Add DoRA training report: llama3.2-1b_C_1k [20260826_203104]

reports/v2.0/dora/reports_training/llama3.2-1b_C_1k_20260826_203104.json

@@ -0,0 +1,116 @@
+{
+  "model_key": "llama3.2-1b",
+  "model_id": "meta-llama/Llama-3.2-1B-Instruct",
+  "lora_config": "C",
+  "lora_config_desc": "Wide — full attention + MLP, rank 16",
+  "dataset_size": "1k",
+  "technique": "DoRA",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_203104",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 11649024,
+  "total_params": 1247463424,
+  "trainable_pct": 0.9338,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 150,
+  "early_stopped": true,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 5737.7,
+  "total_training_time_s": 1434.4,
+  "final_train_loss": 0.160726,
+  "final_eval_loss": 0.196029,
+  "final_val_accuracy": 0.96,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/llama3.2-1b_DoRA_C_1k_20260826-203101",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.43145950317382814,
+      "grad_norm": 0.07242455333471298,
+      "learning_rate": 8.995939984474624e-05,
+      "entropy": 0.38500592762138697,
+      "num_tokens": 318964.0,
+      "mean_token_accuracy": 0.865,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.12781928479671478,
+      "eval_model_preparation_time": 0.0116,
+      "eval_runtime": 12.7802,
+      "eval_samples_per_second": 7.825,
+      "eval_steps_per_second": 1.017,
+      "eval_entropy": 0.04244433325500442,
+      "eval_num_tokens": 318964.0,
+      "eval_mean_token_accuracy": 0.9807692307692307,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.04504281997680664,
+      "grad_norm": 0.051985107362270355,
+      "learning_rate": 5.495227651252315e-05,
+      "entropy": 0.0074617948161903765,
+      "num_tokens": 637928.0,
+      "mean_token_accuracy": 0.994375,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.2221250683069229,
+      "eval_model_preparation_time": 0.0116,
+      "eval_runtime": 12.7891,
+      "eval_samples_per_second": 7.819,
+      "eval_steps_per_second": 1.016,
+      "eval_entropy": 0.01151017749753709,
+      "eval_num_tokens": 637928.0,
+      "eval_mean_token_accuracy": 0.9663461538461539,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.93
+    },
+    {
+      "loss": 0.005674490332603454,
+      "grad_norm": 0.005814565811306238,
+      "learning_rate": 1.6748771394307585e-05,
+      "entropy": 0.004261631539557129,
+      "num_tokens": 956892.0,
+      "mean_token_accuracy": 0.99875,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.19602856040000916,
+      "eval_model_preparation_time": 0.0116,
+      "eval_runtime": 12.7892,
+      "eval_samples_per_second": 7.819,
+      "eval_steps_per_second": 1.016,
+      "eval_entropy": 0.0028350672577149593,
+      "eval_num_tokens": 956892.0,
+      "eval_mean_token_accuracy": 0.9807692307692307,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.96
+    },
+    {
+      "train_runtime": 1433.963,
+      "train_samples_per_second": 2.232,
+      "train_steps_per_second": 0.139,
+      "total_flos": 9146493602168832.0,
+      "train_loss": 0.16072560449441273,
+      "epoch": 3.0,
+      "step": 150
+    }
+  ],
+  "hf_report_path": "reports/v2.0/dora/reports_training/llama3.2-1b_C_1k_20260826_203104.json"
+}