Add DoRA+ training report: llama3.2-1b_A_1k [20260826_053613]

reports/v2.0/doraplus/reports_training/llama3.2-1b_A_1k_20260826_053613.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "llama3.2-1b",
+  "model_id": "meta-llama/Llama-3.2-1B-Instruct",
+  "lora_config": "A",
+  "lora_config_desc": "Light — attention Q/V only, rank 8",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_053613",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 892928,
+  "total_params": 1236707328,
+  "trainable_pct": 0.0722,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 4487.0,
+  "total_training_time_s": 594.3,
+  "final_train_loss": 0.109842,
+  "final_eval_loss": 0.293731,
+  "final_val_accuracy": 0.95,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/llama3.2-1b_DoRA+_A_1k_20260826-053607",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.3441616439819336,
+      "grad_norm": 7.411519527435303,
+      "learning_rate": 0.00011154846369695863,
+      "entropy": 0.2928112076129764,
+      "num_tokens": 318964.0,
+      "mean_token_accuracy": 0.9125,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.2613638937473297,
+      "eval_model_preparation_time": 0.0054,
+      "eval_runtime": 4.7843,
+      "eval_samples_per_second": 20.902,
+      "eval_steps_per_second": 2.717,
+      "eval_entropy": 0.08674019001997434,
+      "eval_num_tokens": 318964.0,
+      "eval_mean_token_accuracy": 0.9567307692307693,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.91
+    },
+    {
+      "loss": 0.05822742462158203,
+      "grad_norm": 0.025751441717147827,
+      "learning_rate": 5.467426590739511e-08,
+      "entropy": 0.019208475212799385,
+      "num_tokens": 637928.0,
+      "mean_token_accuracy": 0.99,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.20248305797576904,
+      "eval_model_preparation_time": 0.0054,
+      "eval_runtime": 4.7935,
+      "eval_samples_per_second": 20.862,
+      "eval_steps_per_second": 2.712,
+      "eval_entropy": 0.02411385614854785,
+      "eval_num_tokens": 637928.0,
+      "eval_mean_token_accuracy": 0.9759615384615384,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.019497088193893432,
+      "grad_norm": 3.946964979171753,
+      "learning_rate": 0.00010495837546732224,
+      "entropy": 0.013751769539667294,
+      "num_tokens": 956892.0,
+      "mean_token_accuracy": 0.995625,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.14664743840694427,
+      "eval_model_preparation_time": 0.0054,
+      "eval_runtime": 4.7874,
+      "eval_samples_per_second": 20.888,
+      "eval_steps_per_second": 2.715,
+      "eval_entropy": 0.010201264235477608,
+      "eval_num_tokens": 956892.0,
+      "eval_mean_token_accuracy": 0.9855769230769231,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.96
+    },
+    {
+      "loss": 0.01748358368873596,
+      "grad_norm": 8.153016090393066,
+      "learning_rate": 0.00019912640693269752,
+      "entropy": 0.006480707966547925,
+      "num_tokens": 1275856.0,
+      "mean_token_accuracy": 0.99625,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.2937309741973877,
+      "eval_model_preparation_time": 0.0054,
+      "eval_runtime": 4.7824,
+      "eval_samples_per_second": 20.91,
+      "eval_steps_per_second": 2.718,
+      "eval_entropy": 0.0037075236121801515,
+      "eval_num_tokens": 1275856.0,
+      "eval_mean_token_accuracy": 0.9759615384615384,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "train_runtime": 593.8936,
+      "train_samples_per_second": 5.388,
+      "train_steps_per_second": 0.337,
+      "total_flos": 1.207124244578304e+16,
+      "train_loss": 0.10984243512153626,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/llama3.2-1b_A_1k_20260826_053613.json"
+}