Add LoRA+ training report: qwen3-0.6b_A_1k [20260826_175503]

reports/v2.0/loraplus/reports_training/qwen3-0.6b_A_1k_20260826_175503.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "qwen3-0.6b",
+  "model_id": "Qwen/Qwen3-0.6B",
+  "lora_config": "A",
+  "lora_config_desc": "Light — attention Q/V only, rank 8",
+  "dataset_size": "1k",
+  "technique": "LoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_175503",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 1146880,
+  "total_params": 597196800,
+  "trainable_pct": 0.192,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 2554.8,
+  "total_training_time_s": 543.3,
+  "final_train_loss": 0.059839,
+  "final_eval_loss": 0.042935,
+  "final_val_accuracy": 0.97,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen3-0.6b_LoRA+_A_1k_20260826-175500",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.21809017181396484,
+      "grad_norm": 0.020474819466471672,
+      "learning_rate": 0.00011154846369695863,
+      "entropy": 0.10545997738372534,
+      "num_tokens": 304665.0,
+      "mean_token_accuracy": 0.9658333295583725,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.06340473145246506,
+      "eval_model_preparation_time": 0.009,
+      "eval_runtime": 3.7006,
+      "eval_samples_per_second": 27.023,
+      "eval_steps_per_second": 3.513,
+      "eval_entropy": 0.011969692611063901,
+      "eval_num_tokens": 304665.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.017622696161270143,
+      "grad_norm": 0.0111448485404253,
+      "learning_rate": 5.467426590739511e-08,
+      "entropy": 0.010140910931513645,
+      "num_tokens": 609330.0,
+      "mean_token_accuracy": 0.9954166650772095,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.06994181126356125,
+      "eval_model_preparation_time": 0.009,
+      "eval_runtime": 3.6711,
+      "eval_samples_per_second": 27.239,
+      "eval_steps_per_second": 3.541,
+      "eval_entropy": 0.009715559534155406,
+      "eval_num_tokens": 609330.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.0036101558804512023,
+      "grad_norm": 0.002565652597695589,
+      "learning_rate": 0.00010495837546732224,
+      "entropy": 0.004121706478181295,
+      "num_tokens": 913995.0,
+      "mean_token_accuracy": 0.9987499994039536,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.022596634924411774,
+      "eval_model_preparation_time": 0.009,
+      "eval_runtime": 3.6846,
+      "eval_samples_per_second": 27.14,
+      "eval_steps_per_second": 3.528,
+      "eval_entropy": 0.009113480245175127,
+      "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": 3.336827969178558e-05,
+      "grad_norm": 0.00034641197999008,
+      "learning_rate": 0.00019912640693269752,
+      "entropy": 0.0003618359156826045,
+      "num_tokens": 1218660.0,
+      "mean_token_accuracy": 1.0,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.04293487221002579,
+      "eval_model_preparation_time": 0.009,
+      "eval_runtime": 3.6841,
+      "eval_samples_per_second": 27.144,
+      "eval_steps_per_second": 3.529,
+      "eval_entropy": 0.005925139234862469,
+      "eval_num_tokens": 1218660.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "train_runtime": 542.9542,
+      "train_samples_per_second": 5.894,
+      "train_steps_per_second": 0.368,
+      "total_flos": 5321481012117504.0,
+      "train_loss": 0.05983909803384449,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/loraplus/reports_training/qwen3-0.6b_A_1k_20260826_175503.json"
+}