arize-phoenix 2.3.0__tar.gz → 2.4.1__tar.gz

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: arize-phoenix
-Version: 2.3.0
+Version: 2.4.1
 Summary: ML Observability in your notebook
 Project-URL: Documentation, https://docs.arize.com/phoenix/
 Project-URL: Issues, https://github.com/Arize-ai/phoenix/issues

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/pyproject.toml

@@ -292,6 +292,7 @@ module = [
   "litellm",
   "nest_asyncio",
   "opentelemetry.*",
+  "pyarrow",
 ]
 ignore_missing_imports = true
 

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/src/phoenix/config.py

@@ -40,6 +40,7 @@ for path in (
     ROOT_DIR := Path.home().resolve() / ".phoenix",
     EXPORT_DIR := ROOT_DIR / "exports",
     DATASET_DIR := ROOT_DIR / "datasets",
+    TRACE_DATASET_DIR := ROOT_DIR / "trace_datasets",
 ):
     path.mkdir(parents=True, exist_ok=True)
 

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/src/phoenix/experimental/evals/evaluators.py

@@ -36,6 +36,9 @@ class LLMEvaluator:
         self._model = model
         self._template = template
 
+    def reload_client(self) -> None:
+        self._model.reload_client()
+
     def evaluate(
         self,
         record: Record,

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/src/phoenix/experimental/evals/functions/classify.py

@@ -127,6 +127,9 @@ def llm_classify(
         from the entries in the rails argument or "NOT_PARSABLE" if the model's output could
         not be parsed.
     """
+    # clients need to be reloaded to ensure that async evals work properly
+    model.reload_client()
+
     tqdm_bar_format = get_tqdm_progress_bar_formatter("llm_classify")
     use_openai_function_call = (
         use_function_calling_if_available
@@ -385,6 +388,9 @@ def run_evals(
         List[DataFrame]: A list of dataframes, one for each evaluator, all of
         which have the same number of rows as the input dataframe.
     """
+    # clients need to be reloaded to ensure that async evals work properly
+    for evaluator in evaluators:
+        evaluator.reload_client()
 
     async def _arun_eval(
         payload: RunEvalsPayload,

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/src/phoenix/experimental/evals/functions/generate.py

@@ -78,6 +78,9 @@ def llm_generate(
         represents the generated output
 
     """
+    # clients need to be reloaded to ensure that async evals work properly
+    model.reload_client()
+
     tqdm_bar_format = get_tqdm_progress_bar_formatter("llm_generate")
     output_parser = output_parser or _no_op_parser
     template = normalize_prompt_template(template)

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/src/phoenix/experimental/evals/models/base.py

@@ -61,6 +61,9 @@ class BaseEvalModel(ABC):
     _verbose: bool = False
     _rate_limiter: RateLimiter = field(default_factory=RateLimiter)
 
+    def reload_client(self) -> None:
+        pass
+
     def _retry(
         self,
         error_types: List[Type[BaseException]],

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/src/phoenix/experimental/evals/models/openai.py

@@ -107,6 +107,9 @@ class OpenAIModel(BaseEvalModel):
         self._init_tiktoken()
         self._init_rate_limiter()
 
+    def reload_client(self) -> None:
+        self._init_open_ai()
+
     def _init_environment(self) -> None:
         try:
             import httpx

{arize_phoenix-2.3.0 → arize_phoenix-2.4.1}/src/phoenix/experimental/evals/models/vertex.py

@@ -4,6 +4,7 @@ from typing import TYPE_CHECKING, Any, Dict, List
 
 from phoenix.experimental.evals.models.base import BaseEvalModel
 from phoenix.experimental.evals.models.rate_limiters import RateLimiter
+from phoenix.utilities.logging import printif
 
 if TYPE_CHECKING:
     from tiktoken import Encoding
@@ -125,8 +126,7 @@ class GeminiModel(BaseEvalModel):
             response = self._model.generate_content(
                 contents=prompt, generation_config=generation_config, **kwargs
             )
-            candidate = response.candidates[0]
-            return candidate.text
+            return self._parse_response_candidates(response)
 
         return _completion_with_retry(**kwargs)
 
@@ -149,7 +149,29 @@ class GeminiModel(BaseEvalModel):
             response = await self._model.generate_content_async(
                 contents=prompt, generation_config=generation_config, **kwargs
             )
-            candidate = response.candidates[0]
-            return candidate.text
+            return self._parse_response_candidates(response)
 
         return await _completion_with_retry(**kwargs)
+
+    def _parse_response_candidates(self, response: Any) -> Any:
+        if hasattr(response, "candidates"):
+            if isinstance(response.candidates, list) and len(response.candidates) > 0:
+                try:
+                    candidate = response.candidates[0].text
+                except ValueError:
+                    printif(
+                        self._verbose, "The 'candidates' object does not have a 'text' attribute."
+                    )
+                    printif(self._verbose, response.candidates[0])
+                    candidate = ""
+            else:
+                printif(
+                    self._verbose,
+                    "The 'candidates' attribute of 'response' is either not a list or is empty.",
+                )
+                printif(self._verbose, response)
+                candidate = ""
+        else:
+            printif(self._verbose, "The 'response' object does not have a 'candidates' attribute.")
+            candidate = ""
+        return candidate