0.2751614101510495,
+      "num_tokens": 904395.0,
+      "mean_token_accuracy": 0.9216666701436043,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.27461162209510803,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 7.2206,
+      "eval_samples_per_second": 13.849,
+      "eval_steps_per_second": 1.8,
+      "eval_entropy": 0.21267771205076805,
+      "eval_num_tokens": 904395.0,
+      "eval_mean_token_accuracy": 0.9262820482254028,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.92,
+      "eval_accuracy": 0.77
+    },
+    {
+      "loss": 0.1371098518371582,
+      "grad_norm": 9.120945930480957,
+      "learning_rate": 4.867325323737765e-05,
+      "entropy": 0.12674232229823246,
+      "num_tokens": 1205860.0,
+      "mean_token_accuracy": 0.9620833370089531,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.25197863578796387,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 7.2173,
+      "eval_samples_per_second": 13.856,
+      "eval_steps_per_second": 1.801,
+      "eval_entropy": 0.1891580602297416,
+      "eval_num_tokens": 1205860.0,
+      "eval_mean_token_accuracy": 0.9358974374257601,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.94,
+      "eval_accuracy": 0.81
+    },
+    {
+      "train_runtime": 767.6735,
+      "train_samples_per_second": 4.168,
+      "train_steps_per_second": 0.261,
+      "total_flos": 3971013289334784.0,
+      "train_loss": 0.5070997858047486,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/qwen2.5-0.5b_D_1k_20260826_044059.json"
+}