Update finetune_flan_t5.py

finetune_flan_t5.py

@@ -16,12 +16,10 @@ model_name = "google/flan-t5-base"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
 
-# 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
+# 3. CORRECTED Formatting function - returns single string per example
+def format_instruction(example):
+    # Return a single formatted string
+    return f"### Instruction:\n{example['input']}\n\n### Response:\n{example['output']}"
 
 # 4. Training arguments
 training_args = TrainingArguments(
@@ -37,20 +35,22 @@ training_args = TrainingArguments(
     report_to="none"
 )
 
-# 5. Initialize SFTTrainer
+# 5. Initialize SFTTrainer with correct parameters
 trainer = SFTTrainer(
     model=model,
     tokenizer=tokenizer,
     train_dataset=dataset,
     args=training_args,
     max_seq_length=512,
-    formatting_func=format_instruction,
+    formatting_func=format_instruction,  # Now returns single string
     data_collator=DataCollatorForSeq2Seq(
         tokenizer,
+        model=model,  # Added model reference
         pad_to_multiple_of=8,
         return_tensors="pt",
         padding=True
-    )
+    ),
+    dataset_text_field="text"  # Explicit field name
 )
 
 # 6. Start training