Upload train_config.json with huggingface_hub

train_config.json

@@ -0,0 +1,54 @@
+{
+  "dataset": {
+    "path": "/root/jb/personas/finance/budgeting/data/qa_pairs_train.json",
+    "num_train_samples": 1148,
+    "dataset_text_field": "text"
+  },
+  "validation_dataset": {
+    "path": "/root/jb/personas/finance/budgeting/data/qa_pairs_val.json",
+    "num_eval_samples": 144
+  },
+  "model": {
+    "name": "Qwen/Qwen3-4B-Instruct-2507",
+    "load_in_4bit": false
+  },
+  "lora": {
+    "r": 16,
+    "lora_alpha": 16,
+    "lora_dropout": 0.05,
+    "bias": "none",
+    "target_modules": [
+      "q_proj",
+      "k_proj",
+      "v_proj",
+      "o_proj",
+      "gate_proj",
+      "up_proj",
+      "down_proj"
+    ],
+    "random_state": 3407
+  },
+  "training": {
+    "output_dir": "/root/jb/personas/finance/budgeting/finetune_lora_output/2026-03-04_04-18-09",
+    "max_seq_length": 2048,
+    "num_train_epochs": 3.0,
+    "max_steps": -1,
+    "per_device_train_batch_size": 2,
+    "gradient_accumulation_steps": 4,
+    "learning_rate": 0.0001,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.03,
+    "lr_scheduler_type": "cosine",
+    "optim": "adamw_torch",
+    "adam_beta1": 0.9,
+    "adam_beta2": 0.95,
+    "bf16": true,
+    "seed": 3407,
+    "logging_steps": 10,
+    "save_strategy": "steps",
+    "save_steps": 200,
+    "save_total_limit": 2,
+    "dataset_num_proc": 4,
+    "packing": false
+  }
+}