payi 0.1.0a82__tar.gz → 0.1.0a83__tar.gz

payi-0.1.0a83/.release-please-manifest.json

@@ -0,0 +1,3 @@
+{
+  ".": "0.1.0-alpha.83"
+}

{payi-0.1.0a82 → payi-0.1.0a83}/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## 0.1.0-alpha.83 (2025-06-06)
+
+Full Changelog: [v0.1.0-alpha.82...v0.1.0-alpha.83](https://github.com/Pay-i/pay-i-python/compare/v0.1.0-alpha.82...v0.1.0-alpha.83)
+
+### Features
+
+* ingest streaming results aggressively ([#316](https://github.com/Pay-i/pay-i-python/issues/316)) ([2d196f7](https://github.com/Pay-i/pay-i-python/commit/2d196f781c7d4ed1317f453e99501d682f684d3c))
+
 ## 0.1.0-alpha.82 (2025-06-06)
 
 Full Changelog: [v0.1.0-alpha.81...v0.1.0-alpha.82](https://github.com/Pay-i/pay-i-python/compare/v0.1.0-alpha.81...v0.1.0-alpha.82)

{payi-0.1.0a82 → payi-0.1.0a83}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: payi
-Version: 0.1.0a82
+Version: 0.1.0a83
 Summary: The official Python library for the payi API
 Project-URL: Homepage, https://github.com/Pay-i/pay-i-python
 Project-URL: Repository, https://github.com/Pay-i/pay-i-python

{payi-0.1.0a82 → payi-0.1.0a83}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "payi"
-version = "0.1.0-alpha.82"
+version = "0.1.0-alpha.83"
 description = "The official Python library for the payi API"
 dynamic = ["readme"]
 license = "Apache-2.0"

{payi-0.1.0a82 → payi-0.1.0a83}/src/payi/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "payi"
-__version__ = "0.1.0-alpha.82"  # x-release-please-version
+__version__ = "0.1.0-alpha.83"  # x-release-please-version

{payi-0.1.0a82 → payi-0.1.0a83}/src/payi/lib/AnthropicInstrumentor.py

@@ -7,7 +7,7 @@ from wrapt import wrap_function_wrapper  # type: ignore
 from payi.lib.helpers import PayiCategories
 from payi.types.ingest_units_params import Units
 
-from .instrument import _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
+from .instrument import _ChunkResult, _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
 
 
 class AnthropicInstrumentor:
@@ -133,7 +133,8 @@ class _AnthropicProviderRequest(_ProviderRequest):
             )
 
     @override
-    def process_chunk(self, chunk: Any) -> bool:
+    def process_chunk(self, chunk: Any) -> _ChunkResult:
+        ingest = False
         if chunk.type == "message_start":
             self._ingest["provider_response_id"] = chunk.message.id
 
@@ -154,9 +155,15 @@ class _AnthropicProviderRequest(_ProviderRequest):
 
         elif chunk.type == "message_delta":
             usage = chunk.usage
+            ingest = True
+
+            # Web search will return an updated input tokens value at the end of streaming
+            if usage.input_tokens > 0:
+                self._ingest["units"]["text"]["input"] = usage.input_tokens
+
             self._ingest["units"]["text"]["output"] = usage.output_tokens
         
-        return True
+        return _ChunkResult(send_chunk_to_caller=True, ingest=ingest)
 
     @override
     def process_synchronous_response(self, response: Any, log_prompt_and_response: bool, kwargs: Any) -> Any:

{payi-0.1.0a82 → payi-0.1.0a83}/src/payi/lib/BedrockInstrumentor.py

@@ -10,7 +10,7 @@ from payi.lib.helpers import PayiCategories, PayiHeaderNames, payi_aws_bedrock_u
 from payi.types.ingest_units_params import Units, IngestUnitsParams
 from payi.types.pay_i_common_models_api_router_header_info_param import PayICommonModelsAPIRouterHeaderInfoParam
 
-from .instrument import _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
+from .instrument import _ChunkResult, _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
 
 _supported_model_prefixes = ["meta.llama3", "anthropic.", "amazon.nova-pro", "amazon.nova-lite", "amazon.nova-micro"]
 
@@ -22,8 +22,6 @@ class BedrockInstrumentor:
         BedrockInstrumentor._instrumentor = instrumentor
 
         try:
-            import boto3  # type: ignore #  noqa: F401  I001
-
             wrap_function_wrapper(
                 "botocore.client",
                 "ClientCreator.create_client",
@@ -43,7 +41,7 @@ class BedrockInstrumentor:
 @_PayiInstrumentor.payi_wrapper
 def create_client_wrapper(instrumentor: _PayiInstrumentor, wrapped: Any, instance: Any, *args: Any, **kwargs: Any) -> Any: #  noqa: ARG001
     if kwargs.get("service_name") != "bedrock-runtime":
-        instrumentor._logger.debug(f"skipping client wrapper creation for {kwargs.get('service_name', '')} service")
+        # instrumentor._logger.debug(f"skipping client wrapper creation for {kwargs.get('service_name', '')} service")
         return wrapped(*args, **kwargs)
 
     try:
@@ -272,13 +270,14 @@ class _BedrockInvokeStreamingProviderRequest(_BedrockProviderRequest):
         self._is_anthropic: bool = model_id.startswith("anthropic.")
 
     @override
-    def process_chunk(self, chunk: Any) -> bool:
+    def process_chunk(self, chunk: Any) -> _ChunkResult:
         if self._is_anthropic:
             return self.process_invoke_streaming_anthropic_chunk(chunk)
         else:
             return self.process_invoke_streaming_llama_chunk(chunk)
 
-    def process_invoke_streaming_anthropic_chunk(self, chunk: str) -> bool:
+    def process_invoke_streaming_anthropic_chunk(self, chunk: str) -> _ChunkResult:
+        ingest = False
         chunk_dict =  json.loads(chunk)
         type = chunk_dict.get("type", "")
 
@@ -301,18 +300,21 @@ class _BedrockInvokeStreamingProviderRequest(_BedrockProviderRequest):
         elif type == "message_delta":
             usage = chunk_dict['usage']
             self._ingest["units"]["text"]["output"] = usage['output_tokens']
+            ingest = True
 
-        return True    
+        return _ChunkResult(send_chunk_to_caller=True, ingest=ingest)    
 
-    def process_invoke_streaming_llama_chunk(self, chunk: str) -> bool:
+    def process_invoke_streaming_llama_chunk(self, chunk: str) -> _ChunkResult:
+        ingest = False
         chunk_dict =  json.loads(chunk)
         metrics = chunk_dict.get("amazon-bedrock-invocationMetrics", {})
         if metrics:
             input = metrics.get("inputTokenCount", 0)
             output = metrics.get("outputTokenCount", 0)
             self._ingest["units"]["text"] = Units(input=input, output=output)
+            ingest = True
 
-        return True
+        return _ChunkResult(send_chunk_to_caller=True, ingest=ingest)    
 
 class _BedrockInvokeSynchronousProviderRequest(_BedrockProviderRequest):
     @override
@@ -374,7 +376,8 @@ class _BedrockConverseSynchronousProviderRequest(_BedrockProviderRequest):
 
 class _BedrockConverseStreamingProviderRequest(_BedrockProviderRequest):
     @override
-    def process_chunk(self, chunk: 'dict[str, Any]') -> bool:
+    def process_chunk(self, chunk: 'dict[str, Any]') -> _ChunkResult:
+        ingest = False
         metadata = chunk.get("metadata", {})
 
         if metadata:
@@ -383,4 +386,6 @@ class _BedrockConverseStreamingProviderRequest(_BedrockProviderRequest):
             output = usage["outputTokens"]
             self._ingest["units"]["text"] = Units(input=input, output=output)
 
-        return True
+            ingest = True
+
+        return _ChunkResult(send_chunk_to_caller=True, ingest=ingest)

{payi-0.1.0a82 → payi-0.1.0a83}/src/payi/lib/GoogleGenAiInstrumentor.py

@@ -8,7 +8,7 @@ from wrapt import wrap_function_wrapper  # type: ignore
 from payi.lib.helpers import PayiCategories
 from payi.types.ingest_units_params import Units
 
-from .instrument import _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
+from .instrument import _ChunkResult, _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
 
 
 class GoogleGenAiInstrumentor:
@@ -248,7 +248,8 @@ class _GoogleGenAiRequest(_ProviderRequest):
             prompt["tool_config"] = tool_config
 
     @override
-    def process_chunk(self, chunk: Any) -> bool:
+    def process_chunk(self, chunk: Any) -> _ChunkResult:
+        ingest = False
         response_dict: dict[str, Any] = chunk.to_json_dict()
         if "provider_response_id" not in self._ingest:
             id = response_dict.get("response_id", None)
@@ -267,8 +268,9 @@ class _GoogleGenAiRequest(_ProviderRequest):
         usage = response_dict.get("usage_metadata", {})
         if usage and "prompt_token_count" in usage and "candidates_token_count" in usage:
             self._compute_usage(response_dict, streaming_candidates_characters=self._candiates_character_count)
+            ingest = True
 
-        return True
+        return _ChunkResult(send_chunk_to_caller=True, ingest=ingest)
     
     @staticmethod
     def _is_character_billing_model(model: str) -> bool:

{payi-0.1.0a82 → payi-0.1.0a83}/src/payi/lib/OpenAIInstrumentor.py

@@ -9,7 +9,7 @@ from wrapt import wrap_function_wrapper  # type: ignore
 from payi.lib.helpers import PayiCategories, PayiHeaderNames
 from payi.types.ingest_units_params import Units
 
-from .instrument import _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
+from .instrument import _ChunkResult, _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
 
 
 class OpenAiInstrumentor:
@@ -22,8 +22,6 @@ class OpenAiInstrumentor:
     @staticmethod
     def instrument(instrumentor: _PayiInstrumentor) -> None:
         try:
-            from openai import OpenAI  # type: ignore #  noqa: F401  I001
-            
             wrap_function_wrapper(
                 "openai.resources.chat.completions",
                 "Completions.create",
@@ -47,7 +45,11 @@ class OpenAiInstrumentor:
                  "AsyncEmbeddings.create",
                 aembeddings_wrapper(instrumentor),
             )
+        except Exception as e:
+            instrumentor._logger.debug(f"Error instrumenting openai: {e}")
 
+        # responses separately as they are relatively new and the client may not be using the latest openai module
+        try:            
             wrap_function_wrapper(
                 "openai.resources.responses",
                 "Responses.create",
@@ -62,8 +64,6 @@ class OpenAiInstrumentor:
 
         except Exception as e:
             instrumentor._logger.debug(f"Error instrumenting openai: {e}")
-            return
-
 
 @_PayiInstrumentor.payi_wrapper
 def embeddings_wrapper(
@@ -338,7 +338,8 @@ class _OpenAiChatProviderRequest(_OpenAiProviderRequest):
         self._include_usage_added = False
 
     @override
-    def process_chunk(self, chunk: Any) -> bool:
+    def process_chunk(self, chunk: Any) -> _ChunkResult:
+        ingest = False
         model = model_to_dict(chunk)
         
         if "provider_response_id" not in self._ingest:
@@ -356,8 +357,9 @@ class _OpenAiChatProviderRequest(_OpenAiProviderRequest):
             # packet which contains the usage to the client as they are not expecting the data
             if self._include_usage_added:
                 send_chunk_to_client = False
+            ingest = True
 
-        return send_chunk_to_client
+        return _ChunkResult(send_chunk_to_caller=send_chunk_to_client, ingest=ingest)
 
     @override
     def process_request(self, instance: Any, extra_headers: 'dict[str, str]', args: Sequence[Any], kwargs: Any) -> bool:
@@ -420,7 +422,8 @@ class _OpenAiResponsesProviderRequest(_OpenAiProviderRequest):
             input_tokens_details_key=_OpenAiProviderRequest.responses_input_tokens_details_key)
 
     @override
-    def process_chunk(self, chunk: Any) -> bool:
+    def process_chunk(self, chunk: Any) -> _ChunkResult:
+        ingest = False
         model = model_to_dict(chunk)
         response: dict[str, Any] = model.get("response", {})
 
@@ -432,8 +435,9 @@ class _OpenAiResponsesProviderRequest(_OpenAiProviderRequest):
         usage = response.get("usage")
         if usage:
             self.add_usage_units(usage)
+            ingest = True
 
-        return True
+        return _ChunkResult(send_chunk_to_caller=True, ingest=ingest)
 
     @override
     def process_request(self, instance: Any, extra_headers: 'dict[str, str]', args: Sequence[Any], kwargs: Any) -> bool:

{payi-0.1.0a82 → payi-0.1.0a83}/src/payi/lib/VertexInstrumentor.py

@@ -8,33 +8,37 @@ from wrapt import wrap_function_wrapper  # type: ignore
 from payi.lib.helpers import PayiCategories
 from payi.types.ingest_units_params import Units
 
-from .instrument import _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
+from .instrument import _ChunkResult, _IsStreaming, _StreamingType, _ProviderRequest, _PayiInstrumentor
 
 
 class VertexInstrumentor:
     @staticmethod
     def instrument(instrumentor: _PayiInstrumentor) -> None:
         try:
-            import vertexai  # type: ignore #  noqa: F401  I001
-
             wrap_function_wrapper(
                 "vertexai.generative_models",
                 "GenerativeModel.generate_content",
                 generate_wrapper(instrumentor),
             )
 
-            wrap_function_wrapper(
-                "vertexai.preview.generative_models",
-                "GenerativeModel.generate_content",
-                generate_wrapper(instrumentor),
-            )
-
             wrap_function_wrapper(
                 "vertexai.generative_models",
                 "GenerativeModel.generate_content_async",
                 agenerate_wrapper(instrumentor),
             )
 
+        except Exception as e:
+            instrumentor._logger.debug(f"Error instrumenting vertex: {e}")
+            return
+
+        # separate instrumetning preview functionality from released in case it fails
+        try:
+            wrap_function_wrapper(
+                "vertexai.preview.generative_models",
+                "GenerativeModel.generate_content",
+                generate_wrapper(instrumentor),
+            )
+
             wrap_function_wrapper(
                 "vertexai.preview.generative_models",
                 "GenerativeModel.generate_content_async",
@@ -93,11 +97,19 @@ class _GoogleVertexRequest(_ProviderRequest):
             )
         self._prompt_character_count = 0
         self._candiates_character_count = 0
+        self._model_name: Optional[str] = None
 
     @override
     def process_request(self, instance: Any, extra_headers: 'dict[str, str]', args: Sequence[Any], kwargs: Any) -> bool:
         from vertexai.generative_models import Content, Image, Part # type: ignore #  noqa: F401  I001
 
+        # Try to extra the model name as a backup if the response does not provide it (older vertexai versions do not)
+        if instance and hasattr(instance, "_model_name"):
+            model = instance._model_name
+            if model and isinstance(model, str):
+                # Extract the model name after the last slash
+                self._model_name = model.split('/')[-1]
+        
         if not args:
             return True
         
@@ -191,17 +203,26 @@ class _GoogleVertexRequest(_ProviderRequest):
             # tool_config does not have to_dict or any other serializable object
             prompt["tool_config"] = str(tool_config)  # type: ignore
 
+    def _get_model_name(self, response: 'dict[str, Any]') -> Optional[str]:
+        model: Optional[str] = response.get("model_version", None)
+        if model:
+            return model
+
+        return self._model_name
+
     @override
-    def process_chunk(self, chunk: Any) -> bool:
+    def process_chunk(self, chunk: Any) -> _ChunkResult:
+        ingest = False
         response_dict: dict[str, Any] = chunk.to_dict()
         if "provider_response_id" not in self._ingest:
             id = response_dict.get("response_id", None)
             if id:
                 self._ingest["provider_response_id"] = id
 
-        model: str = response_dict.get("model_version", "")
-
-        self._ingest["resource"] = "google." + model
+        if "resource" not in self._ingest: 
+            model: Optional[str] = self._get_model_name(response_dict)  # type: ignore[unreachable]
+            if model:
+                self._ingest["resource"] = "google." + model
 
         for candidate in response_dict.get("candidates", []):
             parts = candidate.get("content", {}).get("parts", [])
@@ -211,8 +232,9 @@ class _GoogleVertexRequest(_ProviderRequest):
         usage = response_dict.get("usage_metadata", {})
         if usage and "prompt_token_count" in usage and "candidates_token_count" in usage:
             self._compute_usage(response_dict, streaming_candidates_characters=self._candiates_character_count)
+            ingest = True
 
-        return True
+        return _ChunkResult(send_chunk_to_caller=True, ingest=ingest)
     
     @staticmethod
     def _is_character_billing_model(model: str) -> bool:
@@ -230,7 +252,7 @@ class _GoogleVertexRequest(_ProviderRequest):
         if id:
             self._ingest["provider_response_id"] = id
         
-        model: Optional[str] = response_dict.get("model_version", None)
+        model: Optional[str] = self._get_model_name(response_dict)
         if model:
             self._ingest["resource"] = "google." + model
 
@@ -256,7 +278,9 @@ class _GoogleVertexRequest(_ProviderRequest):
         prompt_tokens_details: list[dict[str, Any]] = usage.get("prompt_tokens_details", [])
         candidates_tokens_details: list[dict[str, Any]] = usage.get("candidates_tokens_details", [])
 
-        model: str = response_dict.get("model_version", "")
+        model: Optional[str] = self._get_model_name(response_dict)
+        if not model:
+            model = ""
         
         # for character billing only
         large_context = "" if input < 128000 else "_large_context"