Edge cases occurring after more than $1000$ training steps could previously cause RL training to fail. These failures remained undetected because the tests used a comparatively small step limit: #573 (comment)
Increase the test step limit and fix the corresponding bugs uncovered by the extended training runs. Mostly implemented by PR #573
Edge cases occurring after more than$1000$ training steps could previously cause RL training to fail. These failures remained undetected because the tests used a comparatively small step limit: #573 (comment)
Increase the test step limit and fix the corresponding bugs uncovered by the extended training runs. Mostly implemented by PR #573