Debug di Modelli ML: Il potere del Logging - Firenze.dev

Transcript

1. firenze.dev
2. What’s the color of the cat? MACHINE LEARNING
3. What’s the color of the cat? MACHINE LEARNING hyperparameters = { batch_size = 128, learning_rate = 0.1, optimizer = “Adam”, num_epochs = 12}
4. What’s the color of the cat? MACHINE LEARNING Hyperparameter Tuning hyperparameters = { batch_size = 128, learning_rate = 0.1, optimizer = “Adam”, num_epochs = 12}
7. WHAT IS WANDB? A platform for ML practitioners that can help simplify the development, monitoring and debugging of Machine Learning Models.
8. WHAT IS WANDB? A platform for developers and researcher that can help simplify the development, monitor and debugging of your experiments
9. WHAT IS WANDB? These tools are open source and you can self-host WANDB on your machine A platform for developers and researcher that can help simplify the development, monitor and debugging of your experiments
10. WHAT IS WANDB? These tools are open source and you can self-host WANDB on your machine A platform for developers and researcher that can help simplify the development, monitor and debugging of your experiments They offer Python and Javascript Libraries to interact with the service from your codebase
11. “Yeah, I know this platform but I’ve already started my project and I cannot waste time implementing this logging mechanism” Cit. Just a random person to whom I suggest using Wandb
12. HOW WANDB CAN BE USEFUL IN MY WORKFLOW?
13. HOW WANDB CAN BE USEFUL IN MY WORKFLOW? Track your experiments: log all the metrics you need to analyse and keep track of
14. Track your experiments: log all the metrics you need to analyse and keep track of Visualisation of the results: It automatically creates plots to visualise the results of your experiments HOW WANDB CAN BE USEFUL IN MY WORKFLOW?
17. Track your experiments: log all the metrics you need to analyse and keep track of Visualisation of the results: It automatically creates plots to visualise the results of your experiments Tune your hyperparameters: the best hyper- parameters for your model are just a yaml file away HOW WANDB CAN BE USEFUL IN MY WORKFLOW?
19. Track your experiments: log all the metrics you need to analyse and keep track of. Visualisation of the results: It automatically creates plots to visualise the results of your experiments. Tune your hyperparameters: the best hyper- parameters for your model are just a yaml file away. Results Reporting: easily create dashboard to report your results. HOW WANDB CAN BE USEFUL IN MY WORKFLOW?
21. OTHER INTERESTING FEATURES Wandb Launch: automation of ML Workflow. Launch your experiment from WANDB on your machine LLMs Monitoring: monitor how users interact with your LLMs LLM Debugging: verify how your LLM changes during the training and identify errors Weave: Tracking and evaluation of LLM Applications (That I haven’t used yet)
22. Thanks 🙏
23. Questions? Luca Corbucci https://lucacorbucci.me/ Repo Slides