Fix prompt stacking in bedrock converse (#17613)

run-llama · Jan 24, 2025 · 19c177c · 19c177c
1 parent 612f670
commit 19c177c
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 13 deletions.
diff --git a/...grations/llms/llama-index-llms-bedrock-converse/llama_index/llms/bedrock_converse/base.py b/...grations/llms/llama-index-llms-bedrock-converse/llama_index/llms/bedrock_converse/base.py
@@ -302,15 +302,13 @@ def _get_content_and_tool_calls(
     def chat(self, messages: Sequence[ChatMessage], **kwargs: Any) -> ChatResponse:
         # convert Llama Index messages to AWS Bedrock Converse messages
         converse_messages, system_prompt = messages_to_converse_messages(messages)
-        if len(system_prompt) > 0 or self.system_prompt is None:
-            self.system_prompt = system_prompt
         all_kwargs = self._get_all_kwargs(**kwargs)
 
         # invoke LLM in AWS Bedrock Converse with retry
         response = converse_with_retry(
             client=self._client,
             messages=converse_messages,
-            system_prompt=self.system_prompt,
+            system_prompt=system_prompt,
             max_retries=self.max_retries,
             stream=False,
             guardrail_identifier=self.guardrail_identifier,
@@ -349,15 +347,13 @@ def stream_chat(
     ) -> ChatResponseGen:
         # convert Llama Index messages to AWS Bedrock Converse messages
         converse_messages, system_prompt = messages_to_converse_messages(messages)
-        if len(system_prompt) > 0 or self.system_prompt is None:
-            self.system_prompt = system_prompt
         all_kwargs = self._get_all_kwargs(**kwargs)
 
         # invoke LLM in AWS Bedrock Converse with retry
         response = converse_with_retry(
             client=self._client,
             messages=converse_messages,
-            system_prompt=self.system_prompt,
+            system_prompt=system_prompt,
             max_retries=self.max_retries,
             stream=True,
             guardrail_identifier=self.guardrail_identifier,
@@ -431,16 +427,14 @@ async def achat(
     ) -> ChatResponse:
         # convert Llama Index messages to AWS Bedrock Converse messages
         converse_messages, system_prompt = messages_to_converse_messages(messages)
-        if len(system_prompt) > 0 or self.system_prompt is None:
-            self.system_prompt = system_prompt
         all_kwargs = self._get_all_kwargs(**kwargs)
 
         # invoke LLM in AWS Bedrock Converse with retry
         response = await converse_with_retry_async(
             session=self._asession,
             config=self._config,
             messages=converse_messages,
-            system_prompt=self.system_prompt,
+            system_prompt=system_prompt,
             max_retries=self.max_retries,
             stream=False,
             guardrail_identifier=self.guardrail_identifier,
@@ -479,16 +473,14 @@ async def astream_chat(
     ) -> ChatResponseAsyncGen:
         # convert Llama Index messages to AWS Bedrock Converse messages
         converse_messages, system_prompt = messages_to_converse_messages(messages)
-        if len(system_prompt) > 0 or self.system_prompt is None:
-            self.system_prompt = system_prompt
         all_kwargs = self._get_all_kwargs(**kwargs)
 
         # invoke LLM in AWS Bedrock Converse with retry
         response_gen = await converse_with_retry_async(
             session=self._asession,
             config=self._config,
             messages=converse_messages,
-            system_prompt=self.system_prompt,
+            system_prompt=system_prompt,
             max_retries=self.max_retries,
             stream=True,
             guardrail_identifier=self.guardrail_identifier,

diff --git a/llama-index-integrations/llms/llama-index-llms-bedrock-converse/pyproject.toml b/llama-index-integrations/llms/llama-index-llms-bedrock-converse/pyproject.toml
@@ -27,7 +27,7 @@ exclude = ["**/BUILD"]
 license = "MIT"
 name = "llama-index-llms-bedrock-converse"
 readme = "README.md"
-version = "0.4.3"
+version = "0.4.4"
 
 [tool.poetry.dependencies]
 python = ">=3.9,<4.0"