starred/ollama-ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-04-17 15:53:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

Add experimental MLX backend and engine with imagegen support (#13648)

Browse Source 
* WIP - MLX backend with gemma3

* MLX: add cmake and go tag build toggles

To build the new MLX backend code:
  cmake --preset MLX
  cmake --build --preset MLX --parallel
  cmake --install build --component MLX
  go build -tags mlx .

Note: the main.go entrypoint for the MLX engine will change in a follow up commit.

* add experimental image generation runtime

* add experimental image generation runtime

* MLX: wire up cuda build for linux

* MLX: get dependencies correct and dedup

This is still too large for a unified github artifact, but is now "correct" for the mlx_cuda_v13
directory.

* fix relative link bug in dedup

* Add darwin build and readme

* add go build tag for mlx dependent code and wire up build_darwin.sh

* lint cleanup

* macos: build mlx for x86

This will be CPU only.

* cuda build instructions and fix drift from mlx bump

* stale comment

* Delete agent helper doc

* Clean up readme.md

* Revise README for tokenizer clarity and details

Updated README to clarify tokenizer functionality and removed correctness section.

---------

Co-authored-by: jmorganca <jmorganca@gmail.com>
This commit is contained in:
Daniel Hiltgen
2026-01-08 16:18:59 -08:00
committed by GitHub
parent 34d0c55ea5
commit 33ee7168ba
123 changed files with 27492 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
Dockerfile
View File

@@ -131,7 +131,39 @@ COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
RUN --mount=type=cache,target=/root/.ccache \
cmake --preset 'Vulkan' \
&& cmake --build --parallel --preset 'Vulkan' \
&& cmake --install build --component Vulkan --strip --parallel 8
&& cmake --install build --component Vulkan --strip --parallel 8
FROM base AS mlx
ARG CUDA13VERSION=13.0
RUN dnf install -y cuda-toolkit-${CUDA13VERSION//./-} \
&& dnf install -y openblas-devel lapack-devel \
&& dnf install -y libcudnn9-cuda-13 libcudnn9-devel-cuda-13 \
&& dnf install -y libnccl libnccl-devel
ENV PATH=/usr/local/cuda-13/bin:$PATH
ENV BLAS_INCLUDE_DIRS=/usr/include/openblas
ENV LAPACK_INCLUDE_DIRS=/usr/include/openblas
ENV CGO_LDFLAGS="-L/usr/local/cuda-13/lib64 -L/usr/local/cuda-13/targets/x86_64-linux/lib/stubs"
ARG PARALLEL
WORKDIR /go/src/github.com/ollama/ollama
COPY CMakeLists.txt CMakePresets.json .
COPY ml/backend/ggml/ggml ml/backend/ggml/ggml
COPY x/ml/backend/mlx x/ml/backend/mlx
COPY go.mod go.sum .
RUN curl -fsSL https://golang.org/dl/go$(awk '/^go/ { print $2 }' go.mod).linux-$(case $(uname -m) in x86_64) echo amd64 ;; aarch64) echo arm64 ;; esac).tar.gz | tar xz -C /usr/local
ENV PATH=/usr/local/go/bin:$PATH
RUN go mod download
RUN --mount=type=cache,target=/root/.ccache \
cmake --preset 'MLX CUDA 13' -DBLAS_INCLUDE_DIRS=/usr/include/openblas -DLAPACK_INCLUDE_DIRS=/usr/include/openblas \
&& cmake --build --parallel ${PARALLEL} --preset 'MLX CUDA 13' \
&& cmake --install build --component MLX --strip --parallel ${PARALLEL}
COPY . .
ARG GOFLAGS="'-ldflags=-w -s'"
ENV CGO_ENABLED=1
ARG CGO_CFLAGS
ARG CGO_CXXFLAGS
# TODO wire up the actual MLX engine here instead of building the main binary...
RUN mkdir -p dist/bin
RUN go build -tags mlx -trimpath -buildmode=pie -o dist/bin/imagegen ./x/imagegen/cmd/engine
FROM base AS build
@@ -153,6 +185,8 @@ FROM --platform=linux/amd64 scratch AS amd64
COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
COPY --from=cuda-13 dist/lib/ollama /lib/ollama/
COPY --from=vulkan dist/lib/ollama /lib/ollama/
COPY --from=mlx /go/src/github.com/ollama/ollama/dist/lib/ollama /lib/ollama/
COPY --from=mlx /go/src/github.com/ollama/ollama/dist/bin/ /bin/
FROM --platform=linux/arm64 scratch AS arm64
# COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/