Tetra: Serverless GPU Computing for AI Workloads

Dynamic GPU provisioning for ML workloads with transparent execution

Overview • Installation • Quick Start • Key Features • Examples • Configuration • Troubleshooting

---

Overview

The Tetra-RunPod integration provides seamless access to on-demand GPU resources through RunPod's serverless platform. With a simple decorator-based API, you can execute functions on powerful GPUs without managing infrastructure, while Tetra handles all the complexity of provisioning, communication, and state management.

Installation

git clone https://github.com/runpod/tetra-rp
cd tetra-rp 
pip install -r requirements.txt

You'll need a RunPod API key to use this integration. Sign up at RunPod.io and generate an API key in your account settings. set it in ENV or save it in a local .env file:

export RUNPOD_API_KEY=<YOUR_API_KEY>

Quick Start

import os
import asyncio
from tetra import remote, ServerlessResource

# Configure RunPod resource
runpod_config = ServerlessResource(
    name="example-diffusion-server",
)

# Define a function to run on RunPod GPU
@remote(
    resource_config=runpod_config,
    dependencies=["torch", "numpy"]
)
def gpu_compute(data):
    import torch
    import numpy as np
    
    # Convert to tensor and perform computation on GPU
    tensor = torch.tensor(data, device="cuda")
    result = tensor.sum().item()
    
    # Get GPU info
    gpu_info = torch.cuda.get_device_properties(0)
    
    return {
        "result": result,
        "gpu_name": gpu_info.name,
        "cuda_version": torch.version.cuda
    }

async def main():
    # Run the function on RunPod GPU
    result = await gpu_compute([1, 2, 3, 4, 5])
    print(f"Result: {result['result']}")
    print(f"Computed on: {result['gpu_name']} with CUDA {result['cuda_version']}")

if __name__ == "__main__":
    try:
        asyncio.run(main())
    except Exception as e:
        print(f"An error occurred: {e}")

Key Features

Dynamic GPU Provisioning

Automatically provision GPUs on demand without any manual setup:

@remote(
    resource_config=runpod_config,
)
def my_gpu_function(data):
    # Runs on GPU when called
    return process(data)

Automatic Dependency Management

Specify dependencies you need, which are automatically installed for you:

@remote(
    resource_config=runpod_config,
    dependencies=["torch==2.0.1", "transformers", "diffusers"]
)
def generate_image(prompt):
    # Dependencies are automatically installed
    from diffusers import StableDiffusionPipeline
    # Generate image...
    return image

Examples

Multi-Stage ML Pipeline

# Feature extraction on GPU
@remote(
    resource_config=runpod_config,
    dependencies=["torch", "transformers"]
)
def extract_features(texts):
    import torch
    from transformers import AutoTokenizer, AutoModel
    
    # Load model
    tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
    model = AutoModel.from_pretrained("bert-base-uncased")
    model.to("cuda")
    
    # Process texts
    features = []
    for text in texts:
        inputs = tokenizer(text, return_tensors="pt").to("cuda")
        with torch.no_grad():
            outputs = model(**inputs)
        features.append(outputs.last_hidden_state[:, 0].cpu().numpy().tolist()[0])
    
    return features

# Classification on GPU
@remote(
    resource_config=runpod_config,
    dependencies=["torch", "sklearn"]
)
def classify(features, labels=None):
    import torch
    import numpy as np
    from sklearn.linear_model import LogisticRegression
    
    # Convert to numpy
    features = np.array(features)
    
    if labels is not None:
        # Training mode
        labels = np.array(labels)
        classifier = LogisticRegression()
        classifier.fit(features, labels)
        
        # Save model coefficients (can't pickle sklearn model easily)
        coefficients = {
            "coef": classifier.coef_.tolist(),
            "intercept": classifier.intercept_.tolist(),
            "classes": classifier.classes_.tolist()
        }
        
        return coefficients
    else:
        # Inference mode (assuming coefficients are passed as first element)
        coefficients = features[0]
        actual_features = features[1:]
        
        # Recreate classifier
        classifier = LogisticRegression()
        classifier.coef_ = np.array(coefficients["coef"])
        classifier.intercept_ = np.array(coefficients["intercept"])
        classifier.classes_ = np.array(coefficients["classes"])
        
        # Predict
        predictions = classifier.predict(actual_features)
        probabilities = classifier.predict_proba(actual_features)
        
        return {
            "predictions": predictions.tolist(),
            "probabilities": probabilities.tolist()
        }

# Complete pipeline
async def text_classification_pipeline(train_texts, train_labels, test_texts):
    # Extract features
    train_features = await extract_features(train_texts)
    test_features = await extract_features(test_texts)
    
    # Train classifier
    model = await classify(train_features, train_labels)
    
    # Predict
    predictions = await classify([model] + test_features)
    
    return predictions

Configuration

Configuration Parameters

Parameter	Description	Default	Example Values
name	(Required) Name for your endpoint	""	"stable-diffusion-server"
gpuIds	Type of GPU to request	"any"	"any" or list of GPU IDs (comma-separated)
gpuCount	Number of GPUs per worker	1	1, 2, 4
workersMin	Minimum number of workers	0	Set to 1 for persistence
workersMax	Maximum number of workers	3	Higher for more concurrency
idleTimeout	Minutes before scaling down	5	10, 30, 60
env	Environment variables	None	{"HF_TOKEN": "xyz"}
networkVolumeId	Persistent storage ID	None	"vol_abc123"
executionTimeoutMs	Max execution time (ms)	0 (no limit)	600000 (10 min)
scalerType	Scaling strategy	QUEUE_DELAY	NONE, QUEUE_SIZE
scalerValue	Scaling parameter value	4	1-10 range typical
locations	Preferred datacenter locations	None	"us-east,eu-central"

Examples

See more examples in the ./examples/* folder

---

License

This project is licensed under the MIT License - see the LICENSE file for details.

---

Tetra • RunPod