I noticed this project uses Ray/RLlib v1.0.1, which is significantly outdated (released in 2020). While the project itself is valuable, relying on such an old version of Ray poses several challenges:
-
API Instability:
Ray has undergone frequent and breaking changes since v1.0.1 (e.g., major updates in Ray 2.0+ for AI/RL workflows). Modern versions (e.g., 2.x) have entirely revamped APIs (e.g., Ray AIR, RLlib Trainer → Algorithm), making migration non-trivial.
-
Unresolved Bugs:
In my testing with recent Ray versions (e.g., 2.4+), I encountered multiple issues (e.g., sporadic crashes during distributed training, inconsistent cluster resource handling). The rapid release cycle seems to introduce regressions.
-
Maintenance Burden:
Sticking with v1.0.1 risks long-term incompatibility with other libraries, while upgrading requires significant effort due to deprecated APIs.
I noticed this project uses
Ray/RLlib v1.0.1, which is significantly outdated (released in 2020). While the project itself is valuable, relying on such an old version of Ray poses several challenges:API Instability:
Ray has undergone frequent and breaking changes since v1.0.1 (e.g., major updates in Ray 2.0+ for AI/RL workflows). Modern versions (e.g., 2.x) have entirely revamped APIs (e.g., Ray AIR, RLlib Trainer → Algorithm), making migration non-trivial.
Unresolved Bugs:
In my testing with recent Ray versions (e.g., 2.4+), I encountered multiple issues (e.g., sporadic crashes during distributed training, inconsistent cluster resource handling). The rapid release cycle seems to introduce regressions.
Maintenance Burden:
Sticking with v1.0.1 risks long-term incompatibility with other libraries, while upgrading requires significant effort due to deprecated APIs.