llama.cpp
8f900abf - CUDA: faster softmax via shared memory + fp16 math (#4742)

Commit

2 years ago

CUDA: faster softmax via shared memory + fp16 math (#4742)

References

#4742 - CUDA: faster softmax via shared memory + fp16 math

Author

JohannesGaessler

JohannesGaessler

Parents

Loading