Cloud Platform Support¶

This playbook supports multiple LLM cloud providers. Learn parameters locally, then test on your preferred cloud.

Supported Platforms¶

Azure OpenAI – Microsoft's managed OpenAI service
OpenAI – Direct OpenAI API access
Google Vertex AI – Google's generative AI platform
Amazon Bedrock – AWS managed LLM service

All platforms support the core parameters documented here. Implementation details vary slightly per platform.

Parameter Equivalence¶

Concept	Azure OpenAI	OpenAI	Vertex AI	Bedrock
Temperature	`temperature` (0-2)	`temperature` (0-2)	`temperature` (0-2)	`temperature` (0-1)
Top-p	`top_p` (0-1)	`top_p` (0-1)	`top_p` (0-1)	`topP` (0-1)
Top-k	(via `top_p`)	(via `top_p`)	`top_k`	`topK`
Max tokens	`max_completion_tokens`	`max_tokens`	`max_output_tokens`	`max_tokens`
Stop sequences	`stop`	`stop`	`stop_sequences`	`stop_sequences`
Frequency penalty	`frequency_penalty`	`frequency_penalty`	(not directly)	(not directly)
Presence penalty	`presence_penalty`	`presence_penalty`	(not directly)	(not directly)
Seed/determinism	`seed`	`seed`	`seed`	(limited)

Quick Setup by Platform¶

Azure OpenAI¶

# Set environment variables
export AZURE_OPENAI_ENDPOINT=https://<name>.openai.azure.com/
export AZURE_OPENAI_API_KEY=<your-key>
export AZURE_OPENAI_DEPLOYMENT=gpt-4

# Run experiment
python3 scripts/cloud/azure_sweep.py --preset summarization --trials 3

Docs: Azure OpenAI Setup

OpenAI¶

# Set environment variables
export OPENAI_API_KEY=<your-key>
export OPENAI_MODEL=gpt-4

# Run experiment
python3 scripts/cloud/openai_sweep.py --preset summarization --trials 3

Docs: OpenAI Setup

Google Vertex AI¶

# Authenticate with Google Cloud
gcloud auth application-default login

# Set environment variables
export GOOGLE_PROJECT_ID=<your-project>
export GOOGLE_REGION=us-central1
export VERTEX_MODEL=gemini-pro

# Run experiment
python3 scripts/cloud/vertex_sweep.py --preset summarization --trials 3

Docs: Google Vertex AI Setup

Amazon Bedrock¶

# Configure AWS credentials (use aws configure or set env vars)
export AWS_REGION=us-east-1
export BEDROCK_MODEL=anthropic.claude-3-sonnet-20240229-v1:0

# Run experiment
python3 scripts/cloud/bedrock_sweep.py --preset summarization --trials 3

Docs: Amazon Bedrock Setup

Which Platform Should I Use?¶

Need	Best Option
Cheapest startup	OpenAI API (pay-as-you-go)
Enterprise+compliance	Azure OpenAI (SOC2, FedRAMP)
Google ecosystem	Vertex AI (integrates with GCP)
AWS ecosystem	Bedrock (integrates with AWS)
Variety of models	Bedrock (Claude, Llama, Mistral)
Cutting-edge models	OpenAI (GPT-4 Turbo, o1)

See also: Parameter Equivalence Guide