Add LoRA+ training report: qwen2.5-0.5b_C_1k [20260826_173600]

reports/v2.0/loraplus/reports_training/qwen2.5-0.5b_C_1k_20260826_173600.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "qwen2.5-0.5b",
+  "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
+  "lora_config": "C",
+  "lora_config_desc": "Wide — full attention + MLP, rank 16",
+  "dataset_size": "1k",
+  "technique": "LoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_173600",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 8798208,
+  "total_params": 502830976,
+  "trainable_pct": 1.7497,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 2283.6,
+  "total_training_time_s": 473.6,
+  "final_train_loss": 0.340094,
+  "final_eval_loss": 0.284944,
+  "final_val_accuracy": 0.76,
+  "final_train_accuracy": 0.88,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen2.5-0.5b_LoRA+_C_1k_20260826-173556",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.7468997955322265,
+      "grad_norm": 5.520288467407227,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.6356516239792108,
+      "num_tokens": 301465.0,
+      "mean_token_accuracy": 0.7970833355188369,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.5062764286994934,
+      "eval_model_preparation_time": 0.0156,
+      "eval_runtime": 3.3452,
+      "eval_samples_per_second": 29.893,
+      "eval_steps_per_second": 3.886,
+      "eval_entropy": 0.5140390510742481,
+      "eval_num_tokens": 301465.0,
+      "eval_mean_token_accuracy": 0.8621794810661902,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.76,
+      "eval_accuracy": 0.56
+    },
+    {
+      "loss": 0.2513236999511719,
+      "grad_norm": 2.579852342605591,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.2212260202690959,
+      "num_tokens": 602930.0,
+      "mean_token_accuracy": 0.9312499964237213,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.2889472544193268,
+      "eval_model_preparation_time": 0.0156,
+      "eval_runtime": 3.3244,
+      "eval_samples_per_second": 30.081,
+      "eval_steps_per_second": 3.91,
+      "eval_entropy": 0.23290404849327528,
+      "eval_num_tokens": 602930.0,
+      "eval_mean_token_accuracy": 0.9391025625742399,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.92,
+      "eval_accuracy": 0.81
+    },
+    {
+      "loss": 0.15021800994873047,
+      "grad_norm": 6.011902332305908,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.13330494993831962,
+      "num_tokens": 904395.0,
+      "mean_token_accuracy": 0.9612499958276749,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.3083612322807312,
+      "eval_model_preparation_time": 0.0156,
+      "eval_runtime": 3.3252,
+      "eval_samples_per_second": 30.073,
+      "eval_steps_per_second": 3.91,
+      "eval_entropy": 0.30045631184027743,
+      "eval_num_tokens": 904395.0,
+      "eval_mean_token_accuracy": 0.9070512881645789,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.88,
+      "eval_accuracy": 0.74
+    },
+    {
+      "loss": 0.21193265914916992,
+      "grad_norm": 12.695247650146484,
+      "learning_rate": 9.956320346634876e-05,
+      "entropy": 0.1474583032936789,
+      "num_tokens": 1205860.0,
+      "mean_token_accuracy": 0.950833330154419,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.28494423627853394,
+      "eval_model_preparation_time": 0.0156,
+      "eval_runtime": 3.3247,
+      "eval_samples_per_second": 30.078,
+      "eval_steps_per_second": 3.91,
+      "eval_entropy": 0.24595708400011063,
+      "eval_num_tokens": 1205860.0,
+      "eval_mean_token_accuracy": 0.9230769322468684,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.88,
+      "eval_accuracy": 0.76
+    },
+    {
+      "train_runtime": 473.1763,
+      "train_samples_per_second": 6.763,
+      "train_steps_per_second": 0.423,
+      "total_flos": 4390551847188480.0,
+      "train_loss": 0.3400935411453247,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/loraplus/reports_training/qwen2.5-0.5b_C_1k_20260826_173600.json"
+}