Instructions to use pytorch/Phi-4-mini-instruct-FP8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use pytorch/Phi-4-mini-instruct-FP8 with Transformers:

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

pipe = pipeline("text-generation", model="pytorch/Phi-4-mini-instruct-FP8", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("pytorch/Phi-4-mini-instruct-FP8", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("pytorch/Phi-4-mini-instruct-FP8", trust_remote_code=True, device_map="auto")
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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use pytorch/Phi-4-mini-instruct-FP8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "pytorch/Phi-4-mini-instruct-FP8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "pytorch/Phi-4-mini-instruct-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/pytorch/Phi-4-mini-instruct-FP8

SGLang

How to use pytorch/Phi-4-mini-instruct-FP8 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 "pytorch/Phi-4-mini-instruct-FP8" \
    --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": "pytorch/Phi-4-mini-instruct-FP8",
		"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 "pytorch/Phi-4-mini-instruct-FP8" \
        --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": "pytorch/Phi-4-mini-instruct-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use pytorch/Phi-4-mini-instruct-FP8 with Docker Model Runner:
```
docker model run hf.co/pytorch/Phi-4-mini-instruct-FP8
```

jerryzh168 commited on Sep 10, 2025

Commit

dd89b44

verified ·

1 Parent(s): 700e10a

Update README.md

Browse files

Files changed (1) hide show

README.md +0 -38

README.md CHANGED Viewed

@@ -295,11 +295,8 @@ print(f"Peak Memory Usage: {mem:.02f} GB")
 |                                  | Phi-4 mini-Ins | Phi-4-mini-instruct-FP8         |
 | latency (batch_size=1)           | 1.61s          | 1.25s (1.29x speedup)           |
 | latency (batch_size=256)         | 5.16s          | 4.89s (1.05x speedup)           |
-| serving (num_prompts=1)          | 1.37 req/s     | 1.66 req/s (1.21x speedup)      |
-| serving (num_prompts=1000)       | 62.55 req/s    | 72.56 req/s (1.16x speedup)     |
 Note the result of latency (benchmark_latency) is in seconds, and serving (benchmark_serving) is in number of requests per second.
-Note the result is not using fbgemm kernels, (no `fbgemm-gpu-genai` installed), fbgemm kernels has less speedup when num_prompts is 1000 currently.
 <details>
 <summary> Reproduce Model Performance Results </summary>
@@ -334,41 +331,6 @@ python benchmarks/benchmark_latency.py --input-len 256 --output-len 256 --model
 VLLM_DISABLE_COMPILE_CACHE=1 python benchmarks/benchmark_latency.py --input-len 256 --output-len 256 --model pytorch/Phi-4-mini-instruct-FP8 --batch-size 1
 ```
-## benchmark_serving
-We benchmarked the throughput in a serving environment.
-Download sharegpt dataset:
-```Shell
-wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
-```
-Other datasets can be found in: https://github.com/vllm-project/vllm/tree/main/benchmarks
-Note: you can change the number of prompts to be benchmarked with `--num-prompts` argument for `benchmark_serving` script.
-### baseline
-Server:
-```Shell
-vllm serve microsoft/Phi-4-mini-instruct --tokenizer microsoft/Phi-4-mini-instruct -O3
-```
-Client:
-```Shell
-python benchmarks/benchmark_serving.py --backend vllm --dataset-name sharegpt --tokenizer microsoft/Phi-4-mini-instruct --dataset-path ./ShareGPT_V3_unfiltered_cleaned_split.json --model microsoft/Phi-4-mini-instruct --num-prompts 1
-```
-### FP8
-Server:
-```Shell
-VLLM_DISABLE_COMPILE_CACHE=1 vllm serve pytorch/Phi-4-mini-instruct-FP8 --tokenizer microsoft/Phi-4-mini-instruct -O3
-```
-Client:
-```Shell
-python benchmarks/benchmark_serving.py --backend vllm --dataset-name sharegpt --tokenizer microsoft/Phi-4-mini-instruct --dataset-path ./ShareGPT_V3_unfiltered_cleaned_split.json --model pytorch/Phi-4-mini-instruct-FP8 --num-prompts 1
-```
 </details>
 # Paper: TorchAO: PyTorch-Native Training-to-Serving Model Optimization

 |                                  | Phi-4 mini-Ins | Phi-4-mini-instruct-FP8         |
 | latency (batch_size=1)           | 1.61s          | 1.25s (1.29x speedup)           |
 | latency (batch_size=256)         | 5.16s          | 4.89s (1.05x speedup)           |
 Note the result of latency (benchmark_latency) is in seconds, and serving (benchmark_serving) is in number of requests per second.
 <details>
 <summary> Reproduce Model Performance Results </summary>
 VLLM_DISABLE_COMPILE_CACHE=1 python benchmarks/benchmark_latency.py --input-len 256 --output-len 256 --model pytorch/Phi-4-mini-instruct-FP8 --batch-size 1
 ```
 </details>
 # Paper: TorchAO: PyTorch-Native Training-to-Serving Model Optimization