Add DoRA+ training report: llama3.2-1b_B_1k [20260826_055035]

reports/v2.0/doraplus/reports_training/llama3.2-1b_B_1k_20260826_055035.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "llama3.2-1b",
+  "model_id": "meta-llama/Llama-3.2-1B-Instruct",
+  "lora_config": "B",
+  "lora_config_desc": "Standard — full attention, rank 16",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_055035",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 3489792,
+  "total_params": 1239304192,
+  "trainable_pct": 0.2816,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 4531.5,
+  "total_training_time_s": 751.2,
+  "final_train_loss": 0.105166,
+  "final_eval_loss": 0.331415,
+  "final_val_accuracy": 0.92,
+  "final_train_accuracy": 0.96,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/llama3.2-1b_DoRA+_B_1k_20260826-055029",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.3394538116455078,
+      "grad_norm": 2.833601951599121,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.26548744392581286,
+      "num_tokens": 318964.0,
+      "mean_token_accuracy": 0.91625,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.24490556120872498,
+      "eval_model_preparation_time": 0.0078,
+      "eval_runtime": 5.58,
+      "eval_samples_per_second": 17.921,
+      "eval_steps_per_second": 2.33,
+      "eval_entropy": 0.05666122456582693,
+      "eval_num_tokens": 318964.0,
+      "eval_mean_token_accuracy": 0.9567307692307693,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.92
+    },
+    {
+      "loss": 0.03891366243362427,
+      "grad_norm": 5.120852947235107,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.02092550735687837,
+      "num_tokens": 637928.0,
+      "mean_token_accuracy": 0.99125,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.13023629784584045,
+      "eval_model_preparation_time": 0.0078,
+      "eval_runtime": 5.5708,
+      "eval_samples_per_second": 17.951,
+      "eval_steps_per_second": 2.334,
+      "eval_entropy": 0.016571327255895503,
+      "eval_num_tokens": 637928.0,
+      "eval_mean_token_accuracy": 0.9807692307692307,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.96
+    },
+    {
+      "loss": 0.00934406280517578,
+      "grad_norm": 2.2951245307922363,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.00907736896071583,
+      "num_tokens": 956892.0,
+      "mean_token_accuracy": 0.9975,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.10635975003242493,
+      "eval_model_preparation_time": 0.0078,
+      "eval_runtime": 5.5915,
+      "eval_samples_per_second": 17.884,
+      "eval_steps_per_second": 2.325,
+      "eval_entropy": 0.0053405584297322025,
+      "eval_num_tokens": 956892.0,
+      "eval_mean_token_accuracy": 0.9855769230769231,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.032953145503997805,
+      "grad_norm": 0.017506301403045654,
+      "learning_rate": 9.956320346634876e-05,
+      "entropy": 0.006816715911409119,
+      "num_tokens": 1275856.0,
+      "mean_token_accuracy": 0.99375,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.33141541481018066,
+      "eval_model_preparation_time": 0.0078,
+      "eval_runtime": 5.5792,
+      "eval_samples_per_second": 17.924,
+      "eval_steps_per_second": 2.33,
+      "eval_entropy": 0.026854831904459458,
+      "eval_num_tokens": 1275856.0,
+      "eval_mean_token_accuracy": 0.9615384615384616,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.92
+    },
+    {
+      "train_runtime": 750.8132,
+      "train_samples_per_second": 4.262,
+      "train_steps_per_second": 0.266,
+      "total_flos": 1.2103425319010304e+16,
+      "train_loss": 0.10516617059707642,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/llama3.2-1b_B_1k_20260826_055035.json"
+}