---
library_name: peft
license: mit
base_model:
- unsloth/QwQ-32B-Preview-GGUF
tags:
- lora
- unsloth
- generated_from_trainer
- text-generation-inference
- gguf
- ollama
- adapter
model-index:
- name: QwQ-32B-Preview-with-Tags-LoRA-GGUF
  results: []
---

# QwQ-32B-Preview LoRA for separating thinking/answer parts

This LoRA file was fine-tuned to make QwQ constantly separate its private thoughts from the final answer using `<THINKING>...</THINKING><ANSWER>...</ANSWER>` tags.

For best results, it's also recommended to add the following to the System Prompt:
> Your private thoughts must be placed inside <THINKING>...</THINKING> XML tags, and your final answer to the user must be placed inside <ANSWER>...</ANSWER> XML tags. These tags MUST appear in all your responses.

This GGUF file can be used with Ollama as an adapter of the [unsloth/QwQ-32B-Preview-GGUF](https://huggingface.co/unsloth/QwQ-32B-Preview-GGUF/tree/main) quantized models. See the attached `Modelfile` for an example.