Add LoRA+ training report: smollm2-1.7b_B_1k [20260826_194429]

reports/v2.0/loraplus/reports_training/smollm2-1.7b_B_1k_20260826_194429.json

@@ -0,0 +1,116 @@
+{
+  "model_key": "smollm2-1.7b",
+  "model_id": "HuggingFaceTB/SmolLM2-1.7B-Instruct",
+  "lora_config": "B",
+  "lora_config_desc": "Standard — full attention, rank 16",
+  "dataset_size": "1k",
+  "technique": "LoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_194429",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 6291456,
+  "total_params": 1717667840,
+  "trainable_pct": 0.3663,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 150,
+  "early_stopped": true,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 5236.5,
+  "total_training_time_s": 717.5,
+  "final_train_loss": 0.088442,
+  "final_eval_loss": 0.088199,
+  "final_val_accuracy": 0.95,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/smollm2-1.7b_LoRA+_B_1k_20260826-194425",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.2426377296447754,
+      "grad_norm": 0.5156200528144836,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.23668485255911947,
+      "num_tokens": 346669.0,
+      "mean_token_accuracy": 0.9383333307504654,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.08092240989208221,
+      "eval_model_preparation_time": 0.0102,
+      "eval_runtime": 8.127,
+      "eval_samples_per_second": 12.305,
+      "eval_steps_per_second": 1.6,
+      "eval_entropy": 0.05753099381063993,
+      "eval_num_tokens": 346669.0,
+      "eval_mean_token_accuracy": 0.9807692261842581,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.94
+    },
+    {
+      "loss": 0.01653146505355835,
+      "grad_norm": 0.6451789140701294,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.01764775813790038,
+      "num_tokens": 693338.0,
+      "mean_token_accuracy": 0.9966666650772095,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.09615980833768845,
+      "eval_model_preparation_time": 0.0102,
+      "eval_runtime": 8.1282,
+      "eval_samples_per_second": 12.303,
+      "eval_steps_per_second": 1.599,
+      "eval_entropy": 0.027609849635225076,
+      "eval_num_tokens": 693338.0,
+      "eval_mean_token_accuracy": 0.9807692261842581,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.94
+    },
+    {
+      "loss": 0.006157773733139038,
+      "grad_norm": 1.5418363809585571,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.009527205278282054,
+      "num_tokens": 1040007.0,
+      "mean_token_accuracy": 0.9987499994039536,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.08819916099309921,
+      "eval_model_preparation_time": 0.0102,
+      "eval_runtime": 8.1268,
+      "eval_samples_per_second": 12.305,
+      "eval_steps_per_second": 1.6,
+      "eval_entropy": 0.016946650450476088,
+      "eval_num_tokens": 1040007.0,
+      "eval_mean_token_accuracy": 0.9839743513327378,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "train_runtime": 717.0733,
+      "train_samples_per_second": 4.463,
+      "train_steps_per_second": 0.279,
+      "total_flos": 1.6666375282753536e+16,
+      "train_loss": 0.08844232281049093,
+      "epoch": 3.0,
+      "step": 150
+    }
+  ],
+  "hf_report_path": "reports/v2.0/loraplus/reports_training/smollm2-1.7b_B_1k_20260826_194429.json"
+}