AsadIsmail/ternary-quant-demo

ternary-quant — post-training quantization to {−1, 0, +1} for VLMs, seq2seq, and audio models, not just LLMs.

Three pre-quantized checkpoints are available:

Select a model, enter a prompt, and click Generate. The first call downloads and loads the checkpoint (may take ~30s); subsequent calls are fast.

Source: github.com/Asad-Ismail/ternary-quant