fix: remove duplicate MAX_SIZE + add Cloudflare response_text support (#26385)

- Remove duplicate MAX_SIZE_PER_ITEM_IN_MEMORY_CACHE_IN_KB definition
  (kept the one with default 1024, removed the one with default 512)
- Add fallback from 'response' to 'response_text' key in Cloudflare
  Workers AI transformation for newer Nemotron models

Co-authored-by: yuneng-jiang <yuneng@berri.ai>
Co-authored-by: Jah-yee <110645028+Jah-yee@users.noreply.github.com>

litellm/constants.py

@@ -409,9 +409,6 @@ CACHED_STREAMING_CHUNK_DELAY = float(os.getenv("CACHED_STREAMING_CHUNK_DELAY", 0
 AUDIO_SPEECH_CHUNK_SIZE = int(
     os.getenv("AUDIO_SPEECH_CHUNK_SIZE", 8192)
 )  # chunk_size for audio speech streaming. Balance between latency and memory usage
-MAX_SIZE_PER_ITEM_IN_MEMORY_CACHE_IN_KB = int(
-    os.getenv("MAX_SIZE_PER_ITEM_IN_MEMORY_CACHE_IN_KB", 512)
-)
 DEFAULT_MAX_TOKENS_FOR_TRITON = int(os.getenv("DEFAULT_MAX_TOKENS_FOR_TRITON", 2000))
 #### Networking settings ####
 # Sentinel used when `REQUEST_TIMEOUT` is unset: `litellm.request_timeout` keeps this

litellm/llms/cloudflare/chat/transformation.py

@@ -147,9 +147,9 @@ class CloudflareChatConfig(BaseConfig):
     ) -> ModelResponse:
         completion_response = raw_response.json()
 
-        model_response.choices[0].message.content = completion_response["result"][  # type: ignore
-            "response"
-        ]
+        # Support both "response" and "response_text" keys (newer models like Nemotron use "response_text")
+        result = completion_response["result"]
+        model_response.choices[0].message.content = result.get("response") if result.get("response") is not None else result.get("response_text", "")  # type: ignore
 
         prompt_tokens = litellm.utils.get_token_count(messages=messages, model=model)
         completion_tokens = len(