4 ms·
I think they are bfloat16, not FP16, but they are both 16bpw formats, so it doesn't make a size difference.
by dragonwriter 1y ago
I think they are bfloat16, not FP16, but they are both 16bpw formats, so it doesn't make a size difference.
- Tokumei-no-hito 1y agopardon the ignorance but it's the first time I've heard of bfloat16. i asked chat for an explanation and it said bfloat has a higher range (like fp32) but less precision. what does that mean for image generation and why was bfloat chosen over fp?
- dragonwriter 1y agoMy fuzzy understanding, and I'm not at all an expert on this, that the main benefit is that bf16 is less prone to overflow/underflow during calculation, which is a source of bigger problems in both training and inference than the simple loss of precision, so once it became widely supported, it became a commonly-preferred format for models (whether image gen or otherwise) over FP16.
- iyn 1y agoWiki article on bfloat16 for reference, since it was new to me: https://en.wikipedia.org/wiki/Bfloat16_floating-point_format https://en.wikipedia.org/wiki/Bfloat16_floating-point_format