ReleaseDatabricks (DBRX)Databricks (DBRX)published Jul 29, 2025seen 5d

databricks/compose-rl v0.8.0

databricks/compose-rl

Open original ↗

Captured source

source ↗
published Jul 29, 2025seen 5dcaptured 10hhttp 200method plain

v0.8.0

Repository: databricks/compose-rl

Tag: v0.8.0

Published: 2025-07-29T21:43:46Z

Prerelease: no

Release notes:

What's Changed

  • Removed Compatibility directories by @jdchang1 in https://github.com/databricks/compose-rl/pull/98
  • Prompts per iteration by @bcui-db in https://github.com/databricks/compose-rl/pull/94
  • Allow reference checkpoint to be both callback and load_path by @dakinggg in https://github.com/databricks/compose-rl/pull/99
  • Added MessagesDataloader so we can just use messages in our datasets rather than tokenized inputs by @SeanKski in https://github.com/databricks/compose-rl/pull/92
  • Add logs around reward process pool recreation by @dakinggg in https://github.com/databricks/compose-rl/pull/101
  • Revert timeout by @dakinggg in https://github.com/databricks/compose-rl/pull/104
  • Added proper temperature scaling of logits by @jdchang1 in https://github.com/databricks/compose-rl/pull/105
  • Wensun/apo by @wensun in https://github.com/databricks/compose-rl/pull/96
  • vLLM Chat Conversion by @jdchang1 in https://github.com/databricks/compose-rl/pull/102
  • hotfix by @jdchang1 in https://github.com/databricks/compose-rl/pull/108
  • STEM Benchmarks and verifiers by @gupta-abhay in https://github.com/databricks/compose-rl/pull/95
  • Add prefix caching by @gupta-abhay in https://github.com/databricks/compose-rl/pull/107
  • Update codeowners by @dakinggg in https://github.com/databricks/compose-rl/pull/113
  • Changes for accumulate flag by @gupta-abhay in https://github.com/databricks/compose-rl/pull/111
  • Adding Token Counter for Online RL by @rithwik-db in https://github.com/databricks/compose-rl/pull/110
  • Use Single Controller design with unit test by [Experimental] @bowenyang008 in https://github.com/databricks/compose-rl/pull/114
  • Update Code Owners by @gupta-abhay in https://github.com/databricks/compose-rl/pull/116
  • refactor ppo callback to move its logic to single controller [Experimental] by @bowenyang008 in https://github.com/databricks/compose-rl/pull/115

New Contributors

  • @SeanKski made their first contribution in https://github.com/databricks/compose-rl/pull/92
  • @wensun made their first contribution in https://github.com/databricks/compose-rl/pull/96
  • @rithwik-db made their first contribution in https://github.com/databricks/compose-rl/pull/110

Full Changelog: https://github.com/databricks/compose-rl/compare/v0.7.0...v0.8.0

Notability

notability 3.0/10

Routine version update of a library, no major traction indicators