vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

tests : more

Browse source
This commit is contained in:
Georgi Gerganov 2024-01-29 18:22:28 +02:00
parent abeaf0d90e
commit c6c1132e5e
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
4 changed files with 22 additions and 24 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml-metal.metal
View file

@ -2326,7 +2326,10 @@ kernel void kernel_flash_attn_ext_f16(
template [[host_name("kernel_flash_attn_ext_f16_h64" )]] kernel flash_attn_ext_f16_t kernel_flash_attn_ext_f16<64, 8, 32>;
template [[host_name("kernel_flash_attn_ext_f16_h80" )]] kernel flash_attn_ext_f16_t kernel_flash_attn_ext_f16<80, 8, 32>;
template [[host_name("kernel_flash_attn_ext_f16_h96" )]] kernel flash_attn_ext_f16_t kernel_flash_attn_ext_f16<96, 8, 32>;
template [[host_name("kernel_flash_attn_ext_f16_h112")]] kernel flash_attn_ext_f16_t kernel_flash_attn_ext_f16<112, 8, 32>;
template [[host_name("kernel_flash_attn_ext_f16_h128")]] kernel flash_attn_ext_f16_t kernel_flash_attn_ext_f16<128, 8, 32>;
template [[host_name("kernel_flash_attn_ext_f16_h256")]] kernel flash_attn_ext_f16_t kernel_flash_attn_ext_f16<256, 8, 32>;
kernel void kernel_cpy_f16_f16(
device const half * src0,