LoRA adapter trained via supervised fine-tuning on
Orpheus 3B using the Expresso conversational speech dataset.
This SFT approach produced a model that closely mimics the Expresso speaker's voice characteristics but showed slight regression in UTMOS naturalness compared to the base Orpheus model. This led us to explore GRPO-based optimization as an alternative — see
orpheus-3b-conversational-grpo.
The lesson: Orpheus was pretrained on ~100k hours of diverse speech. Supervised fine-tuning on a small single-speaker dataset can overfit to that speaker's patterns at the cost of the model's general conversational ability.
1from peft import PeftModel
2from transformers import AutoModelForCausalLM
3
4base = AutoModelForCausalLM.from_pretrained("canopylabs/orpheus-3b-0.1-ft")
5model = PeftModel.from_pretrained(base, "Tachyeon/orpheus-3b-sft-lora")
6model = model.merge_and_unload()
