WORKFLOW TEMPLATES

Pre-Built Workflow Templates

Get started quickly with templates for common GPU compute patterns. Each workflow includes step-by-step guides, code examples, and cost estimates.

Batch RenderingAI TrainingReal-time InferenceScientific SimulationZK Proof GenerationHybrid

Batch Rendering Workflow

Render 3D scenes, animations, and visual effects across distributed GPUs.

How It Works

  1. 1Upload project files (.blend, .ma, scene)
  2. 2Configure render settings (resolution, format, frames)
  3. 3Jobs distributed to available GPU nodes
  4. 4Frames rendered in parallel with progress tracking
  5. 5Results verified and delivered with proof

Use Cases

Feature film VFXAnimation studiosArchitectural visualization
Estimated cost:$0.35/GPU-hr

Quick Start

Python
from bitsage import BatchJob

# Create a rendering job
job = BatchJob.create(
    type="blender",
    project="animation.blend",
    frames="1-240",
    output_format="exr",
    gpu_type="RTX_4090"
)

# Submit and monitor
job.submit()
job.wait()

# Download results
job.download("./renders/")

AI Training Workflow

Distributed model training with checkpointing and verification.

How It Works

  1. 1Upload training script and dataset
  2. 2Configure distributed training parameters
  3. 3Training distributed across GPU cluster
  4. 4Checkpoints saved and verified
  5. 5Final model delivered with training metrics

Use Cases

LLM fine-tuningComputer visionReinforcement learning
Estimated cost:$3.20/GPU-hr

Quick Start

Python
from bitsage import TrainingJob

# Configure distributed training
job = TrainingJob.create(
    script="train.py",
    dataset="s3://bucket/data",
    gpus=8,
    gpu_type="H100",
    checkpoint_interval=1000
)

# Start training
job.submit()

# Monitor progress
for update in job.stream():
    print(f"Step {update.step}: loss={update.loss}")

Real-time Inference Workflow

Deploy models for low-latency serving with auto-scaling.

How It Works

  1. 1Upload model artifacts (ONNX, PyTorch, TensorFlow)
  2. 2Configure serving parameters and scaling
  3. 3Model deployed to edge regions
  4. 4Auto-scaling based on request volume
  5. 5Pay only for actual inference time

Use Cases

Chatbots & assistantsContent generationReal-time translation
Estimated cost:$0.50/1M tokens

Quick Start

Python
from bitsage import Inference

# Deploy a model
deployment = Inference.deploy(
    model="meta-llama/Llama-3-70B",
    gpu="H100",
    min_replicas=1,
    max_replicas=10,
    regions=["us-east", "eu-west"]
)

# Make requests
response = deployment.generate(
    prompt="Explain quantum computing",
    max_tokens=500
)

Scientific Simulation Workflow

Run molecular dynamics, climate models, and physics simulations.

How It Works

  1. 1Upload simulation configuration and input files
  2. 2Configure compute resources and duration
  3. 3Simulation runs on verified GPU cluster
  4. 4Intermediate results checkpointed
  5. 5Final results verified and delivered

Use Cases

Drug discoveryMaterials scienceClimate modeling
Estimated cost:$1.80/GPU-hr

Quick Start

Python
from bitsage import SimulationJob

# Configure molecular dynamics simulation
job = SimulationJob.create(
    type="gromacs",
    config="protein.gro",
    topology="topol.top",
    steps=10000000,
    gpus=4,
    gpu_type="A100"
)

# Submit and monitor
job.submit()
trajectory = job.wait()

# Analyze results
trajectory.download("./results/")

ZK Proof Generation Workflow

Generate zero-knowledge proofs with GPU acceleration.

How It Works

  1. 1Submit computation witness and circuit
  2. 2Proof generation distributed across GPUs
  3. 3Proofs verified on-chain
  4. 4Results delivered with attestation

Use Cases

Private transactionsRollup provingIdentity verification
Estimated cost:$0.35/GPU-hr

Quick Start

Python
from bitsage import ZKJob

# Generate ZK proof
job = ZKJob.create(
    circuit="transfer.r1cs",
    witness="witness.wtns",
    proving_key="proving.key",
    gpu_type="RTX_4090"
)

# Generate proof
proof = job.generate()

# Verify on-chain
tx = proof.verify_on_chain(network="starknet")

Hybrid Workflow

Combine multiple workflow types in a single pipeline.

How It Works

  1. 1Define pipeline stages (training → inference → verification)
  2. 2Configure dependencies and data flow
  3. 3Pipeline executes across workflow types
  4. 4Results from each stage verified
  5. 5Final output delivered with full provenance

Use Cases

ML pipelinesData processingComplex workflows
Estimated cost:Varies by stage

Quick Start

Python
from bitsage import Pipeline

# Define hybrid pipeline
pipeline = Pipeline([
    # Stage 1: Train model
    TrainingJob(script="train.py", gpus=4),

    # Stage 2: Deploy for inference
    InferenceDeployment(model="${stage1.output}"),

    # Stage 3: Generate proofs
    ZKJob(witness="${stage2.output}")
])

# Execute pipeline
results = pipeline.run()

Ready to Build?

Check out our documentation for detailed guides, API references, and SDK downloads.

Get StartedView Full Docs