vibevoice-asr-it-yodas-only-lora

LoRA adapter for microsoft/VibeVoice-ASR, fine-tuned on YODAS-Granary Italian (asr_only ~87k + ast 200k cap, conversational YouTube audio).

Usage

from peft import PeftModel
from vibevoice.modular.modeling_vibevoice_asr import VibeVoiceASRForConditionalGeneration

model = VibeVoiceASRForConditionalGeneration.from_pretrained("microsoft/VibeVoice-ASR", dtype="bfloat16")
model = PeftModel.from_pretrained(model, "LocalAI-io/vibevoice-asr-it-yodas-only-lora")

For a pre-merged model see LocalAI-io/vibevoice-asr-it-yodas-only — same weights, faster to load.