docs/get-started/install/apple-silicon-m1-m2.md
If you own a macOS device with an Apple Silicon M1/M2 chip, you can run Jina-serve natively on it (instead of running under Rosetta) and enjoy up to 10x faster performance. This chapter summarizes how to install Jina-serve.
To ensure you are using the right terminal, run:
uname -m
It should return:
arm64
brew is a package manager for macOS. If you have already installed it, you need to confirm it is actually installed for Apple Silicon not for Rosetta. To check that, run:
which brew
/opt/homebrew/bin/brew
If it's installed under /usr/local/ instead of /opt/homebrew/, it means your brew is installed for Rosetta not for Apple Silicon. You need to reinstall it.
Reinstalling `brew` can be a destructive operation. Ensure you have backed up your data before proceeding.
To (re)install brew, run:
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
You can observe the output to check if it contains /opt/homebrew to ensure you are installing for Apple Silicon.
Python also has to be installed for Apple Silicon. It is possible it is installed for Rosetta, and you are not aware of that. To confirm, run:
import platform
platform.machine()
This should output:
'arm64'
If not, then you are using Python under Rosetta, and you need to install Python for Apple Silicon with brew:
brew install python3
As of August 2022, this will install Python 3.10 natively for Apple Silicon.
Ensure you note down where python and pip are installed. In this example, they are installed to /opt/homebrew/bin/python3 and /opt/homebrew/opt/[email protected]/libexec/bin/pip respectively.
There are some core dependencies that Jina-serve needs to run, whose wheels are not available on PyPI but fortunately are available as wheels. To install them, run:
brew install protobuf numpy
Now we can install Jina-serve via pip. Ensure you use the correct pip:
/opt/homebrew/opt/[email protected]/libexec/bin/pip install jina
grpcio requires building the wheels, it will take some time.
Note: If the previous step fails, adding the environment variables below might solve the problem:
export GRPC_PYTHON_BUILD_SYSTEM_OPENSSL=1
export GRPC_PYTHON_BUILD_SYSTEM_ZLIB=1
After all the dependencies are installed, you can run Jina-serve CLI and check the system information.
jina -vf
---
emphasize-lines: 13-15
---
- jina 3.7.14
- docarray 0.15.4
- jcloud 0.0.35
- jina-hubble-sdk 0.16.1
- jina-proto 0.1.13
- protobuf 3.20.1
- proto-backend python
- grpcio 1.47.0
- pyyaml 6.0
- python 3.10.6
- platform Darwin
- platform-release 21.6.0
- platform-version Darwin Kernel Version 21.6.0: Sat Jun 18 17:07:28 PDT 2022; root:xnu-8020.140.41~1/RELEASE_ARM64_T8110
- architecture arm64
- processor arm
- uid 94731629138370
- session-id 49497356-254e-11ed-9624-56286d1a91c2
- uptime 2022-08-26T16:49:28.279723
- ci-vendor (unset)
* JINA_DEFAULT_HOST (unset)
* JINA_DEFAULT_TIMEOUT_CTRL (unset)
* JINA_DEPLOYMENT_NAME (unset)
* JINA_DISABLE_UVLOOP (unset)
* JINA_EARLY_STOP (unset)
* JINA_FULL_CLI (unset)
* JINA_GATEWAY_IMAGE (unset)
* JINA_GRPC_RECV_BYTES (unset)
* JINA_GRPC_SEND_BYTES (unset)
* JINA_HUB_NO_IMAGE_REBUILD (unset)
* JINA_LOG_CONFIG (unset)
* JINA_LOG_LEVEL (unset)
* JINA_LOG_NO_COLOR (unset)
* JINA_MP_START_METHOD (unset)
* JINA_OPTOUT_TELEMETRY (unset)
* JINA_RANDOM_PORT_MAX (unset)
* JINA_RANDOM_PORT_MIN (unset)
Congratulations! You have successfully installed Jina-serve on Apple Silicon.
To install MPS-enabled PyTorch, run:
```bash
/opt/homebrew/opt/[email protected]/libexec/bin/pip install -U --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu
```