Train multi-node tasks including cycle multi-hop rag and trainable agent#306

Merged

Sylph-AI merged 60 commits intomainfrom li

Jan 29, 2025

+12,100-5,654

Commits on Dec 17, 2024

build cycle multi-hop rag, trainable, added component graph with cycle, output graph as simplified version of the dynamic computation graph
liyin2015
committed

Commits on Dec 18, 2024

add scripts to summarize multiple runs
liyin2015
committed

Commits on Dec 19, 2024

added test to react agent
liyin2015
committed

Commits on Dec 20, 2024

Commits on Dec 21, 2024

Commits on Dec 22, 2024

improve the gradient engine prompt to deal with multi-node and peer nodes
liyin2015
committed

Commits on Dec 23, 2024

Commits on Dec 24, 2024

Commits on Dec 25, 2024

add context in the react agent
liyin2015
committed

Commits on Dec 28, 2024

Commits on Dec 29, 2024

Commits on Jan 2, 2025

use random seed to control the order of training samples, add backward pass setup for the backward engine via the trainer.fit function
liyin2015
committed

Commits on Jan 3, 2025

fix the bug in hotpot qa
liyin2015
committed

Commits on Jan 5, 2025

a high performning meta prompt for the tgd optimizer adding prompt engineering strategy
liyin2015
committed

Commits on Jan 6, 2025

fix the hotpotqa dataaset, add the retriever recall and precision, train multi-hop retriever
liyin2015
committed

Commits on Jan 8, 2025

Commits on Jan 9, 2025

tested on using query to supervise retriever, but does not work well
liyin2015
committed

Commits on Jan 10, 2025

Commits on Jan 16, 2025

Commits on Jan 18, 2025

Commits on Jan 22, 2025

Commits on Jan 24, 2025

Commits on Jan 25, 2025

fix all tests
liyin2015
committed

Commits on Jan 27, 2025

last commit
liyin2015
committed

Commits on Jan 28, 2025

Commits on Jan 29, 2025