diff options
| author | 0cc4m <picard12@live.de> | 2024-05-19 17:19:53 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-05-19 17:19:53 +0200 |
| commit | f030ec1f7a72aa825b2104823946551b9ec5dfc1 (patch) | |
| tree | 705d08bb8df40049ba48bf57629accbd0b4ee953 /ggml_vk_generate_shaders.py | |
| parent | e4e6f67be6a8a697f5f89a28c98934e53c99c359 (diff) | |
Vulkan Embedding Fix (#7360)
* Fix empty Vulkan host buffers
Add fp32 fp16 matmul shader
Fix matmul shader alignment
* Remove deprecated tensor->backend uses
* Fix Vulkan validation errors on embedding models with no offloaded layers
* Fix Vulkan llava segfault when not offloading layers
Diffstat (limited to 'ggml_vk_generate_shaders.py')
| -rw-r--r-- | ggml_vk_generate_shaders.py | 9 |
1 files changed, 6 insertions, 3 deletions
diff --git a/ggml_vk_generate_shaders.py b/ggml_vk_generate_shaders.py index 3c22a986..8096c03b 100644 --- a/ggml_vk_generate_shaders.py +++ b/ggml_vk_generate_shaders.py @@ -2877,13 +2877,16 @@ async def main(): stream.clear() stream.extend((mulmat_head, shader_float_type, mulmat_body1, mulmat_load_scalar, mulmat_body2)) tasks.append(string_to_spv("matmul_f32", "".join(stream), {"A_TYPE": "float", "B_TYPE": "float", "D_TYPE": "float"}, fp16)) - tasks.append(string_to_spv("matmul_f32_aligned", "".join(stream), {"LOAD_VEC_A": 1, "LOAD_VEC_B": load_vec, "A_TYPE": "float", "B_TYPE": vec_type, "D_TYPE": "float"}, fp16)) + tasks.append(string_to_spv("matmul_f32_aligned", "".join(stream), {"LOAD_VEC_A": load_vec, "LOAD_VEC_B": load_vec, "A_TYPE": vec_type, "B_TYPE": vec_type, "D_TYPE": "float"}, fp16)) + + tasks.append(string_to_spv("matmul_f32_f16", "".join(stream), {"A_TYPE": "float", "B_TYPE": "float16_t", "D_TYPE": "float"}, fp16)) + tasks.append(string_to_spv("matmul_f32_f16_aligned", "".join(stream), {"LOAD_VEC_A": load_vec, "LOAD_VEC_B": load_vec, "A_TYPE": vec_type, "B_TYPE": vec_type_f16, "D_TYPE": "float"}, fp16)) tasks.append(string_to_spv("matmul_f16", "".join(stream), {"A_TYPE": "float16_t", "B_TYPE": "float16_t", "D_TYPE": "float"}, fp16)) - tasks.append(string_to_spv("matmul_f16_aligned", "".join(stream), {"LOAD_VEC_A": 1, "LOAD_VEC_B": load_vec, "A_TYPE": "float16_t", "B_TYPE": vec_type_f16, "D_TYPE": "float"}, fp16)) + tasks.append(string_to_spv("matmul_f16_aligned", "".join(stream), {"LOAD_VEC_A": load_vec, "LOAD_VEC_B": load_vec, "A_TYPE": vec_type_f16, "B_TYPE": vec_type_f16, "D_TYPE": "float"}, fp16)) tasks.append(string_to_spv("matmul_f16_f32", "".join(stream), {"A_TYPE": "float16_t", "B_TYPE": "float", "D_TYPE": "float"}, fp16)) - tasks.append(string_to_spv("matmul_f16_f32_aligned", "".join(stream), {"LOAD_VEC_A": 1, "LOAD_VEC_B": load_vec, "A_TYPE": "float16_t", "B_TYPE": vec_type, "D_TYPE": "float"}, fp16)) + tasks.append(string_to_spv("matmul_f16_f32_aligned", "".join(stream), {"LOAD_VEC_A": load_vec, "LOAD_VEC_B": load_vec, "A_TYPE": vec_type_f16, "B_TYPE": vec_type, "D_TYPE": "float"}, fp16)) stream.clear() stream.extend((mulmat_head, shader_int8_ext, shader_float_type, shader_q4_0_defines, mulmat_body1, mulmat_load_q4_0, mulmat_body2)) |