Fix easy baseline errors (1/?)

cl/ai/management/commands/send_gemini_file_batches.py

@@ -191,22 +191,20 @@ def get_s3_file_list(
 
     # Create S3 client with session token support for dev mode
     try:
-        client_kwargs = {
-            "service_name": "s3",
-            "aws_access_key_id": key_id,
-            "aws_secret_access_key": secret,
-        }
-
+        aws_session_token = None
         # In dev mode, add session token if available
         if settings.DEVELOPMENT:
             env = environ.FileAwareEnv()
-            session_token = env("AWS_SESSION_TOKEN", default=None) or env(
+            aws_session_token = env("AWS_SESSION_TOKEN", default=None) or env(
                 "AWS_DEV_SESSION_TOKEN", default=None
             )
-            if session_token:
-                client_kwargs["aws_session_token"] = session_token
 
-        s3_client = boto3.client(**client_kwargs)
+        s3_client = boto3.client(
+            "s3",
+            aws_access_key_id=key_id,
+            aws_secret_access_key=secret,
+            aws_session_token=aws_session_token,
+        )
     except (BotoCoreError, ClientError) as e:
         raise CommandError(f"Failed to create S3 client: {e}")
 

cl/api/pagination.py

@@ -326,19 +326,14 @@ def get_paginated_response(
 
         base_response = {
             "count": self.get_results_count(),
-        }
-        remaining_fields = {
             "next": self.get_next_link(cached_response),
             "previous": self.get_previous_link(cached_response),
             "results": data,
         }
 
         if self.search_type == SEARCH_TYPES.RECAP:
             # Include the document_count for the "r" search type.
-            base_response.update(
-                {"document_count": self.get_child_results_count()}
-            )
-        base_response.update(remaining_fields)
+            base_response["document_count"] = self.get_child_results_count()
         return Response(base_response)
 
     def get_next_link(self, cached_response: bool) -> str | None:

cl/api/utils.py

@@ -2186,6 +2186,7 @@ def fields(self):
             allowed = set(filter(None, filter_fields))
 
         # omit fields in the `omit` argument.
+        # TODO: Typing this correctly involves going through a long chain of things that probably also need to be typed
         omitted = set(filter(None, omit_fields))
 
         for field in existing:

cl/audio/tasks.py

@@ -21,6 +21,7 @@
     OpenAI,
     RateLimitError,
     UnprocessableEntityError,
+    omit,
 )
 from sentry_sdk import capture_exception
 
@@ -152,23 +153,21 @@ def transcribe_from_open_ai_api(self, audio_pk: int, dont_retry: bool = False):
 
         # Prevent default openai client retrying
         client = stack.enter_context(OpenAI(max_retries=0))
-        kwargs = {
-            "file": file,
-            "model": "whisper-1",
-            "language": "en",
-            "response_format": "verbose_json",
-            "timestamp_granularities": ["word", "segment"],
-            "prompt": audio.case_name,
-        }
-
-        # The most common hallucination we have seen is the case name
-        # repeated in a loop. Manual testing showed that not sending
-        # the case name helps to get a clean transcript
-        if audio.stt_status == Audio.STT_HALLUCINATION:
-            kwargs.pop("prompt", "")
 
         try:
-            transcript = client.audio.transcriptions.create(**kwargs)
+            transcript = client.audio.transcriptions.create(
+                file=file,
+                model="whisper-1",
+                language="en",
+                response_format="verbose_json",
+                timestamp_granularities=["word", "segment"],
+                # The most common hallucination we have seen is the case name
+                # repeated in a loop. Manual testing showed that not sending
+                # the case name helps to get a clean transcript
+                prompt=omit
+                if audio.stt_status == Audio.STT_HALLUCINATION
+                else audio.case_name,
+            )
         except APIConnectionError as exc:
             # Transient TCP / DNS blip. Usually resolves in seconds, so a
             # short in-task retry is cheaper than waiting a full daemon
@@ -217,11 +216,9 @@ def transcribe_from_open_ai_api(self, audio_pk: int, dont_retry: bool = False):
             capture_exception(e)
             return
 
-        transcript_dict = transcript.to_dict()
-
         with transaction.atomic():
-            audio.stt_transcript = transcript_dict["text"]
-            audio.duration = ceil(transcript_dict["duration"])
+            audio.stt_transcript = transcript.text
+            audio.duration = ceil(transcript.duration)
             audio.stt_source = Audio.STT_OPENAI_WHISPER
 
             if transcription_was_hallucinated(audio):
@@ -234,6 +231,7 @@ def transcribe_from_open_ai_api(self, audio_pk: int, dont_retry: bool = False):
                 audio.stt_status = Audio.STT_COMPLETE
 
             audio.save()
+            transcript_dict = transcript.to_dict()
             metadata = {
                 "segments": transcript_dict["segments"],
                 "words": transcript_dict["words"],

cl/audio/tests.py

@@ -527,6 +527,10 @@ def setUpTestData(cls) -> None:
         }
 
         class OpenAITranscription:
+            def __init__(self):
+                self.text = cls.open_ai_api_returned_dict["text"]
+                self.duration = cls.open_ai_api_returned_dict["duration"]
+
             def to_dict(self):
                 return cls.open_ai_api_returned_dict