Add DoRA training report: qwen3-0.6b_D_1k [20260826_193928]

reports/v2.0/dora/reports_training/qwen3-0.6b_D_1k_20260826_193928.json

@@ -0,0 +1,140 @@
+{
+  "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": "1k",
+  "technique": "DoRA",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_193928",
+  "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": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 2679.1,
+  "total_training_time_s": 1159.1,
+  "final_train_loss": 0.068808,
+  "final_eval_loss": 0.063022,
+  "final_val_accuracy": 0.98,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen3-0.6b_DoRA_D_1k_20260826-193923",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.2597960090637207,
+      "grad_norm": 0.3098641335964203,
+      "learning_rate": 4.497969992237312e-05,
+      "entropy": 0.08113368359539891,
+      "num_tokens": 304665.0,
+      "mean_token_accuracy": 0.9545833370089531,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.09065386652946472,
+      "eval_model_preparation_time": 0.0206,
+      "eval_runtime": 8.6889,
+      "eval_samples_per_second": 11.509,
+      "eval_steps_per_second": 1.496,
+      "eval_entropy": 0.033362842057473384,
+      "eval_num_tokens": 304665.0,
+      "eval_mean_token_accuracy": 0.9839743559177105,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.015026552677154541,
+      "grad_norm": 0.004639339633285999,
+      "learning_rate": 2.7476138256261575e-05,
+      "entropy": 0.006456450363530166,
+      "num_tokens": 609330.0,
+      "mean_token_accuracy": 0.9958333340287209,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.10122394561767578,
+      "eval_model_preparation_time": 0.0206,
+      "eval_runtime": 8.7028,
+      "eval_samples_per_second": 11.491,
+      "eval_steps_per_second": 1.494,
+      "eval_entropy": 0.015222121542767407,
+      "eval_num_tokens": 609330.0,
+      "eval_mean_token_accuracy": 0.987179485651163,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.96
+    },
+    {
+      "loss": 0.0003817184641957283,
+      "grad_norm": 0.002370664617046714,
+      "learning_rate": 8.374385697153792e-06,
+      "entropy": 0.0012732458028676774,
+      "num_tokens": 913995.0,
+      "mean_token_accuracy": 1.0,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.061382751911878586,
+      "eval_model_preparation_time": 0.0206,
+      "eval_runtime": 8.6901,
+      "eval_samples_per_second": 11.507,
+      "eval_steps_per_second": 1.496,
+      "eval_entropy": 0.0010031604814195396,
+      "eval_num_tokens": 913995.0,
+      "eval_mean_token_accuracy": 0.9935897405330951,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.98
+    },
+    {
+      "loss": 2.6424024254083635e-05,
+      "grad_norm": 0.010112090036273003,
+      "learning_rate": 3.417375188274896e-09,
+      "entropy": 0.00028480165761720857,
+      "num_tokens": 1218660.0,
+      "mean_token_accuracy": 1.0,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.06302230060100555,
+      "eval_model_preparation_time": 0.0206,
+      "eval_runtime": 8.7022,
+      "eval_samples_per_second": 11.491,
+      "eval_steps_per_second": 1.494,
+      "eval_entropy": 0.0008406575420499743,
+      "eval_num_tokens": 1218660.0,
+      "eval_mean_token_accuracy": 0.9935897405330951,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.98
+    },
+    {
+      "train_runtime": 1158.6638,
+      "train_samples_per_second": 2.762,
+      "train_steps_per_second": 0.173,
+      "total_flos": 4906693000691712.0,
+      "train_loss": 0.06880767605733126,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/dora/reports_training/qwen3-0.6b_D_1k_20260826_193928.json"
+}