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

5
scripts/LlamaConfig.cmake.in
View file

@ -5,7 +5,6 @@ set(LLAMA_SHARED_LIB @BUILD_SHARED_LIBS@)
set(LLAMA_BLAS @LLAMA_BLAS@)
set(LLAMA_CUDA @LLAMA_CUDA@)
set(LLAMA_METAL @LLAMA_METAL@)
set(LLAMA_CLBLAST @LLAMA_CLBLAST@)
set(LLAMA_HIPBLAS @LLAMA_HIPBLAS@)
set(LLAMA_ACCELERATE @LLAMA_ACCELERATE@)
@ -36,10 +35,6 @@ if (LLAMA_METAL)
find_library(METALKIT_FRAMEWORK MetalKit REQUIRED)
endif()
if (LLAMA_CLBLAST)
find_package(CLBlast REQUIRED)
endif()
if (LLAMA_HIPBLAS)
find_package(hip REQUIRED)
find_package(hipblas REQUIRED)

6
scripts/compare-llama-bench.py
View file

@ -19,17 +19,17 @@ logger = logging.getLogger("compare-llama-bench")
# Properties by which to differentiate results per commit:
KEY_PROPERTIES = [
"cpu_info", "gpu_info", "n_gpu_layers", "cuda", "opencl", "vulkan", "kompute", "metal", "sycl", "rpc", "gpu_blas",
"cpu_info", "gpu_info", "n_gpu_layers", "cuda", "vulkan", "kompute", "metal", "sycl", "rpc", "gpu_blas",
"blas", "model_filename", "model_type", "model_size", "model_n_params", "n_batch", "n_ubatch", "embeddings", "n_threads",
"type_k", "type_v", "use_mmap", "no_kv_offload", "split_mode", "main_gpu", "tensor_split", "flash_attn", "n_prompt", "n_gen"
]
# Properties that are boolean and are converted to Yes/No for the table:
BOOL_PROPERTIES = ["cuda", "opencl", "vulkan", "kompute", "metal", "sycl", "gpu_blas", "blas", "embeddings", "use_mmap", "no_kv_offload", "flash_attn"]
BOOL_PROPERTIES = ["cuda", "vulkan", "kompute", "metal", "sycl", "gpu_blas", "blas", "embeddings", "use_mmap", "no_kv_offload", "flash_attn"]
# Header names for the table:
PRETTY_NAMES = {
"cuda": "CUDA", "opencl": "OpenCL", "vulkan": "Vulkan", "kompute": "Kompute", "metal": "Metal", "sycl": "SYCL", "rpc": "RPC",
"cuda": "CUDA", "vulkan": "Vulkan", "kompute": "Kompute", "metal": "Metal", "sycl": "SYCL", "rpc": "RPC",
"gpu_blas": "GPU BLAS", "blas": "BLAS", "cpu_info": "CPU", "gpu_info": "GPU", "model_filename": "File", "model_type": "Model",
"model_size": "Model Size [GiB]", "model_n_params": "Num. of Par.", "n_batch": "Batch size", "n_ubatch": "Microbatch size",
"n_threads": "Threads", "type_k": "K type", "type_v": "V type", "n_gpu_layers": "GPU layers", "split_mode": "Split mode",

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

4
scripts/sync-ggml-am.sh
View file

@ -106,8 +106,6 @@ if [ -f $SRC_LLAMA/ggml-src.patch ]; then
# src/ggml-kompute.h -> ggml-kompute.h
# src/ggml-metal.h -> ggml-metal.h
# src/ggml-metal.m -> ggml-metal.m
# src/ggml-opencl.cpp -> ggml-opencl.cpp
# src/ggml-opencl.h -> ggml-opencl.h
# src/ggml-quants.c -> ggml-quants.c
# src/ggml-quants.h -> ggml-quants.h
# src/ggml-rpc.cpp -> ggml-rpc.cpp
@ -143,8 +141,6 @@ if [ -f $SRC_LLAMA/ggml-src.patch ]; then
-e 's/src\/ggml-kompute\.h/ggml-kompute.h/g' \
-e 's/src\/ggml-metal\.h/ggml-metal.h/g' \
-e 's/src\/ggml-metal\.m/ggml-metal.m/g' \
-e 's/src\/ggml-opencl\.cpp/ggml-opencl.cpp/g' \
-e 's/src\/ggml-opencl\.h/ggml-opencl.h/g' \
-e 's/src\/ggml-quants\.c/ggml-quants.c/g' \
-e 's/src\/ggml-quants\.h/ggml-quants.h/g' \
-e 's/src\/ggml-rpc\.cpp/ggml-rpc.cpp/g' \

2
scripts/sync-ggml.sh
View file

@ -14,8 +14,6 @@ cp -rpv ../ggml/src/ggml-kompute.h ./ggml-kompute.h
cp -rpv ../ggml/src/ggml-metal.h ./ggml-metal.h
cp -rpv ../ggml/src/ggml-metal.m ./ggml-metal.m
cp -rpv ../ggml/src/ggml-metal.metal ./ggml-metal.metal
cp -rpv ../ggml/src/ggml-opencl.cpp ./ggml-opencl.cpp
cp -rpv ../ggml/src/ggml-opencl.h ./ggml-opencl.h
cp -rpv ../ggml/src/ggml-quants.c ./ggml-quants.c
cp -rpv ../ggml/src/ggml-quants.h ./ggml-quants.h
cp -rpv ../ggml/src/ggml-rpc.cpp ./ggml-rpc.cpp