ai-pipeline-core 0.2.8__py3-none-any.whl → 0.2.9__py3-none-any.whl

ai_pipeline_core/__init__.py

@@ -118,7 +118,7 @@ from .prompt_manager import PromptManager
 from .settings import Settings
 from .tracing import TraceInfo, TraceLevel, set_trace_cost, trace
 
-__version__ = "0.2.8"
+__version__ = "0.2.9"
 
 __all__ = [
     # Config/Settings

ai_pipeline_core/llm/client.py

@@ -45,31 +45,30 @@ def _process_messages(
 
     Internal function that combines context and messages into a single
     list of API-compatible messages. Applies caching directives to
-    context messages for efficiency.
+    system prompt and context messages for efficiency.
 
     Args:
         context: Messages to be cached (typically expensive/static content).
         messages: Regular messages without caching (dynamic queries).
         system_prompt: Optional system instructions for the model.
-        cache_ttl: Cache TTL for context messages (e.g. "120s", "300s", "1h").
+        cache_ttl: Cache TTL for system and context messages (e.g. "120s", "300s", "1h").
                    Set to None or empty string to disable caching.
 
     Returns:
         List of formatted messages ready for API calls, with:
-        - System prompt at the beginning (if provided)
-        - Context messages with cache_control on the last one (if cache_ttl)
+        - System prompt at the beginning with cache_control (if provided and cache_ttl set)
+        - Context messages with cache_control on all messages (if cache_ttl set)
         - Regular messages without caching
 
     System Prompt Location:
         The system prompt parameter is always injected as the FIRST message
-        with role="system". It is NOT cached with context, allowing dynamic
-        system prompts without breaking cache efficiency.
+        with role="system". It is cached along with context when cache_ttl is set.
 
     Cache behavior:
-        The last context message gets ephemeral caching with specified TTL
+        All system and context messages get ephemeral caching with specified TTL
         to reduce token usage on repeated calls with same context.
         If cache_ttl is None or empty string (falsy), no caching is applied.
-        Only the last context message receives cache_control to maximize efficiency.
+        All system and context messages receive cache_control to maximize cache efficiency.
 
     Note:
         This is an internal function used by _generate_with_retry().
@@ -79,26 +78,28 @@ def _process_messages(
 
     # Add system prompt if provided
     if system_prompt:
-        processed_messages.append({"role": "system", "content": system_prompt})
+        processed_messages.append({
+            "role": "system",
+            "content": [{"type": "text", "text": system_prompt}],
+        })
 
     # Process context messages with caching if provided
     if context:
         # Use AIMessages.to_prompt() for context
         context_messages = context.to_prompt()
+        processed_messages.extend(context_messages)
 
-        # Apply caching to last context message and last content part if cache_ttl is set
-        if cache_ttl:
-            context_messages[-1]["cache_control"] = {  # type: ignore
-                "type": "ephemeral",
-                "ttl": cache_ttl,
-            }
-            assert isinstance(context_messages[-1]["content"], list)  # type: ignore
-            context_messages[-1]["content"][-1]["cache_control"] = {  # type: ignore
+    if cache_ttl:
+        for message in processed_messages:
+            message["cache_control"] = {  # type: ignore
                 "type": "ephemeral",
                 "ttl": cache_ttl,
             }
-
-        processed_messages.extend(context_messages)
+            if isinstance(message["content"], list):  # type: ignore
+                message["content"][-1]["cache_control"] = {  # type: ignore
+                    "type": "ephemeral",
+                    "ttl": cache_ttl,
+                }
 
     # Process regular messages without caching
     if messages:
@@ -156,7 +157,7 @@ def _model_name_to_openrouter_model(model: ModelName) -> str:
     if model == "grok-4-fast-search":
         return "x-ai/grok-4-fast:online"
     if model == "sonar-pro-search":
-        return "perplexity/sonar-reasoning-pro"
+        return "perplexity/sonar-pro-search"
     if model.startswith("gemini"):
         return f"google/{model}"
     elif model.startswith("gpt"):

ai_pipeline_core/llm/model_response.py

@@ -88,10 +88,13 @@ class ModelResponse(ChatCompletion):
         data = chat_completion.model_dump()
 
         # fixes issue where the role is "assistantassistant" instead of "assistant"
+        valid_finish_reasons = {"stop", "length", "tool_calls", "content_filter", "function_call"}
         for i in range(len(data["choices"])):
-            if role := data["choices"][i]["message"].get("role"):
-                if role.startswith("assistant") and role != "assistant":
-                    data["choices"][i]["message"]["role"] = "assistant"
+            data["choices"][i]["message"]["role"] = "assistant"
+            # Only update finish_reason if it's not already a valid value
+            current_finish_reason = data["choices"][i].get("finish_reason")
+            if current_finish_reason not in valid_finish_reasons:
+                data["choices"][i]["finish_reason"] = "stop"
 
         super().__init__(**data)
 

{ai_pipeline_core-0.2.8.dist-info → ai_pipeline_core-0.2.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ai-pipeline-core
-Version: 0.2.8
+Version: 0.2.9
 Summary: Core utilities for AI-powered processing pipelines using prefect
 Project-URL: Homepage, https://github.com/bbarwik/ai-pipeline-core
 Project-URL: Repository, https://github.com/bbarwik/ai-pipeline-core

{ai_pipeline_core-0.2.8.dist-info → ai_pipeline_core-0.2.9.dist-info}/RECORD

@@ -1,4 +1,4 @@
-ai_pipeline_core/__init__.py,sha256=NMpeyF-rJdNnNWfi9eoOgWNnk7kOVHqvSZq3cnyDuCI,5720
+ai_pipeline_core/__init__.py,sha256=o0J4DdxbwUbZ6I07G0OVuCnc28NsywvNqEeiYuCmYI0,5720
 ai_pipeline_core/exceptions.py,sha256=vx-XLTw2fJSPs-vwtXVYtqoQUcOc0JeI7UmHqRqQYWU,1569
 ai_pipeline_core/pipeline.py,sha256=fWTVmrnOEIFge6o2NUYW2ndGef5UurpL8_fK5tkXbzI,28700
 ai_pipeline_core/prefect.py,sha256=91ZgLJHsDsRUW77CpNmkKxYs3RCJuucPM3pjKmNBeDg,2199
@@ -19,9 +19,9 @@ ai_pipeline_core/flow/config.py,sha256=3PCDph2n8dj-txqAvd9Wflbi_6lmfXFR9rUhM-szG
 ai_pipeline_core/flow/options.py,sha256=zn3N5DgYtlxLq0AvXfana3UOhym7A3XCheQSBIIarZE,2295
 ai_pipeline_core/llm/__init__.py,sha256=3B_vtEzxrzidP1qOUNQ4RxlUmxZ2MBKQcUhQiTybM9g,661
 ai_pipeline_core/llm/ai_messages.py,sha256=Onin3UPdbJQNl3WfY3-_jE5KRmF-ciXsa5K6UPOiy5s,14410
-ai_pipeline_core/llm/client.py,sha256=Da1NgGzfIkFRw_aDASK36MMdKe3DXaj5_3wMg0gR-Hk,24999
+ai_pipeline_core/llm/client.py,sha256=VQOUxGT8bUoKiD3-XX0VY5OHUX80Xdz6esrN9j2KUZ4,25013
 ai_pipeline_core/llm/model_options.py,sha256=uRNIHfVeh2sgt1mZBiOUx6hPQ6GKjB8b7TytZJ6afKg,11768
-ai_pipeline_core/llm/model_response.py,sha256=6kEr9ss3UGlykvtWAvh1l55rGw2-wyVup3QJhm0Oggc,13264
+ai_pipeline_core/llm/model_response.py,sha256=-fKJcblDP_Z6NV9CGp4bm_hitb0Z0jyy0ZndCQUpRkQ,13493
 ai_pipeline_core/llm/model_types.py,sha256=2J4Qsb1x21I4eo_VPeaMMOW8shOGPqzJuoGjTLcBFPM,2791
 ai_pipeline_core/logging/__init__.py,sha256=Nz6-ghAoENsgNmLD2ma9TW9M0U2_QfxuQ5DDW6Vt6M0,651
 ai_pipeline_core/logging/logging.yml,sha256=YTW48keO_K5bkkb-KXGM7ZuaYKiquLsjsURei8Ql0V4,1353
@@ -35,7 +35,7 @@ ai_pipeline_core/storage/storage.py,sha256=ClMr419Y-eU2RuOjZYd51dC0stWQk28Vb56Pv
 ai_pipeline_core/utils/__init__.py,sha256=TJSmEm1Quf-gKwXrxM96u2IGzVolUyeNNfLMPoLstXI,254
 ai_pipeline_core/utils/deploy.py,sha256=rAtRuwkmGkc-fqvDMXpt08OzLrD7KTDMAmLDC9wYg7Y,13147
 ai_pipeline_core/utils/remote_deployment.py,sha256=cPTgnS5InK08qiWnuPz3e8YKjoT3sPBloSaDfNTzghs,10137
-ai_pipeline_core-0.2.8.dist-info/METADATA,sha256=Sl_8s24ar0lwa7iio4d0QYDhzsmAIGHHqOKuXnFxP7s,15159
-ai_pipeline_core-0.2.8.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-ai_pipeline_core-0.2.8.dist-info/licenses/LICENSE,sha256=kKj8mfbdWwkyG3U6n7ztB3bAZlEwShTkAsvaY657i3I,1074
-ai_pipeline_core-0.2.8.dist-info/RECORD,,
+ai_pipeline_core-0.2.9.dist-info/METADATA,sha256=8p4PXSJqP5j4XV4cxjuncSN3i8914ZMupaU3EKs6Qpk,15159
+ai_pipeline_core-0.2.9.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+ai_pipeline_core-0.2.9.dist-info/licenses/LICENSE,sha256=kKj8mfbdWwkyG3U6n7ztB3bAZlEwShTkAsvaY657i3I,1074
+ai_pipeline_core-0.2.9.dist-info/RECORD,,