Spaces:
Runtime error
Runtime error
Commit History
-
Implementation of the FinRL workflow (#241)
MackofiUnverified signature* Barebones setup from LLM training. * Training workflow for FinRL. * Training works, evaluation work in progress * Moved everyhing into a module, reward functions now customizable, evaluation works a bit better now. * Created new lynxkite-llm module, mostly working evaluation workflow, refactored fin_rl specific functions. * Evaluation works, memory management is much better. * Fallback to unsloth generation * Everything should work. * Fixes FinRL workflow, working example, evaluation with JudgeLLM works, but training possibly fails. * Update dependencies to follow other packages. * Fixed Demo workspace, when training on responses only auto detect instruction and response part, new filter long prompts by token box.f583080