Upload train_question_generator.py with huggingface_hub

train_question_generator.py

@@ -57,7 +57,7 @@ peft_config = LoraConfig(
     task_type="CAUSAL_LM",
 )
 
-# Training config - fewer epochs due to larger dataset
+# Training config - removed max_seq_length, gradient_checkpointing from SFTConfig
 training_args = SFTConfig(
     output_dir="./question-generator-output",
     num_train_epochs=2,
@@ -73,8 +73,6 @@ training_args = SFTConfig(
     push_to_hub=True,
     hub_model_id="KevinKeller/cognitive-question-generator-qwen2.5-7b",
     report_to="none",
-    max_seq_length=8192,
-    gradient_checkpointing=True,
 )
 
 print("Starting training...")
@@ -85,6 +83,7 @@ trainer = SFTTrainer(
     peft_config=peft_config,
     tokenizer=tokenizer,
     args=training_args,
+    max_seq_length=8192,  # Moved here
 )
 
 trainer.train()