Add DoRA training report: smollm2-360m_A_1k [20260826_163548]

reports/v2.0/dora/reports_training/smollm2-360m_A_1k_20260826_163548.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "smollm2-360m",
+  "model_id": "HuggingFaceTB/SmolLM2-360M-Instruct",
+  "lora_config": "A",
+  "lora_config_desc": "Light — attention Q/V only, rank 8",
+  "dataset_size": "1k",
+  "technique": "DoRA",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_163548",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 860160,
+  "total_params": 362681280,
+  "trainable_pct": 0.2372,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 1611.8,
+  "total_training_time_s": 437.5,
+  "final_train_loss": 0.691192,
+  "final_eval_loss": 0.554405,
+  "final_val_accuracy": 0.49,
+  "final_train_accuracy": 0.62,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/smollm2-360m_DoRA_A_1k_20260826-163545",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 1.1354335021972657,
+      "grad_norm": 0.671927273273468,
+      "learning_rate": 0.0001799187996894925,
+      "entropy": 1.1237754690647126,
+      "num_tokens": 346669.0,
+      "mean_token_accuracy": 0.7004166656732559,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.8116267323493958,
+      "eval_model_preparation_time": 0.0098,
+      "eval_runtime": 2.7523,
+      "eval_samples_per_second": 36.333,
+      "eval_steps_per_second": 4.723,
+      "eval_entropy": 0.8381677040686975,
+      "eval_num_tokens": 346669.0,
+      "eval_mean_token_accuracy": 0.7371794810661902,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.24,
+      "eval_accuracy": 0.2
+    },
+    {
+      "loss": 0.6749427032470703,
+      "grad_norm": 1.1144015789031982,
+      "learning_rate": 0.0001099045530250463,
+      "entropy": 0.7632645869255066,
+      "num_tokens": 693338.0,
+      "mean_token_accuracy": 0.7795833325386048,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.6584625840187073,
+      "eval_model_preparation_time": 0.0098,
+      "eval_runtime": 2.7414,
+      "eval_samples_per_second": 36.478,
+      "eval_steps_per_second": 4.742,
+      "eval_entropy": 0.6928549225513752,
+      "eval_num_tokens": 693338.0,
+      "eval_mean_token_accuracy": 0.7852564041431134,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.5,
+      "eval_accuracy": 0.36
+    },
+    {
+      "loss": 0.5062131881713867,
+      "grad_norm": 1.3099958896636963,
+      "learning_rate": 3.349754278861517e-05,
+      "entropy": 0.5743581032752991,
+      "num_tokens": 1040007.0,
+      "mean_token_accuracy": 0.8245833313465118,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.5587055683135986,
+      "eval_model_preparation_time": 0.0098,
+      "eval_runtime": 2.7339,
+      "eval_samples_per_second": 36.577,
+      "eval_steps_per_second": 4.755,
+      "eval_entropy": 0.5502871779295114,
+      "eval_num_tokens": 1040007.0,
+      "eval_mean_token_accuracy": 0.8237179563595698,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.62,
+      "eval_accuracy": 0.5
+    },
+    {
+      "loss": 0.44817920684814455,
+      "grad_norm": 1.6124805212020874,
+      "learning_rate": 1.3669500753099585e-08,
+      "entropy": 0.49049379721283914,
+      "num_tokens": 1386676.0,
+      "mean_token_accuracy": 0.8433333343267441,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.5544049739837646,
+      "eval_model_preparation_time": 0.0098,
+      "eval_runtime": 2.7353,
+      "eval_samples_per_second": 36.558,
+      "eval_steps_per_second": 4.753,
+      "eval_entropy": 0.533219766158324,
+      "eval_num_tokens": 1386676.0,
+      "eval_mean_token_accuracy": 0.8269230815080496,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.62,
+      "eval_accuracy": 0.49
+    },
+    {
+      "train_runtime": 437.1436,
+      "train_samples_per_second": 7.32,
+      "train_steps_per_second": 0.458,
+      "total_flos": 4338949634933760.0,
+      "train_loss": 0.6911921501159668,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/dora/reports_training/smollm2-360m_A_1k_20260826_163548.json"
+}