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.
Top comments (2)
Thanks for this!
Thank you :)