Laurent Mazare eb26e2467e Add the cuda dequantize f16 kernels. (#2137) ...

* Add the cuda dequantize f16 kernels.

* Expose the cuda kernels.

* Add some testing + fix.

* Test the other cases too.

* A few more tests.

* Add an environment variable to enable the dequantize f16 + matmul behavior.

2024-04-28 20:05:05 +02:00

..

affine.cu

Cuda backend optimization (#1886)

2024-03-20 18:32:55 +01:00

binary_op_macros.cuh

Cuda backend optimization (#1886)

2024-03-20 18:32:55 +01:00

binary.cu

Add support for i64 (#563)

2023-08-23 10:42:19 +01:00

cast.cu

Add cast_bf16_x/cast_x_bf16 when CUDA_ARCH<800 but CUDA_VERSION >= 11000 (#1919)

2024-03-23 13:44:10 +01:00

compatibility.cuh

Compat windows.

2023-08-10 17:46:47 +02:00

conv.cu

ConvTranspose1d cuda support. (#1697)

2024-02-12 15:03:18 +01:00

cuda_utils.cuh

Relax the contiguous check for cuda kernels. (#2000)

2024-04-03 09:02:38 +02:00

fill.cu

Optimize the cat operation on contiguous tensors (#1855)

2024-03-17 10:49:13 +01:00

indexing.cu

Support scatter/index_add with i64 indices for f16 (#1915)

2024-03-22 11:51:41 +01:00

lib.rs

Add argsort. (#2132)

2024-04-27 20:17:35 +02:00

quantized.cu

Add the cuda dequantize f16 kernels. (#2137)

2024-04-28 20:05:05 +02:00

reduce.cu

Add the rope THD kernel. (#2014)

2024-04-05 08:32:58 +02:00

sort.cu

Add argsort. (#2132)

2024-04-27 20:17:35 +02:00

ternary.cu

Add support for i64 (#563)

2023-08-23 10:42:19 +01:00

unary.cu

Add support for "sign" on tensors (#2012)

2024-04-04 22:32:47 +02:00