aiguy110/VibeVoice-Large-f16

VibeVoice F16 Model (Incremental Conversion)

This model has been converted to float16 (f16) precision using incremental processing to avoid RAM exhaustion.

Conversion Details

• Original model: sheliak/VibeVoice-Large_Mirror
• Conversion method: Incremental shard-by-shard processing
• Mixed precision: True
• Total parameters: 9,343,355,363
• F16 parameters: 8,073,169,443 (86.4%)
• F32 parameters: 0 (0.0%)
• Model size: 15.04GB
• Memory savings: ~0.0%

Usage

from vibevoice.modular.modeling_vibevoice_inference import VibeVoiceForConditionalGenerationInference
from vibevoice.processor.vibevoice_processor import VibeVoiceProcessor

# Load with f16 precision
model = VibeVoiceForConditionalGenerationInference.from_pretrained(
    "./VibeVoice-Large-f16",
    torch_dtype=torch.float16,
    device_map="cpu"
)

processor = VibeVoiceProcessor.from_pretrained("./VibeVoice-Large-f16")

# Use --use_f16 flag with demo scripts
python demo/inference_from_file.py --model_path ./VibeVoice-Large-f16 --use_f16 --device cpu

RAM-Friendly Conversion

This model was converted using incremental processing, making it possible to convert large models on systems with limited RAM.