agent-lightning

agent-lightning copied to clipboard

Is the training on-policy?

3

Each step will conduct rollout, so I guess the training is on-policy

XianglongTan

multi-agent system

1

I have read your paper and code, the work is beautiful and practical. My question is as follows: Taking the SQL Agent as an example: if, in a multi-agent system,...

boardman0

trnasition-level reward design

3

In a fixed workflow with multiple roles (each defined by a distinct system prompt), AgentLightning models each role’s I/O as transitions and may group them. Without role-level rewards, is training...

boardman0

The server training ran out of memory (OOM) and crashed.

1

When I was using the local server for model training, I found that the server crashed and couldn't be connected to, showing ray OOM and other logs.

ccp123456789

I find agent lightning only use 1 actor during rollout, can we launch multiple actors?

3

Agent rollout takes a long time, can we create multiple actor instances to accelerate rollout. Currently I have to set TP = [number of GPUs] to use all gpus during...

Thisislegit

How to run agentic rl with agent lightning in the distributed training environment with multi-nodes and multi-gpus ?

4

How to run agentic rl with agent lightning in the distributed training environment with multi-nodes and multi-gpus (e.g., 6 machines and 8 gpus for each machine)

wendongbi

Is is possible to support Google Agent Development Kit (ADK) in the future?

15

jiaxi-xu-fsx

JiahangXu

Will trajectory modifications that bypass the LLM API be lost from the trajectory and thus unavailable for RL training?

2

In my RAG agent’s “think → search → think…” chain of operations, I sometimes edit the tokens generated by the LLM (e.g., deleting a few tokens, reordering them, etc.) before...

GitMonkey0

Rollout phase takes lots of time, and I find that only 1 of 8 GPU is working when rollout.

1

Is it possible to accelerate rollout by taking advantage of all the GPU?

XianglongTan