Add DoRA training report: smollm2-360m_C_1k [20260826_170911]

reports/v2.0/dora/reports_training/smollm2-360m_C_1k_20260826_170911.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_170911",
+  "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": 1165.7,
+  "final_train_loss": 0.564821,
+  "final_eval_loss": 0.469289,
+  "final_val_accuracy": 0.57,
+  "final_train_accuracy": 0.7,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/smollm2-360m_DoRA_C_1k_20260826-170907",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.9826673126220703,
+      "grad_norm": 0.730368971824646,
+      "learning_rate": 8.995939984474624e-05,
+      "entropy": 0.9718597215414048,
+      "num_tokens": 346669.0,
+      "mean_token_accuracy": 0.7254166656732559,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.7610267400741577,
+      "eval_model_preparation_time": 0.0225,
+      "eval_runtime": 6.8831,
+      "eval_samples_per_second": 14.528,
+      "eval_steps_per_second": 1.889,
+      "eval_entropy": 0.808483939904433,
+      "eval_num_tokens": 346669.0,
+      "eval_mean_token_accuracy": 0.7532051297334524,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.32,
+      "eval_accuracy": 0.28
+    },
+    {
+      "loss": 0.5506010437011719,
+      "grad_norm": 1.1063673496246338,
+      "learning_rate": 5.495227651252315e-05,
+      "entropy": 0.6121571019291878,
+      "num_tokens": 693338.0,
+      "mean_token_accuracy": 0.8195833373069763,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.5113226771354675,
+      "eval_model_preparation_time": 0.0225,
+      "eval_runtime": 6.8906,
+      "eval_samples_per_second": 14.513,
+      "eval_steps_per_second": 1.887,
+      "eval_entropy": 0.4823713187987988,
+      "eval_num_tokens": 693338.0,
+      "eval_mean_token_accuracy": 0.852564105620751,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.66,
+      "eval_accuracy": 0.57
+    },
+    {
+      "loss": 0.3879635238647461,
+      "grad_norm": 1.1636685132980347,
+      "learning_rate": 1.6748771394307585e-05,
+      "entropy": 0.3966414794325829,
+      "num_tokens": 1040007.0,
+      "mean_token_accuracy": 0.8599999988079071,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.46326887607574463,
+      "eval_model_preparation_time": 0.0225,
+      "eval_runtime": 6.8849,
+      "eval_samples_per_second": 14.525,
+      "eval_steps_per_second": 1.888,
+      "eval_entropy": 0.41631388434996974,
+      "eval_num_tokens": 1040007.0,
+      "eval_mean_token_accuracy": 0.8589743650876559,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.7,
+      "eval_accuracy": 0.58
+    },
+    {
+      "loss": 0.3380523681640625,
+      "grad_norm": 1.2447329759597778,
+      "learning_rate": 6.834750376549792e-09,
+      "entropy": 0.3553758755326271,
+      "num_tokens": 1386676.0,
+      "mean_token_accuracy": 0.8762499976158142,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.46928858757019043,
+      "eval_model_preparation_time": 0.0225,
+      "eval_runtime": 6.8917,
+      "eval_samples_per_second": 14.51,
+      "eval_steps_per_second": 1.886,
+      "eval_entropy": 0.3996601872719251,
+      "eval_num_tokens": 1386676.0,
+      "eval_mean_token_accuracy": 0.8525641102057236,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.7,
+      "eval_accuracy": 0.57
+    },
+    {
+      "train_runtime": 1165.2423,
+      "train_samples_per_second": 2.746,
+      "train_steps_per_second": 0.172,
+      "total_flos": 4450345244083200.0,
+      "train_loss": 0.5648210620880127,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/dora/reports_training/smollm2-360m_C_1k_20260826_170911.json"
+}