[Spheron] Add Spheron cloud provider support

[rekpero]

This PR adds support for Spheron as a new cloud provider in SkyPilot. Spheron is a unified GPU cloud platform that aggregates enterprise-grade GPU compute from certified global data centers into a single platform — providing on-demand access to NVIDIA GPUs (H100, A100, B200, B300, and more) across multiple providers with no lock-in, pay-per-hour pricing, and costs 40–60% below major hyperscalers.

We currently have an open PR for our catalog here.

YAML Config Example

Use cheapest GPU across all providers in Spheron's marketplace

resources:
    infra: spheron

Look across all providers that have a specific accelerator and use the cheapest one

resources:
    infra: spheron
    accelerators: A100:1

Use a specific instance type (offer ID) from Spheron

resources:
    infra: spheron
    accelerators: A100:1
    instance_type: <spheron_offer_id>

NOTE: accelerators is optional when instance_type is specified.

Setup

1. Get Spheron API Key: Sign up at https://app.spheron.ai and obtain an API key.
2. Configure credentials:

   mkdir ~/.spheron
   echo "your-api-key-here" > ~/.spheron/api_key

3. Fetch the catalog:

   python sky/catalog/data_fetchers/fetch_spheron.py

4. Verify setup:

   sky check

I have ran a lot of local tests across different Spheron GPU instance types. Note that code formatting has only been fixed for Spheron-related changes — there are pre-existing formatting issues in other parts of the codebase that have not been touched.

Tested

• Code formatting: bash format.sh (Spheron-related files only; pre-existing issues in other files left untouched)
• Unit tests: pytest tests/unit_tests/test_spheron.py
• Manual tests: launched clusters across different Spheron GPU instance types locally
• All smoke tests: /smoke-test (CI) or pytest tests/test_smoke.py (local)
• Relevant individual tests: /smoke-test -k test_name (CI) or pytest tests/test_smoke.py::test_name (local)
• Backward compatibility: /quicktest-core (CI) or pytest tests/smoke_tests/test_backward_compat.py (local)

[gemini-code-assist]

Code Review

This pull request introduces support for Spheron Cloud as a new provider in SkyPilot. The implementation includes a new cloud adaptor, authentication setup for SSH keys, catalog fetching scripts, and provisioning logic for instance management. It also integrates Spheron with the Ray backend using a dedicated template and updates the test suite to include Spheron-specific configurations. Feedback was provided to improve the robustness of the catalog fetching script by replacing a debug assertion with an explicit exception for missing API keys.

[gemini-code-assist]

The assert statement here is typically used for debugging and can be optimized out in production. For robust error handling, it's better to explicitly raise a ValueError or FileNotFoundError if the API key is not found or is empty. This aligns with the error handling in sky/provision/spheron/spheron_utils.py.

Suggested change

	assert api_key is not None, (
	f'API key not found. Please provide via --api-key or place in '
	f'{DEFAULT_SPHERON_API_KEY_PATH}')
	if api_key is None:
	raise ValueError(
	f'API key not found. Please provide via --api-key or place in '
	f'{DEFAULT_SPHERON_API_KEY_PATH}')

[rekpero]

@Michaelvll @aylei can you please check this PR