Distribute and run LLMs with a single file.
C++ cross-platformggufllama-cpplocal-ailocal-inferencelocal-llmopen-source-aisingle-file-executablespeech-to-text
Release llamafile v0.9.1
CJ Paisdd75ecebf386
diff --git a/llamafile/rocm.sh b/llamafile/rocm.shindex 5df26ed7..b0b1dfb7 100755--- a/llamafile/rocm.sh+++ b/llamafile/rocm.sh@@ -5,6 +5,27 @@# NVCUDA DLLs are provided by the installation of the windows GPU# driver on a Windows system that has a CUDA-capable GPU installed.+TMP=$(mktemp -d) || exit++cp llama.cpp/ggml-cuda.cu \+ llama.cpp/ggml-cuda.h \+ llama.cpp/ggml-impl.h \+ llama.cpp/ggml-alloc.h \+ llama.cpp/ggml-common.h \+ llama.cpp/ggml-backend.h \+ llama.cpp/ggml-backend-impl.h \+ llama.cpp/ggml.h \+ llamafile/tinyblas.h \+ llamafile/tinyblas.cu \+ llamafile/llamafile.h \+ llamafile/rocm.bat \+ llamafile/rocm.sh \+ llamafile/cuda.bat \+ llamafile/cuda.sh \+ "$TMP" || exit++cd "$TMP"+hipcc \-O2 \-fPIC \@@ -23,5 +44,5 @@ hipcc \-DGGML_MINIMIZE_CODE_SIZE=1 \-DGGML_CUDA_PEER_MAX_BATCH_SIZE=128 \--amdgpu-target=gfx1100,gfx1031,gfx1030,gfx1032,gfx906,gfx1101,gfx1102,gfx1103 \- -o ggml-rocm.so \+ -o ~/ggml-rocm.so \ggml-cuda.cudiff --git a/llamafile/version.h b/llamafile/version.hindex 3447c5b1..29d9083b 100644--- a/llamafile/version.h+++ b/llamafile/version.h@@ -2,7 +2,7 @@#define LLAMAFILE_MAJOR 0#define LLAMAFILE_MINOR 9-#define LLAMAFILE_PATCH 0+#define LLAMAFILE_PATCH 1#define LLAMAFILE_VERSION \(100000000 * LLAMAFILE_MAJOR + 1000000 * LLAMAFILE_MINOR + LLAMAFILE_PATCH)