Comment by pclmulqdq
2 months ago
Oh yeah, I did that math not assuming any quantization. I think if you can get a 3-4 bit quant working + int8 math, ~80 might be achievable.
2 months ago
Oh yeah, I did that math not assuming any quantization. I think if you can get a 3-4 bit quant working + int8 math, ~80 might be achievable.
No comments yet
Contribute on Hacker News ↗