Testing scenarios
This document goes over various testing scenarios involved in making the ai3-hackathon GitHub repository on MSI laptop.
- Hardware check
- Compare MSI server specs with MSI laptop.
- MSI laptop: 4070 with 8GB memory (4GB physical + 4GB virtual) and Intel-i7 13th Gen
- MSI server: 4070 with 16 GB memory and Intel-i9 13th Gen
- Required files
input_data.zipextracts to/datafolder which has lookup DB and model files to load the model for inference.PSIAP_AI3.zipextracts to/PSIAP_AI3folder which has files for the publisher, required to test local pub/sub.
- Model loading:
- Faced issues loading the model due to Hugging face credentials issue which made it look online for the models even though local files are present.
- Fixed it by saving Hugging Face credentials in environment variables, as follows:
- Open a terminal window.
- Run the command:
python - Run the command:
import huggingface_hub as hf - Run the command:
hf.login() - Paste the key in when prompted.
- The model takes about 5-8 seconds on MSI server to load the shards.
- Once the shards are downloaded from the internet, it takes about 8-15 seconds to load on MSI laptop in offline mode.
- Model Inference speed
- On MSI server, the model takes about 2-5 seconds for inference.
- On MSI laptop, the model takes about 2-10 seconds for inference.
- Internet access
- Need to verify the app functionality with and without internet access.
- The model loading and inference speeds in these cases are documented in the above sections.