Nikhil1581/qwen3.5-0.8b-intent-classification

Qwen3.5-0.8B Intent Classification

A lightweight, quantized GGUF model fine-tuned on top of Qwen3.5-0.8B for conversational intent classification. Designed to run efficiently on consumer hardware with no GPU required.

---

Model Details

Property	Value
Base Model	Qwen3.5-0.8B
Format	GGUF (Q4_K_M quantization)
Parameters	~0.8 Billion
File Size	529 MB
Architecture	Qwen 3.5
License	MIT
Task	Intent Classification

---

Intended Use

This model is designed to classify user intents from conversational text. It is suitable for:

• Chatbot routing and intent detection
• Virtual assistant pipelines
• Customer support automation
• NLU (Natural Language Understanding) systems

---

Quickstart

Using llama.cpp

# Download the model
huggingface-cli download Nikhil1581/qwen3.5-0.8b-intent-classification Qwen3.5-0.8B.Q4_K_M.gguf --local-dir ./models

# Run inference
./llama-cli -m ./models/Qwen3.5-0.8B.Q4_K_M.gguf \
  -p "Classify the intent of the following message: 'What is the weather like today?'" \
  -n 128

Using llama-cpp-python

from llama_cpp import Llama

llm = Llama(
    model_path="./models/Qwen3.5-0.8B.Q4_K_M.gguf",
    n_ctx=2048,
)

prompt = """You are an intent classification assistant. 
Classify the intent of the user message below into a single intent label.

User message: "Book me a flight to New York for next Monday."
Intent:"""

output = llm(prompt, max_tokens=64, stop=["\n"])
print(output["choices"][0]["text"].strip())

Using Ollama

# Create a Modelfile
cat <<EOF > Modelfile
FROM ./models/Qwen3.5-0.8B.Q4_K_M.gguf
SYSTEM "You are an intent classification assistant. Given a user message, respond with the most appropriate intent label."
EOF

ollama create qwen-intent -f Modelfile
ollama run qwen-intent "Cancel my subscription"

---

Quantization Details

This model uses Q4_K_M quantization, which offers a good balance between size, speed, and accuracy.

Format	Size	Notes
Q4_K_M	529 MB	Recommended — balanced

---

Hardware Requirements

Setup	Minimum RAM
CPU only	4 GB RAM
GPU offload	2 GB VRAM

---

Limitations

• Output quality depends on prompt formatting — clear, structured prompts yield better results.
• As a 0.8B parameter model, performance on complex or ambiguous intents may be limited compared to larger models.
• Primarily optimized for English-language inputs.

---

Citation

If you use this model in your work, please cite:

@misc{nikhil1581-qwen3.5-intent,
  author       = {Nikhil1581},
  title        = {Qwen3.5-0.8B Intent Classification (GGUF)},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/Nikhil1581/qwen3.5-0.8b-intent-classification}
}

---

Acknowledgements

Built on top of Qwen3.5 by Alibaba Cloud. Quantized using llama.cpp.