ML Ops is an ML engineering culture and practice that aims ate unifying ML development(Dev) and ML operation(Ops)
Automation and Monitoring at all steps of ML system construction, including:
- integration(data...)
- testing
- releasing
- deployment
- infrastructure management(i.e. GPU, CPU...)
- Data Ingestion
- Data Validation
- Data Transformation - LLM can understand - *
- Model - *
- Model Analysis
- Serving *
- Logging
With * are belong to Job Ochestration, otherwise Job Management, Monitoring
It's one dementional, one model per use case(i.e. summarization)
Focus on the LLM development and managing the model in production.
Note about LLMOps:
- experiment on multiple foundation models for your use case
- think promopt design and managment for experiment and prod
- pre-precessing includes chunking ...
- grounding includes put the chunked user input into a prompt to feed the model and compare the LLM resp to the fact.
