Skip to content

Release CoD v2 - #623

Merged
yanxi-chen merged 1 commit into
agentscope-ai:research/cod_v2from
shiweijiezero:release/cod-v2
Sep 20, 2026
Merged

yanxi-chen merged 1 commit into
agentscope-ai:research/cod_v2from
shiweijiezero:release/cod-v2

Conversation

@shiweijiezero

@shiweijiezero shiweijiezero commented Sep 18, 2026

Copy link
Copy Markdown
Contributor

Summary

Connect the Dots (CoD) trains long-lifecycle agents to learn from experience across related tasks. Agents alternate between solving tasks and updating their context, carrying useful knowledge into subsequent tasks.

CoD v2 supports:

  • End-to-end reinforcement learning with fine-grained credit assignment and mixed-domain on-policy distillation.
  • FrozenLake, Alchemy, Terminal, Learn2Ask, Optimal Control, PDE Discovery, and Grid Navigation environments.
  • Data generation, training, and checkpoint evaluation through the provided example configurations.

See the English guide or 中文指南 for setup, commands, and experimental results.

Add CoD workflows, credit assignment, environments, and mixed-domain on-policy distillation. Include release training and evaluation configs, data generators, and bilingual documentation.
@yanxi-chen
yanxi-chen merged commit 139d1c4 into agentscope-ai:research/cod_v2 Sep 20, 2026
1 check failed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants