Add LoRA training report: smollm2-1.7b_D_1k [20260825_222419]

reports/v2.0/lora/reports_training/smollm2-1.7b_D_1k_20260825_222419.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "smollm2-1.7b",
+  "model_id": "HuggingFaceTB/SmolLM2-1.7B-Instruct",
+  "lora_config": "D",
+  "lora_config_desc": "Heavy — full attention + MLP, rank 32",
+  "dataset_size": "1k",
+  "technique": "LoRA",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260825_222419",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 36175872,
+  "total_params": 1747552256,
+  "trainable_pct": 2.0701,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 5076.4,
+  "total_training_time_s": 1082.8,
+  "final_train_loss": 0.151113,
+  "final_eval_loss": 0.117639,
+  "final_val_accuracy": 0.94,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/smollm2-1.7b_LoRA_D_1k_20260825-222405",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.5047369384765625,
+      "grad_norm": 0.6044948697090149,
+      "learning_rate": 4.497969992237312e-05,
+      "entropy": 0.5048886869382113,
+      "num_tokens": 346669.0,
+      "mean_token_accuracy": 0.8637500032782555,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.19356133043766022,
+      "eval_model_preparation_time": 0.0153,
+      "eval_runtime": 10.7927,
+      "eval_samples_per_second": 9.265,
+      "eval_steps_per_second": 1.205,
+      "eval_entropy": 0.10314230592204975,
+      "eval_num_tokens": 346669.0,
+      "eval_mean_token_accuracy": 0.9583333272200364,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.94,
+      "eval_accuracy": 0.87
+    },
+    {
+      "loss": 0.07209014892578125,
+      "grad_norm": 0.21705107390880585,
+      "learning_rate": 2.7476138256261575e-05,
+      "entropy": 0.05274912181717809,
+      "num_tokens": 693338.0,
+      "mean_token_accuracy": 0.9841666692495346,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.15963640809059143,
+      "eval_model_preparation_time": 0.0153,
+      "eval_runtime": 10.79,
+      "eval_samples_per_second": 9.268,
+      "eval_steps_per_second": 1.205,
+      "eval_entropy": 0.048771885295326896,
+      "eval_num_tokens": 693338.0,
+      "eval_mean_token_accuracy": 0.9647435821019686,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.89
+    },
+    {
+      "loss": 0.02051819324493408,
+      "grad_norm": 0.30775290727615356,
+      "learning_rate": 8.374385697153792e-06,
+      "entropy": 0.023366771423025057,
+      "num_tokens": 1040007.0,
+      "mean_token_accuracy": 0.9954166677594185,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.11255333572626114,
+      "eval_model_preparation_time": 0.0153,
+      "eval_runtime": 10.7818,
+      "eval_samples_per_second": 9.275,
+      "eval_steps_per_second": 1.206,
+      "eval_entropy": 0.03178400227513451,
+      "eval_num_tokens": 1040007.0,
+      "eval_mean_token_accuracy": 0.9807692261842581,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.007108409404754639,
+      "grad_norm": 0.030860496684908867,
+      "learning_rate": 3.417375188274896e-09,
+      "entropy": 0.013284350297908532,
+      "num_tokens": 1386676.0,
+      "mean_token_accuracy": 0.9987500002980232,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.1176392063498497,
+      "eval_model_preparation_time": 0.0153,
+      "eval_runtime": 10.7672,
+      "eval_samples_per_second": 9.287,
+      "eval_steps_per_second": 1.207,
+      "eval_entropy": 0.028779175580264285,
+      "eval_num_tokens": 1386676.0,
+      "eval_mean_token_accuracy": 0.9807692261842581,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.94
+    },
+    {
+      "train_runtime": 1082.3454,
+      "train_samples_per_second": 2.957,
+      "train_steps_per_second": 0.185,
+      "total_flos": 1.99936931033088e+16,
+      "train_loss": 0.1511134225130081,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/lora/reports_training/smollm2-1.7b_D_1k_20260825_222419.json"
+}