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

Libraries

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

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

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

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

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

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "openbmb/MiniCPM5-2B"
# 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",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

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

SGLang

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

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

Add KV-BSS (Key-Value Binding Softmax Sharpening) attention hook

#11

by F-Labs - opened 7 days ago

base: refs/heads/main

←

from: refs/pr/11

Discussion Files changed

+920

-0

F-Labs

7 days ago

KV-BSS: Key-Value Binding Softmax Sharpening for 128K Context Inference

This Pull Request adds the standalone KV-BSS (Key-Value Binding Softmax Sharpening) attention hook module for MiniCPM5-2B.

Background & Motivation

In ultra-long context inference (up to 131,072 tokens), standard attention logits suffer from entropy dispersion (attention haze), which reduces associative retrieval accuracy on structured bindings (e.g. ["key"] => "value", AST parsing, schema extraction).

What KV-BSS does:

Focus Factor Scaling (τ_focus = 1.10): Steeping attention logits concentrates probability mass around exact antecedent keys.
Attention Haze Floor Filtering: Masks logits falling more than 12.0 nats below the maximum attention score to -inf, eliminating associative noise.

Full compressed model release: F-Labs/MiniCPM5-2B-Hadamard-GSQ
Core framework: GitHub: dsadawq3/FQuant

Add KV-BSS (Key-Value Binding Softmax Sharpening) attention hook752d3173

Add complete production modeling, configuration, test suite, and integration documentationdcea2e53

Update to RoPE+KV-cache modeling (LF normalization, GenerationMixin, honest test docs)ff0399b8

Harden KV-BSS masking and add numerical cache tests954c497a

Harden KV-BSS masking and add numerical cache testsac793215

xcjthu

OpenBMB org 3 days ago

Thank you for the contribution and for the detailed explanation of KV-BSS.

The idea looks interesting, especially for reducing attention haze and improving associative retrieval in ultra-long-context inference. For this functionality, we would currently prefer to rely on your own repository as the recommended way to use and maintain KV-BSS, rather than merging the standalone hook module directly into this repo.

Thanks again for sharing this work and for contributing to the community.

xcjthu changed pull request status to closed 3 days ago

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