3 ms·
This reads like parody (from llama.cpp, to it being a beacon of where industry is going (!?), to GPUs are overkill for what is effectively a lot of dot products
by refulgentis 3y ago
This reads like parody (from llama.cpp, to it being a beacon of where industry is going (!?), to GPUs are overkill for what is effectively a lot of dot products)
- NavinF 3y agoYeah using CPUs for inference or training is ridiculous. We're talking 1/20th the performance for 1/4th the energy