My open-source CLI for managing project rules and agent skills across AI coding tools. Deterministic exports and drift checks keep configurations consistent.
Contributor to a PyTorch toolkit for LLM training and fine-tuning. 40+ merged PRs.
Fixed assistant-only loss masking in SFT packing.
Added device-side checks for non-finite distillation losses.
Unified trainer callbacks across SFT, DPO and GRPO.
