From 8f86959c3298d8be87710d2270ff5e28722c41ca Mon Sep 17 00:00:00 2001
From: Krish Dholakia <krrishdholakia@gmail.com>
Date: Fri, 28 Feb 2025 14:34:17 -0800
Subject: [PATCH] Litellm dev 02 27 2025 p6 (#8891)

* fix(http_parsing_utils.py): orjson can throw errors on some emoji's in text, default to json.loads

* fix(sagemaker/handler.py): support passing model id on async streaming

* fix(litellm_pre_call_utils.py): Fixes https://github.com/BerriAI/litellm/issues/7237
---
 litellm/llms/sagemaker/completion/handler.py  |  4 ++++
 .../proxy/common_utils/http_parsing_utils.py  | 21 ++++++++++++++++++-
 litellm/proxy/litellm_pre_call_utils.py       |  6 +++---
 3 files changed, 27 insertions(+), 4 deletions(-)

diff --git a/litellm/llms/sagemaker/completion/handler.py b/litellm/llms/sagemaker/completion/handler.py
index ae40dd26d9a..4aff5f5d71b 100644
--- a/litellm/llms/sagemaker/completion/handler.py
+++ b/litellm/llms/sagemaker/completion/handler.py
@@ -433,6 +433,10 @@ class SagemakerLLM(BaseAWSLLM):
             "messages": messages,
         }
         prepared_request = await asyncified_prepare_request(**prepared_request_args)
+        if model_id is not None:  # Fixes https://github.com/BerriAI/litellm/issues/8889
+            prepared_request.headers.update(
+                {"X-Amzn-SageMaker-Inference-Component": model_id}
+            )
         completion_stream = await self.make_async_call(
             api_base=prepared_request.url,
             headers=prepared_request.headers,  # type: ignore
diff --git a/litellm/proxy/common_utils/http_parsing_utils.py b/litellm/proxy/common_utils/http_parsing_utils.py
index ce8f1661c1d..5736ee21527 100644
--- a/litellm/proxy/common_utils/http_parsing_utils.py
+++ b/litellm/proxy/common_utils/http_parsing_utils.py
@@ -42,7 +42,26 @@ async def _read_request_body(request: Optional[Request]) -> Dict:
             if not body:
                 parsed_body = {}
             else:
-                parsed_body = orjson.loads(body)
+                try:
+                    parsed_body = orjson.loads(body)
+                except orjson.JSONDecodeError:
+                    # Fall back to the standard json module which is more forgiving
+                    # First decode bytes to string if needed
+                    body_str = body.decode("utf-8") if isinstance(body, bytes) else body
+
+                    # Replace invalid surrogate pairs
+                    import re
+
+                    # This regex finds incomplete surrogate pairs
+                    body_str = re.sub(
+                        r"[\uD800-\uDBFF](?![\uDC00-\uDFFF])", "", body_str
+                    )
+                    # This regex finds low surrogates without high surrogates
+                    body_str = re.sub(
+                        r"(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]", "", body_str
+                    )
+
+                    parsed_body = json.loads(body_str)
 
         # Cache the parsed result
         _safe_set_request_parsed_body(request=request, parsed_body=parsed_body)
diff --git a/litellm/proxy/litellm_pre_call_utils.py b/litellm/proxy/litellm_pre_call_utils.py
index 29fcddf6a21..c15bbbd4a7d 100644
--- a/litellm/proxy/litellm_pre_call_utils.py
+++ b/litellm/proxy/litellm_pre_call_utils.py
@@ -425,9 +425,9 @@ class LiteLLMProxyRequestSetup:
                     tags = [tag.strip() for tag in _tags]
                 elif isinstance(headers["x-litellm-tags"], list):
                     tags = headers["x-litellm-tags"]
-            # Check request body for tags
-            if "tags" in data and isinstance(data["tags"], list):
-                tags = data["tags"]
+        # Check request body for tags
+        if "tags" in data and isinstance(data["tags"], list):
+            tags = data["tags"]
 
         return tags