Currently, if a model fails to load, the provider is marked as failed. Instead, the combination should be marked as failed and the next item in the failover chain should be tried. This provides better support for e.g. Google AI Studio models with individual rate limits.
Currently, if a model fails to load, the provider is marked as failed. Instead, the combination should be marked as failed and the next item in the failover chain should be tried. This provides better support for e.g. Google AI Studio models with individual rate limits.