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

ggml : remove OpenCL (#7735)

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-06-04 21:23:20 +03:00 • committed by GitHub
parent 0cd6bd3483
commit 554c247caf
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
21 changed files with 29 additions and 2639 deletions
Download patch file Download diff file Expand all files Collapse all files

11
scripts/server-llm.sh
View file

@ -3,7 +3,7 @@
# Helper script for deploying llama.cpp server with a single Bash command
#
# - Works on Linux and macOS
# - Supports: CPU, CUDA, Metal, OpenCL
# - Supports: CPU, CUDA, Metal
# - Can run all GGUF models from HuggingFace
# - Can serve requests in parallel
# - Always builds latest llama.cpp from GitHub
@ -19,7 +19,7 @@
# --port: port number, default is 8888
# --repo: path to a repo containing GGUF model files
# --wtype: weights type (f16, q8_0, q4_0, q4_1), default is user-input
# --backend: cpu, cuda, metal, opencl, depends on the OS
# --backend: cpu, cuda, metal, depends on the OS
# --gpu-id: gpu id, default is 0
# --n-parallel: number of parallel requests, default is 8
# --n-kv: KV cache size, default is 4096
@ -72,7 +72,7 @@ function print_usage {
printf " --port: port number, default is 8888\n"
printf " --repo: path to a repo containing GGUF model files\n"
printf " --wtype: weights type (f16, q8_0, q4_0, q4_1), default is user-input\n"
printf " --backend: cpu, cuda, metal, opencl, depends on the OS\n"
printf " --backend: cpu, cuda, metal, depends on the OS\n"
printf " --gpu-id: gpu id, default is 0\n"
printf " --n-parallel: number of parallel requests, default is 8\n"
printf " --n-kv: KV cache size, default is 4096\n"
@ -387,9 +387,6 @@ elif [[ "$backend" == "cpu" ]]; then
elif [[ "$backend" == "metal" ]]; then
printf "[+] Building with Metal backend\n"
make -j server $log
elif [[ "$backend" == "opencl" ]]; then
printf "[+] Building with OpenCL backend\n"
LLAMA_CLBLAST=1 make -j server $log
else
printf "[-] Unknown backend: %s\n" "$backend"
exit 1
@ -407,8 +404,6 @@ elif [[ "$backend" == "cpu" ]]; then
args="-ngl 0"
elif [[ "$backend" == "metal" ]]; then
args="-ngl 999"
elif [[ "$backend" == "opencl" ]]; then
args="-ngl 999"
else
printf "[-] Unknown backend: %s\n" "$backend"
exit 1