What's Changed
- Revert regression of multi-turn templating problem by @kalomaze in #497
- V0.1.6.post0 staging by @willccbb in #499
- Small refactors to make using
eval_utilseasier by @mikasenghaas in #496 - fix bug in overlong exception during get_model_response by @keatonelvins in #502
- Allow passing explicit semaphore by @mikasenghaas in #504
- RLTrainer overhaul, replacing GRPOTrainer by @willccbb in #505
- Fix environment.py generate function deepcopying state by @reachv in #503
- fix: vf-eval save state columns by @ercbot in #517
- Added bulk_delete functionality to bulk delete sandboxes with global ID by @ParamThakkar123 in #487
- docs(overview): correct wording by @hensapir in #495
- fix: pull request template agents.md link doesn't work by @ercbot in #518
- Implement MaybeThinkParser by @mikasenghaas in #522
- Allow setting env args in pyproject for num_examples + rollouts_per_example by @willccbb in #523
- fix: update metadata after generation by @anakin87 in #509
- Overwrite
tasktoenv_nameinEnvGroupby @mikasenghaas in #526 - Allow passing map_kwargs by @mikasenghaas in #530
- route
process_env_results_vllmforEnvGroups by @rasdani in #533 - Pass map_kwargs to base env by @mikasenghaas in #535
- Make example_id unique in EnvGroup by @mikasenghaas in #531
- fix envgroup edge case; rename generator to orchestrator by @willccbb in #539
- fix state columns parsing in
vf-evalcli by @rasdani in #544 - V0.1.7 staging by @willccbb in #545
New Contributors
- @keatonelvins made their first contribution in #502
- @ercbot made their first contribution in #517
- @ParamThakkar123 made their first contribution in #487
- @hensapir made their first contribution in #495
- @rasdani made their first contribution in #533
Full Changelog: v0.1.6...v0.1.7