different original_max_position_embeddings

Thank you for your amazing work on the Qwen23 models.

I have a question about the configuration for the Qwen3-32B model. In the config.json file, the value for **original_max_position_embeddings** is set to 40960. However, the mirothinker was trained on a **original_max_position_embeddings** of 32k tokens using YaRN.

Will this setting affect the model's performance?