Thanks for the PR, @JinYue-GitHub! Adding NVIDIA provider support is a useful addition since NVIDIA's NIM API hosts models from multiple namespaces (e.g. qwen/, meta/, etc.) which requires preserving the full model path.

The changes look straightforward:

• Adding nvidia to _PROVIDER_DISPLAY, _PROVIDER_MODELS, _PROVIDER_ALIASES follows the existing provider registration pattern.
• Including nvidia in _PORTAL_PROVIDERS is the key fix for the 404 errors — portal providers pass the full model path through rather than stripping the namespace prefix.
• NVIDIA_API_KEY in _PROVIDER_ENV_VAR enables the standard key management flow.

A couple of questions to round this out:

1. Does the PR include any tests for the NVIDIA provider registration? A test verifying that model paths are preserved (not stripped) for NVIDIA would match the pattern of existing provider tests.
2. What's the base URL for NVIDIA NIM? Is it set via an env var or hardcoded somewhere? Worth confirming the endpoint config is complete.

The fix for the 404 issue is clear and the approach is consistent with how other portal providers are handled. Thanks for the contribution!

Merged in v0.50.237 via #1243. Thank you @JinYue-GitHub! 🎉