Update finetune_flan_t5.py

finetune_flan_t5.py

@@ -16,9 +16,12 @@ model_name = "google/flan-t5-base"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
 
-# 3. Formatting function for SFTTrainer
-def format_instruction(example):
-    return f"### Instruction:\n{example['input']}\n\n### Response:\n{example['output']}"
+# 3. CORRECTED Formatting function - must return a list
+def format_instruction(examples):
+    texts = []
+    for input_text, output_text in zip(examples["input"], examples["output"]):
+        texts.append(f"### Instruction:\n{input_text}\n\n### Response:\n{output_text}")
+    return {"text": texts}  # Return dict with "text" key containing list
 
 # 4. Training arguments
 training_args = TrainingArguments(
@@ -34,7 +37,7 @@ training_args = TrainingArguments(
     report_to="none"
 )
 
-# 5. Initialize SFTTrainer correctly
+# 5. Initialize SFTTrainer
 trainer = SFTTrainer(
     model=model,
     tokenizer=tokenizer,