microsoft/agent-lightning

View on GitHub

The absolute trainer to light up AI agents.

Stars: 18.4k
Forks: 1.6k
Open beginner issues: 2
Indexed issues: 100
Avg merge: 2d 8h
Merged PRs (30d): 27

Dominant language: Python
License: MIT
Last GitHub push: Sep 17, 2026
Latest indexed: Sep 19, 2026
Contributing guide: No contributing guide
Code of conduct: Code of conduct
Beginner labels: help wanted

Beginner friendly Everything

Open All

100 open issues indexed

per_rollout_mean divides by training rows instead of rollouts, so the loss scale changes when a rollout spans several rows Open

Difficulty 3/5 1-2 days Newbie friendliness 72/100

microsoft/agent-lightning#593 · 6 comments · Opened Sep 12, 2026
verl 0.7.x embedded vLLM server cannot load custom tool parser plugins (tool_parser_plugin import step skipped) Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

microsoft/agent-lightning#592 · 1 comment · Opened Sep 11, 2026
Documentation menu hides under warning banner Open

Difficulty 3/5 1-2 days Newbie friendliness 35/100

microsoft/agent-lightning#570 · Opened Aug 27, 2026
Add CPU-side Kubernetes image-readiness preflight for VERL training Open

Difficulty 5/5 Over a week Newbie friendliness 45/100

microsoft/agent-lightning#567 · 2 comments · Opened Aug 26, 2026
Design: ThinkingBox x Agent Lightning stateful MCP training example Open

Difficulty 5/5 Over a week Newbie friendliness 42/100

microsoft/agent-lightning#566 · 1 comment · Opened Aug 25, 2026
inquiry Open

Difficulty 3/5 1-2 days Newbie friendliness 45/100

microsoft/agent-lightning#535 · 1 comment · Opened Aug 1, 2026
Support for agents built in .Net using microsoft agent framework Open

Difficulty 5/5 Over a week Newbie friendliness 35/100

microsoft/agent-lightning#526 · 2 comments · Opened May 1, 2026
Add interaction scripts and token utilities Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

microsoft/agent-lightning#514 · 1 reaction · Opened Apr 1, 2026
Question about Code Availability for EMPO^2 Paper Open

v0.x

Difficulty 1/5 Under an hour Newbie friendliness 35/100

microsoft/agent-lightning#511 · 1 comment · 3 reactions · Opened Mar 30, 2026
Please add a Claude Code plugin for agent-lightning. Open

v0.x

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/agent-lightning#510 · Opened Mar 29, 2026
about RL algorithms Open

Difficulty 2/5 1-3 hours Newbie friendliness 45/100

microsoft/agent-lightning#505 · Opened Mar 23, 2026
RAG training becomes slow over time with frequent MCP retrieve timeouts Open

v0.x

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/agent-lightning#504 · 1 comment · Opened Mar 13, 2026
microsoft/agent-lightning has a Nerq Trust Score of 87.6/100 — verified badge available Open

Difficulty 1/5 Under an hour Newbie friendliness 25/100

microsoft/agent-lightning#503 · Opened Mar 11, 2026
calc-x example fails on next Open

bug v0.x

microsoft/agent-lightning#498 · 3 comments · 1 reaction · 2 assignees · Opened Feb 28, 2026
pyright on lint next fails Open

bug v0.x

microsoft/agent-lightning#496 · 1 reaction · 2 assignees · Opened Feb 28, 2026
Release EMPO2 checkpoints on Hugging Face Open

examples question

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/agent-lightning#495 · Opened Feb 28, 2026
RayPPOTrainer._dump_generations() missing gts argument in _train_step call Open

bug v0.x verl

Difficulty 1/5 Under an hour Newbie friendliness 68/100

microsoft/agent-lightning#492 · Opened Feb 18, 2026
Exploration collapse: agents converge to fixed policies, losing novel behavior discovery Open

enhancement v0.x

Difficulty 5/5 Over a week Newbie friendliness 20/100

microsoft/agent-lightning#490 · Opened Feb 13, 2026
GRPO grouping in multi-turn agent RL: is it valid to mix samples with different prompts in the same group? Open

question v0.x verl

Difficulty 5/5 Over a week Newbie friendliness 35/100

microsoft/agent-lightning#489 · 2 comments · Opened Feb 13, 2026
Encountered a problem "Unclosed client session" Open

bug examples topic/rag v0.x

Difficulty 4/5 3-5 days Newbie friendliness 42/100

microsoft/agent-lightning#471 · 4 comments · Opened Feb 1, 2026
Any examples using local models/OpenRouter? Open

question

Difficulty 2/5 1-3 hours Newbie friendliness 25/100

microsoft/agent-lightning#470 · 1 comment · Opened Jan 29, 2026
Which hyper-parameters in the standard VERL configuration do not take effect? Open

documentation enhancement help wanted v0.x

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/agent-lightning#464 · 1 comment · Opened Jan 23, 2026
about llm endpoint Open

proxy question v0.x verl

Difficulty 3/5 1-2 days Newbie friendliness 25/100

microsoft/agent-lightning#456 · 2 comments · Opened Jan 20, 2026
how to use verl-recipe Open

examples question v0.x verl

Difficulty 3/5 1-2 days Newbie friendliness 25/100

microsoft/agent-lightning#455 · 1 comment · Opened Jan 19, 2026
The reward on the validation set of the CALX example has been staying around 0.5 and is unable to improve or increase. Open

examples question v0.x

Difficulty 4/5 3-5 days Newbie friendliness 28/100

microsoft/agent-lightning#454 · 1 comment · Opened Jan 19, 2026
RFC: Integrate Slime RL backend into Agent Lightning (tinker-style service vs native integration) Open

enhancement help wanted v0.x

Difficulty 5/5 Over a week Newbie friendliness 24/100

microsoft/agent-lightning#453 · 1 comment · 3 reactions · Opened Jan 16, 2026
aboue "Cannot log to store." Open

tracer v0.x waiting for reply

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/agent-lightning#449 · 5 comments · Opened Jan 14, 2026
How to deal with failed rollouts? Open

documentation help wanted question v0.x

Difficulty 5/5 Over a week Newbie friendliness 30/100

microsoft/agent-lightning#448 · 6 comments · Opened Jan 13, 2026
Why Tinker Open

question tinker v0.x

Difficulty 3/5 1-2 days Newbie friendliness 20/100

microsoft/agent-lightning#447 · 1 comment · Opened Jan 12, 2026
RuntimeError: max(): Expected reduction dim to be specified for input.numel() == 0. Specify the reduction dim with the 'dim' argument. Open

v0.x verl waiting for reply

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/agent-lightning#446 · 1 comment · Opened Jan 12, 2026
When training Qwen2.5-VL-3B, an error occurred in `get_rope_index`: RuntimeError: shape mismatch (mRoPE position_ids / attention_mask lengths do not match). Open

need investigation topic/multimodality v0.x

Difficulty 4/5 3-5 days Newbie friendliness 32/100

microsoft/agent-lightning#441 · 3 comments · Opened Dec 31, 2025
Would it be possible to allow custom text gradient computation and editing templates for APO in the form of SDK usage? Open

apo enhancement v0.x

Difficulty 5/5 Over a week Newbie friendliness 25/100

microsoft/agent-lightning#439 · 1 comment · Opened Dec 31, 2025
[BUG] Potential Memory Leak Issue Open

out-of-memory v0.x verl waiting for reply

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/agent-lightning#438 · 12 comments · Opened Dec 29, 2025
Solution: No module named 'vllm.lora.models' Open

blocked by upstream v0.x verl

Difficulty 2/5 1-3 hours Newbie friendliness 35/100

microsoft/agent-lightning#437 · 2 comments · Opened Dec 27, 2025
Regarding the server's lifecycle Open

question store v0.x

Difficulty 5/5 Over a week Newbie friendliness 25/100

microsoft/agent-lightning#436 · 1 comment · Opened Dec 27, 2025
multi-node multi-GPU training Open

duplicate verl

Difficulty 5/5 Over a week Newbie friendliness 25/100

microsoft/agent-lightning#434 · 2 comments · Opened Dec 25, 2025
What is the expected behavior of the adapter TracerTraceToTriplet? Open

adapter question v0.x

Difficulty 4/5 3-5 days Newbie friendliness 35/100

microsoft/agent-lightning#432 · 2 comments · Opened Dec 24, 2025
APO's TraceToMessages adapter fails with multi-turn agent rollouts (KeyError: 'role') Open

adapter apo bug need investigation v0.x

Difficulty 4/5 3-5 days Newbie friendliness 52/100

microsoft/agent-lightning#425 · 14 comments · Opened Dec 17, 2025
agentops tracer Open

examples tracer v0.x

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/agent-lightning#424 · 1 comment · Opened Dec 17, 2025
How can agent built with Java (Spring AI) integrate with agent lightning? Open

help wanted

Difficulty 5/5 Over a week Newbie friendliness 25/100

microsoft/agent-lightning#419 · 2 comments · 1 reaction · Opened Dec 17, 2025
Is that reward must be a float between 0~1? Open

v0.x waiting for reply

Difficulty 3/5 1-2 days Newbie friendliness 35/100

microsoft/agent-lightning#405 · 1 comment · Opened Dec 11, 2025
ValueError: Extracted completion from trace is not a list Open

v0.x waiting for reply

Difficulty 2/5 1-3 hours Newbie friendliness 45/100

microsoft/agent-lightning#404 · 2 comments · Opened Dec 11, 2025
How to save val_data predictions during each evaluation of `trainer.fit`? Open

enhancement question v0.x verl

Difficulty 3/5 1-2 days Newbie friendliness 35/100

microsoft/agent-lightning#401 · 14 comments · Opened Dec 11, 2025
Cleaner termination Open

help wanted store v0.x

Difficulty 4/5 3-5 days Newbie friendliness 30/100

microsoft/agent-lightning#397 · 2 comments · Opened Dec 11, 2025
RuntimeError: Server failed to start within the 10 seconds. Open

store v0.x waiting for reply

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/agent-lightning#395 · 6 comments · Opened Dec 10, 2025
Question: What is the linking rule between intermediate rewards and the triplet? Open

credit assignment examples question v0.x

Difficulty 4/5 3-5 days Newbie friendliness 30/100

microsoft/agent-lightning#392 · 7 comments · Opened Dec 10, 2025
Question: How will intermediate rewards be used? Open

credit assignment question v0.x verl

Difficulty 3/5 1-2 days Newbie friendliness 35/100

microsoft/agent-lightning#390 · 5 comments · Opened Dec 10, 2025
Question: Any way to reduce memory for spider example Open

examples question v0.x verl

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/agent-lightning#389 · 3 comments · Opened Dec 10, 2025
How to Integrate Process Rewards in LangGraph Agents and Handle Node Execution Failures? Open

examples question v0.x

Difficulty 4/5 3-5 days Newbie friendliness 25/100

microsoft/agent-lightning#387 · 4 comments · Opened Dec 9, 2025
Redundant logic after using aiologic Open

enhancement store v0.x

Difficulty 2/5 1-3 hours Newbie friendliness 62/100

microsoft/agent-lightning#372 · 10 comments · Opened Dec 5, 2025

microsoft/agent-lightning

Get new issues in your inbox