Add LoRA+ training report: smollm2-1.7b_A_1k [20260826_193123]

reports/v2.0/loraplus/reports_training/smollm2-1.7b_A_1k_20260826_193123.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "smollm2-1.7b",
+  "model_id": "HuggingFaceTB/SmolLM2-1.7B-Instruct",
+  "lora_config": "A",
+  "lora_config_desc": "Light — attention Q/V only, rank 8",
+  "dataset_size": "1k",
+  "technique": "LoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_193123",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 1572864,
+  "total_params": 1712949248,
+  "trainable_pct": 0.0918,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 5075.2,
+  "total_training_time_s": 867.1,
+  "final_train_loss": 0.07806,
+  "final_eval_loss": 0.148283,
+  "final_val_accuracy": 0.95,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/smollm2-1.7b_LoRA+_A_1k_20260826-193120",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.2709200477600098,
+      "grad_norm": 3.384413480758667,
+      "learning_rate": 0.00011154846369695863,
+      "entropy": 0.26733902214094996,
+      "num_tokens": 346669.0,
+      "mean_token_accuracy": 0.9287499976158142,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.08273226022720337,
+      "eval_model_preparation_time": 0.0071,
+      "eval_runtime": 7.6588,
+      "eval_samples_per_second": 13.057,
+      "eval_steps_per_second": 1.697,
+      "eval_entropy": 0.07552582655961697,
+      "eval_num_tokens": 346669.0,
+      "eval_mean_token_accuracy": 0.9775641010357783,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.92,
+      "eval_accuracy": 0.93
+    },
+    {
+      "loss": 0.02210291862487793,
+      "grad_norm": 0.6268931031227112,
+      "learning_rate": 5.467426590739511e-08,
+      "entropy": 0.02679341256734915,
+      "num_tokens": 693338.0,
+      "mean_token_accuracy": 0.9949999976158143,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.09820631146430969,
+      "eval_model_preparation_time": 0.0071,
+      "eval_runtime": 7.6513,
+      "eval_samples_per_second": 13.07,
+      "eval_steps_per_second": 1.699,
+      "eval_entropy": 0.0260245819361164,
+      "eval_num_tokens": 693338.0,
+      "eval_mean_token_accuracy": 0.9839743559177105,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.011228382587432861,
+      "grad_norm": 0.04244871065020561,
+      "learning_rate": 0.00010495837546732224,
+      "entropy": 0.015659121192293242,
+      "num_tokens": 1040007.0,
+      "mean_token_accuracy": 0.9970833325386047,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.07785949856042862,
+      "eval_model_preparation_time": 0.0071,
+      "eval_runtime": 7.6596,
+      "eval_samples_per_second": 13.055,
+      "eval_steps_per_second": 1.697,
+      "eval_entropy": 0.01488687565478568,
+      "eval_num_tokens": 1040007.0,
+      "eval_mean_token_accuracy": 0.987179485651163,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.96
+    },
+    {
+      "loss": 0.0079898202419281,
+      "grad_norm": 0.08075383305549622,
+      "learning_rate": 0.00019912640693269752,
+      "entropy": 0.006461533498950302,
+      "num_tokens": 1386676.0,
+      "mean_token_accuracy": 0.9983333325386048,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.14828304946422577,
+      "eval_model_preparation_time": 0.0071,
+      "eval_runtime": 7.6498,
+      "eval_samples_per_second": 13.072,
+      "eval_steps_per_second": 1.699,
+      "eval_entropy": 0.012998863947219573,
+      "eval_num_tokens": 1386676.0,
+      "eval_mean_token_accuracy": 0.9839743559177105,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "train_runtime": 866.7291,
+      "train_samples_per_second": 3.692,
+      "train_steps_per_second": 0.231,
+      "total_flos": 2.217347203724083e+16,
+      "train_loss": 0.07806029230356216,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/loraplus/reports_training/smollm2-1.7b_A_1k_20260826_193123.json"
+}