Show HN: We Westernized Qwen for US enterprise use
Category: ai-ml
Tags: llm, fine-tuning, model-weights, unlearning, qwen
Score: 5.7/10 (Innovation: 6, Technical: 6, Documentation: 5, Utility: 5)
A merged fine-tuned checkpoint of Qwen3.6-35B-A3B that uses behavioral unlearning to reduce censorship, refusal, and geopolitical bias in the base model, aimed at Western enterprise deployments. It reports significant improvements on custom benchmarks (CCPC-500, DECCP, ChinaBench) while claiming to preserve general safety behavior on XSTest and OR-Bench.
Target audience: AI/ML engineers and enterprises deploying open-weight LLMs
Repository: https://huggingface.co/hirundo-io/Qwen3.6-35B-A3B-Westernized
View on Hacker News