Add DoRA+ training report: qwen3-0.6b_D_10k [20260827_213058]

reports/v2.0/doraplus/reports_training/qwen3-0.6b_D_10k_20260827_213058.json

@@ -0,0 +1,164 @@
+{
+  "model_key": "qwen3-0.6b",
+  "model_id": "Qwen/Qwen3-0.6B",
+  "lora_config": "D",
+  "lora_config_desc": "Heavy — full attention + MLP, rank 32",
+  "dataset_size": "10k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260827_213058",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 20529152,
+  "total_params": 616579072,
+  "trainable_pct": 3.3295,
+  "steps_per_epoch": 500,
+  "total_steps_trained": 1250,
+  "early_stopped": true,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 4593.3,
+  "total_training_time_s": 9873.2,
+  "final_train_loss": 0.03305,
+  "final_eval_loss": 0.002246,
+  "final_val_accuracy": 1.0,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen3-0.6b_DoRA+_D_10k_20260827-213047",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.12966786193847657,
+      "grad_norm": 2.364213466644287,
+      "learning_rate": 4.9245121086159674e-05,
+      "entropy": 0.05329545314458665,
+      "num_tokens": 1516296.0,
+      "mean_token_accuracy": 0.9799166703224182,
+      "epoch": 0.5,
+      "step": 250
+    },
+    {
+      "eval_loss": 0.04242685064673424,
+      "eval_model_preparation_time": 0.0208,
+      "eval_runtime": 83.6812,
+      "eval_samples_per_second": 11.95,
+      "eval_steps_per_second": 1.494,
+      "eval_entropy": 0.02787096245959401,
+      "eval_num_tokens": 1516296.0,
+      "eval_mean_token_accuracy": 0.9943333320617675,
+      "epoch": 0.5,
+      "step": 250,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.014683025360107422,
+      "grad_norm": 3.3569235801696777,
+      "learning_rate": 4.4753875309392266e-05,
+      "entropy": 0.007536330722155981,
+      "num_tokens": 3035617.0,
+      "mean_token_accuracy": 0.9975833353996277,
+      "epoch": 1.0,
+      "step": 500
+    },
+    {
+      "eval_loss": 0.012298869900405407,
+      "eval_model_preparation_time": 0.0208,
+      "eval_runtime": 83.5578,
+      "eval_samples_per_second": 11.968,
+      "eval_steps_per_second": 1.496,
+      "eval_entropy": 0.011986315403133632,
+      "eval_num_tokens": 3035617.0,
+      "eval_mean_token_accuracy": 0.9966666650772095,
+      "epoch": 1.0,
+      "step": 500,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.99
+    },
+    {
+      "loss": 0.016262353897094725,
+      "grad_norm": 0.004455740563571453,
+      "learning_rate": 3.693502312666304e-05,
+      "entropy": 0.007805552490521222,
+      "num_tokens": 4550488.0,
+      "mean_token_accuracy": 0.99775000166893,
+      "epoch": 1.5,
+      "step": 750
+    },
+    {
+      "eval_loss": 0.0012079532025381923,
+      "eval_model_preparation_time": 0.0208,
+      "eval_runtime": 83.6801,
+      "eval_samples_per_second": 11.95,
+      "eval_steps_per_second": 1.494,
+      "eval_entropy": 0.0037327693440020082,
+      "eval_num_tokens": 4550488.0,
+      "eval_mean_token_accuracy": 0.999666666507721,
+      "epoch": 1.5,
+      "step": 750,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 1.0
+    },
+    {
+      "loss": 0.003704591512680054,
+      "grad_norm": 0.0007531999726779759,
+      "learning_rate": 2.7105676355426248e-05,
+      "entropy": 0.003422930711181834,
+      "num_tokens": 6071234.0,
+      "mean_token_accuracy": 0.9995000004768372,
+      "epoch": 2.0,
+      "step": 1000
+    },
+    {
+      "eval_loss": 0.0042778607457876205,
+      "eval_model_preparation_time": 0.0208,
+      "eval_runtime": 83.828,
+      "eval_samples_per_second": 11.929,
+      "eval_steps_per_second": 1.491,
+      "eval_entropy": 0.0005781573285348713,
+      "eval_num_tokens": 6071234.0,
+      "eval_mean_token_accuracy": 0.999666666507721,
+      "epoch": 2.0,
+      "step": 1000,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 1.0
+    },
+    {
+      "loss": 0.00093178790807724,
+      "grad_norm": 0.000455356523161754,
+      "learning_rate": 1.6921621355072144e-05,
+      "entropy": 0.0006047176675638184,
+      "num_tokens": 7586814.0,
+      "mean_token_accuracy": 0.9999166667461395,
+      "epoch": 2.5,
+      "step": 1250
+    },
+    {
+      "eval_loss": 0.002245673444122076,
+      "eval_model_preparation_time": 0.0208,
+      "eval_runtime": 83.7139,
+      "eval_samples_per_second": 11.945,
+      "eval_steps_per_second": 1.493,
+      "eval_entropy": 0.000366206516395323,
+      "eval_num_tokens": 7586814.0,
+      "eval_mean_token_accuracy": 0.999666666507721,
+      "epoch": 2.5,
+      "step": 1250,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 1.0
+    },
+    {
+      "train_runtime": 9872.6045,
+      "train_samples_per_second": 3.241,
+      "train_steps_per_second": 0.203,
+      "total_flos": 3.82456210218025e+16,
+      "train_loss": 0.0330499241232872,
+      "epoch": 2.5,
+      "step": 1250
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/qwen3-0.6b_D_10k_20260827_213058.json"
+}