Chirag Jain 50896ec574 Make nvcc threads configurable via environment variable (#885) пре 9 месеци
..
fused_softmax.cpp ed553e9238 Add Megatron attention implementation for benchmarking пре 2 година
scaled_masked_softmax.h ed553e9238 Add Megatron attention implementation for benchmarking пре 2 година
scaled_masked_softmax_cuda.cu ed553e9238 Add Megatron attention implementation for benchmarking пре 2 година
scaled_upper_triang_masked_softmax.h ed553e9238 Add Megatron attention implementation for benchmarking пре 2 година
scaled_upper_triang_masked_softmax_cuda.cu ed553e9238 Add Megatron attention implementation for benchmarking пре 2 година
setup.py 50896ec574 Make nvcc threads configurable via environment variable (#885) пре 9 месеци
type_shim.h ed553e9238 Add Megatron attention implementation for benchmarking пре 2 година