diff options
| author | 0cc4m <picard12@live.de> | 2024-02-01 19:25:24 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-02-01 19:25:24 +0100 |
| commit | 4d0924a8902010d31bd737b6f1f594943d120d0f (patch) | |
| tree | 091227c1265488e6a528f280304b6ad92d6e8e17 /ggml_vk_generate_shaders.py | |
| parent | 8ca511cadee2c67f0bd8c7034a2513778ee9a1b7 (diff) | |
Vulkan Phi Fix for AMD Proprietary Drivers (#5260)
* Replace tanh to avoid NaN in gelu shader on AMD proprietary driver
* Fix another Vulkan CPY buffer size bug
Diffstat (limited to 'ggml_vk_generate_shaders.py')
| -rw-r--r-- | ggml_vk_generate_shaders.py | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/ggml_vk_generate_shaders.py b/ggml_vk_generate_shaders.py index 6b1b82bf..67981a75 100644 --- a/ggml_vk_generate_shaders.py +++ b/ggml_vk_generate_shaders.py @@ -1689,7 +1689,8 @@ void main() { } const float xi = float(data_a[i]); - data_d[i] = D_TYPE(0.5f*xi*(1.0f + tanh(SQRT_2_OVER_PI*xi*(1.0f + GELU_COEF_A*xi*xi)))); + const float val = SQRT_2_OVER_PI*xi*(1.0f + GELU_COEF_A*xi*xi); + data_d[i] = D_TYPE(0.5f*xi*(2.0f - 2.0f / (exp(2 * val) + 1))); } """ |