models: - model: Qwen/QwQ-32B - model: qihoo360/Light-R1-32B - model: maldv/Qwentile2.5-32B-Instruct - model: Qwen/Qwen2.5-32B-Instruct - model: maldv/Awqward2.5-32B-Instruct - model: OpenPipe/Deductive-Reasoning-Qwen-32B - model: Rombo-Org/Rombo-LLM-V3.0-Qwen-32b merge_method: model_stock base_model: Qwen/Qwen2.5-32B-Instruct tokenizer_source: base normalize: true int8_mask: true dtype: bfloat16