Adding `safetensors` variant of this model

Upload fDPO‑trained Qwen3‑0.6B (100k samples) — no local weight load

Upload fDPO trained Qwen3-0.6B model on MNLP M3 dataset (69k samples)

initial commit