Instructions to use openbmb/MiniCPM5-2B-MLX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use openbmb/MiniCPM5-2B-MLX with Transformers:

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

pipe = pipeline("text-generation", model="openbmb/MiniCPM5-2B-MLX")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("openbmb/MiniCPM5-2B-MLX")
model = AutoModelForCausalLM.from_pretrained("openbmb/MiniCPM5-2B-MLX", 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 openbmb/MiniCPM5-2B-MLX with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "openbmb/MiniCPM5-2B-MLX"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/MiniCPM5-2B-MLX",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/openbmb/MiniCPM5-2B-MLX

SGLang

How to use openbmb/MiniCPM5-2B-MLX 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 "openbmb/MiniCPM5-2B-MLX" \
    --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": "openbmb/MiniCPM5-2B-MLX",
		"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 "openbmb/MiniCPM5-2B-MLX" \
        --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": "openbmb/MiniCPM5-2B-MLX",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use openbmb/MiniCPM5-2B-MLX with Docker Model Runner:
```
docker model run hf.co/openbmb/MiniCPM5-2B-MLX
```

oMLX 0.6.4 (2529) can't yet handle this model's toolcalls

by hf260814 - opened 9 days ago

Discussion

hf260814

9 days ago

Just a heads up. A trivial query "is rust installed?" generates request like this
data: {... "choices":[{"index":0,"delta":{"content":"\n\n<function name=\"bash\">"}}]}
which is just returned instead of being executed.

For comparison gemma4 generates a request like this
data: {... "choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"id":"call_45b52d0e","type":"function","function":{"name":"bash","arguments"...

See https://github.com/jundot/omlx/issues/3429 and proposed https://github.com/jundot/omlx/pull/3530

OpenBMB-LLM

OpenBMB org 1 day ago

Thanks for the heads up — this has been fixed.

PR jundot/omlx#3530 was merged into main on Sept 16 (issue #3429 now closed). The root cause was that oMLX's tool_call parsers only recognized wrapped dialects like <tool_call>…</tool_call>, so MiniCPM5's attribute-style <function name="bash"><param name="...">…</param></function> output fell through and got streamed back as plain content.

The merge adds _parse_attribute_function_tool_calls (bare form) plus _parse_xml_tool_calls for the wrapped form, handles CDATA-wrapped values, accepts both <param> and <parameter>, and gates parsing on declared tools so prose mentioning a function name isn't misread as a call. Stream filtering also suppresses the markup while leaving undeclared tags alone.

So on an oMLX build newer than 0.6.4 (2529) this should now surface as a proper tool_calls delta instead of content.

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