Show HN: We compressed a 31B video+audio diffusion model to 14B, 6 steps, INT8
Compression of a large video and audio diffusion model to 14B parameters using INT8 quantization over 6 steps.
Hacker News submitter: jiant
Submitted to Hacker News:
Compression of a large video and audio diffusion model to 14B parameters using INT8 quantization over 6 steps.
Hacker News submitter: jiant
Submitted to Hacker News: