agent-lightning
agent-lightning copied to clipboard
microsoft
The absolute trainer to light up AI agents.
Add a `trace_agg_mode` option to control whether the trace aggregation is performed per transition or per trajectory (by adding response mask for multi-turn). Default remains 'transition' for backward compatibility.
## Summary Added comprehensive troubleshooting documentation to help users resolve common issues with multi-GPU rollout and response_mask contiguity. ## Changes - Created `docs/troubleshooting.md` with detailed troubleshooting guide - Added Multi-GPU...
## Summary - expand documentation across the SQLite span/resource helper methods to describe JSON rehydration and validation flows - add inline commentary around span persistence, running-rollout fetches, and health check...
## Summary - stream wait_for_rollouts responses over SSE so the client can wait for longer timeouts - add client-side SSE handling with 60s retry chunks and document the new behavior...
**Vote this issue if you believe it may help you.** This proposal is introduce a general Env into agent-lightning, with this Env, you will no longer need to write new...
Let's vote on this one. Make the verl integration compatible with VERL 0.6 could need some efforts and I'm not sure how urgent it is. Also, as verl 0.5 is...
### Summary This PR upgrades the uv dependency manager requirement to version >=0.9.5 and adds a version constraint to ensure compatibility across the team. ### Changes - ✅ Added `required-version...
This issue serves as a central backlog & roadmap tracker for Agent Lightning. Our core members come from a research and engineering team at MSRA. As our team is still...
We introduce a suffix to distinguish between metrics computed before and after AgentLightning’s post-processing. "Before" refers to raw reward and advantage values. "After" refers to values computed following post-processing, which...
Promote AI-Driven Spec-Driven Development Practices - [ ] Review Microsoft’s Spec Kit and related resources. - [ ] Create guidance for integrating spec-driven workflows with GitHub copilot. - [ ]...
