Add LoRA+ training report: llama3.2-1b_A_1k [20260826_183836]

reports/v2.0/loraplus/reports_training/llama3.2-1b_A_1k_20260826_183836.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": "LoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_183836",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 851968,
+  "total_params": 1236666368,
+  "trainable_pct": 0.0689,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 4486.4,
+  "total_training_time_s": 493.5,
+  "final_train_loss": 0.114604,
+  "final_eval_loss": 0.141388,
+  "final_val_accuracy": 0.95,
+  "final_train_accuracy": 0.94,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/llama3.2-1b_LoRA+_A_1k_20260826-183834",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.36762428283691406,
+      "grad_norm": 21.581533432006836,
+      "learning_rate": 0.00011154846369695863,
+      "entropy": 0.29958768947049974,
+      "num_tokens": 318964.0,
+      "mean_token_accuracy": 0.906875,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.2956034541130066,
+      "eval_model_preparation_time": 0.0048,
+      "eval_runtime": 4.3335,
+      "eval_samples_per_second": 23.076,
+      "eval_steps_per_second": 3.0,
+      "eval_entropy": 0.13585213772379434,
+      "eval_num_tokens": 318964.0,
+      "eval_mean_token_accuracy": 0.9278846153846154,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.92,
+      "eval_accuracy": 0.87
+    },
+    {
+      "loss": 0.049195146560668944,
+      "grad_norm": 8.14330005645752,
+      "learning_rate": 5.467426590739511e-08,
+      "entropy": 0.02369907493935898,
+      "num_tokens": 637928.0,
+      "mean_token_accuracy": 0.988125,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.14864620566368103,
+      "eval_model_preparation_time": 0.0048,
+      "eval_runtime": 4.3042,
+      "eval_samples_per_second": 23.233,
+      "eval_steps_per_second": 3.02,
+      "eval_entropy": 0.027470061304764107,
+      "eval_num_tokens": 637928.0,
+      "eval_mean_token_accuracy": 0.9807692307692307,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.96
+    },
+    {
+      "loss": 0.015695568323135376,
+      "grad_norm": 0.008895198814570904,
+      "learning_rate": 0.00010495837546732224,
+      "entropy": 0.01068563232489396,
+      "num_tokens": 956892.0,
+      "mean_token_accuracy": 0.9975,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.0870068296790123,
+      "eval_model_preparation_time": 0.0048,
+      "eval_runtime": 4.3165,
+      "eval_samples_per_second": 23.167,
+      "eval_steps_per_second": 3.012,
+      "eval_entropy": 0.025836414003816362,
+      "eval_num_tokens": 956892.0,
+      "eval_mean_token_accuracy": 0.9807692307692307,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.02589911699295044,
+      "grad_norm": 23.856914520263672,
+      "learning_rate": 0.00019912640693269752,
+      "entropy": 0.013229600353806745,
+      "num_tokens": 1275856.0,
+      "mean_token_accuracy": 0.99375,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.14138758182525635,
+      "eval_model_preparation_time": 0.0048,
+      "eval_runtime": 4.3161,
+      "eval_samples_per_second": 23.169,
+      "eval_steps_per_second": 3.012,
+      "eval_entropy": 0.02055562370403025,
+      "eval_num_tokens": 1275856.0,
+      "eval_mean_token_accuracy": 0.9807692307692307,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.94,
+      "eval_accuracy": 0.95
+    },
+    {
+      "train_runtime": 493.1206,
+      "train_samples_per_second": 6.489,
+      "train_steps_per_second": 0.406,
+      "total_flos": 1.207073482948608e+16,
+      "train_loss": 0.11460352867841721,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/loraplus/reports_training/llama3.2-1b_A_1k_20260826_183836.json"
+}