### On-Device Inference with transformers.js

Source: https://github.com/google-gemma/gemma-skills/blob/main/skills/gemma-dev/SKILL.md

Run inference directly on-device or in the browser using transformers.js. Ensure you install the Hugging Face version (`npm i @huggingface/transformers`).

```javascript
import { AutoTokenizer, AutoModelForCausalLM } from '@huggingface/transformers';

const modelName = 'google/gemma-2b-it';

async function runInference() {
    const tokenizer = await AutoTokenizer.from_pretrained(modelName);
    const model = await AutoModelForCausalLM.from_pretrained(modelName);

    const prompt = "Hello, Gemma!";
    const inputIds = tokenizer(prompt, { returnTensors: 'pt' }).input_ids;
    const outputs = await model.generate(inputIds, { maxNewTokens: 50 });
    const result = tokenizer.decode(outputs[0], { skipSpecialTokens: true });
    console.log(result);
}

runInference();
```

--------------------------------

### Rapid Prototyping with Gradio and Transformers

Source: https://github.com/google-gemma/gemma-skills/blob/main/skills/gemma-dev/SKILL.md

Use Gradio and Transformers for rapid, interactive UI prototyping in Python. Follow the best practices outlined in the assets/gradio-app.py.

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

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

def predict(prompt):
    input_ids = tokenizer.encode(prompt, return_tensors='pt')
    outputs = model.generate(input_ids, max_new_tokens=100)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

iface = gr.Interface(fn=predict, inputs="text", outputs="text")
iface.launch()
```

--------------------------------

### Enterprise Cloud Deployment with Vertex AI

Source: https://github.com/google-gemma/gemma-skills/blob/main/skills/gemma-dev/SKILL.md

Use Vertex AI for containerized, scalable, cloud-native production deployments. This requires setting specific Google Cloud environment variables.

```python
import google.cloud.aiplatform as aiplatform

# Set environment variables before running:
# GOOGLE_CLOUD_PROJECT='your-project-id'
# GOOGLE_CLOUD_LOCATION='us-central1'
# GOOGLE_CLOUD_ENDPOINT_ID='your-endpoint-id'

# Initialize Vertex AI SDK
aiplatform.init()

# Example: Deploying a model (details depend on specific model and deployment)
# endpoint = aiplatform.Endpoint("projects/your-project-id/locations/us-central1/endpoints/your-endpoint-id")
# prediction = endpoint.predict(instances=[{"prompt": "Your input prompt here"}])
# print(prediction.predictions)
```

--------------------------------

### Generate Embeddings with EmbeddingGemma

Source: https://github.com/google-gemma/gemma-skills/blob/main/skills/gemma-dev/SKILL.md

Use EmbeddingGemma for RAG and Vector Search workflows. Fetch the recommended best practices for generating embeddings.

```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer('google/embeddinggemma-300m')

# Example usage:
text = "This is a sample sentence."
embeddings = model.encode(text)
print(embeddings)
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.