Add DoRA training report: llama3.2-1b_D_1k [20260826_205916]

reports/v2.0/dora/reports_training/llama3.2-1b_D_1k_20260826_205916.json

@@ -0,0 +1,116 @@
+{
+  "model_key": "llama3.2-1b",
+  "model_id": "meta-llama/Llama-3.2-1B-Instruct",
+  "lora_config": "D",
+  "lora_config_desc": "Heavy — full attention + MLP, rank 32",
+  "dataset_size": "1k",
+  "technique": "DoRA",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_205916",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 22921216,
+  "total_params": 1258735616,
+  "trainable_pct": 1.821,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 150,
+  "early_stopped": true,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 4603.5,
+  "total_training_time_s": 1402.2,
+  "final_train_loss": 0.15843,
+  "final_eval_loss": 0.198895,
+  "final_val_accuracy": 0.97,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/llama3.2-1b_DoRA_D_1k_20260826-205911",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.45010025024414063,
+      "grad_norm": 0.03857315704226494,
+      "learning_rate": 4.497969992237312e-05,
+      "entropy": 0.39982086431002245,
+      "num_tokens": 318964.0,
+      "mean_token_accuracy": 0.861875,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.15361879765987396,
+      "eval_model_preparation_time": 0.0116,
+      "eval_runtime": 12.8099,
+      "eval_samples_per_second": 7.806,
+      "eval_steps_per_second": 1.015,
+      "eval_entropy": 0.02806645130308775,
+      "eval_num_tokens": 318964.0,
+      "eval_mean_token_accuracy": 0.9759615384615384,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.96
+    },
+    {
+      "loss": 0.024006786346435546,
+      "grad_norm": 0.12987594306468964,
+      "learning_rate": 2.7476138256261575e-05,
+      "entropy": 0.005113552697876003,
+      "num_tokens": 637928.0,
+      "mean_token_accuracy": 0.99625,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.24818845093250275,
+      "eval_model_preparation_time": 0.0116,
+      "eval_runtime": 12.7968,
+      "eval_samples_per_second": 7.814,
+      "eval_steps_per_second": 1.016,
+      "eval_entropy": 0.006707652906278292,
+      "eval_num_tokens": 637928.0,
+      "eval_mean_token_accuracy": 0.9759615384615384,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.0011827129870653152,
+      "grad_norm": 0.001537011587060988,
+      "learning_rate": 8.374385697153792e-06,
+      "entropy": 0.0018033458654826973,
+      "num_tokens": 956892.0,
+      "mean_token_accuracy": 0.999375,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.1988954395055771,
+      "eval_model_preparation_time": 0.0116,
+      "eval_runtime": 12.8087,
+      "eval_samples_per_second": 7.807,
+      "eval_steps_per_second": 1.015,
+      "eval_entropy": 0.0012766208025053716,
+      "eval_num_tokens": 956892.0,
+      "eval_mean_token_accuracy": 0.9855769230769231,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "train_runtime": 1401.7748,
+      "train_samples_per_second": 2.283,
+      "train_steps_per_second": 0.143,
+      "total_flos": 8214280386625536.0,
+      "train_loss": 0.15842991652588048,
+      "epoch": 3.0,
+      "step": 150
+    }
+  ],
+  "hf_report_path": "reports/v2.0/dora/reports_training/llama3.2-1b_D_1k_20260826_205916.json"
+}