Instructions to use FastFlowLM/Qwen3-8B-NPU2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use FastFlowLM/Qwen3-8B-NPU2 with Transformers:

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

pipe = pipeline("text-generation", model="FastFlowLM/Qwen3-8B-NPU2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("FastFlowLM/Qwen3-8B-NPU2")
model = AutoModelForCausalLM.from_pretrained("FastFlowLM/Qwen3-8B-NPU2", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use FastFlowLM/Qwen3-8B-NPU2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "FastFlowLM/Qwen3-8B-NPU2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "FastFlowLM/Qwen3-8B-NPU2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/FastFlowLM/Qwen3-8B-NPU2

SGLang

How to use FastFlowLM/Qwen3-8B-NPU2 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 "FastFlowLM/Qwen3-8B-NPU2" \
    --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": "FastFlowLM/Qwen3-8B-NPU2",
		"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 "FastFlowLM/Qwen3-8B-NPU2" \
        --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": "FastFlowLM/Qwen3-8B-NPU2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use FastFlowLM/Qwen3-8B-NPU2 with Docker Model Runner:
```
docker model run hf.co/FastFlowLM/Qwen3-8B-NPU2
```

FastFlowLM commited on Jul 31, 2025

Commit

070b3b9

verified ·

1 Parent(s): 62a857e

Create README.md

Browse files

Files changed (1) hide show

README.md +69 -0

README.md ADDED Viewed

	@@ -0,0 +1,69 @@

+---
+license: mit
+language:
+- en
+library_name: transformers
+tags:
+- qwen
+- qwen3
+- qwen3-8b
+- text-generation
+- AMD
+- Ryzen
+- NPU
+pipeline_tag: text-generation
+base_model:
+- Qwen/Qwen3-7B-Instruct
+---
+# 🐉 Qwen3 8B – Optimized for FastFlowLM on AMD Ryzen™ AI NPU (XDNA2 Only)
+## Model Summary
+This model is based on **Qwen3 8B Instruct** from Alibaba Cloud. It uses the original Qwen3-7B architecture and weights, potentially with enhancements such as quantization or kernel-level acceleration for NPU efficiency via the FastFlowLM runtime.
+> ✅ **Licensed under the permissive MIT License.**
+## 📝 License & Usage Terms
+### Base Model License
+- Released under the MIT License by Alibaba Cloud:
+  👉 https://huggingface.co/Qwen/Qwen3-7B-Instruct
+- Permissions include:
+  - Free commercial and non-commercial use
+  - Permission to modify and redistribute
+  - Attribution not required, but appreciated
+### Redistribution Notice
+- This repository **does not** contain original base weights.
+- You must acquire the official weights from Qwen’s Hugging Face page:
+  👉 https://huggingface.co/Qwen/Qwen3-7B-Instruct
+### If Fine-tuned
+If this model has been modified (e.g., quantized, fine-tuned):
+- **Base Model License**: MIT
+- **Derivative Weights License**: [e.g., MIT, CC-BY-NC-4.0, custom]
+- **Training Dataset License(s)**:
+  - [Dataset A] – [license]
+  - [Dataset B] – [license]
+Ensure all datasets used are appropriately licensed for redistribution and use.
+## Intended Use
+- **Recommended For**: High-performance on-device inference, private LLM workloads, local chat assistants, research
+- **Not Recommended For**: Critical systems or regulated applications without thorough evaluation
+## Limitations & Risks
+- Large model may require tuning for low-latency use
+- Possibility of hallucination, toxicity, or factual errors
+- May encode pretraining biases
+## Citation
+```bibtex
+@misc{qwen32024,
+  title={Qwen3: Smaller, Smarter, and More Open},
+  author={Alibaba Cloud},
+  year={2024},
+  url={https://huggingface.co/Qwen}
+```