mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-01 07:03:02 +00:00
c0ded7b282
* fix(responses): send a dict reasoning_effort the same way on every Responses bridge Azure and ChatGPT bridges dropped reasoning mode and context unless the dict also carried summary. Every bridge now runs the reasoning step the OpenAI bridge uses, and each Responses config declares the reasoning values its API accepts, so drop_params drops what a backend refuses. Perplexity ignores mode and context, so it answers 400 for them without drop_params. * fix(responses): refuse ignored reasoning fields on xAI and Fireworks, resolve openai/ names for OpenRouter * fix(responses): refuse a reasoning mode Copilot ignores, keep catalog modes, return refusals as values Apply additional_drop_params before the refusal check, keep a null reasoning field instead of refusing it, let a cost-map row's reasoning_modes and reasoning_contexts override a backend's declared values, and declare GitHub Copilot's accepted values from a live probe * fix(responses): read Azure base_model and deployment rows, accept standard mode, refuse only mode and context * fix(responses): send only effort and summary to an Azure deployment litellm cannot identify under drop_params Azure and Azure AI deployment names are user-chosen, so when neither the name, model_info.base_model, nor a cost map row resolves the model, drop_params now drops mode and context instead of forwarding a pro the model may refuse. Names resolve by their last path segment, so a regional base_model like azure/eu/gpt-5.4 follows gpt-5.4's rows * fix(responses): resolve deployment names by model family and keep reasoning_modes per deployment A custom Azure deployment name now resolves to the longest OpenAI catalog prefix of its base_model or name, an unresolved name forwards mode and context, and a deployment's model_info reasoning_modes and reasoning_contexts apply only to that deployment instead of every sibling sharing its model name. litellm_proxy forwards every reasoning field. * fix(responses): find the model family anywhere in an Azure deployment name and keep only effort and summary for unknown gpt-5 or gpt-6 names under drop_params * fix(responses): look up OpenAI model names inside deployment names only on Azure Databricks endpoints like databricks-gpt-5-5-pro spell versions with hyphens and other backends name local models freely, so only Azure and Azure AI search inside the name, and a run that cuts a hyphen-spelled version never matches * fix(responses): judge Azure AI reasoning fields by the request api_base --------- Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com> |
||
|---|---|---|
| .. | ||
| _support | ||
| audio_tests | ||
| base_sdk_tests | ||
| benchmarks | ||
| code_coverage_tests | ||
| documentation_tests | ||
| e2e | ||
| e2e_harness | ||
| fixtures | ||
| harness_e2e | ||
| integration | ||
| rust-python-harness | ||
| test_litellm_rust | ||
| typing | ||
| unit | ||
| windows_tests | ||
| __init__.py | ||
| _fake_openai_endpoint_server.py | ||
| _flush_vcr_cache.py | ||
| _live_test_helpers.py | ||
| _master_key.py | ||
| _openai_record_replay_proxy.py | ||
| _process_helpers.py | ||
| _vcr_conftest_common.py | ||
| _vcr_redis_persister.py | ||
| _wait_helpers.py | ||
| _ws_vcr.py | ||
| AGENTS.md | ||
| capturing_transport.py | ||
| eval_swe_bench.py | ||
| fake_openai_endpoint.py | ||
| gettysburg.wav | ||
| large_text.py | ||
| openai_batch_completions.jsonl | ||
| pyrightconfig.json | ||
| README.MD | ||
| test_logging.conf | ||
| test_rust_python_harness.py | ||
| VCR_TESTING.md | ||
| white_100x100.png | ||
In total litellm runs 1000+ tests
[02/20/2025] Update:
To make it easier to contribute and map what behavior is tested,
we've started mapping the litellm directory in tests/unit
This folder can only run mock tests.