YellowRoseCx
diff --git a/‎.devops/full-cuda.Dockerfile
+33 b/‎.devops/full-cuda.Dockerfile
+33
diff --git a/‎.devops/lamma-cpp-clblast.srpm.spec
+58 b/‎.devops/lamma-cpp-clblast.srpm.spec
+58
diff --git a/‎.devops/lamma-cpp-cublas.srpm.spec
+59 b/‎.devops/lamma-cpp-cublas.srpm.spec
+59
diff --git a/‎.devops/llama-cpp.srpm.spec
+58 b/‎.devops/llama-cpp.srpm.spec
+58
diff --git a/‎.devops/main-cuda.Dockerfile
+32 b/‎.devops/main-cuda.Dockerfile
+32
diff --git a/‎.github/workflows/c-cpp.yml
+17 b/‎.github/workflows/c-cpp.yml
+17
diff --git a/‎.gitignore
+40-24 b/‎.gitignore
+40-24
@@ -0,0 +1,33 @@
+ARG UBUNTU_VERSION=22.04
+
+# This needs to generally match the container host's environment.
+ARG CUDA_VERSION=11.7.1
+
+# Target the CUDA build image
+ARG BASE_CUDA_DEV_CONTAINER=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION}
+
+FROM ${BASE_CUDA_DEV_CONTAINER} as build
+
+# Unless otherwise specified, we make a fat build.
+ARG CUDA_DOCKER_ARCH=all
+
+RUN apt-get update && \
+    apt-get install -y build-essential python3 python3-pip
+
+COPY requirements.txt requirements.txt
+
+RUN pip install --upgrade pip setuptools wheel \
+    && pip install -r requirements.txt
+
+WORKDIR /app
+
+COPY . .
+
+# Set nvcc architecture
+ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
+# Enable cuBLAS
+ENV LLAMA_CUBLAS=1
+
+RUN make
+
+ENTRYPOINT ["/app/.devops/tools.sh"]
@@ -0,0 +1,58 @@
+# SRPM for building from source and packaging an RPM for RPM-based distros.
+# https://fedoraproject.org/wiki/How_to_create_an_RPM_package
+# Built and maintained by John Boero - [email protected]
+# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal
+
+# Notes for llama.cpp:
+# 1. Tags are currently based on hash - which will not sort asciibetically.
+#    We need to declare standard versioning if people want to sort latest releases.
+# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies.
+# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed.
+#    Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo
+# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries.
+#    It is up to the user to install the correct vendor-specific support.
+
+Name:           llama.cpp-clblast
+Version:        master
+Release:        1%{?dist}
+Summary:        OpenCL Inference of LLaMA model in pure C/C++
+License:        MIT
+Source0:        https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz
+BuildRequires:  coreutils make gcc-c++ git mesa-libOpenCL-devel
+URL:            https://github.com/ggerganov/llama.cpp
+
+%define debug_package %{nil}
+%define source_date_epoch_from_changelog 0
+
+%description
+CPU inference for Meta's Lllama2 models using default options.
+
+%prep
+%setup -n llama.cpp-master
+
+%build
+make -j LLAMA_CLBLAST=1
+
+%install
+mkdir -p %{buildroot}%{_bindir}/
+cp -p main %{buildroot}%{_bindir}/llamacppclblast
+cp -p server %{buildroot}%{_bindir}/llamacppclblastserver
+cp -p simple %{buildroot}%{_bindir}/llamacppclblastsimple
+
+%clean
+rm -rf %{buildroot}
+rm -rf %{_builddir}/*
+
+%files
+%{_bindir}/llamacppclblast
+%{_bindir}/llamacppclblastserver
+%{_bindir}/llamacppclblastsimple
+
+%pre
+
+%post
+
+%preun
+%postun
+
+%changelog
@@ -0,0 +1,59 @@
+# SRPM for building from source and packaging an RPM for RPM-based distros.
+# https://fedoraproject.org/wiki/How_to_create_an_RPM_package
+# Built and maintained by John Boero - [email protected]
+# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal
+
+# Notes for llama.cpp:
+# 1. Tags are currently based on hash - which will not sort asciibetically.
+#    We need to declare standard versioning if people want to sort latest releases.
+# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies.
+# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed.
+#    Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo
+# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries.
+#    It is up to the user to install the correct vendor-specific support.
+
+Name:           llama.cpp-cublas
+Version:        master
+Release:        1%{?dist}
+Summary:        CPU Inference of LLaMA model in pure C/C++ (no CUDA/OpenCL)
+License:        MIT
+Source0:        https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz
+BuildRequires:  coreutils make gcc-c++ git cuda-toolkit
+Requires:       cuda-toolkit
+URL:            https://github.com/ggerganov/llama.cpp
+
+%define debug_package %{nil}
+%define source_date_epoch_from_changelog 0
+
+%description
+CPU inference for Meta's Lllama2 models using default options.
+
+%prep
+%setup -n llama.cpp-master
+
+%build
+make -j LLAMA_CUBLAS=1
+
+%install
+mkdir -p %{buildroot}%{_bindir}/
+cp -p main %{buildroot}%{_bindir}/llamacppcublas
+cp -p server %{buildroot}%{_bindir}/llamacppcublasserver
+cp -p simple %{buildroot}%{_bindir}/llamacppcublassimple
+
+%clean
+rm -rf %{buildroot}
+rm -rf %{_builddir}/*
+
+%files
+%{_bindir}/llamacppcublas
+%{_bindir}/llamacppcublasserver
+%{_bindir}/llamacppcublassimple
+
+%pre
+
+%post
+
+%preun
+%postun
+
+%changelog
@@ -0,0 +1,58 @@
+# SRPM for building from source and packaging an RPM for RPM-based distros.
+# https://fedoraproject.org/wiki/How_to_create_an_RPM_package
+# Built and maintained by John Boero - [email protected]
+# In honor of Seth Vidal https://www.redhat.com/it/blog/thank-you-seth-vidal
+
+# Notes for llama.cpp:
+# 1. Tags are currently based on hash - which will not sort asciibetically.
+#    We need to declare standard versioning if people want to sort latest releases.
+# 2. Builds for CUDA/OpenCL support are separate, with different depenedencies.
+# 3. NVidia's developer repo must be enabled with nvcc, cublas, clblas, etc installed.
+#    Example: https://developer.download.nvidia.com/compute/cuda/repos/fedora37/x86_64/cuda-fedora37.repo
+# 4. OpenCL/CLBLAST support simply requires the ICD loader and basic opencl libraries.
+#    It is up to the user to install the correct vendor-specific support.
+
+Name:           llama.cpp
+Version:        master
+Release:        1%{?dist}
+Summary:        CPU Inference of LLaMA model in pure C/C++ (no CUDA/OpenCL)
+License:        MIT
+Source0:        https://github.com/ggerganov/llama.cpp/archive/refs/heads/master.tar.gz
+BuildRequires:  coreutils make gcc-c++ git
+URL:            https://github.com/ggerganov/llama.cpp
+
+%define debug_package %{nil}
+%define source_date_epoch_from_changelog 0
+
+%description
+CPU inference for Meta's Lllama2 models using default options.
+
+%prep
+%autosetup
+
+%build
+make -j
+
+%install
+mkdir -p %{buildroot}%{_bindir}/
+cp -p main %{buildroot}%{_bindir}/llamacpp
+cp -p server %{buildroot}%{_bindir}/llamacppserver
+cp -p simple %{buildroot}%{_bindir}/llamacppsimple
+
+%clean
+rm -rf %{buildroot}
+rm -rf %{_builddir}/*
+
+%files
+%{_bindir}/llamacpp
+%{_bindir}/llamacppserver
+%{_bindir}/llamacppsimple
+
+%pre
+
+%post
+
+%preun
+%postun
+
+%changelog
@@ -0,0 +1,32 @@
+ARG UBUNTU_VERSION=22.04
+# This needs to generally match the container host's environment.
+ARG CUDA_VERSION=11.7.1
+# Target the CUDA build image
+ARG BASE_CUDA_DEV_CONTAINER=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VERSION}
+# Target the CUDA runtime image
+ARG BASE_CUDA_RUN_CONTAINER=nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu${UBUNTU_VERSION}
+
+FROM ${BASE_CUDA_DEV_CONTAINER} as build
+
+# Unless otherwise specified, we make a fat build.
+ARG CUDA_DOCKER_ARCH=all
+
+RUN apt-get update && \
+    apt-get install -y build-essential
+
+WORKDIR /app
+
+COPY . .
+
+# Set nvcc architecture
+ENV CUDA_DOCKER_ARCH=${CUDA_DOCKER_ARCH}
+# Enable cuBLAS
+ENV LLAMA_CUBLAS=1
+
+RUN make
+
+FROM ${BASE_CUDA_RUN_CONTAINER} as runtime
+
+COPY --from=build /app/main /main
+
+ENTRYPOINT [ "/main" ]
@@ -0,0 +1,17 @@
+name: C/C++ CI
+
+on:
+  push:
+    branches: [ "main" ]
+  pull_request:
+    branches: [ "main" ]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v3
+    - name: make
+      run: make
@@ -1,6 +1,6 @@
 *.o
 *.a
-*.so
+*.bin
 .DS_Store
 .build/
 .cache/
@@ -12,18 +12,12 @@
 .vs/
 .vscode/
 
-build/
-build-em/
-build-debug/
-build-release/
-build-static/
-build-cublas/
-build-opencl/
-build-metal/
-build-no-accel/
-build-sanitize-addr/
-build-sanitize-thread/
+build*/
 out/
+tmp/
+
+models/*
+models-mnt
 
 /main
 /quantize
@@ -32,14 +26,18 @@ out/
 /perplexity
 /embedding
 /train-text-from-scratch
+/convert-llama2c-to-ggml
 /simple
 /benchmark-matmult
 /vdot
 /server
 /Pipfile
 /embd-input-test
+/gguf
+/gguf-llama-simple
 /libllama.so
-
+/llama-bench
+build-info.h
 arm_neon.h
 compile_commands.json
 CMakeSettings.json
@@ -57,14 +55,32 @@ qnt-*.txt
 perf-*.txt
 
 examples/jeopardy/results.txt
-koboldcpp.so
-koboldcpp_failsafe.so
-koboldcpp_openblas.so
-koboldcpp_openblas_noavx2.so
-koboldcpp_clblast.so
-koboldcpp.dll
-koboldcpp_failsafe.dll
-koboldcpp_openblas.dll
-koboldcpp_openblas_noavx2.dll
-koboldcpp_clblast.dll
-koboldcpp_cublas.dll
+
+pyproject.toml
+poetry.lock
+poetry.toml
+
+# Test binaries
+tests/test-grammar-parser
+tests/test-double-float
+tests/test-grad0
+tests/test-opt
+tests/test-quantize-fns
+tests/test-quantize-perf
+tests/test-sampling
+tests/test-tokenizer-0
+
+/koboldcpp_default.so
+/koboldcpp_failsafe.so
+/koboldcpp_openblas.so
+/koboldcpp_noavx2.so
+/koboldcpp_clblast.so
+/koboldcpp_cublas.so
+/koboldcpp_default.dll
+/koboldcpp_failsafe.dll
+/koboldcpp_openblas.dll
+/koboldcpp_noavx2.dll
+/koboldcpp_clblast.dll
+/koboldcpp_cublas.dll
+/cublas64_11.dll
+/cublasLt64_11.dll