[Multi-hop RAG opt + improvement] [V0.3.0 Cycle Graph]#223

Merged

Sylph-AI merged 26 commits intomainfrom li

Dec 16, 2024

+2,984-538

Commits on Sep 23, 2024

fix the api in the word count code and generate better diagnose report
liyin2015
committed

Commits on Sep 24, 2024

multi-hop retriever, design choice with cycle graph
liyin2015
committed
structure the component trace such as input_args, full_response, and api_kwargs for generator/retriever for debug in the parameter and to visualize in the graph in the parameter
liyin2015
committed
replace the retriever in vanilla rag with the multi-hop rag works well
liyin2015
committed

Commits on Oct 1, 2024

delete type_extension
liyin2015
committed

Commits on Oct 6, 2024

commit
liyin2015
committed

Commits on Oct 26, 2024

Commits on Oct 27, 2024

Commits on Oct 28, 2024

able to train multi-hop rag with demo, the backtrace of dataclass and score is correct at the demo parameter, need to check demo optimizer
liyin2015
committed

Commits on Dec 5, 2024

Commits on Dec 10, 2024

add rag in the tutorials code
liyin2015
committed

Commits on Dec 11, 2024

make the role of the loss component clear
liyin2015
committed

Commits on Dec 12, 2024

Commits on Dec 13, 2024

first end to end optimization on the task description using text grad on multi-hop rag, from 0.58 to 0.68, still 6% lower than few shot learning
liyin2015
committed

Commits on Dec 14, 2024

rebase on main
liyin2015
committed

Commits on Dec 15, 2024