zisisbatzos/3SFTs_synthetic_emobench_llama3.2-3B

training_arguments = SFTConfig( output_dir=output_dir, per_device_train_batch_size=16, gradient_accumulation_steps=1, optim="adamw_torch", save_steps=100, logging_steps=1, learning_rate=1e-4, bf16=True, # max_grad_norm=0.3, num_train_epochs=3, save_strategy="epoch", warmup_ratio=0.05, group_by_length=True, # lr_scheduler_type="cosine", gradient_checkpointing=True, gradient_checkpointing_kwargs = {"use_reentrant": True}, dataset_text_field="text", max_seq_length=1024, packing=False # report_to="wandb", )

baseline_model = 'zisisbatzos/2SFTs_llama3.2-3B'

Metrics: emotion_accuracy = 0.36554621848739494 correct_format = 0.7058823529411765