Sequence-to-sequence protein language model bridging sequence understanding and generation.
Distilled a 1.3B-parameter teacher into a 400M-parameter student, then trained with SFT and GRPO (RLVR) against structural/functional reward models.
Structured UniProt into a queryable biomedical knowledge graph powering instruction-dataset construction for LLM fine-tuning.
Multi-model serving observability CLI in Python that manages LLM deployment and configuration through a unified API, replacing vLLM's full CLI option set.