# Gemma 2

{% hint style="info" %}
**Newer version available!** Google released [**Gemma 3**](https://docs.clore.ai/guides/language-models/gemma3) in March 2025 — the 27B model beats Llama 3.1 405B and adds native multimodal support. Consider upgrading.
{% endhint %}

Run Google's Gemma 2 models for efficient inference.

{% hint style="success" %}
All examples can be run on GPU servers rented through [CLORE.AI Marketplace](https://clore.ai/marketplace).
{% endhint %}

## Renting on CLORE.AI

1. Visit [CLORE.AI Marketplace](https://clore.ai/marketplace)
2. Filter by GPU type, VRAM, and price
3. Choose **On-Demand** (fixed rate) or **Spot** (bid price)
4. Configure your order:
   * Select Docker image
   * Set ports (TCP for SSH, HTTP for web UIs)
   * Add environment variables if needed
   * Enter startup command
5. Select payment: **CLORE**, **BTC**, or **USDT/USDC**
6. Create order and wait for deployment

### Access Your Server

* Find connection details in **My Orders**
* Web interfaces: Use the HTTP port URL
* SSH: `ssh -p <port> root@<proxy-address>`

## What is Gemma 2?

Gemma 2 from Google offers:

* Models from 2B to 27B parameters
* Excellent performance per size
* Strong instruction following
* Efficient architecture

## Model Variants

| Model       | Parameters | VRAM | Context |
| ----------- | ---------- | ---- | ------- |
| Gemma-2-2B  | 2B         | 3GB  | 8K      |
| Gemma-2-9B  | 9B         | 12GB | 8K      |
| Gemma-2-27B | 27B        | 32GB | 8K      |

## Quick Deploy

**Docker Image:**

```
pytorch/pytorch:2.5.1-cuda12.4-cudnn9-runtime
```

**Ports:**

```
22/tcp
8000/http
```

**Command:**

```bash
pip install vllm && \
vllm serve google/gemma-2-9b-it --port 8000
```

## Accessing Your Service

After deployment, find your `http_pub` URL in **My Orders**:

1. Go to **My Orders** page
2. Click on your order
3. Find the `http_pub` URL (e.g., `abc123.clorecloud.net`)

Use `https://YOUR_HTTP_PUB_URL` instead of `localhost` in examples below.

## Using Ollama

```bash

# Run Gemma 2
ollama run gemma2

# Specific sizes
ollama run gemma2:2b
ollama run gemma2:9b
ollama run gemma2:27b
```

## Installation

```bash
pip install transformers accelerate torch
```

## Basic Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "google/gemma-2-9b-it"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

messages = [
    {"role": "user", "content": "Explain how neural networks learn."}
]

inputs = tokenizer.apply_chat_template(
    messages,
    return_tensors="pt",
    add_generation_prompt=True
).to("cuda")

outputs = model.generate(
    inputs,
    max_new_tokens=512,
    temperature=0.7,
    do_sample=True
)

response = tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)
print(response)
```

## Gemma 2 2B (Lightweight)

For edge/mobile deployment:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "google/gemma-2-2b-it"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

# Fast inference for simple tasks
messages = [{"role": "user", "content": "Summarize in one sentence: AI is transforming industries."}]
```

## Gemma 2 27B (Best Quality)

```python
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch

model_id = "google/gemma-2-27b-it"

# Use 4-bit to fit in 24GB VRAM
quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16
)

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=quantization_config,
    device_map="auto"
)
```

## vLLM Server

```bash
vllm serve google/gemma-2-9b-it \
    --port 8000 \
    --dtype bfloat16 \
    --max-model-len 8192
```

### OpenAI-Compatible API

```python
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="x")

response = client.chat.completions.create(
    model="google/gemma-2-9b-it",
    messages=[
        {"role": "user", "content": "Write a haiku about programming"}
    ],
    temperature=0.8
)

print(response.choices[0].message.content)
```

## Streaming

```python
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="x")

stream = client.chat.completions.create(
    model="google/gemma-2-9b-it",
    messages=[{"role": "user", "content": "Tell me a short story"}],
    stream=True
)

for chunk in stream:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
```

## Gradio Interface

```python
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "google/gemma-2-9b-it"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype=torch.bfloat16, device_map="auto"
)

def chat(message, history, temperature):
    messages = []
    for h in history:
        messages.append({"role": "user", "content": h[0]})
        messages.append({"role": "assistant", "content": h[1]})
    messages.append({"role": "user", "content": message})

    inputs = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda")
    outputs = model.generate(inputs, max_new_tokens=512, temperature=temperature, do_sample=True)

    return tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True)

demo = gr.ChatInterface(
    fn=chat,
    additional_inputs=[gr.Slider(0.1, 1.5, value=0.7, label="Temperature")],
    title="Gemma 2 Chat"
)

demo.launch(server_name="0.0.0.0", server_port=7860)
```

## Batch Processing

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "google/gemma-2-9b-it"
tokenizer = AutoTokenizer.from_pretrained(model_id, padding_side="left")
tokenizer.pad_token = tokenizer.eos_token

model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype=torch.bfloat16, device_map="auto"
)

prompts = [
    "Explain gravity in one sentence.",
    "What is photosynthesis?",
    "Define machine learning.",
    "What is the speed of light?"
]

messages_batch = [[{"role": "user", "content": p}] for p in prompts]

inputs = tokenizer.apply_chat_template(
    messages_batch,
    return_tensors="pt",
    padding=True,
    add_generation_prompt=True
).to("cuda")

outputs = model.generate(inputs, max_new_tokens=128, pad_token_id=tokenizer.pad_token_id)

for i, output in enumerate(outputs):
    response = tokenizer.decode(output, skip_special_tokens=True)
    print(f"Q: {prompts[i]}")
    print(f"A: {response.split('<start_of_turn>model')[-1].strip()}\n")
```

## Performance

| Model               | GPU      | Tokens/sec |
| ------------------- | -------- | ---------- |
| Gemma-2-2B          | RTX 3060 | \~100      |
| Gemma-2-9B          | RTX 3090 | \~60       |
| Gemma-2-9B          | RTX 4090 | \~85       |
| Gemma-2-27B         | A100     | \~45       |
| Gemma-2-27B (4-bit) | RTX 4090 | \~30       |

## Comparison

| Model        | MMLU  | Quality | Speed |
| ------------ | ----- | ------- | ----- |
| Gemma-2-9B   | 71.3% | Great   | Fast  |
| Llama-3.1-8B | 69.4% | Good    | Fast  |
| Mistral-7B   | 62.5% | Good    | Fast  |

## Troubleshooting

{% hint style="danger" %}
**CUDA out of memory**
{% endhint %}

for 27B - Use 4-bit quantization with BitsAndBytesConfig - Reduce \`max\_new\_tokens\` - Clear GPU cache: \`torch.cuda.empty\_cache()\`

### Slow generation

* Use vLLM for production deployment
* Enable Flash Attention
* Try 9B model for faster inference

### Output quality issues

* Use instruction-tuned version (`-it` suffix)
* Adjust temperature (0.7-0.9 recommended)
* Add system prompt for context

### Tokenizer warnings

* Update transformers to latest version
* Use `padding_side="left"` for batch inference

## Cost Estimate

Typical CLORE.AI marketplace rates (as of 2024):

| GPU       | Hourly Rate | Daily Rate | 4-Hour Session |
| --------- | ----------- | ---------- | -------------- |
| RTX 3060  | \~$0.03     | \~$0.70    | \~$0.12        |
| RTX 3090  | \~$0.06     | \~$1.50    | \~$0.25        |
| RTX 4090  | \~$0.10     | \~$2.30    | \~$0.40        |
| A100 40GB | \~$0.17     | \~$4.00    | \~$0.70        |
| A100 80GB | \~$0.25     | \~$6.00    | \~$1.00        |

*Prices vary by provider and demand. Check* [*CLORE.AI Marketplace*](https://clore.ai/marketplace) *for current rates.*

**Save money:**

* Use **Spot** market for flexible workloads (often 30-50% cheaper)
* Pay with **CLORE** tokens
* Compare prices across different providers

## Next Steps

* Llama 3.2 - Meta's model
* Qwen2.5 - Alibaba's model
* vLLM Inference - Production serving