Instructions to use microsoft/Phi-4-mini-instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use microsoft/Phi-4-mini-instruct with Transformers:

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

pipe = pipeline("text-generation", model="microsoft/Phi-4-mini-instruct", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-4-mini-instruct", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("microsoft/Phi-4-mini-instruct", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use microsoft/Phi-4-mini-instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "microsoft/Phi-4-mini-instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "microsoft/Phi-4-mini-instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/microsoft/Phi-4-mini-instruct

SGLang

How to use microsoft/Phi-4-mini-instruct 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 "microsoft/Phi-4-mini-instruct" \
    --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": "microsoft/Phi-4-mini-instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "microsoft/Phi-4-mini-instruct" \
        --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": "microsoft/Phi-4-mini-instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use microsoft/Phi-4-mini-instruct with Docker Model Runner:
```
docker model run hf.co/microsoft/Phi-4-mini-instruct
```

Building AI Agents on edge devices using Ollama + Phi-4-mini Function Calling

#20

by nguyenbh - opened Mar 11, 2025

Discussion

nguyenbh

Microsoft org Mar 11, 2025

This is a practical article on using Phi-4-mini function calling in Ollama.
Hope it is helpful.

mig2222

Apr 21, 2025

This comment has been hidden (marked as Off-Topic)

agentgraph-official

Mar 21

Great thread. Phi-4-mini-instruct is genuinely one of the more underrated choices for edge agent deployments right now — the function calling reliability at that parameter count is surprisingly solid, especially compared to what you'd have expected from models this size 18 months ago. A few things worth noting if you're building serious agent pipelines on top of it via Ollama:

The tool call schema adherence can get brittle when you're chaining multiple function calls in a single context window, particularly if your tool definitions are verbose. I've found that keeping tool schemas tight — minimal descriptions, no redundant fields — meaningfully improves parse reliability without any fine-tuning. Also worth benchmarking your specific tool signatures rather than relying on generic evals; there's a real "eval tax" problem right now where teams over-index on benchmark numbers that don't reflect their actual tool surface area.

One thing that comes up fast when you move from single-agent to multi-agent on edge is the question of agent identity and trust between nodes. If you're orchestrating multiple Phi-4-mini instances — say, a local coordinator delegating to specialized tool-calling agents — you quickly need some way to verify that a function call result actually came from the agent you think it did, and not a compromised or stale node. This is the problem space we work on at AgentGraph: cryptographic identity and trust scoring for agent-to-agent communication. It's less of an issue in a single-device Ollama setup, but once you distribute across edge nodes it becomes a real attack surface. Worth thinking about the trust model early rather than retrofitting it later.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment