mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-05 02:41:56 +00:00
da3f58aeed
Azure OpenAI native /responses must dispatch to the native responses handler, never the chat-completions bridge. Routing it through the bridge historically collapsed streaming to delta-only events (dropping response.created, output_item.added, content_part.added and the matching .done events) and broke clients like Codex CLI that depend on the full response lifecycle. The existing bridge-flag tests all mock get_provider_responses_api_config, so none covered the real Azure resolution that was the original trigger (config is None -> bridge). These exercise the real ProviderConfigManager and assert azure/* (gpt and o-series) resolves a native responses config and dispatches to the native handler for both streaming and non-streaming, failing if Azure ever falls back to the bridge again. |
||
|---|---|---|
| .. | ||
| litellm_completion_transformation | ||
| mcp | ||
| test_metadata_codex_callback.py | ||
| test_no_duplicate_spend_logs.py | ||
| test_null_test_fix.py | ||
| test_responses_api_bridge_flag.py | ||
| test_responses_api_request_body.py | ||
| test_responses_prompt_management.py | ||
| test_responses_router_cooldown.py | ||
| test_responses_utils.py | ||
| test_responses_websocket_all_providers.py | ||
| test_sse_output_recovery.py | ||
| test_text_format_conversion.py | ||