3 ms·
Memory. I don't know the equation, but its very easy to see when you load a 128k context model at 8K vs 80K. The quant I am running would double VRAM requiremen
by monkmartinez 2y ago
Memory. I don't know the equation, but its very easy to see when you load a 128k context model at 8K vs 80K. The quant I am running would double VRAM requirements when loading 80K.
- thomasfromcdnjs 2y agoThis was my understanding too. Would love more people to chime in on the limits and costs of larger contexts.