Add DoRA+ training report: smollm2-360m_C_1k [20260826_144617]

reports/v2.0/doraplus/reports_training/smollm2-360m_C_1k_20260826_144617.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "smollm2-360m",
+  "model_id": "HuggingFaceTB/SmolLM2-360M-Instruct",
+  "lora_config": "C",
+  "lora_config_desc": "Wide — full attention + MLP, rank 16",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_144617",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 8960000,
+  "total_params": 370781120,
+  "trainable_pct": 2.4165,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 2344.3,
+  "total_training_time_s": 1170.7,
+  "final_train_loss": 0.341787,
+  "final_eval_loss": 0.226764,
+  "final_val_accuracy": 0.86,
+  "final_train_accuracy": 0.86,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/smollm2-360m_DoRA+_C_1k_20260826-144608",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.6746440887451172,
+      "grad_norm": 1.0517995357513428,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.6457112693786621,
+      "num_tokens": 346669.0,
+      "mean_token_accuracy": 0.7979166632890702,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.5325514674186707,
+      "eval_model_preparation_time": 0.0224,
+      "eval_runtime": 6.9393,
+      "eval_samples_per_second": 14.411,
+      "eval_steps_per_second": 1.873,
+      "eval_entropy": 0.5285803240079147,
+      "eval_num_tokens": 346669.0,
+      "eval_mean_token_accuracy": 0.8301282066565293,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.56,
+      "eval_accuracy": 0.51
+    },
+    {
+      "loss": 0.28994916915893554,
+      "grad_norm": 0.7273637056350708,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.3092053846269846,
+      "num_tokens": 693338.0,
+      "mean_token_accuracy": 0.8966666662693024,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.36715853214263916,
+      "eval_model_preparation_time": 0.0224,
+      "eval_runtime": 6.9296,
+      "eval_samples_per_second": 14.431,
+      "eval_steps_per_second": 1.876,
+      "eval_entropy": 0.35181020773374116,
+      "eval_num_tokens": 693338.0,
+      "eval_mean_token_accuracy": 0.8942307692307693,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.74,
+      "eval_accuracy": 0.67
+    },
+    {
+      "loss": 0.22105234146118163,
+      "grad_norm": 1.690340518951416,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.25370809972286223,
+      "num_tokens": 1040007.0,
+      "mean_token_accuracy": 0.9250000005960465,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.38307881355285645,
+      "eval_model_preparation_time": 0.0224,
+      "eval_runtime": 6.9281,
+      "eval_samples_per_second": 14.434,
+      "eval_steps_per_second": 1.876,
+      "eval_entropy": 0.28170177799004775,
+      "eval_num_tokens": 1040007.0,
+      "eval_mean_token_accuracy": 0.8814102594669049,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.74,
+      "eval_accuracy": 0.67
+    },
+    {
+      "loss": 0.18150375366210938,
+      "grad_norm": 0.3076472580432892,
+      "learning_rate": 9.956320346634876e-05,
+      "entropy": 0.1771650810074061,
+      "num_tokens": 1386676.0,
+      "mean_token_accuracy": 0.940416664481163,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.2267642766237259,
+      "eval_model_preparation_time": 0.0224,
+      "eval_runtime": 6.9307,
+      "eval_samples_per_second": 14.429,
+      "eval_steps_per_second": 1.876,
+      "eval_entropy": 0.25207306903142196,
+      "eval_num_tokens": 1386676.0,
+      "eval_mean_token_accuracy": 0.9583333272200364,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.86,
+      "eval_accuracy": 0.86
+    },
+    {
+      "train_runtime": 1170.2262,
+      "train_samples_per_second": 2.735,
+      "train_steps_per_second": 0.171,
+      "total_flos": 4450345244083200.0,
+      "train_loss": 0.34178733825683594,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/smollm2-360m_C_1k_20260826_144617.json"
+}