# Explicitly set to None so hf_config_to_provider_kwargs skips them. # Qwen3-Next is not an MLA model and not a DeepSeek-style MoE / MTP model.