Aditya Kanekar
Posted on July 11, 2022
Background
I was working on developing a API based on SciSpacy to search for medical terms. Looking at the slow performance of loading the model and entity linker on the Intel i5 chip on my Lenovo laptop, I was curious to check how the Apple M1 chip would perform since I own a MacBook Air with M1 chip with following specifications,
- RAM - 16GB
- Storage - 512GB SSD
- Apple M1 chip
- 8-core CPU with 4 performance cores and 4 efficiency cores
- 8-core GPU
- 16-core Neural Engine
The Process
Following are the dependencies from requirement.txt file for SciSpacy and the en_core_sci_lg model,
scispacy==0.5.0
https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.5.0/en_core_sci_lg-0.5.0.tar.gz
The ** pip install -r requirement.txt ** failed with lot of errors while installing a dependency ** nmslib ** with an error ** clang compiler does not support '-march=native' **. Since I already had the Xcode installed on my Mac the version of C was 11 but somehow the mentioned command was failing, since the said argument was not supported on the C version. With a lot of digging around I finally found a solution mentioned on SciSpacy GitHub page.
Install the nmslib using following command,
CFLAGS="-mavx -DWARN(a)=(a)" pip install nmslib
After running this command all the packages got installed as expected and I was able to load and run the queries on SciSpacy model.
Next I will be coming up with the performance comparison on Apple M1 vs Intel i5 chip. I understand the comparison isn't fair but it just shows how much improved the M1 Chip is over the Intel processors.
Posted on July 11, 2022
Join Our Newsletter. No Spam, Only the good stuff.
Sign up to receive the latest update from our blog.