scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-10 03:28:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 84

fix(cost-map): azure/gpt-6-astra accepts reasoning_effort none on Foundry

Browse source
This commit is contained in:
mateo-berri 2026-09-04 17:29:55 -07:00
parent 3202963f25
commit 51514b9123
4 changed files with 37 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
litellm/model_prices_and_context_window_backup.json
View file

@ -7192,7 +7192,7 @@
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": false,
"supports_native_streaming": true,
"supports_none_reasoning_effort": false,
"supports_none_reasoning_effort": true,
"supports_parallel_function_calling": true,
"supports_pdf_input": true,
"supports_prompt_caching": true,
@ -7458,7 +7458,7 @@
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": false,
"supports_native_streaming": true,
"supports_none_reasoning_effort": false,
"supports_none_reasoning_effort": true,
"supports_parallel_function_calling": true,
"supports_pdf_input": true,
"supports_prompt_caching": true,

4
model_prices_and_context_window.json
View file

@ -7192,7 +7192,7 @@
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": false,
"supports_native_streaming": true,
"supports_none_reasoning_effort": false,
"supports_none_reasoning_effort": true,
"supports_parallel_function_calling": true,
"supports_pdf_input": true,
"supports_prompt_caching": true,
@ -7458,7 +7458,7 @@
"supports_max_reasoning_effort": true,
"supports_minimal_reasoning_effort": false,
"supports_native_streaming": true,
"supports_none_reasoning_effort": false,
"supports_none_reasoning_effort": true,
"supports_parallel_function_calling": true,
"supports_pdf_input": true,
"supports_prompt_caching": true,

28
tests/test_litellm/llms/azure/chat/test_azure_gpt5_transformation.py
View file

@ -348,3 +348,31 @@ def test_azure_gpt_6_astra_takes_the_reasoning_series_request_shape():
assert params["max_completion_tokens"] == 100
assert "max_tokens" not in params
assert params["reasoning_effort"] == "max"
@pytest.mark.parametrize("model", ["azure/gpt-6-astra", "azure/us/gpt-6-astra"])
def test_azure_gpt6_astra_reasoning_effort_none_unlocks_temperature(config: AzureOpenAIGPT5Config, model: str):
"""Foundry's gpt-6-astra accepts reasoning_effort='none' and, only then, a non-default
temperature (verified live against a Foundry deployment), unlike OpenAI's gpt-6-astra."""
params = config.map_openai_params(
non_default_params={"temperature": 0.2, "reasoning_effort": "none"},
optional_params={},
model=model,
drop_params=False,
api_version="2025-04-01-preview",
)
assert params["temperature"] == 0.2
assert params["reasoning_effort"] == "none"
@pytest.mark.parametrize("model", ["azure/gpt-6-astra", "azure/us/gpt-6-astra"])
def test_azure_gpt6_astra_rejects_reasoning_effort_minimal(config: AzureOpenAIGPT5Config, model: str):
"""Foundry's gpt-6-astra lists none, low, medium, high, xhigh and max but not minimal."""
with pytest.raises(litellm.utils.UnsupportedParamsError):
config.map_openai_params(
non_default_params={"reasoning_effort": "minimal"},
optional_params={},
model=model,
drop_params=False,
api_version="2025-04-01-preview",
)

8
tests/test_litellm/router_utils/test_reasoning_effort_capability.py
View file

@ -390,14 +390,16 @@ class TestGpt6AstraAdvertisesItsDocumentedLevels:
)
@pytest.mark.parametrize("model", ["azure/gpt-6-astra", "azure/us/gpt-6-astra"])
def test_a_foundry_deployment_advertises_the_same_levels(self, local_model_cost_map, model):
"""Microsoft Foundry serves the same model, so an Azure deployment must offer low
through max and never none, exactly like the OpenAI entry."""
def test_a_foundry_deployment_also_advertises_none(self, local_model_cost_map, model):
"""Microsoft Foundry serves the same model but its API accepts reasoning_effort none
(verified live: 200 with zero reasoning tokens, and it unlocks temperature), which
OpenAI's rejects, so an Azure deployment offers none on top of low through max."""
from litellm.utils import _get_model_info_helper
model_info = dict(_get_model_info_helper(model=model, custom_llm_provider="azure"))
assert resolve_supported_reasoning_efforts(model_info, deployment_is_mapped=True) == (
"none",
"low",
"medium",
"high",