llama-index-llms-bedrock-converse 0.12.2__tar.gz → 0.12.4__tar.gz

{llama_index_llms_bedrock_converse-0.12.2 → llama_index_llms_bedrock_converse-0.12.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llama-index-llms-bedrock-converse
-Version: 0.12.2
+Version: 0.12.4
 Summary: llama-index llms bedrock converse integration
 Author-email: Your Name <you@example.com>
 License-Expression: MIT

{llama_index_llms_bedrock_converse-0.12.2 → llama_index_llms_bedrock_converse-0.12.4}/llama_index/llms/bedrock_converse/base.py

@@ -1,5 +1,6 @@
 import warnings
 from typing import (
+    TYPE_CHECKING,
     Any,
     Callable,
     Dict,
@@ -9,16 +10,21 @@ from typing import (
     Sequence,
     Tuple,
     Union,
-    TYPE_CHECKING,
 )
 
+from llama_index.core.base.llms.generic_utils import (
+    achat_to_completion_decorator,
+    astream_chat_to_completion_decorator,
+    chat_to_completion_decorator,
+    stream_chat_to_completion_decorator,
+)
 from llama_index.core.base.llms.types import (
     ChatMessage,
     ChatResponse,
-    ChatResponseGen,
     ChatResponseAsyncGen,
-    CompletionResponseAsyncGen,
+    ChatResponseGen,
     CompletionResponse,
+    CompletionResponseAsyncGen,
     CompletionResponseGen,
     LLMMetadata,
     MessageRole,
@@ -33,26 +39,20 @@ from llama_index.core.llms.callbacks import (
     llm_chat_callback,
     llm_completion_callback,
 )
-from llama_index.core.base.llms.generic_utils import (
-    achat_to_completion_decorator,
-    astream_chat_to_completion_decorator,
-    chat_to_completion_decorator,
-    stream_chat_to_completion_decorator,
-)
 from llama_index.core.llms.function_calling import FunctionCallingLLM, ToolSelection
 from llama_index.core.llms.utils import parse_partial_json
 from llama_index.core.types import BaseOutputParser, PydanticProgramMode
 from llama_index.llms.bedrock_converse.utils import (
+    ThinkingDict,
     bedrock_modelname_to_context_size,
     converse_with_retry,
     converse_with_retry_async,
     force_single_tool_call,
     is_bedrock_function_calling_model,
+    is_reasoning,
     join_two_dicts,
     messages_to_converse_messages,
     tools_to_converse_tools,
-    is_reasoning,
-    ThinkingDict,
 )
 
 if TYPE_CHECKING:
@@ -364,7 +364,9 @@ class BedrockConverse(FunctionCallingLLM):
         }
 
     def _get_content_and_tool_calls(
-        self, response: Optional[Dict[str, Any]] = None, content: Dict[str, Any] = None
+        self,
+        response: Optional[Dict[str, Any]] = None,
+        content: Optional[Dict[str, Any]] = None,
     ) -> Tuple[
         List[Union[TextBlock, ThinkingBlock, ToolCallBlock]], List[str], List[str]
     ]:
@@ -505,10 +507,13 @@ class BedrockConverse(FunctionCallingLLM):
                     content_delta = content_block_delta["delta"]
                     content = join_two_dicts(content, content_delta)
 
+                    thinking_delta_value = None
                     if "reasoningContent" in content_delta:
-                        thinking += content_delta.get("reasoningContent", {}).get(
+                        reasoning_text = content_delta.get("reasoningContent", {}).get(
                             "text", ""
                         )
+                        thinking += reasoning_text
+                        thinking_delta_value = reasoning_text
                         thinking_signature += content_delta.get(
                             "reasoningContent", {}
                         ).get("signature", "")
@@ -566,6 +571,14 @@ class BedrockConverse(FunctionCallingLLM):
                                 )
                             )
 
+                    response_additional_kwargs = self._get_response_token_counts(
+                        dict(chunk)
+                    )
+                    if thinking_delta_value is not None:
+                        response_additional_kwargs["thinking_delta"] = (
+                            thinking_delta_value
+                        )
+
                     yield ChatResponse(
                         message=ChatMessage(
                             role=role,
@@ -579,7 +592,7 @@ class BedrockConverse(FunctionCallingLLM):
                         ),
                         delta=content_delta.get("text", ""),
                         raw=chunk,
-                        additional_kwargs=self._get_response_token_counts(dict(chunk)),
+                        additional_kwargs=response_additional_kwargs,
                     )
                 elif content_block_start := chunk.get("contentBlockStart"):
                     # New tool call starting
@@ -777,10 +790,13 @@ class BedrockConverse(FunctionCallingLLM):
                     content_delta = content_block_delta["delta"]
                     content = join_two_dicts(content, content_delta)
 
+                    thinking_delta_value = None
                     if "reasoningContent" in content_delta:
-                        thinking += content_delta.get("reasoningContent", {}).get(
+                        reasoning_text = content_delta.get("reasoningContent", {}).get(
                             "text", ""
                         )
+                        thinking += reasoning_text
+                        thinking_delta_value = reasoning_text
                         thinking_signature += content_delta.get(
                             "reasoningContent", {}
                         ).get("signature", "")
@@ -838,6 +854,14 @@ class BedrockConverse(FunctionCallingLLM):
                                 )
                             )
 
+                    response_additional_kwargs = self._get_response_token_counts(
+                        dict(chunk)
+                    )
+                    if thinking_delta_value is not None:
+                        response_additional_kwargs["thinking_delta"] = (
+                            thinking_delta_value
+                        )
+
                     yield ChatResponse(
                         message=ChatMessage(
                             role=role,
@@ -851,7 +875,7 @@ class BedrockConverse(FunctionCallingLLM):
                         ),
                         delta=content_delta.get("text", ""),
                         raw=chunk,
-                        additional_kwargs=self._get_response_token_counts(dict(chunk)),
+                        additional_kwargs=response_additional_kwargs,
                     )
                 elif content_block_start := chunk.get("contentBlockStart"):
                     # New tool call starting

{llama_index_llms_bedrock_converse-0.12.2 → llama_index_llms_bedrock_converse-0.12.4}/pyproject.toml

@@ -29,7 +29,7 @@ dev = [
 
 [project]
 name = "llama-index-llms-bedrock-converse"
-version = "0.12.2"
+version = "0.12.4"
 description = "llama-index llms bedrock converse integration"
 authors = [{name = "Your Name", email = "you@example.com"}]
 requires-python = ">=3.9,<4.0"