When running the LiteLLM proxy server with Fireworks AI models configured, the proxy's /v1/models endpoint triggers an internal call to fetch available models from Fireworks AI. This call constructs a malformed URL with a duplicated /v1 path segment, resulting in a 404 error. LiteLLM:WARNING: utils.py:7202 - Error getting valid models: Failed to fetch models from Fireworks AI. Status code: 404, Response: {'error': {'message': 'Path not found: /v1/v1/accounts/fireworks/models', 'param': None, 'code': 'NOT_FOUND', 'type': 'error'}}

Fix Plan

To fix the issue, we need to remove the duplicated /v1 path segment from the URL constructed in the get_models method.

• Update the get_models method in litellm/llms/fireworks_ai/chat/transformation.py to remove the extra /v1:

url = f"{api_base}/accounts/{account_id}/models"

Alternatively, you can use the urllib.parse.urljoin function to construct the URL, which will handle the path joining correctly:

import urllib.parse

url = urllib.parse.urljoin(api_base, f"/accounts/{account_id}/models")

Verification

To verify that the fix worked, follow these steps:

• Restart the LiteLLM proxy server
• Request the models endpoint again:

curl http://localhost:4000/v1/models

• Check the proxy server logs for any warnings or errors related to the duplicated /v1 path segment. The warning should be gone, and the models should be fetched successfully.

Extra Tips

• Make sure to test the fix thoroughly to ensure that it doesn't introduce any new issues.
• Consider adding a test case to cover this scenario to prevent regressions in the future.