diff options
Diffstat (limited to 'scripts')
| -rw-r--r-- | scripts/LlamaConfig.cmake.in | 4 | ||||
| -rwxr-xr-x | scripts/compare-commits.sh | 2 | ||||
| -rw-r--r-- | scripts/pod-llama.sh | 16 | ||||
| -rw-r--r-- | scripts/server-llm.sh | 2 |
4 files changed, 12 insertions, 12 deletions
diff --git a/scripts/LlamaConfig.cmake.in b/scripts/LlamaConfig.cmake.in index 6a6d8e39..f842c713 100644 --- a/scripts/LlamaConfig.cmake.in +++ b/scripts/LlamaConfig.cmake.in @@ -3,7 +3,7 @@ set(LLAMA_BUILD_COMMIT @LLAMA_BUILD_COMMIT@) set(LLAMA_BUILD_NUMBER @LLAMA_BUILD_NUMBER@) set(LLAMA_SHARED_LIB @BUILD_SHARED_LIBS@) set(LLAMA_BLAS @LLAMA_BLAS@) -set(LLAMA_CUBLAS @LLAMA_CUBLAS@) +set(LLAMA_CUDA @LLAMA_CUDA@) set(LLAMA_METAL @LLAMA_METAL@) set(LLAMA_MPI @LLAMA_MPI@) set(LLAMA_CLBLAST @LLAMA_CLBLAST@) @@ -27,7 +27,7 @@ if (LLAMA_BLAS) find_package(BLAS REQUIRED) endif() -if (LLAMA_CUBLAS) +if (LLAMA_CUDA) find_package(CUDAToolkit REQUIRED) endif() diff --git a/scripts/compare-commits.sh b/scripts/compare-commits.sh index 331c4b9c..d1272506 100755 --- a/scripts/compare-commits.sh +++ b/scripts/compare-commits.sh @@ -23,7 +23,7 @@ fi make_opts="" if [[ "$backend" == "cuda" ]]; then - make_opts="LLAMA_CUBLAS=1" + make_opts="LLAMA_CUDA=1" fi git checkout $1 diff --git a/scripts/pod-llama.sh b/scripts/pod-llama.sh index 6cf1ab4f..2058ceab 100644 --- a/scripts/pod-llama.sh +++ b/scripts/pod-llama.sh @@ -42,7 +42,7 @@ git clone https://github.com/ggerganov/llama.cpp cd llama.cpp -LLAMA_CUBLAS=1 make -j +LLAMA_CUDA=1 make -j ln -sfn /workspace/TinyLlama-1.1B-Chat-v0.3 ./models/tinyllama-1b ln -sfn /workspace/CodeLlama-7b-hf ./models/codellama-7b @@ -60,7 +60,7 @@ cd /workspace/llama.cpp mkdir build-cublas cd build-cublas -cmake -DLLAMA_CUBLAS=1 ../ +cmake -DLLAMA_CUDA=1 ../ make -j if [ "$1" -eq "0" ]; then @@ -186,17 +186,17 @@ if [ "$1" -eq "1" ]; then # batched cd /workspace/llama.cpp - LLAMA_CUBLAS=1 make -j && ./batched ./models/tinyllama-1b/ggml-model-f16.gguf "Hello, my name is" 8 128 999 + LLAMA_CUDA=1 make -j && ./batched ./models/tinyllama-1b/ggml-model-f16.gguf "Hello, my name is" 8 128 999 # batched-bench cd /workspace/llama.cpp - LLAMA_CUBLAS=1 make -j && ./batched-bench ./models/tinyllama-1b/ggml-model-f16.gguf 4608 1 99 0 512 128 1,2,3,4,5,6,7,8,16,32 + LLAMA_CUDA=1 make -j && ./batched-bench ./models/tinyllama-1b/ggml-model-f16.gguf 4608 1 99 0 512 128 1,2,3,4,5,6,7,8,16,32 # parallel cd /workspace/llama.cpp - LLAMA_CUBLAS=1 make -j && ./parallel -m ./models/tinyllama-1b/ggml-model-f16.gguf -t 1 -ngl 100 -c 4096 -b 512 -s 1 -np 8 -ns 128 -n 100 -cb + LLAMA_CUDA=1 make -j && ./parallel -m ./models/tinyllama-1b/ggml-model-f16.gguf -t 1 -ngl 100 -c 4096 -b 512 -s 1 -np 8 -ns 128 -n 100 -cb fi @@ -204,10 +204,10 @@ fi #if [ "$1" -eq "7" ]; then # cd /workspace/llama.cpp # -# LLAMA_CUBLAS=1 make -j && ./speculative -m ./models/codellama-34b-instruct/ggml-model-f16.gguf -md ./models/codellama-7b-instruct/ggml-model-q4_0.gguf -p "# Dijkstra's shortest path algorithm in Python (4 spaces indentation) + complexity analysis:\n\n" -e -ngl 999 -ngld 999 -t 4 -n 512 -c 4096 -s 21 --draft 16 -np 1 --temp 0.0 +# LLAMA_CUDA=1 make -j && ./speculative -m ./models/codellama-34b-instruct/ggml-model-f16.gguf -md ./models/codellama-7b-instruct/ggml-model-q4_0.gguf -p "# Dijkstra's shortest path algorithm in Python (4 spaces indentation) + complexity analysis:\n\n" -e -ngl 999 -ngld 999 -t 4 -n 512 -c 4096 -s 21 --draft 16 -np 1 --temp 0.0 #fi # more benches -#LLAMA_CUBLAS=1 make -j && ./batched-bench ./models/codellama-7b/ggml-model-q4_k.gguf 4096 1 99 1 512,3200 128,128,800 1 -#LLAMA_CUBLAS=1 make -j && ./batched-bench ./models/codellama-13b/ggml-model-q4_k.gguf 4096 1 99 1 512,3200 128,128,800 1 +#LLAMA_CUDA=1 make -j && ./batched-bench ./models/codellama-7b/ggml-model-q4_k.gguf 4096 1 99 1 512,3200 128,128,800 1 +#LLAMA_CUDA=1 make -j && ./batched-bench ./models/codellama-13b/ggml-model-q4_k.gguf 4096 1 99 1 512,3200 128,128,800 1 diff --git a/scripts/server-llm.sh b/scripts/server-llm.sh index 30bbac32..eb6ce458 100644 --- a/scripts/server-llm.sh +++ b/scripts/server-llm.sh @@ -380,7 +380,7 @@ fi if [[ "$backend" == "cuda" ]]; then printf "[+] Building with CUDA backend\n" - LLAMA_CUBLAS=1 make -j server $log + LLAMA_CUDA=1 make -j server $log elif [[ "$backend" == "cpu" ]]; then printf "[+] Building with CPU backend\n" make -j server $log |