Abstract out generate validation and error handling into BaseProvider, eliminating duplicated logic across Anthropic, OpenRouter, and OpenAI providers.
Add token limit error detection and formatting across all provider modules to surface clear diagnostics when context windows are exceeded, improving error messages and preventing silent failures.
New Features
- Introduce `is_token_limit_error` checks in Anthropic, Gemini, Hugging Face, Ollama, OpenAI, and OpenRouter providers.
- Use `format_token_limit_error` to generate user‑friendly error messages for token limit overs.
Enhancements
- Centralize token limit error handling for consistent diagnostics across providers.
- Explicitly re‑raise `RuntimeError` in Hugging Face to preserve original behavior.
Bug Fixes
- None
Enhancements:
- Extend error message parsing to capture additional throttling indicators such as "RESOURCE_EXHAUSTED", "QUOTA_EXCEEDED", and "THROTTLED".
- Refactor the conditional logic for better readability and maintainability.
- Keep existing retry behavior while increasing resilience to API rate‑limit errors.
The update changes the default model, refactors imports, and implements robust retry logic with exponential back‑off and clearer error messages. This improves resilience and clarifies usage.
New Features
- Exponential back‑off retry on 429/quota exhausted responses.
- Configurable max_retries and retry_delay parameters.
Enhancements
- Default model set to gemini-2.5-flash.
- Improved error output with actionable troubleshooting steps.
sankalp