chore: minor updates on logging and resource allocation (#73)

feat: add fp16 training (#70)

feat: support LoRA RL training (#64)

* feat: support LoRA RL training

* minor

fix: truncated importance sampling to handle precision mismatch (#62)

* support tis

* make tis default

update readme

feat: reduce vram footprint (#56)

* feat: add fused lm head to reduce vram usage

* fix sft slicing and add dry run

* minor

* bump version

Add one file for k8s job launcher

feat: fix deps, refactor apis, allow resume training (#39)

* fix deps, refactor apis

* bump version

* updates

* actor identity

* fix ref offload

* training resume

* bump version

Upgrade to vllm V1 (0.8.4) and use actor api init() (#38)

* updates

* bump version

Upgrade vllm for more efficient collocation (#34)

* upgrade vllm & adopt collective_rpc

* use .float() for kl & increase timeout to 60m

* speed up minibatch training

* add constant lr scheduler

* update

* updates

* fix non_eos detection

* changes

* minor

* update

* ratio

* updates