fix(types): keep text completion, batch ids and guardrail callback on main's code

litellm/caching/redis_semantic_cache.py

@@ -407,16 +407,14 @@ class RedisSemanticCache(BaseCache):
 
         # Convert string representation to Python object
         try:
-            cached_response = json.loads(response_text)
+            return json.loads(response_text)
         except json.JSONDecodeError:
             try:
-                cached_response = ast.literal_eval(response_text)
+                return ast.literal_eval(response_text)
             except (ValueError, SyntaxError) as e:
                 print_verbose(f"Error parsing cached response: {e}")
                 return None
 
-        return cached_response
-
     def set_cache(self, key: str, value: object, **kwargs) -> None:
         """
         Store a value in the semantic cache.

litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py

@@ -618,12 +618,12 @@ class LiteLLMResponseObjectHandler:
         choices_list: Final[list[TextChoices]] = []
 
         # Convert each choice to TextChoices
-        for choice in response.choices:
+        for choice in response["choices"]:
             text_choices = TextChoices()
-            text_choices["text"] = choice.message.content
-            text_choices["index"] = choice.index
+            text_choices["text"] = choice["message"]["content"]
+            text_choices["index"] = choice["index"]
             text_choices["logprobs"] = transformed_logprobs
-            text_choices["finish_reason"] = choice.finish_reason
+            text_choices["finish_reason"] = choice["finish_reason"]
             choices_list.append(text_choices)
 
         text_completion_response["choices"] = choices_list

litellm/proxy/openai_files_endpoints/common_utils.py

@@ -1380,10 +1380,7 @@ async def resolve_input_file_id_to_unified(response, prisma_client) -> None:
             pass
 
 
-async def resolve_output_file_ids_to_unified(
-    response: "LiteLLMBatch",
-    prisma_client: "PrismaClient | None",
-) -> None:
+async def resolve_output_file_ids_to_unified(response, prisma_client) -> None:
     """
     If the batch response contains raw provider output_file_id or error_file_id
     (not already unified IDs), look up the corresponding unified file IDs from
@@ -1392,9 +1389,7 @@ async def resolve_output_file_ids_to_unified(
     if not prisma_client:
         return
     for attr in ("output_file_id", "error_file_id"):
-        if not hasattr(response, attr):
-            continue
-        raw_id = response.output_file_id if attr == "output_file_id" else response.error_file_id
+        raw_id = getattr(response, attr, None)
         if not raw_id or is_base64_encoded_unified_file_id(raw_id):
             continue
         try:

litellm/types/router.py

@@ -35,7 +35,6 @@ from litellm.types.llms.base import LiteLLMBaseModel
 from litellm.types.router_weights import RouterWeights
 
 if TYPE_CHECKING:
-    from litellm.integrations.custom_guardrail import CustomGuardrail
     from litellm.router import Router
 
 from .completion import CompletionRequest
@@ -1090,7 +1089,7 @@ class GuardrailTypedDict(TypedDict, total=False):
 
     guardrail_name: Required[str]
     litellm_params: Required[GuardrailLiteLLMParams]
-    callback: "ReadOnly[CustomGuardrail | None]"  # The CustomGuardrail instance
+    callback: Any  # The CustomGuardrail instance
     id: str | None  # Unique identifier for the guardrail deployment
 
 

tests/unit/litellm_core_utils/test_text_completion_conversion.py

@@ -5,6 +5,7 @@ import pytest
 
 import litellm
 from litellm import text_completion
+from litellm.llms.custom_llm import CustomLLM
 from litellm.types.utils import ModelResponse, TextCompletionResponse, Usage
 from litellm.utils import LiteLLMResponseObjectHandler
 
@@ -1298,3 +1299,30 @@ def mock_post(*args, **kwargs):
         "usage": {"prompt_tokens": 2, "total_tokens": 18, "completion_tokens": 16},
     }
     return mock_response
+
+
+def test_text_completion_reads_a_custom_provider_dict_reply(monkeypatch: pytest.MonkeyPatch) -> None:
+    class DictReplyLLM(CustomLLM):
+        def completion(self, *args: object, **kwargs: object) -> dict[str, object]:
+            return {
+                "id": "dict-reply",
+                "object": "chat.completion",
+                "created": 1,
+                "model": "x",
+                "choices": [
+                    {
+                        "index": 0,
+                        "message": {"role": "assistant", "content": "hi from a dict reply"},
+                        "finish_reason": "stop",
+                    }
+                ],
+            }
+
+    monkeypatch.setattr(litellm, "custom_provider_map", [{"provider": "dictreply", "custom_handler": DictReplyLLM()}])
+    monkeypatch.setattr(litellm, "provider_list", [*litellm.provider_list, "dictreply"])
+    monkeypatch.setattr(litellm, "_custom_providers", [*litellm._custom_providers, "dictreply"])
+
+    response = text_completion(model="dictreply/x", prompt="hi")
+
+    assert response.choices[0].text == "hi from a dict reply"
+    assert response.choices[0].finish_reason == "stop"