Fix trl version and padding in trl qlora example (#12368)

* Change trl to 0.9.6
* Enable padding to avoid padding related errors.

python/llm/example/GPU/LLM-Finetuning/QLoRA/trl-example/README.md

@@ -19,7 +19,7 @@ conda activate llm
 pip install --pre --upgrade ipex-llm[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
 pip install transformers==4.36.0 datasets
 pip install peft==0.10.0
-pip install bitsandbytes scipy "trl<0.12.0"
+pip install bitsandbytes scipy trl==0.9.6
 ```
 
 ### 2. Configures OneAPI environment variables

python/llm/example/GPU/LLM-Finetuning/QLoRA/trl-example/qlora_finetuning.py

@@ -105,6 +105,9 @@ if __name__ == "__main__":
             gradient_checkpointing=True, # can further reduce memory but slower
         ),
         dataset_text_field="instruction",
+        data_collator=transformers.DataCollatorForSeq2Seq(
+            tokenizer, pad_to_multiple_of=8, return_tensors="pt", padding=True
+        ),
     )
     model.config.use_cache = False  # silence the warnings. Please re-enable for inference!
     result = trainer.train()