Instructions to use khazarai/Qwen3.5-4B-Agentic-Coding with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use khazarai/Qwen3.5-4B-Agentic-Coding with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="khazarai/Qwen3.5-4B-Agentic-Coding")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("khazarai/Qwen3.5-4B-Agentic-Coding")
model = AutoModelForMultimodalLM.from_pretrained("khazarai/Qwen3.5-4B-Agentic-Coding", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use khazarai/Qwen3.5-4B-Agentic-Coding with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "khazarai/Qwen3.5-4B-Agentic-Coding"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "khazarai/Qwen3.5-4B-Agentic-Coding",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/khazarai/Qwen3.5-4B-Agentic-Coding

SGLang

How to use khazarai/Qwen3.5-4B-Agentic-Coding with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "khazarai/Qwen3.5-4B-Agentic-Coding" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "khazarai/Qwen3.5-4B-Agentic-Coding",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "khazarai/Qwen3.5-4B-Agentic-Coding" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "khazarai/Qwen3.5-4B-Agentic-Coding",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Unsloth Desktop
Docker Model Runner
How to use khazarai/Qwen3.5-4B-Agentic-Coding with Docker Model Runner:
```
docker model run hf.co/khazarai/Qwen3.5-4B-Agentic-Coding
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Qwen3.5-4B-Agentic-Coding

Model Summary

Qwen3.5-4B-Agentic-Coding is a fine-tuned, 4-billion parameter language model optimized for agentic coding tasks, structured technical reasoning, and automated tool interaction. Built upon the Qwen3.5 architecture, this model excels at multi-step problem solving, refactoring codebases without logic gaps, and seamlessly executing tool calls (Bash, Python, File Editor, Browser).

Unlike general-purpose models that may produce loose code syntax or lose track of historical code patterns during refactoring, Qwen3.5-4B-Agentic-Coding maintains a strict internal chain of thought to preserve execution correctness (e.g., accurately demonstrating legacy Callback-to-Async transformations).

Intended Use

Primary Use Cases

Agentic Coding Assistants: Autonomous software development agents using external tool execution.
Code Refactoring & Optimization: Transforming legacy paradigms (e.g., callback hell) into modern asynchronous or modular paradigms with strict logical checks.
Structured Technical Reasoning: Explaining architectural decisions and step-by-step logic before generating executable output.
Interactive Tool Execution: Generating structured commands for environments supporting Bash, Python, or file operations.

Out-of-Scope / Non-Recommended Use

Non-English programming inquiries or multi-lingual conversational tasks.
Direct deployment in production without human-in-the-loop validation for critical code infrastructure.

Training Dataset & Tools

The model was fine-tuned on a high-quality dataset designed to boost agentic problem-solving capabilities.

Integrated Dataset Tools

Tool	Purpose
`editor`	File creation, updating, and repository-wide modifications.
`bash`	Executing terminal commands and managing execution pipelines.
`python`	In-line code validation, sandboxed testing, and mathematical checks.
`browser`	Real-time technical documentation lookup and validation.

Strengths & Behavioral Characteristics

Precise Problem Framing: Validates legacy structures before refactoring, preventing logical contradictions (such as confusing Promises with Callbacks).
Minimal Injected Friction: Keeps core implementations clean and standalone without adding unnecessary, non-functional code abstractions.
Structured System Thought: Embeds step-by-step reasoning blocks (<think>) prior to code delivery for complete transparency.

Limitations & Biases

Synthetically Generated: The fine-tuning dataset relies on synthetic distillation techniques, which may occasionally echo systemic patterns from the teacher models.
Language Restriction: Optimizations are strictly centered around English code annotations and prompt instructions.
Model Biases: Inherits standard biases from the core Qwen3.5 architecture.othai/unsloth)