llama.cpp
03d46982 - CUDA: use mma FA kernel for gqa > 4 on RTX 4000 (#15035)

Commit

39 days ago

CUDA: use mma FA kernel for gqa > 4 on RTX 4000 (#15035)

References

#15035 - CUDA: use mma FA kernel for gqa > 4 on RTX 4000

Author

JohannesGaessler

JohannesGaessler

Parents

Loading