Instructions to use Qwen/Qwen2.5-32B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Qwen/Qwen2.5-32B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="Qwen/Qwen2.5-32B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-32B-Instruct")
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-32B-Instruct")
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
Local Apps Settings

vLLM

How to use Qwen/Qwen2.5-32B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Qwen/Qwen2.5-32B-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": "Qwen/Qwen2.5-32B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Qwen/Qwen2.5-32B-Instruct

SGLang

How to use Qwen/Qwen2.5-32B-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 "Qwen/Qwen2.5-32B-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": "Qwen/Qwen2.5-32B-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 "Qwen/Qwen2.5-32B-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": "Qwen/Qwen2.5-32B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Qwen/Qwen2.5-32B-Instruct with Docker Model Runner:
```
docker model run hf.co/Qwen/Qwen2.5-32B-Instruct
```

max_window_layers is 70?

by haili-tian - opened Sep 19, 2024

Discussion

haili-tian

Sep 19, 2024

The config.json is just copied from other place, and even did not modify?

"max_window_layers": 70,
"num_hidden_layers": 64,

The total layer is 64, will it take effect to set the max_window_layers to 70?

jklj077

Qwen org Sep 19, 2024

Hi, use_sliding_window, sliding_window, and max_window_layers control how silding window attention (SWA) is applied. If use_sliding_window, and layer index is not greater than max_window_layers, SWA will be used with a context length of sliding_window. However, we don't recommend enabling SWA for Qwen2.5 (or for Qwen2) as they were not trained with SWA in mind, and we have no idea how SWA will perform. You can still enable SWA and max_window_layers=70 is valid, which means all layers will use SWA, but you should adjust the value based on your own tests.

haili-tian

Sep 19, 2024

Hi, use_sliding_window, sliding_window, and max_window_layers control how silding window attention (SWA) is applied. If use_sliding_window, and layer index is not greater than max_window_layers, SWA will be used with a context length of sliding_window. However, we don't recommend enabling SWA for Qwen2.5 (or for Qwen2) as they were not trained with SWA in mind, and we have no idea how SWA will perform. You can still enable SWA and max_window_layers=70 is valid, which means all layers will use SWA, but you should adjust the value based on your own tests.

got it.

haili-tian changed discussion status to closed Sep 19, 2024

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