GGUF release of Qwen3-4B-Computer-Science, a specialized language model fine-tuned for computer science, software engineering, Python programming, debugging, code generation, and technical reasoning.

This repository provides multiple GGUF quantizations for use with llama.cpp, LM Studio, Ollama, Jan, KoboldCpp, and other GGUF-compatible inference engines.

Model Overview

Base Model: Qwen/Qwen3-4B
Architecture: Qwen3
Format: GGUF
License: Apache-2.0
Language: English
Domain: Computer Science & Software Engineering

The model is designed to provide strong performance across software engineering tasks while remaining efficient enough to run locally on modern CPUs and GPUs.

Training Data

This model was fine-tuned using openly licensed datasets:

Dataset	License
HuggingFaceTB/smoltalk (smol-magpie-ultra)	Apache-2.0
agentica-org/DeepCoder-Preview-Dataset (primeintellect)	MIT

Training split:

60,989 samples

Evaluation split:

512 samples

Available Quantizations

File	Recommended Use
BF16.gguf	Highest quality, requires significant memory
Q8_0.gguf	Near-BF16 quality
Q6_K.gguf	Excellent quality/performance balance
Q5_K_M.gguf	Recommended for most users
Q4_K_M.gguf	Best memory efficiency

Recommended Quantization

For most systems:

Q5_K_M offers the best balance between:

Quality
Memory usage
Speed

If memory is limited, use Q4_K_M.

If maximum quality is desired, use Q8_0 or BF16.

Example (llama.cpp)

bash

./llama-cli \
  -m Qwen3-4B-Computer-Science-Q5_K_M.gguf \
  -c 8192

Example (LM Studio)

Download one of the GGUF files.
Import the model into LM Studio.
Select the model.
Start chatting.

Example (Ollama)

Create a Modelfile:

FROM Qwen3-4B-Computer-Science-Q5_K_M.gguf

Then run:

bash

ollama create qwen3-cs -f Modelfile
ollama run qwen3-cs

Intended Use

This model is intended for:

Software Engineering
Python Programming
Debugging
Code Review
Code Generation
Technical Question Answering
Algorithm Design
Computer Science Education

Limitations

This model is specialized for computer science tasks.

Performance outside software engineering domains may differ from the original base model.

As with all language models:

outputs may contain mistakes
generated code should be reviewed
security-critical code should always be validated

Integrity

SHA-256 hashes for every GGUF file are included in:

SHA256SUMS

Users are encouraged to verify downloaded files before use.

License

This repository is released under the Apache-2.0 License.

The fine-tuning datasets are compatible with commercial use:

Apache-2.0
MIT

Acknowledgements

Alibaba Qwen Team
Hugging Face
HuggingFaceTB
Agentica
llama.cpp contributors

Citation

If you use this model in your work, please cite this repository.

@misc{qwen3_4b_computer_science_gguf,
  title={Qwen3-4B-Computer-Science-GGUF},
  author={Irfanuruchi},
  year={2026},
  publisher={Hugging Face}
}

Irfanuruchi/Qwen3-4B-Computer-Science-GGUF

Qwen3-4B-Computer-Science-GGUF