Instructions to use google/gemma-2-27b-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use google/gemma-2-27b-it with Transformers:

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

pipe = pipeline("text-generation", model="google/gemma-2-27b-it")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-27b-it")
model = AutoModelForCausalLM.from_pretrained("google/gemma-2-27b-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]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use google/gemma-2-27b-it with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "google/gemma-2-27b-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/gemma-2-27b-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/google/gemma-2-27b-it

SGLang

How to use google/gemma-2-27b-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/gemma-2-27b-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/gemma-2-27b-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/gemma-2-27b-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/gemma-2-27b-it",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use google/gemma-2-27b-it with Docker Model Runner:
```
docker model run hf.co/google/gemma-2-27b-it
```

JAX/Flax Implementation

#13

by lemon-mint - opened Jun 30, 2024

Discussion

lemon-mint

Jun 30, 2024

DeepMind's Gemma implementation does not seem to have been updated in accordance with the new release.

Are there any plans to release the JAX/Flax implementation and model?

lemon-mint changed discussion title from JAX/Flax implementation to JAX/Flax Implementation Jun 30, 2024

canyon289

Jul 1, 2024

There is! Our focus was on getting the weights out properly. For my own curiosity why are you interested in flax/jax in particular?

lemon-mint

Jul 1, 2024

For my own curiosity why are you interested in flax/jax in particular?

I think using TPU is the most cost-effective way to full fine-tune the 27B model.

Additionally, the JAX/Flax implementation is good to use as a reference implementation. Last time, in Gemma 1, DeepMind's implementation was the only one without bugs.

anique

Jul 9, 2024

•

edited Jul 9, 2024

There is! Our focus was on getting the weights out properly. For my own curiosity why are you interested in flax/jax in particular?

@canyon289 This would be very convenient. I want to integrate with our JORA library (Jax centered LLM PEFT finetuning). I believe the only differences from Gemma 1/1.1 are

Logit softcaps,
Sliding Window Attention, and
query normalization

Plus, the weights in Flax format (i.e. orbax.checkpoint)

canyon289

Jul 10, 2024

Thank you both for the answers. There's a couple of other changes such as GQA! Regardless its still being worked on, it should be out soonish. My apologies for the delay

RCrowe

Google org Jul 10, 2024

•

edited Jul 10, 2024

JORA looks interesting! I'd suggest adding a link to the paper in the readme.

canyon289

Jul 24, 2024

We haven't forgotten about this. We're making some final changes and its on its way to release

RCrowe

Google org Jul 24, 2024

I'd also suggest sending a PR to add it to https://github.com/n2cholas/awesome-jax

canyon289

Jul 30, 2024

•

edited Jul 30, 2024

Its updated! Check out it folks. Hope you enjoy the models

beomi

Sep 9, 2024

@canyon289 Hi, could you check where the implementation with jax/flax of the model? I couldn't find python code related with gemma 2 implementation, rather, there are only weight files on Kaggle.

gusthema

Google org Sep 9, 2024

The official JAX repo has the configurations for Gemma 2: https://github.com/google-deepmind/gemma/blob/main/gemma/transformer.py

lkv

Google org Jan 22, 2025

This comment has been hidden

xujfcn

Feb 24

For those asking about API access — I've been using Crazyrouter as a unified gateway. One API key, OpenAI SDK compatible. Works well for testing different models without managing multiple accounts.

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