Upload train_qwen3_codeforces.py with huggingface_hub

train_qwen3_codeforces.py

@@ -22,8 +22,25 @@ full_dataset = load_dataset("open-r1/codeforces-cots", split="train")
 dataset = full_dataset.select(range(20))
 print(f"✅ Dataset loaded: {len(dataset)} examples")
 
-# Use messages column directly - TRL SFT supports this format
-# No need for train/eval split for quick demo - use full dataset
+# Format the dataset - convert messages to text format for SFT
+def format_for_sft(example):
+    """Convert messages to a single text format for training."""
+    messages = example.get("messages", [])
+    text = ""
+    for msg in messages:
+        role = msg.get("role", "unknown")
+        content = msg.get("content", "")
+        if role == "system":
+            text += f"System: {content}\n\n"
+        elif role == "user":
+            text += f"User: {content}\n\n"
+        elif role == "assistant":
+            text += f"Assistant: {content}\n\n"
+    return {"text": text.strip()}
+
+print("🔄 Formatting dataset...")
+dataset = dataset.map(format_for_sft, remove_columns=dataset.column_names)
+print(f"   Formatted to text: {dataset[0]['text'][:200]}...")
 
 # Training configuration
 config = SFTConfig(
@@ -70,6 +87,7 @@ print("🎯 Initializing trainer...")
 trainer = SFTTrainer(
     model="Qwen/Qwen3-0.6B",
     train_dataset=dataset,
+    dataset_text_field="text",
     args=config,
     peft_config=peft_config,
 )