Add QLoRA training report: qwen3-0.6b_C_1k [20260827_020538]

reports/v2.0/qlora/reports_training/qwen3-0.6b_C_1k_20260827_020538.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "qwen3-0.6b",
+  "model_id": "Qwen/Qwen3-0.6B",
+  "lora_config": "C",
+  "lora_config_desc": "Wide — full attention + MLP, rank 16",
+  "dataset_size": "1k",
+  "technique": "QLoRA",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260827_020538",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": false,
+  "trainable_params": 10092544,
+  "total_params": 385941504,
+  "trainable_pct": 2.615,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 20385.7,
+  "total_training_time_s": 828.3,
+  "final_train_loss": 0.11485,
+  "final_eval_loss": 0.032436,
+  "final_val_accuracy": 0.99,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/qwen3-0.6b_QLoRA_C_1k_20260827-020531",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.43555850982666017,
+      "grad_norm": 0.40234375,
+      "learning_rate": 8.995939984474624e-05,
+      "entropy": 0.15142830192577095,
+      "num_tokens": 304665.0,
+      "mean_token_accuracy": 0.9220833313465119,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.057093601673841476,
+      "eval_model_preparation_time": 0.0278,
+      "eval_runtime": 7.2182,
+      "eval_samples_per_second": 13.854,
+      "eval_steps_per_second": 1.801,
+      "eval_entropy": 0.04345604917034507,
+      "eval_num_tokens": 304665.0,
+      "eval_mean_token_accuracy": 0.9807692261842581,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.95
+    },
+    {
+      "loss": 0.015201172828674316,
+      "grad_norm": 0.00146484375,
+      "learning_rate": 5.495227651252315e-05,
+      "entropy": 0.00799851529984153,
+      "num_tokens": 609330.0,
+      "mean_token_accuracy": 0.9962499988079071,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.05451250448822975,
+      "eval_model_preparation_time": 0.0278,
+      "eval_runtime": 7.2232,
+      "eval_samples_per_second": 13.844,
+      "eval_steps_per_second": 1.8,
+      "eval_entropy": 0.015081649177143564,
+      "eval_num_tokens": 609330.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.008558189272880554,
+      "grad_norm": 0.0030059814453125,
+      "learning_rate": 1.6748771394307585e-05,
+      "entropy": 0.001733155381953111,
+      "num_tokens": 913995.0,
+      "mean_token_accuracy": 0.9983333325386048,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.032598163932561874,
+      "eval_model_preparation_time": 0.0278,
+      "eval_runtime": 7.2275,
+      "eval_samples_per_second": 13.836,
+      "eval_steps_per_second": 1.799,
+      "eval_entropy": 0.0024861693415844524,
+      "eval_num_tokens": 913995.0,
+      "eval_mean_token_accuracy": 0.9967948702665476,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.99
+    },
+    {
+      "loss": 8.032183162868023e-05,
+      "grad_norm": 0.001251220703125,
+      "learning_rate": 6.834750376549792e-09,
+      "entropy": 0.0006823884158075089,
+      "num_tokens": 1218660.0,
+      "mean_token_accuracy": 1.0,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.032435718923807144,
+      "eval_model_preparation_time": 0.0278,
+      "eval_runtime": 7.225,
+      "eval_samples_per_second": 13.841,
+      "eval_steps_per_second": 1.799,
+      "eval_entropy": 0.002079444402345241,
+      "eval_num_tokens": 1218660.0,
+      "eval_mean_token_accuracy": 0.9967948702665476,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.99
+    },
+    {
+      "train_runtime": 827.8794,
+      "train_samples_per_second": 3.865,
+      "train_steps_per_second": 0.242,
+      "total_flos": 5429276835840000.0,
+      "train_loss": 0.11484954843996092,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/qlora/reports_training/qwen3-0.6b_C_1k_20260827_020538.json"
+}