OMEGA-HET-SFT-RL
v1 (interim, superseded) checkpoint. Base: Qwen/Qwen3-4B-Base, full-FT SFT then GRPO.
NOTE: v1 SFT data was ~50%% Chinese-contaminated; this run is superseded by the English-clean
v2.
Provided for completeness. See dataset:
https://huggingface.co/datasets/shizhuo2/omega-het-sft-rl