Chapter 3
Is the biggest model always the safest choice?
Model lớn nhất có phải lúc nào cũng an toàn nhất?
The chapter's question
Which of my tasks use a bigger model than needed?
Big idea
Give simple tasks to a small model and the results are no worse while the price is many times lower.
In this chapter
Part 3.1
Listing Each Task
Which model-calling steps make up my feature?
Part 3.2
Trying a Smaller Model
How many questions in the test set does a smaller model pass on this step?
Part 3.3
Routing by Difficulty
Is there a way to detect hard questions and route them to a larger model?
Part 3.4
The Cost of Latency
How many extra seconds will users wait before giving up?
The reading for this chapter is inside Conan School
This chapter has 4 readings, with case stories and exercises. Chapter 1 of The Bill for One Answer is free to read right here.
Read in Conan School