arize-phoenix 5.11.0__tar.gz → 6.0.0__tar.gz

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: arize-phoenix
-Version: 5.11.0
+Version: 6.0.0
 Summary: AI Observability and Evaluation
 Project-URL: Documentation, https://docs.arize.com/phoenix/
 Project-URL: Issues, https://github.com/Arize-ai/phoenix/issues

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/src/phoenix/experiments/evaluators/llm_evaluators.py

@@ -232,7 +232,7 @@ class RelevanceEvaluator(LLMEvaluator):
         "query. First, explain step-by-step why you think the text is or is not relevant. "
         "Then provide a single word label; 'true' if the text is relevant or 'false' if the text "
         "is not relevant. "
-        "Here is an example template for your reponse:\n\n"
+        "Here is an example template for your response:\n\n"
         "CRITERIA: the response is 'relevant' to the query\n"
         "QUERY: *text that contains a query*\n"
         "RESPONSE: *a response that may or may not be relevant to the query*\n"
@@ -367,7 +367,7 @@ class LLMRelationalEvaluator(LLMEvaluator):
         "First, explain step-by-step why you think the response '{relation}' the reference. "
         "Then provide a single word label; 'true' if the response '{relation}' the reference or "
         "'false' if the text is not '{relation}' to the reference. "
-        "Here is an example template for your reponse:\n\n"
+        "Here is an example template for your response:\n\n"
         "CRITERIA: the response '{relation}' the reference\n"
         "REFERENCE: *text that contains a reference*\n"
         "RESPONSE: *a response that may or may not be '{relation}' to the reference*\n"

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/src/phoenix/server/api/helpers/dataset_helpers.py

@@ -1,54 +1,43 @@
 import json
 from collections.abc import Mapping
-from typing import Any, Literal, Optional, Protocol
+from typing import Any, Literal, Optional
 
 from openinference.semconv.trace import (
     MessageAttributes,
     OpenInferenceMimeTypeValues,
     OpenInferenceSpanKindValues,
+    SpanAttributes,
     ToolCallAttributes,
 )
 
+from phoenix.db.models import Span
 from phoenix.trace.attributes import get_attribute_value
 
 
-class HasSpanIO(Protocol):
-    """
-    An interface that contains the information needed to extract dataset example
-    input and output values from a span.
-    """
-
-    span_kind: Optional[str]
-    input_value: Any
-    input_mime_type: Optional[str]
-    output_value: Any
-    output_mime_type: Optional[str]
-    llm_prompt_template_variables: Any
-    llm_input_messages: Any
-    llm_output_messages: Any
-    retrieval_documents: Any
-
-
-def get_dataset_example_input(span: HasSpanIO) -> dict[str, Any]:
+def get_dataset_example_input(span: Span) -> dict[str, Any]:
     """
     Extracts the input value from a span and returns it as a dictionary. Input
     values from LLM spans are extracted from the input messages and prompt
     template variables (if present). For other span kinds, the input is
     extracted from the input value and input mime type attributes.
     """
-    input_value = span.input_value
-    input_mime_type = span.input_mime_type
-    if span.span_kind == OpenInferenceSpanKindValues.LLM.value:
+    span_kind = span.span_kind
+    attributes = span.attributes
+    input_value = get_attribute_value(attributes, INPUT_VALUE)
+    input_mime_type = get_attribute_value(attributes, INPUT_MIME_TYPE)
+    prompt_template_variables = get_attribute_value(attributes, LLM_PROMPT_TEMPLATE_VARIABLES)
+    input_messages = get_attribute_value(attributes, LLM_INPUT_MESSAGES)
+    if span_kind == LLM:
         return _get_llm_span_input(
-            input_messages=span.llm_input_messages,
+            input_messages=input_messages,
             input_value=input_value,
             input_mime_type=input_mime_type,
-            prompt_template_variables=span.llm_prompt_template_variables,
+            prompt_template_variables=prompt_template_variables,
         )
     return _get_generic_io_value(io_value=input_value, mime_type=input_mime_type, kind="input")
 
 
-def get_dataset_example_output(span: HasSpanIO) -> dict[str, Any]:
+def get_dataset_example_output(span: Span) -> dict[str, Any]:
     """
     Extracts the output value from a span and returns it as a dictionary. Output
     values from LLM spans are extracted from the output messages (if present).
@@ -56,18 +45,21 @@ def get_dataset_example_output(span: HasSpanIO) -> dict[str, Any]:
     present). For other span kinds, the output is extracted from the output
     value and output mime type attributes.
     """
-
-    output_value = span.output_value
-    output_mime_type = span.output_mime_type
-    if (span_kind := span.span_kind) == OpenInferenceSpanKindValues.LLM.value:
+    span_kind = span.span_kind
+    attributes = span.attributes
+    output_value = get_attribute_value(attributes, OUTPUT_VALUE)
+    output_mime_type = get_attribute_value(attributes, OUTPUT_MIME_TYPE)
+    output_messages = get_attribute_value(attributes, LLM_OUTPUT_MESSAGES)
+    retrieval_documents = get_attribute_value(attributes, RETRIEVAL_DOCUMENTS)
+    if span_kind == LLM:
         return _get_llm_span_output(
-            output_messages=span.llm_output_messages,
+            output_messages=output_messages,
             output_value=output_value,
             output_mime_type=output_mime_type,
         )
     if span_kind == OpenInferenceSpanKindValues.RETRIEVER.value:
         return _get_retriever_span_output(
-            retrieval_documents=span.retrieval_documents,
+            retrieval_documents=retrieval_documents,
             output_value=output_value,
             output_mime_type=output_mime_type,
         )
@@ -90,8 +82,8 @@ def _get_llm_span_input(
         input["messages"] = messages
     if not input:
         input = _get_generic_io_value(io_value=input_value, mime_type=input_mime_type, kind="input")
-    if prompt_template_variables:
-        input = {**input, "prompt_template_variables": prompt_template_variables}
+    if prompt_template_variables_data := _safely_json_decode(prompt_template_variables):
+        input["prompt_template_variables"] = prompt_template_variables_data
     return input
 
 
@@ -118,7 +110,7 @@ def _get_retriever_span_output(
     Extracts the output value from a retriever span and returns it as a dictionary.
     The output is extracted from the retrieval documents (if present).
     """
-    if retrieval_documents is not None:
+    if (retrieval_documents := _parse_retrieval_documents(retrieval_documents)) is not None:
         return {"documents": retrieval_documents}
     return _get_generic_io_value(io_value=output_value, mime_type=output_mime_type, kind="output")
 
@@ -130,12 +122,14 @@ def _get_generic_io_value(
     Makes a best-effort attempt to extract the input or output value from a span
     and returns it as a dictionary.
     """
-    if mime_type == OpenInferenceMimeTypeValues.JSON.value:
-        parsed_value = json.loads(io_value)
-        if isinstance(parsed_value, dict):
-            return parsed_value
+    if (
+        mime_type == OpenInferenceMimeTypeValues.JSON.value
+        and (io_value_data := _safely_json_decode(io_value)) is not None
+    ):
+        if isinstance(io_value_data, dict):
+            return io_value_data
         else:
-            return {kind: parsed_value}
+            return {kind: io_value_data}
     if isinstance(io_value, str):
         return {kind: io_value}
     return {}
@@ -169,6 +163,35 @@ def _get_message(message: Mapping[str, Any]) -> dict[str, Any]:
     }
 
 
+def _parse_retrieval_documents(retrieval_documents: Any) -> Optional[list[dict[str, Any]]]:
+    """
+    Safely un-nests a list of retrieval documents.
+
+    Example: [{"document": {"content": "..."}}] -> [{"content": "..."}]
+    """
+    if not isinstance(retrieval_documents, list):
+        return None
+    docs = []
+    for retrieval_doc in retrieval_documents:
+        if not isinstance(retrieval_doc, dict) or not (doc := retrieval_doc.get("document")):
+            return None
+        docs.append(doc)
+    return docs
+
+
+def _safely_json_decode(value: Any) -> Any:
+    """
+    Safely decodes a JSON-encoded value.
+    """
+    if not isinstance(value, str):
+        return None
+    try:
+        return json.loads(value)
+    except json.JSONDecodeError:
+        return None
+
+
+# MessageAttributes
 MESSAGE_CONTENT = MessageAttributes.MESSAGE_CONTENT
 MESSAGE_FUNCTION_CALL_ARGUMENTS_JSON = MessageAttributes.MESSAGE_FUNCTION_CALL_ARGUMENTS_JSON
 MESSAGE_FUNCTION_CALL_NAME = MessageAttributes.MESSAGE_FUNCTION_CALL_NAME
@@ -176,5 +199,19 @@ MESSAGE_NAME = MessageAttributes.MESSAGE_NAME
 MESSAGE_ROLE = MessageAttributes.MESSAGE_ROLE
 MESSAGE_TOOL_CALLS = MessageAttributes.MESSAGE_TOOL_CALLS
 
-TOOL_CALL_FUNCTION_NAME = ToolCallAttributes.TOOL_CALL_FUNCTION_NAME
+# OpenInferenceSpanKindValues
+LLM = OpenInferenceSpanKindValues.LLM.value
+
+# SpanAttributes
+INPUT_MIME_TYPE = SpanAttributes.INPUT_MIME_TYPE
+INPUT_VALUE = SpanAttributes.INPUT_VALUE
+LLM_INPUT_MESSAGES = SpanAttributes.LLM_INPUT_MESSAGES
+LLM_OUTPUT_MESSAGES = SpanAttributes.LLM_OUTPUT_MESSAGES
+LLM_PROMPT_TEMPLATE_VARIABLES = SpanAttributes.LLM_PROMPT_TEMPLATE_VARIABLES
+OUTPUT_MIME_TYPE = SpanAttributes.OUTPUT_MIME_TYPE
+OUTPUT_VALUE = SpanAttributes.OUTPUT_VALUE
+RETRIEVAL_DOCUMENTS = SpanAttributes.RETRIEVAL_DOCUMENTS
+
+# ToolCallAttributes
 TOOL_CALL_FUNCTION_ARGUMENTS_JSON = ToolCallAttributes.TOOL_CALL_FUNCTION_ARGUMENTS_JSON
+TOOL_CALL_FUNCTION_NAME = ToolCallAttributes.TOOL_CALL_FUNCTION_NAME

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/src/phoenix/server/api/helpers/playground_clients.py

@@ -131,8 +131,8 @@ class PlaygroundRateLimiter(RateLimiter, KeyedSingleton):
                 await self._throttler.async_wait_until_ready()
                 request_start_time = time.time()
                 maybe_coroutine = fn(*args, **kwargs)
-                if inspect.iscoroutine(maybe_coroutine):
-                    return await maybe_coroutine  # type: ignore
+                if inspect.isawaitable(maybe_coroutine):
+                    return await maybe_coroutine  # type: ignore[no-any-return]
                 else:
                     return maybe_coroutine
             except self._rate_limit_error:
@@ -144,10 +144,11 @@ class PlaygroundRateLimiter(RateLimiter, KeyedSingleton):
                             try:
                                 request_start_time = time.time()
                                 await self._throttler.async_wait_until_ready()
-                                if inspect.iscoroutinefunction(fn):
-                                    return await fn(*args, **kwargs)  # type: ignore
+                                maybe_coroutine = fn(*args, **kwargs)
+                                if inspect.isawaitable(maybe_coroutine):
+                                    return await maybe_coroutine  # type: ignore[no-any-return]
                                 else:
-                                    return fn(*args, **kwargs)
+                                    return maybe_coroutine
                             except self._rate_limit_error:
                                 self._throttler.on_rate_limit_error(
                                     request_start_time, verbose=self._verbose
@@ -283,7 +284,7 @@ class OpenAIStreamingClient(PlaygroundStreamingClient):
                 invocation_name="temperature",
                 canonical_name=CanonicalParameterName.TEMPERATURE,
                 label="Temperature",
-                default_value=0.0,
+                default_value=1.0,
                 min_value=0.0,
                 max_value=2.0,
             ),
@@ -295,12 +296,14 @@ class OpenAIStreamingClient(PlaygroundStreamingClient):
             BoundedFloatInvocationParameter(
                 invocation_name="frequency_penalty",
                 label="Frequency Penalty",
+                default_value=0.0,
                 min_value=-2.0,
                 max_value=2.0,
             ),
             BoundedFloatInvocationParameter(
                 invocation_name="presence_penalty",
                 label="Presence Penalty",
+                default_value=0.0,
                 min_value=-2.0,
                 max_value=2.0,
             ),
@@ -313,6 +316,7 @@ class OpenAIStreamingClient(PlaygroundStreamingClient):
                 invocation_name="top_p",
                 canonical_name=CanonicalParameterName.TOP_P,
                 label="Top P",
+                default_value=1.0,
                 min_value=0.0,
                 max_value=1.0,
             ),
@@ -345,7 +349,11 @@ class OpenAIStreamingClient(PlaygroundStreamingClient):
         from openai.types.chat import ChatCompletionStreamOptionsParam
 
         # Convert standard messages to OpenAI messages
-        openai_messages = [self.to_openai_chat_completion_param(*message) for message in messages]
+        openai_messages = []
+        for message in messages:
+            openai_message = self.to_openai_chat_completion_param(*message)
+            if openai_message is not None:
+                openai_messages.append(openai_message)
         tool_call_ids: dict[int, str] = {}
         token_usage: Optional["CompletionUsage"] = None
         throttled_create = self.rate_limiter._alimit(self.client.chat.completions.create)
@@ -395,7 +403,7 @@ class OpenAIStreamingClient(PlaygroundStreamingClient):
         content: JSONScalarType,
         tool_call_id: Optional[str] = None,
         tool_calls: Optional[list[JSONScalarType]] = None,
-    ) -> "ChatCompletionMessageParam":
+    ) -> Optional["ChatCompletionMessageParam"]:
         from openai.types.chat import (
             ChatCompletionAssistantMessageParam,
             ChatCompletionSystemMessageParam,
@@ -495,65 +503,7 @@ class OpenAIO1StreamingClient(OpenAIStreamingClient):
             ),
         ]
 
-    async def chat_completion_create(
-        self,
-        messages: list[
-            tuple[ChatCompletionMessageRole, str, Optional[str], Optional[list[JSONScalarType]]]
-        ],
-        tools: list[JSONScalarType],
-        **invocation_parameters: Any,
-    ) -> AsyncIterator[ChatCompletionChunk]:
-        from openai import NOT_GIVEN
-
-        # Convert standard messages to OpenAI messages
-        unfiltered_openai_messages = [
-            self.to_openai_o1_chat_completion_param(*message) for message in messages
-        ]
-
-        # filter out unsupported messages
-        openai_messages: list[ChatCompletionMessageParam] = [
-            message for message in unfiltered_openai_messages if message is not None
-        ]
-
-        tool_call_ids: dict[int, str] = {}
-
-        throttled_create = self.rate_limiter._alimit(self.client.chat.completions.create)
-        response = await throttled_create(
-            messages=openai_messages,
-            model=self.model_name,
-            tools=tools or NOT_GIVEN,
-            **invocation_parameters,
-        )
-
-        choice = response.choices[0]
-        message = choice.message
-        content = message.content
-
-        text_chunk = TextChunk(content=content)
-        yield text_chunk
-
-        if (tool_calls := message.tool_calls) is not None:
-            for tool_call_index, tool_call in enumerate(tool_calls):
-                tool_call_id = (
-                    tool_call.id
-                    if tool_call.id is not None
-                    else tool_call_ids.get(tool_call_index, f"tool_call_{tool_call_index}")
-                )
-                tool_call_ids[tool_call_index] = tool_call_id
-                if (function := tool_call.function) is not None:
-                    tool_call_chunk = ToolCallChunk(
-                        id=tool_call_id,
-                        function=FunctionCallChunk(
-                            name=function.name or "",
-                            arguments=function.arguments or "",
-                        ),
-                    )
-                    yield tool_call_chunk
-
-        if (usage := response.usage) is not None:
-            self._attributes.update(dict(self._llm_token_counts(usage)))
-
-    def to_openai_o1_chat_completion_param(
+    def to_openai_chat_completion_param(
         self,
         role: ChatCompletionMessageRole,
         content: JSONScalarType,
@@ -625,6 +575,8 @@ class AzureOpenAIStreamingClient(OpenAIStreamingClient):
         super().__init__(model=model, api_key=api_key)
         self._attributes[LLM_PROVIDER] = OpenInferenceLLMProviderValues.AZURE.value
         self._attributes[LLM_SYSTEM] = OpenInferenceLLMSystemValues.OPENAI.value
+        if not (api_key := api_key or os.environ.get("AZURE_OPENAI_API_KEY")):
+            raise BadRequest("An Azure API key is required for Azure OpenAI models")
         if not (endpoint := model.endpoint or os.environ.get("AZURE_OPENAI_ENDPOINT")):
             raise BadRequest("An Azure endpoint is required for Azure OpenAI models")
         if not (api_version := model.api_version or os.environ.get("OPENAI_API_VERSION")):
@@ -640,8 +592,12 @@ class AzureOpenAIStreamingClient(OpenAIStreamingClient):
     provider_key=GenerativeProviderKey.ANTHROPIC,
     model_names=[
         PROVIDER_DEFAULT,
+        "claude-3-5-sonnet-latest",
+        "claude-3-5-haiku-latest",
+        "claude-3-5-sonnet-20241022",
+        "claude-3-5-haiku-20241022",
         "claude-3-5-sonnet-20240620",
-        "claude-3-opus-20240229",
+        "claude-3-opus-latest",
         "claude-3-sonnet-20240229",
         "claude-3-haiku-20240307",
     ],
@@ -674,12 +630,14 @@ class AnthropicStreamingClient(PlaygroundStreamingClient):
                 invocation_name="max_tokens",
                 canonical_name=CanonicalParameterName.MAX_COMPLETION_TOKENS,
                 label="Max Tokens",
+                default_value=1024,
                 required=True,
             ),
             BoundedFloatInvocationParameter(
                 invocation_name="temperature",
                 canonical_name=CanonicalParameterName.TEMPERATURE,
                 label="Temperature",
+                default_value=1.0,
                 min_value=0.0,
                 max_value=1.0,
             ),
@@ -692,6 +650,7 @@ class AnthropicStreamingClient(PlaygroundStreamingClient):
                 invocation_name="top_p",
                 canonical_name=CanonicalParameterName.TOP_P,
                 label="Top P",
+                default_value=1.0,
                 min_value=0.0,
                 max_value=1.0,
             ),
@@ -850,7 +809,7 @@ class GeminiStreamingClient(PlaygroundStreamingClient):
                 invocation_name="temperature",
                 canonical_name=CanonicalParameterName.TEMPERATURE,
                 label="Temperature",
-                default_value=0.0,
+                default_value=1.0,
                 min_value=0.0,
                 max_value=2.0,
             ),
@@ -860,35 +819,31 @@ class GeminiStreamingClient(PlaygroundStreamingClient):
                 label="Max Output Tokens",
             ),
             StringListInvocationParameter(
-                invocation_name="stop",
+                invocation_name="stop_sequences",
                 canonical_name=CanonicalParameterName.STOP_SEQUENCES,
                 label="Stop Sequences",
             ),
             FloatInvocationParameter(
                 invocation_name="presence_penalty",
                 label="Presence Penalty",
+                default_value=0.0,
             ),
             FloatInvocationParameter(
                 invocation_name="frequency_penalty",
                 label="Frequency Penalty",
+                default_value=0.0,
             ),
             BoundedFloatInvocationParameter(
                 invocation_name="top_p",
                 canonical_name=CanonicalParameterName.TOP_P,
                 label="Top P",
+                default_value=1.0,
                 min_value=0.0,
                 max_value=1.0,
             ),
-            BoundedFloatInvocationParameter(
+            IntInvocationParameter(
                 invocation_name="top_k",
                 label="Top K",
-                min_value=0.0,
-                max_value=1.0,
-            ),
-            IntInvocationParameter(
-                invocation_name="seed",
-                canonical_name=CanonicalParameterName.RANDOM_SEED,
-                label="Seed",
             ),
         ]
 
@@ -923,6 +878,13 @@ class GeminiStreamingClient(PlaygroundStreamingClient):
         chat = client.start_chat(history=gemini_message_history)
         stream = await chat.send_message_async(**gemini_params)
         async for event in stream:
+            self._attributes.update(
+                {
+                    LLM_TOKEN_COUNT_PROMPT: event.usage_metadata.prompt_token_count,
+                    LLM_TOKEN_COUNT_COMPLETION: event.usage_metadata.candidates_token_count,
+                    LLM_TOKEN_COUNT_TOTAL: event.usage_metadata.total_token_count,
+                }
+            )
             yield TextChunk(content=event.text)
 
     def _build_gemini_messages(

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/src/phoenix/server/api/helpers/playground_spans.py

@@ -31,6 +31,7 @@ from typing_extensions import Self, TypeAlias, assert_never
 
 from phoenix.datetime_utils import local_now, normalize_datetime
 from phoenix.db import models
+from phoenix.server.api.helpers.dataset_helpers import get_dataset_example_output
 from phoenix.server.api.input_types.ChatCompletionInput import (
     ChatCompletionInput,
     ChatCompletionOverDatasetInput,
@@ -220,7 +221,7 @@ def get_db_experiment_run(
         dataset_example_id=example_id,
         trace_id=db_trace.trace_id,
         output=models.ExperimentRunOutput(
-            task_output=get_attribute_value(db_span.attributes, LLM_OUTPUT_MESSAGES),
+            task_output=get_dataset_example_output(db_span),
         ),
         repetition_number=1,
         start_time=db_span.start_time,

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/src/phoenix/server/api/mutations/chat_mutations.py

@@ -25,8 +25,10 @@ from typing_extensions import assert_never
 from phoenix.datetime_utils import local_now, normalize_datetime
 from phoenix.db import models
 from phoenix.db.helpers import get_dataset_example_revisions
+from phoenix.server.api.auth import IsNotReadOnly
 from phoenix.server.api.context import Context
 from phoenix.server.api.exceptions import BadRequest, CustomGraphQLError, NotFound
+from phoenix.server.api.helpers.dataset_helpers import get_dataset_example_output
 from phoenix.server.api.helpers.playground_clients import (
     PlaygroundStreamingClient,
     initialize_playground_clients,
@@ -61,12 +63,13 @@ from phoenix.server.api.types.node import from_global_id_with_expected_type
 from phoenix.server.api.types.Span import Span, to_gql_span
 from phoenix.server.api.types.TemplateLanguage import TemplateLanguage
 from phoenix.server.dml_event import SpanInsertEvent
-from phoenix.trace.attributes import get_attribute_value, unflatten
+from phoenix.trace.attributes import unflatten
 from phoenix.trace.schemas import SpanException
 from phoenix.utilities.json import jsonify
 from phoenix.utilities.template_formatters import (
     FStringTemplateFormatter,
     MustacheTemplateFormatter,
+    NoOpFormatter,
     TemplateFormatter,
 )
 
@@ -117,7 +120,7 @@ class ChatCompletionOverDatasetMutationPayload:
 
 @strawberry.type
 class ChatCompletionMutationMixin:
-    @strawberry.mutation
+    @strawberry.mutation(permission_classes=[IsNotReadOnly])  # type: ignore
     @classmethod
     async def chat_completion_over_dataset(
         cls,
@@ -242,7 +245,7 @@ class ChatCompletionMutationMixin:
                     dataset_example_id=revision.dataset_example_id,
                     trace_id=str(result.span.context.trace_id),
                     output=models.ExperimentRunOutput(
-                        task_output=get_attribute_value(db_span.attributes, LLM_OUTPUT_MESSAGES),
+                        task_output=get_dataset_example_output(db_span),
                     ),
                     prompt_token_count=db_span.cumulative_llm_token_count_prompt,
                     completion_token_count=db_span.cumulative_llm_token_count_completion,
@@ -272,7 +275,7 @@ class ChatCompletionMutationMixin:
             payload.examples.append(example_payload)
         return payload
 
-    @strawberry.mutation
+    @strawberry.mutation(permission_classes=[IsNotReadOnly])  # type: ignore
     @classmethod
     async def chat_completion(
         cls, info: Info[Context, None], input: ChatCompletionInput
@@ -481,6 +484,8 @@ def _template_formatter(template_language: TemplateLanguage) -> TemplateFormatte
         return MustacheTemplateFormatter()
     if template_language is TemplateLanguage.F_STRING:
         return FStringTemplateFormatter()
+    if template_language is TemplateLanguage.NONE:
+        return NoOpFormatter()
     assert_never(template_language)
 
 

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/src/phoenix/server/api/mutations/dataset_mutations.py

@@ -136,23 +136,7 @@ class DatasetMutationMixin:
                 .returning(models.DatasetVersion.id)
             )
             spans = (
-                await session.execute(
-                    select(
-                        models.Span.id,
-                        models.Span.span_kind,
-                        models.Span.attributes,
-                        _span_attribute(INPUT_MIME_TYPE),
-                        _span_attribute(INPUT_VALUE),
-                        _span_attribute(OUTPUT_MIME_TYPE),
-                        _span_attribute(OUTPUT_VALUE),
-                        _span_attribute(LLM_PROMPT_TEMPLATE_VARIABLES),
-                        _span_attribute(LLM_INPUT_MESSAGES),
-                        _span_attribute(LLM_OUTPUT_MESSAGES),
-                        _span_attribute(RETRIEVAL_DOCUMENTS),
-                    )
-                    .select_from(models.Span)
-                    .where(models.Span.id.in_(span_rowids))
-                )
+                await session.scalars(select(models.Span).where(models.Span.id.in_(span_rowids)))
             ).all()
             if missing_span_rowids := span_rowids - {span.id for span in spans}:
                 raise ValueError(
@@ -160,18 +144,10 @@ class DatasetMutationMixin:
                 )  # todo: implement error handling types https://github.com/Arize-ai/phoenix/issues/3221
 
             span_annotations = (
-                await session.execute(
-                    select(
-                        models.SpanAnnotation.span_rowid,
-                        models.SpanAnnotation.name,
-                        models.SpanAnnotation.label,
-                        models.SpanAnnotation.score,
-                        models.SpanAnnotation.explanation,
-                        models.SpanAnnotation.metadata_,
-                        models.SpanAnnotation.annotator_kind,
+                await session.scalars(
+                    select(models.SpanAnnotation).where(
+                        models.SpanAnnotation.span_rowid.in_(span_rowids)
                     )
-                    .select_from(models.SpanAnnotation)
-                    .where(models.SpanAnnotation.span_rowid.in_(span_rowids))
                 )
             ).all()
 
@@ -214,8 +190,12 @@ class DatasetMutationMixin:
                         DatasetExampleRevision.input.key: get_dataset_example_input(span),
                         DatasetExampleRevision.output.key: get_dataset_example_output(span),
                         DatasetExampleRevision.metadata_.key: {
-                            **span.attributes,
-                            "annotations": span_annotations_by_span[span.id],
+                            "span_kind": span.span_kind,
+                            **(
+                                {"annotations": annotations}
+                                if (annotations := span_annotations_by_span[span.id])
+                                else {}
+                            ),
                         },
                         DatasetExampleRevision.revision_kind.key: "CREATE",
                     }

{arize_phoenix-5.11.0 → arize_phoenix-6.0.0}/src/phoenix/server/api/routers/v1/evaluations.py

@@ -40,7 +40,7 @@ from .utils import add_errors_to_responses
 
 EvaluationName: TypeAlias = str
 
-router = APIRouter(tags=["traces"], include_in_schema=False)
+router = APIRouter(tags=["traces"], include_in_schema=True)
 
 
 @router.post(