Upload training_config.json with huggingface_hub

training_config.json

@@ -0,0 +1,38 @@
+{
+  "base_model": "unsloth/DeepSeek-R1-Distill-Llama-8B",
+  "task": "text-generation",
+  "language": "en",
+  "license": "apache-2.0",
+  "tags": [
+    "dyck-language",
+    "bracket-completion",
+    "reasoning",
+    "lora",
+    "fine-tuned"
+  ],
+  "training": {
+    "dataset_size": 60000,
+    "train_split": 0.95,
+    "lora_rank": 32,
+    "lora_alpha": 64,
+    "lora_dropout": 0.25,
+    "learning_rate": 3e-06,
+    "batch_size": 4,
+    "gradient_accumulation_steps": 32,
+    "effective_batch_size": 128,
+    "num_epochs": 4,
+    "warmup_ratio": 0.3,
+    "max_grad_norm": 0.05,
+    "optimizer": "adamw_torch",
+    "lr_scheduler": "linear"
+  },
+  "model_info": {
+    "lora_target_modules": [
+      "q_proj",
+      "k_proj",
+      "v_proj",
+      "o_proj"
+    ],
+    "max_seq_length": 2048
+  }
+}