Add DoRA+ training report: smollm2-1.7b_C_1k [20260826_084936]

reports/v2.0/doraplus/reports_training/smollm2-1.7b_C_1k_20260826_084936.json

@@ -0,0 +1,140 @@
+{
+  "model_key": "smollm2-1.7b",
+  "model_id": "HuggingFaceTB/SmolLM2-1.7B-Instruct",
+  "lora_config": "C",
+  "lora_config_desc": "Wide — full attention + MLP, rank 16",
+  "dataset_size": "1k",
+  "technique": "DoRA+",
+  "device": "cuda",
+  "dtype": "bfloat16",
+  "timestamp": "20260826_084936",
+  "prompt_format": "tool_id_v1",
+  "tool_id_scheme": "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ",
+  "no_tool_id": "-",
+  "gradient_checkpointing": true,
+  "trainable_params": 18726912,
+  "total_params": 1730103296,
+  "trainable_pct": 1.0824,
+  "steps_per_epoch": 50,
+  "total_steps_trained": 200,
+  "early_stopped": false,
+  "early_stopping_patience": 2,
+  "early_stopping_threshold": 0.001,
+  "peak_memory_mb": 7418.5,
+  "total_training_time_s": 3371.4,
+  "final_train_loss": 0.059765,
+  "final_eval_loss": 0.113412,
+  "final_val_accuracy": 0.96,
+  "final_train_accuracy": 1.0,
+  "hf_repo": "kon172verma/intent-classifier-experiments",
+  "hf_subfolder": "v2.0/smollm2-1.7b_DoRA+_C_1k_20260826-084923",
+  "version": "v2.0",
+  "log_history": [
+    {
+      "loss": 0.21614213943481445,
+      "grad_norm": 2.1305623054504395,
+      "learning_rate": 5.577423184847932e-05,
+      "entropy": 0.19348831116687507,
+      "num_tokens": 346669.0,
+      "mean_token_accuracy": 0.9474999970197677,
+      "epoch": 1.0,
+      "step": 50
+    },
+    {
+      "eval_loss": 0.1063322052359581,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 22.5899,
+      "eval_samples_per_second": 4.427,
+      "eval_steps_per_second": 0.575,
+      "eval_entropy": 0.07486873478270493,
+      "eval_num_tokens": 346669.0,
+      "eval_mean_token_accuracy": 0.977564091865833,
+      "epoch": 1.0,
+      "step": 50,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.93
+    },
+    {
+      "loss": 0.0165695858001709,
+      "grad_norm": 0.020507918670773506,
+      "learning_rate": 2.7337132953697554e-08,
+      "entropy": 0.024118779718410224,
+      "num_tokens": 693338.0,
+      "mean_token_accuracy": 0.9949999982118607,
+      "epoch": 2.0,
+      "step": 100
+    },
+    {
+      "eval_loss": 0.09716735035181046,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 22.5846,
+      "eval_samples_per_second": 4.428,
+      "eval_steps_per_second": 0.576,
+      "eval_entropy": 0.025209982091417678,
+      "eval_num_tokens": 693338.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 2.0,
+      "step": 100,
+      "train_accuracy": 0.98,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.0044663786888122555,
+      "grad_norm": 0.0033515747636556625,
+      "learning_rate": 5.247918773366112e-05,
+      "entropy": 0.00982705016154796,
+      "num_tokens": 1040007.0,
+      "mean_token_accuracy": 0.9987499994039536,
+      "epoch": 3.0,
+      "step": 150
+    },
+    {
+      "eval_loss": 0.07785006612539291,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 22.5974,
+      "eval_samples_per_second": 4.425,
+      "eval_steps_per_second": 0.575,
+      "eval_entropy": 0.014512143617209334,
+      "eval_num_tokens": 1040007.0,
+      "eval_mean_token_accuracy": 0.9903846107996427,
+      "epoch": 3.0,
+      "step": 150,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.97
+    },
+    {
+      "loss": 0.0018799801170825959,
+      "grad_norm": 0.0017396783223375678,
+      "learning_rate": 9.956320346634876e-05,
+      "entropy": 0.0041385530192928855,
+      "num_tokens": 1386676.0,
+      "mean_token_accuracy": 0.9991666662693024,
+      "epoch": 4.0,
+      "step": 200
+    },
+    {
+      "eval_loss": 0.1134122759103775,
+      "eval_model_preparation_time": 0.0175,
+      "eval_runtime": 22.5752,
+      "eval_samples_per_second": 4.43,
+      "eval_steps_per_second": 0.576,
+      "eval_entropy": 0.006670154774418244,
+      "eval_num_tokens": 1386676.0,
+      "eval_mean_token_accuracy": 0.9871794810661902,
+      "epoch": 4.0,
+      "step": 200,
+      "train_accuracy": 1.0,
+      "eval_accuracy": 0.96
+    },
+    {
+      "train_runtime": 3370.8887,
+      "train_samples_per_second": 0.949,
+      "train_steps_per_second": 0.059,
+      "total_flos": 2.240938850304e+16,
+      "train_loss": 0.05976452101022005,
+      "epoch": 4.0,
+      "step": 200
+    }
+  ],
+  "hf_report_path": "reports/v2.0/doraplus/reports_training/smollm2-1.7b_C_1k_20260826_084936.json"
+}