fix(realtime): keep transcription validation compatible with cost map guard

litellm/main.py

@@ -23,7 +23,7 @@ from collections.abc import AsyncIterator, Callable, Coroutine, Iterable, Mappin
 from concurrent import futures
 from concurrent.futures import FIRST_COMPLETED, ThreadPoolExecutor, wait
 from copy import deepcopy
-from functools import lru_cache, partial
+from functools import partial
 from types import MappingProxyType
 from typing import TYPE_CHECKING, Any, Final, Literal, Optional, Protocol, Union, cast, get_args
 
@@ -8248,23 +8248,6 @@ async def atranscription(
         )
 
 
-@lru_cache(maxsize=1)
-def _bundled_transcription_response_formats() -> Mapping[str, tuple[str, ...]]:
-    from litellm.litellm_core_utils.get_model_cost_map import GetModelCostMap
-
-    catalog: Final = TypeAdapter(dict[str, dict[str, object]]).validate_python(
-        GetModelCostMap.load_local_model_cost_map()
-    )
-    formats_adapter: Final = TypeAdapter(tuple[str, ...])
-    return MappingProxyType(
-        {
-            model: formats_adapter.validate_python(entry["supported_transcription_response_formats"])
-            for model, entry in catalog.items()
-            if "supported_transcription_response_formats" in entry
-        }
-    )
-
-
 def _validate_gpt_transcription_request(
     model: str,
     custom_llm_provider: str,
@@ -8275,9 +8258,6 @@ def _validate_gpt_transcription_request(
 ) -> str | None:
     model_info: Final = get_model_info(model=model) if model in litellm.model_cost else {}
     supported_endpoints: Final = model_info.get("supported_endpoints")
-    supported_formats: Final = model_info.get("supported_transcription_response_formats") or (
-        _bundled_transcription_response_formats().get(model)
-    )
     if language is not None and languages is not None:
         raise litellm.UnsupportedParamsError(
             message="language and languages cannot be used together",
@@ -8290,13 +8270,13 @@ def _validate_gpt_transcription_request(
             model=model,
             llm_provider=custom_llm_provider,
         )
-    if supported_formats is not None and response_format is not None and response_format not in supported_formats:
+    if model == "gpt-transcribe" and response_format not in (None, "json"):
         raise litellm.UnsupportedParamsError(
-            message=f"{model} only supports response_format={', '.join(repr(fmt) for fmt in supported_formats)}",
+            message="gpt-transcribe only supports response_format='json'",
             model=model,
             llm_provider=custom_llm_provider,
         )
-    if custom_llm_provider == "azure" and supported_formats is not None:
+    if custom_llm_provider == "azure" and model == "gpt-transcribe":
         if api_version in (None, "v1", "latest", "preview"):
             return litellm.AZURE_DEFAULT_API_VERSION
         return api_version

litellm/model_prices_and_context_window_backup.json

@@ -214,9 +214,6 @@
             "/v1/realtime",
             "/v1/realtime/transcription_sessions"
         ],
-        "supported_transcription_response_formats": [
-            "json"
-        ],
         "supported_modalities": [
             "audio",
             "text"
@@ -63374,9 +63371,6 @@
             "/v1/audio/transcriptions",
             "/v1/realtime/transcription_sessions"
         ],
-        "supported_transcription_response_formats": [
-            "json"
-        ],
         "supported_modalities": [
             "text",
             "audio"

litellm/types/utils.py

@@ -465,7 +465,6 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False):
         ]
     ]
     supported_endpoints: list[str] | None
-    supported_transcription_response_formats: ReadOnly[Sequence[str] | None]
     use_openai_responses_path: bool | None
     tpm: int | None
     rpm: int | None

litellm/utils.py

@@ -6940,9 +6940,6 @@ def get_model_info_helper(
                 litellm_provider=_model_info.get("litellm_provider", custom_llm_provider),
                 mode=_model_info.get("mode"),
                 supported_endpoints=_model_info.get("supported_endpoints", None),
-                supported_transcription_response_formats=_model_info.get(
-                    "supported_transcription_response_formats", None
-                ),
                 supports_system_messages=_model_info.get("supports_system_messages", None),
                 supports_response_schema=_model_info.get("supports_response_schema", None),
                 supports_vision=_model_info.get("supports_vision", None),

model_prices_and_context_window.json

@@ -214,9 +214,6 @@
             "/v1/realtime",
             "/v1/realtime/transcription_sessions"
         ],
-        "supported_transcription_response_formats": [
-            "json"
-        ],
         "supported_modalities": [
             "audio",
             "text"
@@ -63374,9 +63371,6 @@
             "/v1/audio/transcriptions",
             "/v1/realtime/transcription_sessions"
         ],
-        "supported_transcription_response_formats": [
-            "json"
-        ],
         "supported_modalities": [
             "text",
             "audio"