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

Libraries

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

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

pipe = pipeline("text-generation", model="openbmb/MiniCPM5-2B-DSpark", trust_remote_code=True)

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

tokenizer = AutoTokenizer.from_pretrained("openbmb/MiniCPM5-2B-DSpark", trust_remote_code=True)
model = AutoModel.from_pretrained("openbmb/MiniCPM5-2B-DSpark", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use openbmb/MiniCPM5-2B-DSpark with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "openbmb/MiniCPM5-2B-DSpark"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/MiniCPM5-2B-DSpark",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

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

SGLang

How to use openbmb/MiniCPM5-2B-DSpark 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-DSpark" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/MiniCPM5-2B-DSpark",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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-DSpark" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/MiniCPM5-2B-DSpark",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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

What is the difference between T0 and T1, and why is there such a large discrepancy in acceptance rate?

by sunny-infra - opened 25 days ago

Discussion

sunny-infra

25 days ago

•

edited 25 days ago

In Evaluation, what is the difference between T=0 and T=1, and why is there such a large discrepancy in acceptance rate?

zhiyaowang

25 days ago

T denotes the sampling temperature. T=0 uses greedy decoding; T=1.0 uses sampling at temperature 1.0. Both use the same draft checkpoint and preserve the target's output distribution under the respective decoding settings.

Under rejection sampling, expected per-token acceptance is:

$\alpha = \sum_x \min(p(x), q(x))$

where p and q are the target and draft distributions. This measures their overlapping probability mass: greater overlap means higher acceptance.

At T=0, all probability mass is concentrated on the top token, so the distributions overlap completely whenever their top choices agree. T=1.0 retains the sampling entropy, with multiple candidates carrying non-negligible probability mass. Even when the models prefer the same token, they can assign different probabilities to the candidates. Rejection sampling corrects these differences by rejecting some proposals, so the same draft can achieve strong greedy agreement yet substantially lower acceptance under sampling.

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