Update unsloth-cli.py

[qgallouedec]

• We now use SFTConfig (subclass of TrainingArguments)
• max_seq_length was renamed max_length
• tokenizer was renamed processing_class
• dataset_text_field is "text" by default
• max_length, dataset_text_field, dataset_num_proc and packing are all now arguments of SFTConfig, not SFTTrainer

[danielhanchen]

Thank you!