arize-phoenix 5.12.0__py3-none-any.whl → 6.1.0__py3-none-any.whl

{arize_phoenix-5.12.0.dist-info → arize_phoenix-6.1.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: arize-phoenix
-Version: 5.12.0
+Version: 6.1.0
 Summary: AI Observability and Evaluation
 Project-URL: Documentation, https://docs.arize.com/phoenix/
 Project-URL: Issues, https://github.com/Arize-ai/phoenix/issues
@@ -196,8 +196,8 @@ Phoenix container images are available via [Docker Hub](https://hub.docker.com/r
 | Projects                                                                                                         | ✅             |
 | [Self-Hosting](https://docs.arize.com/phoenix/deployment)                                                        | ✅             |
 | Jupyter Notebooks                                                                                                | ✅             |
+| [Prompt Playground](https://docs.arize.com/phoenix/prompt-engineering/overview-prompts)                          | ✅             |
 | [Sessions](https://github.com/Arize-ai/phoenix/issues/2619)                                                      | In Progress 🚧 |
-| [Prompt Playground](https://github.com/Arize-ai/phoenix/issues/3435)                                             | In Progress 🚧 |
 | Prompt Management                                                                                                | Coming soon ⏱️ |
 
 ## Tracing Integrations

{arize_phoenix-5.12.0.dist-info → arize_phoenix-6.1.0.dist-info}/RECORD

@@ -6,7 +6,7 @@ phoenix/exceptions.py,sha256=n2L2KKuecrdflB9MsCdAYCiSEvGJptIsfRkXMoJle7A,169
 phoenix/py.typed,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
 phoenix/services.py,sha256=kpW1WL0kiB8XJsO6XycvZVJ-lBkNoenhQ7atCvBoSe8,5365
 phoenix/settings.py,sha256=ht-0oN-sMV6SPXrk7Tu1EZlngpAYkGNLYPhO8DyrdQI,661
-phoenix/version.py,sha256=9dFDHPiqLy0dTf00acCCS0XDudNj7hSnMr9G2iMDI6A,23
+phoenix/version.py,sha256=PD5agZ0_DXqsFLMBR0hacZV3LbzGbH8qn-j7MIAm3Xs,22
 phoenix/core/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/core/embedding_dimension.py,sha256=zKGbcvwOXgLf-yrJBpQyKtd-LEOPRKHnUToyAU8Owis,87
 phoenix/core/model.py,sha256=qBFraOtmwCCnWJltKNP18DDG0mULXigytlFsa6YOz6k,4837
@@ -47,7 +47,7 @@ phoenix/experiments/utils.py,sha256=MZ1-OnTcavk_KUtbfGqt55Fk9TGtJpYG_K71WsN-zDk,
 phoenix/experiments/evaluators/__init__.py,sha256=j63fi3fa3U7-itVPHa82GowhjQRU-wO6yhO34u_lhsA,714
 phoenix/experiments/evaluators/base.py,sha256=zefFLqyYLMxJnZxDs1S0QCrnb43YJraoadCMqRMU72A,5576
 phoenix/experiments/evaluators/code_evaluators.py,sha256=JIu_8GezA3UKA7isUmZvBxjcJeXllIDESEgdF2zeDck,6755
-phoenix/experiments/evaluators/llm_evaluators.py,sha256=6pvCWP1a3tLMyufaU_U8TkItJO2Q03hkEf7kcI8gdGY,20549
+phoenix/experiments/evaluators/llm_evaluators.py,sha256=1HsZQsKNWA_19EHrjI6OXqm1aBwvIKS-qHYc3R0bOqc,20551
 phoenix/experiments/evaluators/utils.py,sha256=W-k2bHL67ST-B7ne8t5LvCFHGBaKCPHbOv3x1SlDENU,9330
 phoenix/inferences/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/inferences/errors.py,sha256=AOsCs-xR1LV469Gygje16EOK3qGheiPGW0Ylg-MakhQ,8282
@@ -124,10 +124,10 @@ phoenix/server/api/dataloaders/users.py,sha256=Uh86Kny_xpqDdEvEklcTBUA_MELgu4Z0j
 phoenix/server/api/dataloaders/cache/__init__.py,sha256=SYoOM9n8FJaMdQarma5d1blu-jIg2GB8Shqg5ezSzZ8,106
 phoenix/server/api/dataloaders/cache/two_tier_cache.py,sha256=cmo8FUT3E91R139IEzh4yCga-6nTamc5KPXAfMrzNDM,2315
 phoenix/server/api/helpers/__init__.py,sha256=m2-xaSPqUiSs91k62JaRDjFNfl-1byxBfY-m_Vxw16U,272
-phoenix/server/api/helpers/dataset_helpers.py,sha256=14mldZp9to3rr9BdvvoFqEwZHHV_k2e7jPm8q9z2OdQ,6896
-phoenix/server/api/helpers/playground_clients.py,sha256=hIxAM09HexNyKk0HQYbAD-X5axyw7my7B0dhPFzwueo,36383
+phoenix/server/api/helpers/dataset_helpers.py,sha256=AMlKY9_e0wnTrTSSQemM5NHfnpwARSytx-m9YK6f6bY,8421
+phoenix/server/api/helpers/playground_clients.py,sha256=zgpYRn7c8Yi6-0ZW7miKTDKnKor9nq4U40-mjgtqIGY,36164
 phoenix/server/api/helpers/playground_registry.py,sha256=CPLMziFB2wmr-dfbx7VbzO2f8YIG_k5RftzvGXYGQ1w,2570
-phoenix/server/api/helpers/playground_spans.py,sha256=LJjndkVxr6DsfVFerrtMZWlRd5YL5AIbcM2CvFe8ajc,16489
+phoenix/server/api/helpers/playground_spans.py,sha256=qGk7V7IZK7EkRE1mvZyROpLN5kgOahOZifFzUWmqYFc,16546
 phoenix/server/api/input_types/AddExamplesToDatasetInput.py,sha256=mIQz0S_z8YdrktKIY6RCvtNJ2yZF9pYvTGgasUsI-54,430
 phoenix/server/api/input_types/AddSpansToDatasetInput.py,sha256=-StSstyMAVrba3tG1U30b-srkKCtu_svflQuSM19iJA,362
 phoenix/server/api/input_types/ChatCompletionInput.py,sha256=g_5ARuwylt-uCVAsGyZPEVtidEQiOhbKakvDQsZumzw,1451
@@ -164,8 +164,8 @@ phoenix/server/api/input_types/UserRoleInput.py,sha256=xxhFe0ITZOgRVEJbVem_W6F1I
 phoenix/server/api/input_types/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/server/api/mutations/__init__.py,sha256=1wClieLNA3_Tin4Ah67rkrQvwSSZAdPU0EPsRiUxyAA,1103
 phoenix/server/api/mutations/api_key_mutations.py,sha256=OmPipsmlQIb6DKvAgO58mZUwkYJihlJB2N4lTyeUlAA,6164
-phoenix/server/api/mutations/chat_mutations.py,sha256=Nz7pz6cTvxub33kW6J0KgYmf7Lhb1ZmzF65KeUqYjiQ,22795
-phoenix/server/api/mutations/dataset_mutations.py,sha256=siwsSmozKRIK8ZhPrfWl-GxKCL4lAmRPms862tG2KXY,27064
+phoenix/server/api/mutations/chat_mutations.py,sha256=YUZGgoUoA-652UVo5p7qWNnN4u7tdU6XfhDnTIxvbqU,22831
+phoenix/server/api/mutations/dataset_mutations.py,sha256=OkVek-GtDf_1eloI7Sdz6W6l5FFmwU_41LJM8-fw3q4,26088
 phoenix/server/api/mutations/experiment_mutations.py,sha256=p3CoLAa8nFPa3D759Y2A7De_PVJNGOL98mA3HoZBrRQ,3188
 phoenix/server/api/mutations/export_events_mutations.py,sha256=xoDnVWC7eA_8wNQP0-oyiHojyUZ0EhVVSrsAnztetC0,3993
 phoenix/server/api/mutations/project_mutations.py,sha256=KdcA3c6tgLZ8QhpIv37qJESTBPxs56SbfeYdDKzU7-8,2263
@@ -182,7 +182,7 @@ phoenix/server/api/routers/oauth2.py,sha256=bSrTZAAWW4WgZVwkr39xbo5jZEYL4w4wCbEe
 phoenix/server/api/routers/utils.py,sha256=M41BoH-fl37izhRuN2aX7lWm7jOC20A_3uClv9TVUUY,583
 phoenix/server/api/routers/v1/__init__.py,sha256=aLEHzzU8kQo4Oqsv2an35lH5VYUxAZQrcG7CXZA_Lx4,2214
 phoenix/server/api/routers/v1/datasets.py,sha256=tNh0CxAvSkWh-_5AwisGN1degQlUNGU3uufGa7MIbOw,36985
-phoenix/server/api/routers/v1/evaluations.py,sha256=g-Z5Dk33Epk7WrxlCyFsHHwneHvI2KI7x4Tb5nGw9m4,12648
+phoenix/server/api/routers/v1/evaluations.py,sha256=RpOkTylp5Da6BvPZGuN8ksnxz_BVXRIwyOvwX9Iko8U,12647
 phoenix/server/api/routers/v1/experiment_evaluations.py,sha256=1XuKqE5F6CC6R4kF4g4YdBtcYvch1rCk8Am5CGyqfJ4,4838
 phoenix/server/api/routers/v1/experiment_runs.py,sha256=xKXhol1-G4zAFECAQK9lAjKtSJsvB0Bp6bAXJYqJ7eI,6387
 phoenix/server/api/routers/v1/experiments.py,sha256=sBxOIE4k4eIt-sQ6lPo2HUv9Tf1aURXtTJTq8J-Rfto,11808
@@ -241,7 +241,7 @@ phoenix/server/api/types/Retrieval.py,sha256=OhMK2ncjoyp5h1yjKhjlKpoTbQrMHuxmgSF
 phoenix/server/api/types/ScalarDriftMetricEnum.py,sha256=IUAcRPpgL41WdoIgK6cNk2Te38SspXGyEs-S1fY23_A,232
 phoenix/server/api/types/Segments.py,sha256=vT2v0efoa5cuBKxLtxTnsUP5YJJCZfTloM71Spu0tMI,2915
 phoenix/server/api/types/SortDir.py,sha256=OUpXhlCzCxPoXSDkJJygEs9Rw9pMymfaZUG5zPTrw4Y,152
-phoenix/server/api/types/Span.py,sha256=6GS6MpJ3f8P2LrQUe2TWPrPf7ENxmde_wisQkJguphw,16919
+phoenix/server/api/types/Span.py,sha256=wJGvenSL6d1oDmpJt4hsLjIVTxmemJBRCBKHy4rYztc,15355
 phoenix/server/api/types/SpanAnnotation.py,sha256=6b5G-b_OoRvDL2ayWk7MkbqarLK-F-pQMx21CpUuNGY,1168
 phoenix/server/api/types/SystemApiKey.py,sha256=2ym8EgsTBIvxx1l9xZ-2YMovz58ZwYb_MaHBTJ9NH2E,166
 phoenix/server/api/types/TemplateLanguage.py,sha256=6j_0uwO_GZIeCpR7sTOnxySXudT7qBSC6LFsjzbvW1o,160
@@ -273,15 +273,15 @@ phoenix/server/static/apple-touch-icon-76x76.png,sha256=CT_xT12I0u2i0WU8JzBZBuOQ
 phoenix/server/static/apple-touch-icon.png,sha256=fOfpjqGpWYbJ0eAurKsyoZP1EAs6ZVooBJ_SGk2ZkDs,3801
 phoenix/server/static/favicon.ico,sha256=bY0vvCKRftemZfPShwZtE93DiiQdaYaozkPGwNFr6H8,34494
 phoenix/server/static/modernizr.js,sha256=mvK-XtkNqjOral-QvzoqsyOMECXIMu5BQwSVN_wcU9c,2564
-phoenix/server/static/.vite/manifest.json,sha256=MbaJGZ9fqq-pJiyZWU0xbb2FXhmggvZc6DNnIWBeJow,1929
-phoenix/server/static/assets/components-72cQL1d1.js,sha256=wSbuZXCJY152KhyfYItsBNWOU8LqshboNmVlCof-r70,306866
-phoenix/server/static/assets/index-BowjltW-.js,sha256=Zgvw87yavo4A9jEyPQAzFKNBiAFF_SWzZopNsbfwMig,7290
-phoenix/server/static/assets/pages-DFAkBAUh.js,sha256=SMq23Vg-OZSvgSdKOkogJBhNXzlRfyl_oZYKJPyNqvY,636153
-phoenix/server/static/assets/vendor-DexmGnha.js,sha256=sieuT9Y_X4RXRfqPAXRBpPPPwdb3KCfkEXwCfKwyJFc,10899238
+phoenix/server/static/.vite/manifest.json,sha256=R621q8gnowayj2q8CpWgoj2oHbsTB8H_T7mU1UfiydI,1929
+phoenix/server/static/assets/components-CdiZ1Osh.js,sha256=iG9VJFu5MxrXzhYpAwM1Q9OsKOkOxJcFJZujcT55JNo,314108
+phoenix/server/static/assets/index-D_sCOjlG.js,sha256=5EarOgtXkqjy1ALckGtmpGJCiPo1xWdn-6ygjFoiqB8,8305
+phoenix/server/static/assets/pages-FArMEfgg.js,sha256=9CY_Ri-LnqY6xk2V95TN0NlvgbTmm4kCWdtVfa6oLqw,631219
+phoenix/server/static/assets/vendor-Bnv1dNRQ.js,sha256=sHBast25yiz3Tg7DMNCKx_KpPHkxi94PKJz_2-hD0_Y,10975957
 phoenix/server/static/assets/vendor-DxkFTwjz.css,sha256=nZrkr0u6NNElFGvpWHk9GTHeGoibCXCli1bE7mXZGZg,1816
-phoenix/server/static/assets/vendor-arizeai--Q3ol330.js,sha256=DmwyVdLagWoGvOKRRAlwDibT1Qk4-IPmTDaLlXFjRFQ,308483
-phoenix/server/static/assets/vendor-codemirror-B4bYvWa6.js,sha256=u5U1cT5PopN6AbYt4S55bI9X0REQ35Pa9BXZmb5Piww,392709
-phoenix/server/static/assets/vendor-recharts-B4ZzJhNh.js,sha256=gS2dHFlC9A4tMk1tmNhtQo04SDcD2gQSkS4N6QS7d_0,282859
+phoenix/server/static/assets/vendor-arizeai-BG6iwyLC.js,sha256=rHkAk1Q2wkNV_3MirT0MYdA-oyCKpu5bFhpsaISg9fs,308489
+phoenix/server/static/assets/vendor-codemirror-BotnVFFX.js,sha256=yZnTXw9fvnqCrz53jYGv1pXM2QX-9W_ioLS9417Mijw,392709
+phoenix/server/static/assets/vendor-recharts-Dy5gEFzQ.js,sha256=z5_pZ5IJNbdA7EjAYdIeGxd1L7HifEzKmxXvq7EVIBw,282859
 phoenix/server/static/assets/vendor-three-DwGkEfCM.js,sha256=0D12ZgKzfKCTSdSTKJBFR2RZO_xxeMXrqDp0AszZqHY,620972
 phoenix/server/templates/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/server/templates/index.html,sha256=ram6sfy2obf_F053ay35V30v-mnRWZ86rK-PstXLy1c,4457
@@ -322,9 +322,9 @@ phoenix/utilities/project.py,sha256=auVpARXkDb-JgeX5f2aStyFIkeKvGwN9l7qrFeJMVxI,
 phoenix/utilities/re.py,sha256=x8Xbk-Wa6qDMAtUd_7JtZvKtrYEuMY-bchB0n163_5c,2006
 phoenix/utilities/span_store.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoenix/utilities/template_formatters.py,sha256=gh9PJD6WEGw7TEYXfSst1UR4pWWwmjxMLrDVQ_CkpkQ,2779
-arize_phoenix-5.12.0.dist-info/METADATA,sha256=DTW4JHWstCwzlLmXkne9Ls3JG9uGXXpGjSNqqz2BfqI,22671
-arize_phoenix-5.12.0.dist-info/WHEEL,sha256=C2FUgwZgiLbznR-k0b_5k3Ai_1aASOXDss3lzCUsUug,87
-arize_phoenix-5.12.0.dist-info/entry_points.txt,sha256=Pgpn8Upxx9P8z8joPXZWl2LlnAlGc3gcQoVchb06X1Q,94
-arize_phoenix-5.12.0.dist-info/licenses/IP_NOTICE,sha256=JBqyyCYYxGDfzQ0TtsQgjts41IJoa-hiwDrBjCb9gHM,469
-arize_phoenix-5.12.0.dist-info/licenses/LICENSE,sha256=HFkW9REuMOkvKRACuwLPT0hRydHb3zNg-fdFt94td18,3794
-arize_phoenix-5.12.0.dist-info/RECORD,,
+arize_phoenix-6.1.0.dist-info/METADATA,sha256=8ecHY7M9HSPtGBPqVfj5np5ZbUEJ2ijGIHVGiw_u_Ak,22669
+arize_phoenix-6.1.0.dist-info/WHEEL,sha256=C2FUgwZgiLbznR-k0b_5k3Ai_1aASOXDss3lzCUsUug,87
+arize_phoenix-6.1.0.dist-info/entry_points.txt,sha256=Pgpn8Upxx9P8z8joPXZWl2LlnAlGc3gcQoVchb06X1Q,94
+arize_phoenix-6.1.0.dist-info/licenses/IP_NOTICE,sha256=JBqyyCYYxGDfzQ0TtsQgjts41IJoa-hiwDrBjCb9gHM,469
+arize_phoenix-6.1.0.dist-info/licenses/LICENSE,sha256=HFkW9REuMOkvKRACuwLPT0hRydHb3zNg-fdFt94td18,3794
+arize_phoenix-6.1.0.dist-info/RECORD,,

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"

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

phoenix/server/api/helpers/playground_clients.py

@@ -284,7 +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,
             ),
@@ -323,7 +323,6 @@ class OpenAIStreamingClient(PlaygroundStreamingClient):
             IntInvocationParameter(
                 invocation_name="seed",
                 canonical_name=CanonicalParameterName.RANDOM_SEED,
-                default_value=0,
                 label="Seed",
             ),
             JSONInvocationParameter(
@@ -496,7 +495,6 @@ class OpenAIO1StreamingClient(OpenAIStreamingClient):
                 invocation_name="seed",
                 canonical_name=CanonicalParameterName.RANDOM_SEED,
                 label="Seed",
-                default_value=0,
             ),
             JSONInvocationParameter(
                 invocation_name="tool_choice",
@@ -577,6 +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")):
@@ -637,7 +637,7 @@ class AnthropicStreamingClient(PlaygroundStreamingClient):
                 invocation_name="temperature",
                 canonical_name=CanonicalParameterName.TEMPERATURE,
                 label="Temperature",
-                default_value=0.0,
+                default_value=1.0,
                 min_value=0.0,
                 max_value=1.0,
             ),
@@ -809,7 +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,
             ),
@@ -819,7 +819,7 @@ class GeminiStreamingClient(PlaygroundStreamingClient):
                 label="Max Output Tokens",
             ),
             StringListInvocationParameter(
-                invocation_name="stop",
+                invocation_name="stop_sequences",
                 canonical_name=CanonicalParameterName.STOP_SEQUENCES,
                 label="Stop Sequences",
             ),
@@ -841,18 +841,9 @@ class GeminiStreamingClient(PlaygroundStreamingClient):
                 min_value=0.0,
                 max_value=1.0,
             ),
-            BoundedFloatInvocationParameter(
+            IntInvocationParameter(
                 invocation_name="top_k",
                 label="Top K",
-                default_value=1.0,
-                min_value=0.0,
-                max_value=1.0,
-            ),
-            IntInvocationParameter(
-                invocation_name="seed",
-                canonical_name=CanonicalParameterName.RANDOM_SEED,
-                label="Seed",
-                default_value=0,
             ),
         ]
 

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,

phoenix/server/api/mutations/chat_mutations.py

@@ -28,6 +28,7 @@ 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,
@@ -62,7 +63,7 @@ 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 (
@@ -244,7 +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,

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",
                     }

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(

phoenix/server/api/types/Span.py

@@ -1,13 +1,12 @@
 import json
 from collections.abc import Mapping, Sized
-from dataclasses import dataclass
 from datetime import datetime
 from enum import Enum
 from typing import TYPE_CHECKING, Any, Optional, cast
 
 import numpy as np
 import strawberry
-from openinference.semconv.trace import EmbeddingAttributes, SpanAttributes
+from openinference.semconv.trace import SpanAttributes
 from strawberry import ID, UNSET
 from strawberry.relay import Node, NodeID
 from strawberry.types import Info
@@ -39,18 +38,6 @@ from .SpanAnnotation import SpanAnnotation
 if TYPE_CHECKING:
     from phoenix.server.api.types.Project import Project
 
-EMBEDDING_EMBEDDINGS = SpanAttributes.EMBEDDING_EMBEDDINGS
-EMBEDDING_VECTOR = EmbeddingAttributes.EMBEDDING_VECTOR
-INPUT_MIME_TYPE = SpanAttributes.INPUT_MIME_TYPE
-INPUT_VALUE = SpanAttributes.INPUT_VALUE
-LLM_PROMPT_TEMPLATE_VARIABLES = SpanAttributes.LLM_PROMPT_TEMPLATE_VARIABLES
-LLM_INPUT_MESSAGES = SpanAttributes.LLM_INPUT_MESSAGES
-LLM_OUTPUT_MESSAGES = SpanAttributes.LLM_OUTPUT_MESSAGES
-METADATA = SpanAttributes.METADATA
-OUTPUT_MIME_TYPE = SpanAttributes.OUTPUT_MIME_TYPE
-OUTPUT_VALUE = SpanAttributes.OUTPUT_VALUE
-RETRIEVAL_DOCUMENTS = SpanAttributes.RETRIEVAL_DOCUMENTS
-
 
 @strawberry.enum
 class SpanKind(Enum):
@@ -236,21 +223,7 @@ class Span(Node):
         description="The span's attributes translated into an example revision for a dataset",
     )  # type: ignore
     async def as_example_revision(self, info: Info[Context, None]) -> SpanAsExampleRevision:
-        db_span = self.db_span
-        attributes = db_span.attributes
-        span_io = _SpanIO(
-            span_kind=db_span.span_kind,
-            input_value=get_attribute_value(attributes, INPUT_VALUE),
-            input_mime_type=get_attribute_value(attributes, INPUT_MIME_TYPE),
-            output_value=get_attribute_value(attributes, OUTPUT_VALUE),
-            output_mime_type=get_attribute_value(attributes, OUTPUT_MIME_TYPE),
-            llm_prompt_template_variables=get_attribute_value(
-                attributes, LLM_PROMPT_TEMPLATE_VARIABLES
-            ),
-            llm_input_messages=get_attribute_value(attributes, LLM_INPUT_MESSAGES),
-            llm_output_messages=get_attribute_value(attributes, LLM_OUTPUT_MESSAGES),
-            retrieval_documents=get_attribute_value(attributes, RETRIEVAL_DOCUMENTS),
-        )
+        span = self.db_span
 
         # Fetch annotations associated with this span
         span_annotations = await self.span_annotations(info)
@@ -265,13 +238,13 @@ class Span(Node):
             }
         # Merge annotations into the metadata
         metadata = {
-            **attributes,
-            "annotations": annotations,
+            "span_kind": span.span_kind,
+            **({"annotations": annotations} if annotations else {}),
         }
 
         return SpanAsExampleRevision(
-            input=get_dataset_example_input(span_io),
-            output=get_dataset_example_output(span_io),
+            input=get_dataset_example_input(span),
+            output=get_dataset_example_output(span),
             metadata=metadata,
         )
 
@@ -435,19 +408,9 @@ def _convert_metadata_to_string(metadata: Any) -> Optional[str]:
         return str(metadata)
 
 
-@dataclass
-class _SpanIO:
-    """
-    An class 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
+INPUT_MIME_TYPE = SpanAttributes.INPUT_MIME_TYPE
+INPUT_VALUE = SpanAttributes.INPUT_VALUE
+METADATA = SpanAttributes.METADATA
+OUTPUT_MIME_TYPE = SpanAttributes.OUTPUT_MIME_TYPE
+OUTPUT_VALUE = SpanAttributes.OUTPUT_VALUE
+RETRIEVAL_DOCUMENTS = SpanAttributes.RETRIEVAL_DOCUMENTS