Instructions to use google/shieldgemma-2b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use google/shieldgemma-2b with Transformers:

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

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

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("google/shieldgemma-2b")
model = AutoModelForCausalLM.from_pretrained("google/shieldgemma-2b")
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
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use google/shieldgemma-2b with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/google/shieldgemma-2b

SGLang

How to use google/shieldgemma-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 "google/shieldgemma-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": "google/shieldgemma-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 "google/shieldgemma-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": "google/shieldgemma-2b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use google/shieldgemma-2b with Docker Model Runner:
```
docker model run hf.co/google/shieldgemma-2b
```

Suggestion: Better KV cache usage

by ngxson - opened Jul 31, 2024

Discussion

ngxson

Jul 31, 2024

Hello and thanks for the great work!

I had a look on the prompt format for this model, but I'm wondering if we can re-arrange the format, so that some LLM inference engine can take advantage of KV cache.

For example, with the current format, only this part of the prompt is fixed, and thus can be re-used across multiple requests:

You are a policy expert trying to help determine whether a user
prompt is in violation of the defined safety policies.

<start_of_turn>
Human Question:

Instead of putting user message to the beginning, could we for example put it to the bottom:

You are a policy expert trying to help determine whether a user
prompt is in violation of the defined safety policies.

Our safety principle is defined in the below:

{safety_policy.strip()}

Does the human question violate the above principle? Your answer must start
with 'Yes' or 'No'. And then walk through step by step to be sure we answer
correctly.

<start_of_turn>
Human Question: {user_prompt.strip()}
<end_of_turn>

Your answer:

So that all the tokens from the beginning until <start_of_turn> can be re-used across multiple requests.

Xenova

Jul 31, 2024

@RyanMullins will be able to answer :)

RyanMullins

Google org Aug 1, 2024

Thanks for the suggestion @ngxson ! I'm going to talk to @fgtnsky about this today and we'll get back to you.

WillBankes

Aug 29, 2024

@RyanMullins are there any further updates on this? Thanks!

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