Instructions to use Qwen/Qwen2.5-Math-1.5B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Qwen/Qwen2.5-Math-1.5B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="Qwen/Qwen2.5-Math-1.5B-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-Math-1.5B-Instruct")
model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-Math-1.5B-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-Math-1.5B-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-Math-1.5B-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-Math-1.5B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Qwen/Qwen2.5-Math-1.5B-Instruct

SGLang

How to use Qwen/Qwen2.5-Math-1.5B-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-Math-1.5B-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-Math-1.5B-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-Math-1.5B-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-Math-1.5B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Qwen/Qwen2.5-Math-1.5B-Instruct with Docker Model Runner:
```
docker model run hf.co/Qwen/Qwen2.5-Math-1.5B-Instruct
```

WHERE IS Qwen2.5-Math-1.5B-Instruct.py

by MartialTerran - opened Feb 6, 2025

Discussion

MartialTerran

Feb 6, 2025

In order for these small models to be useful they have to be ported to edge devices, including raspberry pi, analog MatrixMultipliers, and smart phones with minimized "script" overhead and minimized script complexity.
The multiple files have to be reduced to a single Qwen2_model.py having an internal set of hyperparameters (config.json converted to internal hyperparameters dictionary, then ported to a torchscript and pytorch checkpoint)

When these Qwen models are released, the authors should also construct and include in the same "files" section a working Qwen2.5-Math-1.5B-Instruct.py that reliably loads the pretrained weights, loads the config.json etc. and just provides a prompt and delivers a model response (e.g. in CMD console of Windows 10).

It seems that a functioning Qwen2.5-Math-1.5B-Instruct.py for this Qwen model can be derived from this complexified scripts found on an entirely different website (github):

https://raw.githubusercontent.com/huggingface/transformers/refs/heads/main/src/transformers/models/qwen2/modeling_qwen2.py

https://raw.githubusercontent.com/huggingface/transformers/refs/heads/main/src/transformers/models/qwen2/configuration_qwen2.py

https://raw.githubusercontent.com/huggingface/transformers/refs/heads/main/src/transformers/models/qwen2/__init__.py

I might attempt to use Generative AI to filter out all the useless complexity to distill these fragmented code scripts into a functional Qwen2_model.py It is so burdensom to try to reconstruct a working consolidated Qwen2.5-Math-1.5B-Instruct.py from all these complexified fragments.

I am here to tell you that it does not make any sense at all to publish a set of weights for Qwen2.5-Math-1.5B-Instruct without also publishing a simple standalone python script Qwen2.5-Math-1.5B-Instruct.py that will reliably load the weights and deliver inference performance, and supporting user optimizations and hardware adaptations.

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