forked from tinygrad/tinygrad
it made gpt2 slower (2ms -> 2.5ms on 3090, 7ms -> 8ms on M1 Max with BEAM=2). disabled it in gpt2 benchmark before understanding the full issue
it made gpt2 slower (2ms -> 2.5ms on 3090, 7ms -> 8ms on M1 Max with BEAM=2). disabled it in gpt2 benchmark before understanding the full issue