databricks/compose-rl v0.7.0

v0.7.0

Repository: databricks/compose-rl

Tag: v0.7.0

Published: 2025-06-23T16:58:19Z

Prerelease: no

Release notes:

What's Changed

• Added verified answers to the logging by @abaheti95 in https://github.com/databricks/compose-rl/pull/63
• Adding GPU CI back by @dakinggg in https://github.com/databricks/compose-rl/pull/64
• Fix args propagation by @dakinggg in https://github.com/databricks/compose-rl/pull/65
• Fix weight propagation by @bcui-db in https://github.com/databricks/compose-rl/pull/66
• Microbatching fixes by @dakinggg in https://github.com/databricks/compose-rl/pull/71
• Make myself admin by @gupta-abhay in https://github.com/databricks/compose-rl/pull/72
• Update ci-testing to latest version by @dakinggg in https://github.com/databricks/compose-rl/pull/70
• Move generate to be done via prompt_token_ids by @bcui-db in https://github.com/databricks/compose-rl/pull/73
• Add GRPO assert that we need more than one generation by @bcui-db in https://github.com/databricks/compose-rl/pull/74
• Adding a Math format verifier by @gupta-abhay in https://github.com/databricks/compose-rl/pull/75
• Ping foundry version and hash to prepare foundry upgrade by @bowenyang008 in https://github.com/databricks/compose-rl/pull/76
• Bump to torch 2.7 by @bowenyang008 in https://github.com/databricks/compose-rl/pull/77
• Allow DPO reference model to be loaded from LoadCheckpoint callback by @dakinggg in https://github.com/databricks/compose-rl/pull/80
• Set default value as this is only used for local debugging by @gupta-abhay in https://github.com/databricks/compose-rl/pull/84
• Add More Codeowners by @bcui-db in https://github.com/databricks/compose-rl/pull/86
• Fix reward timeouts by @dakinggg in https://github.com/databricks/compose-rl/pull/87
• Remove llama models as defaults by @gupta-abhay in https://github.com/databricks/compose-rl/pull/88
• Skip initial vLLM weight load. by @dakinggg in https://github.com/databricks/compose-rl/pull/89
• Fix memory leak by @dakinggg in https://github.com/databricks/compose-rl/pull/90
• Renaming and Organization of RL algorithms in preparation for Development by @jdchang1 in https://github.com/databricks/compose-rl/pull/83
• Causal classifier by @alextrott16 in https://github.com/databricks/compose-rl/pull/8
• Vllm import Hotfix by @jdchang1 in https://github.com/databricks/compose-rl/pull/91
• Fixing entropy calculation by @abaheti95 in https://github.com/databricks/compose-rl/pull/85

Full Changelog: https://github.com/databricks/compose-rl/compare/v0.5.0...v0.7.0