Instructions to use google/recurrentgemma-9b-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use google/recurrentgemma-9b-it with Transformers:

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

pipe = pipeline("text-generation", model="google/recurrentgemma-9b-it")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("google/recurrentgemma-9b-it")
model = AutoModelForCausalLM.from_pretrained("google/recurrentgemma-9b-it")
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 google/recurrentgemma-9b-it with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "google/recurrentgemma-9b-it"
# 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/recurrentgemma-9b-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/google/recurrentgemma-9b-it

SGLang

How to use google/recurrentgemma-9b-it 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/recurrentgemma-9b-it" \
    --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/recurrentgemma-9b-it",
		"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/recurrentgemma-9b-it" \
        --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/recurrentgemma-9b-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use google/recurrentgemma-9b-it with Docker Model Runner:
```
docker model run hf.co/google/recurrentgemma-9b-it
```

Evaluation Result

#15

by tanliboy - opened Aug 23, 2024

Discussion

tanliboy

Aug 23, 2024

I conducted an evaluation of the model, and while the inference speed is impressive, I was unable to replicate the performance results reported in the paper. Below are the results I got:

Groups	Version	Filter	Metric		Value		Stderr
mmlu	1	none	acc	↑	0.5486	±	0.0040
- humanities	1	none	acc	↑	0.4997	±	0.0069
- other	1	none	acc	↑	0.6308	±	0.0083
- social sciences	1	none	acc	↑	0.6406	±	0.0085
- stem	1	none	acc	↑	0.4510	±	0.0086

Tasks	Version	Filter	n-shot	Metric		Value		Stderr
hellaswag	1	none	0	acc	↑	0.6188	±	0.0048
		none	0	acc_norm	↑	0.8026	±	0.0040

The difference might be due to differences in evaluation settings. Overall, the model's performance seems outdated compared to the latest models. Do we have any plans to release an updated version on Griffin architecture?

lkv

Google org Feb 6, 2025

Hi @tanliboy , As for your question about releasing an updated version of the Griffin architecture, I currently don’t have direct information regarding upcoming releases or updates to this specific architecture.
If possible, Kindly try fine-tuning Griffin on the specific tasks you are working on. Fine-tuning can lead to significant improvements over the base model.

Thank you.

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