telestes 0.1.6

transformer-based reinforcment learning agents.

telestes

telestes is a library providing transformer-based rl agents, that is extremely customizable, but also comes with sane defaults.

source code: https://erga.apotheke.earth/apotheke/alectors
license: GPLv3 or later

Description

telestes serves as the rl base for alectors, our natural language reinforcement learning (nlrl) agents.

We decided to seperate the rl part into its own module, so that we may explore non-nlp rl tasks that might benefit from a transformer architecture.

At the same time, we are interested in expanding the different available networks; so far we support Transformers and Linear models[^1].

Supported Architectures

The currently supported architecture is PPO. Plans exist to include SAC, GRPO, and maybe DDQN.

[^1]: we might split our network architectures into a seperate module also down the road, in order to use it on non rl research as well.