Skip to content

Comments

Migrate to verl v0.5.0#193

Merged
jeffreysijuntan merged 18 commits intov0.2from
v0.2-verl-stable
Aug 21, 2025
Merged

Migrate to verl v0.5.0#193
jeffreysijuntan merged 18 commits intov0.2from
v0.2-verl-stable

Conversation

@kylemontgomery1
Copy link
Collaborator

@kylemontgomery1 kylemontgomery1 commented Aug 20, 2025

Replaces verl submodule with stable v0.5.0 verl release. To support this, the following major changes were made:

  • refactor the rollout engine and chat/tool parsers
  • update/refactor training config to align with verl; update examples/trainer to use new config keys
  • improve the workflow design; add basic single and multilturn workflows
  • update dependencies; add verl installation script
  • update ppo trainers (including support for Megatron)
  • fix strands/smolagents integration

To test:

  1. reinstall rllm + dependencies (see readme for updated instructions)
  2. run solver judge workflow example

@kylemontgomery1 kylemontgomery1 marked this pull request as ready for review August 21, 2025 04:41
@jeffreysijuntan jeffreysijuntan merged commit 4b45a70 into v0.2 Aug 21, 2025
2 checks passed
@jeffreysijuntan jeffreysijuntan deleted the v0.2-verl-stable branch September 5, 2025 06:46
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