Spaces:
Running
Running
| """What can go wrong on the way out to the model. | |
| The distinction that matters is between an error another key can fix and one it cannot. Quota is | |
| the only error worth rotating on: a bad model name or a dead network burns all five keys and then | |
| reports the wrong problem. | |
| """ | |
| class LlmError(Exception): | |
| """Base for everything in this module.""" | |
| class NoKeyConfigured(LlmError): | |
| def __init__(self, provider: str = "gemini") -> None: | |
| super().__init__(f"No API key configured for {provider}") | |
| self.provider = provider | |
| class QuotaExhausted(LlmError): | |
| """429 / RESOURCE_EXHAUSTED. The one error that rotates to the next key.""" | |
| def __init__(self, status: int, message: str) -> None: | |
| super().__init__(f"{status}: {message}") | |
| self.status = status | |
| class ModelUnavailable(LlmError): | |
| """404 / NOT_FOUND on the model. Falls through to the next model without burning keys.""" | |
| def __init__(self, status: int, message: str) -> None: | |
| super().__init__(f"{status}: {message}") | |
| self.status = status | |
| class Rejected(LlmError): | |
| """Anything else the API said no to. Fails fast.""" | |
| def __init__(self, status: int, message: str) -> None: | |
| super().__init__(f"{status}: {message}") | |
| self.status = status | |
| class AllKeysExhausted(LlmError): | |
| def __init__(self, count: int) -> None: | |
| super().__init__(f"All {count} configured keys are out of quota") | |
| self.count = count | |
| class EmptyReply(LlmError): | |
| """The call succeeded and the model returned nothing usable.""" | |