Add DoRA+ training report: qwen2.5-0.5b_C_1k [20260826_150641]

reports/v2.0/doraplus/reports_training/qwen2.5-0.5b_C_1k_20260826_150641.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": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_150641",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 9102336,
+  "total_params": 503135104,
+  "trainable_pct": 1.8091,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 2730.7,
+  "total_training_time_s": 1091.3,
+  "final_train_loss": 0.320833,
+  "final_eval_loss": 0.254256,
+  "final_val_accuracy": 0.79,
+  "final_train_accuracy": 0.84,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen2.5-0.5b_DoRA+_C_1k_20260826-150632",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.7828556823730469,
+      "grad_norm": 7.804486274719238,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.650215854421258,
+      "num_tokens": 301465.0,
+      "mean_token_accuracy": 0.7858333331346512,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.4703764021396637,
+      "eval_model_preparation_time": 0.0176,
+      "eval_runtime": 7.2096,
+      "eval_samples_per_second": 13.87,
+      "eval_steps_per_second": 1.803,
+      "eval_entropy": 0.4753659986532651,
+      "eval_num_tokens": 301465.0,
+      "eval_mean_token_accuracy": 0.8685897359481225,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.78,
+      "eval_accuracy": 0.6
+    },
+    {
+      "loss": 0.2075057792663574,
+      "grad_norm": 3.9430124759674072,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.19689129885286094,
+      "num_tokens": 602930.0,
+      "mean_token_accuracy": 0.9512499970197678,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.20672902464866638,
+      "eval_model_preparation_time": 0.0176,
+      "eval_runtime": 7.211,
+      "eval_samples_per_second": 13.868,
+      "eval_steps_per_second": 1.803,
+      "eval_entropy": 0.2282913401722908,
+      "eval_num_tokens": 602930.0,
+      "eval_mean_token_accuracy": 0.9455128174561721,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.83
+    },
+    {
+      "loss": 0.11423568725585938,
+      "grad_norm": 3.2578463554382324,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.10342155083548278,
+      "num_tokens": 904395.0,
+      "mean_token_accuracy": 0.9729166615009308,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.22538158297538757,
+      "eval_model_preparation_time": 0.0176,
+      "eval_runtime": 7.2158,
+      "eval_samples_per_second": 13.859,
+      "eval_steps_per_second": 1.802,
+      "eval_entropy": 0.297334047464224,
+      "eval_num_tokens": 904395.0,
+      "eval_mean_token_accuracy": 0.9326923076923077,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 0.96,
+      "eval_accuracy": 0.8
+    },
+    {
+      "loss": 0.17873613357543947,
+      "grad_norm": 6.641326904296875,
+      "learning_rate": 9.956320346634876e-05,
+      "entropy": 0.12517021249979735,
+      "num_tokens": 1205860.0,
+      "mean_token_accuracy": 0.9558333307504654,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.25425586104393005,
+      "eval_model_preparation_time": 0.0176,
+      "eval_runtime": 7.2124,
+      "eval_samples_per_second": 13.865,
+      "eval_steps_per_second": 1.802,
+      "eval_entropy": 0.18964181324610344,
+      "eval_num_tokens": 1205860.0,
+      "eval_mean_token_accuracy": 0.9262820528103755,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 0.84,
+      "eval_accuracy": 0.79
+    },
+    {
+      "train_runtime": 1090.8859,
+      "train_samples_per_second": 2.933,
+      "train_steps_per_second": 0.183,
+      "total_flos": 4394193252022272.0,
+      "train_loss": 0.3208333206176758,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/qwen2.5-0.5b_C_1k_20260826_150641.json"
+}