qwen3-0.6b-nothink-gguf

Qwen3-0.6B-nothink

A modified (not fine-tuned) version of Qwen3-0.6B with Chain-of-Thought (CoT) completely disabled for faster responses, since enable_thinking=False (or /no_think) is not perfect and does not completely prevent reasoning in certain contexts.
This model supports llama.cpp and other compatible apps using llama.cpp as the backend.

Modification Details

The model has been modified (just two tokens) to prevent spontaneous Chain-of-Thought reasoning:

Changes:

• lm_head[151667] (token <think>): Set to a zero vector
• lm_head[151668] (token </think>): Set to the weights of original <think>

Thus the model never outputs <think>.
Manual modification is used since fine-tuning may negatively affect some model capabilities (especially for very small models).

Effect:

The <think> token generated by models are automatically converted to </think>, thus the model will generate non-reasoning contents after </think>.

An example output using the user prompt What is the .safetensors format, with T=0.5:

</think>

The `.safetensors` format is a high-performance, memory-efficient format used primarily in **TensorFlow** and **PyTorch** for storing model parameters. It is designed to be both fast and compact, making it suitable for large models.

### Key Features of the `.safetensors` Format:

1. **Compact Memory Usage**: 
   - Uses a binary format that takes up less memory than JSON or XML.
   - Stores weights in a way that allows efficient memory usage during inference.

2. **Fast Access**:
   - Enables fast access to model parameters, which is crucial for performance in large models.

3. **High Performance**:
   - Optimized for speed and efficiency, making it suitable for training and deployment of large models.

4. **Portable and Cross-Platform**:
   - Works on any platform that supports the `.safetensors` format (e.g., TensorFlow, PyTorch).

...

The modification tool

See ekcbw/llm-nothink.