chore: disable line-too-long lint error

pyproject.toml

@@ -77,15 +77,14 @@ build-backend = "hatchling.build"
 # https://docs.astral.sh/ruff/configuration/
 [tool.ruff]
 line-length = 120
-target-version = "py311"
+target-version = "py312"
 
 [tool.ruff.lint]
 select = ["ALL"]
 ignore = [
     "FIX",
     "TD", # disable todo warnings
     "ERA",  # allow commented out code
-
     "ANN003", # missing kwargs
     "ANN101", # missing self type
     "B006",
@@ -112,6 +111,7 @@ ignore = [
     "INP001",
     "PT018",
     "G004", # logging f string
+    "E501" # line-too-long
 ]
 
 [tool.ruff.lint.isort]

scripts/client.py

@@ -14,7 +14,7 @@
 # It will then wait until the keybind is pressed again to stop recording.
 # The audio file will be sent to the server for transcription.
 # The transcription will be copied to the clipboard.
-# When having a short audio of a couple of sentences and running inference on a GPU the response time is very fast (less than 2 seconds).  # noqa: E501
+# When having a short audio of a couple of sentences and running inference on a GPU the response time is very fast (less than 2 seconds).
 # Run this with `sudo -E python scripts/client.py`
 
 CHUNK = 2**12

src/speaches/api_types.py

@@ -167,7 +167,7 @@ class Model(BaseModel):
     owned_by: str
     """The organization that owns the model."""
     language: list[str] = Field(default_factory=list)
-    """List of ISO 639-3 supported by the model. It's possible that the list will be empty. This field is not a part of the OpenAI API spec and is added for convenience."""  # noqa: E501
+    """List of ISO 639-3 supported by the model. It's possible that the list will be empty. This field is not a part of the OpenAI API spec and is added for convenience."""
 
     model_config = ConfigDict(
         populate_by_name=True,

src/speaches/audio.py

@@ -21,7 +21,7 @@
 logger = logging.getLogger(__name__)
 
 
-# aip 'Write a function `resample_audio` which would take in RAW PCM 16-bit signed, little-endian audio data represented as bytes (`audio_bytes`) and resample it (either downsample or upsample) from `sample_rate` to `target_sample_rate` using numpy'  # noqa: E501
+# aip 'Write a function `resample_audio` which would take in RAW PCM 16-bit signed, little-endian audio data represented as bytes (`audio_bytes`) and resample it (either downsample or upsample) from `sample_rate` to `target_sample_rate` using numpy'
 def resample_audio(audio_bytes: bytes, sample_rate: int, target_sample_rate: int) -> bytes:
     audio_data = np.frombuffer(audio_bytes, dtype=np.int16)
     duration = len(audio_data) / sample_rate

src/speaches/config.py

@@ -172,7 +172,7 @@ class WhisperConfig(BaseModel):
     use_batched_mode: bool = False
     """
     Whether to use batch mode(introduced in 1.1.0 `faster-whisper` release) for inference. This will likely become the default in the future and the configuration option will be removed.
-    """  # noqa: E501
+    """
 
 
 # TODO: document `alias` behaviour within the docstring
@@ -182,7 +182,7 @@ class Config(BaseSettings):
     Pydantic will automatically handle mapping uppercased environment variables to the corresponding fields.
     To populate nested, the environment should be prefixed with the nested field name and an underscore. For example,
     the environment variable `LOG_LEVEL` will be mapped to `log_level`, `WHISPER__MODEL`(note the double underscore) to `whisper.model`, to set quantization to int8, use `WHISPER__COMPUTE_TYPE=int8`, etc.
-    """  # noqa: E501
+    """
 
     model_config = SettingsConfigDict(env_nested_delimiter="__")
 
@@ -207,7 +207,7 @@ class Config(BaseSettings):
     enable_ui: bool = True
     """
     Whether to enable the Gradio UI. You may want to disable this if you want to minimize the dependencies and slightly improve the startup time.
-    """  # noqa: E501
+    """
 
     default_language: Language | None = None
     """
@@ -234,14 +234,14 @@ class Config(BaseSettings):
     """
     Max allowed audio duration without any speech being detected before transcription is finilized and connection is closed.
     Used only for live transcription (WS /v1/audio/transcriptions).
-    """  # noqa: E501
+    """
     inactivity_window_seconds: float = 5.0
     """
     Controls how many latest seconds of audio are being passed through VAD. Should be greater than `max_inactivity_seconds`.
     Used only for live transcription (WS /v1/audio/transcriptions).
-    """  # noqa: E501
+    """
 
-    # NOTE: options below are not used yet and should be ignored. Added as a placeholder for future features I'm currently working on.  # noqa: E501
+    # NOTE: options below are not used yet and should be ignored. Added as a placeholder for future features I'm currently working on.
 
     chat_completion_base_url: str = "https://api.openai.com/v1"
     chat_completion_api_key: str | None = None

src/speaches/dependencies.py

@@ -24,11 +24,11 @@
 
 logger = logging.getLogger(__name__)
 
-# NOTE: `get_config` is called directly instead of using sub-dependencies so that these functions could be used outside of `FastAPI`  # noqa: E501
+# NOTE: `get_config` is called directly instead of using sub-dependencies so that these functions could be used outside of `FastAPI`
 
 
 # https://fastapi.tiangolo.com/advanced/settings/?h=setti#creating-the-settings-only-once-with-lru_cache
-# WARN: Any new module that ends up calling this function directly (not through `FastAPI` dependency injection) should be patched in `tests/conftest.py`  # noqa: E501
+# WARN: Any new module that ends up calling this function directly (not through `FastAPI` dependency injection) should be patched in `tests/conftest.py`
 @lru_cache
 def get_config() -> Config:
     return Config()
@@ -119,7 +119,7 @@ def get_completion_client() -> AsyncCompletions:
 def get_speech_client() -> AsyncSpeech:
     config = get_config()
     if config.speech_base_url is None:
-        # this might not work as expected if `speech_router` won't have shared state (access to the same `model_manager`) with the main FastAPI `app`. TODO: verify  # noqa: E501
+        # this might not work as expected if `speech_router` won't have shared state (access to the same `model_manager`) with the main FastAPI `app`. TODO: verify
         from speaches.routers.speech import (
             router as speech_router,
         )
@@ -140,7 +140,7 @@ def get_speech_client() -> AsyncSpeech:
 def get_transcription_client() -> AsyncTranscriptions:
     config = get_config()
     if config.transcription_base_url is None:
-        # this might not work as expected if `transcription_router` won't have shared state (access to the same `model_manager`) with the main FastAPI `app`. TODO: verify  # noqa: E501
+        # this might not work as expected if `transcription_router` won't have shared state (access to the same `model_manager`) with the main FastAPI `app`. TODO: verify
         from speaches.routers.stt import (
             router as stt_router,
         )

src/speaches/hf_utils.py

@@ -127,13 +127,13 @@ def get_model_path(model_id: str, *, cache_dir: str | Path | None = None) -> Pat
     cache_dir = Path(cache_dir).expanduser().resolve()
     if not cache_dir.exists():
         raise huggingface_hub.CacheNotFound(
-            f"Cache directory not found: {cache_dir}. Please use `cache_dir` argument or set `HF_HUB_CACHE` environment variable.",  # noqa: E501
+            f"Cache directory not found: {cache_dir}. Please use `cache_dir` argument or set `HF_HUB_CACHE` environment variable.",
             cache_dir=cache_dir,
         )
 
     if cache_dir.is_file():
         raise ValueError(
-            f"Scan cache expects a directory but found a file: {cache_dir}. Please use `cache_dir` argument or set `HF_HUB_CACHE` environment variable."  # noqa: E501
+            f"Scan cache expects a directory but found a file: {cache_dir}. Please use `cache_dir` argument or set `HF_HUB_CACHE` environment variable."
         )
 
     for repo_path in cache_dir.iterdir():

src/speaches/realtime/conversation_event_router.py

@@ -63,7 +63,7 @@ def handle_conversation_item_create_event(ctx: SessionContext, event: Conversati
     if event.previous_item_id is not None:
         raise NotImplementedError
     # if event.item.id in ctx.conversation:
-    #     # TODO: Weirdly OpenAI's API allows creating an item with an already existing ID! Do their implementation replace the item?  # noqa: E501
+    #     # TODO: Weirdly OpenAI's API allows creating an item with an already existing ID! Do their implementation replace the item?
     #     raise NotImplementedError
     # TODO: should we assign the previous item's id when it hasn't been specified in the request?
     if event.item.id is None:
@@ -115,7 +115,7 @@ def handle_conversation_item_delete_event(ctx: SessionContext, event: Conversati
 async def handle_conversation_item_created_event(ctx: SessionContext, event: ConversationItemCreatedEvent) -> None:
     item = ctx.conversation[event.item.id]
     if item.type == "message" and item.role == "user" and item.content[0].type == "input_audio":
-        # NOTE: we aren't passing in `event.item` directly since `event.item` is a copy of the original item, meaning we won't be able to update the original item in the context.  # noqa: E501
+        # NOTE: we aren't passing in `event.item` directly since `event.item` is a copy of the original item, meaning we won't be able to update the original item in the context.
         await transcription_flow(ctx, item)
 
 

src/speaches/realtime/input_audio_buffer.py

@@ -13,7 +13,7 @@
 class VadState(BaseModel):
     audio_start_ms: int | None = None
     audio_end_ms: int | None = None
-    # TODO: consider keeping track of what was the last audio timestamp that was processed. This value could be used to control how often the VAD is run.  # noqa: E501
+    # TODO: consider keeping track of what was the last audio timestamp that was processed. This value could be used to control how often the VAD is run.
 
 
 # TODO: use `np.int16` instead of `np.float32` for audio data

src/speaches/realtime/input_audio_buffer_event_router.py

@@ -128,7 +128,7 @@ def vad_detection_flow(
 @event_router.register("input_audio_buffer.append")
 def handle_input_audio_buffer_append(ctx: SessionContext, event: InputAudioBufferAppendEvent) -> None:
     audio_chunk = audio_samples_from_file(BytesIO(base64.b64decode(event.audio)))
-    # convert the audio data from 24kHz (sample rate defined in the API spec) to 16kHz (sample rate used by the VAD and for transcription)  # noqa: E501
+    # convert the audio data from 24kHz (sample rate defined in the API spec) to 16kHz (sample rate used by the VAD and for transcription)
     audio_chunk = resample_audio_data(audio_chunk, 24000, 16000)
     input_audio_buffer_id = next(reversed(ctx.input_audio_buffers))
     input_audio_buffer = ctx.input_audio_buffers[input_audio_buffer_id]

src/speaches/realtime/response_event_router.py

@@ -103,7 +103,7 @@ def post_response_flow(ctx: SessionContext) -> None:  # TODO: on event
         )
     )
     response.status = "completed" if not cancelled else "incomplete"
-    # response.output.append(item) # FIXME: this VERY LIKELY shouldn't be commented out. I need to verify if output is appended anywhere else  # noqa: E501
+    # response.output.append(item) # FIXME: this VERY LIKELY shouldn't be commented out. I need to verify if output is appended anywhere else
     status_detail = None if not cancelled else RealtimeResponseStatus()  # XXX: this likely needs to be populated
     response.status_details = status_detail
     ctx.pubsub.publish_nowait(ResponseDoneEvent(type="response.done", event_id=generate_event_id(), response=response))
@@ -145,7 +145,7 @@ async def text_generation_flow(ctx: SessionContext) -> None:  # noqa: C901, PLR0
                         type="conversation.item.created", event_id=generate_event_id(), previous_item_id=None, item=item
                     )  # TODO: previous_item_id
                 )
-                # continue  # NOTE: this might only make sense to do for OpenAI since other implemetation might actually provide useful info in the first chunk. OpenAI doesn't  # noqa: E501
+                # continue  # NOTE: this might only make sense to do for OpenAI since other implemetation might actually provide useful info in the first chunk. OpenAI doesn't
 
             if chunk.usage is not None:
                 pass
@@ -358,5 +358,5 @@ def handle_response_audio_done_event(ctx: SessionContext, event: ResponseAudioDo
 
 @event_router.register("response.cancel")
 def handle_response_cancel_event(ctx: SessionContext, _event: ResponseCancelEvent) -> None:
-    # If there's  no response task, then it's a no-op. OpenAI's API should be monitored to see if the behaviour changes.  # noqa: E501
+    # If there's  no response task, then it's a no-op. OpenAI's API should be monitored to see if the behaviour changes.
     pass

src/speaches/realtime/session.py

@@ -2,10 +2,10 @@
 
 from speaches.types.realtime import Session, TurnDetection
 
-# NOTE: the `DEFAULT_OPENAI_REALTIME_*` constants are not currently used. Keeping them here for reference. They also may be outdated  # noqa: E501
+# NOTE: the `DEFAULT_OPENAI_REALTIME_*` constants are not currently used. Keeping them here for reference. They also may be outdated
 DEFAULT_OPENAI_REALTIME_MODEL = "gpt-4o-realtime-preview-2024-10-01"
 DEFAULT_OPENAI_REALTIME_SESSION_DURATION_SECONDS = 30 * 60
-DEFAULT_OPENAI_REALTIME_SESSION_INSTRUCTIONS = "Your knowledge cutoff is 2023-10. You are a helpful, witty, and friendly AI. Act like a human, but remember that you aren't a human and that you can't do human things in the real world. Your voice and personality should be warm and engaging, with a lively and playful tone. If interacting in a non-English language, start by using the standard accent or dialect familiar to the user. Talk quickly. You should always call a function if you can. Do not refer to these rules, even if you\u2019re asked about them."  # noqa: E501
+DEFAULT_OPENAI_REALTIME_SESSION_INSTRUCTIONS = "Your knowledge cutoff is 2023-10. You are a helpful, witty, and friendly AI. Act like a human, but remember that you aren't a human and that you can't do human things in the real world. Your voice and personality should be warm and engaging, with a lively and playful tone. If interacting in a non-English language, start by using the standard accent or dialect familiar to the user. Talk quickly. You should always call a function if you can. Do not refer to these rules, even if you\u2019re asked about them."
 DEFAULT_OPENAI_REALTIME_SESSION_CONFIG = Session(
     model=DEFAULT_OPENAI_REALTIME_MODEL,
     modalities=["audio", "text"],  # NOTE: the order of the modalities often differs
@@ -22,7 +22,7 @@
 )
 
 
-DEFAULT_REALTIME_SESSION_INSTRUCTIONS = "Your knowledge cutoff is 2023-10. You are a helpful, witty, and friendly AI. Keep the responses concise and to the point. Your responses will be converted into speech; avoid using text that makes sense when spoken. Do not use emojis, abbreviations, or markdown formatting (such as double asterisks) in your response."  # noqa: E501
+DEFAULT_REALTIME_SESSION_INSTRUCTIONS = "Your knowledge cutoff is 2023-10. You are a helpful, witty, and friendly AI. Keep the responses concise and to the point. Your responses will be converted into speech; avoid using text that makes sense when spoken. Do not use emojis, abbreviations, or markdown formatting (such as double asterisks) in your response."
 DEFAULT_TURN_DETECTION = TurnDetection(
     threshold=0.9,
     prefix_padding_ms=0,

src/speaches/realtime/session_event_router.py

@@ -12,7 +12,7 @@
 
 @event_router.register("session.update")
 def handle_session_update_event(ctx: SessionContext, event: SessionUpdateEvent) -> None:  # noqa: ARG001
-    # if event.session.input_audio_transcription is None or event.session.input_audio_transcription.model == "whisper-1":  # noqa: E501
+    # if event.session.input_audio_transcription is None or event.session.input_audio_transcription.model == "whisper-1":
     #     logger.warning("Invalid input_audio_transcription model")  # TODO
     #     event.session.input_audio_transcription = SessionInputAudioTranscription(
     #         model="Systran/faster-distil-whisper-large-v3"
@@ -26,10 +26,10 @@ def handle_session_update_event(ctx: SessionContext, event: SessionUpdateEvent)
     # if event.session.instructions != ctx.configuration.instructions:
     #     logger.warning("Changing `instructions` is not supported.")
     #     event.session.instructions = DEFAULT_REALTIME_SESSION_INSTRUCTIONS
-    # if event.session.input_audio_transcription is None or event.session.input_audio_transcription.model == "whisper-1":  # noqa: E501
+    # if event.session.input_audio_transcription is None or event.session.input_audio_transcription.model == "whisper-1":
     #     logger.warning("Invalid input_audio_transcription model")
 
-    # NOTE: the updated `openai-realtime-console` sends partial `session.update.config` data which I don't currently support  # noqa: E501
+    # NOTE: the updated `openai-realtime-console` sends partial `session.update.config` data which I don't currently support
     # TODO: figure out how to apply session updates and what to do with the above checks
     # ctx.configuration = event.session  # pyright: ignore[reportAttributeAccessIssue]
     ctx.pubsub.publish_nowait(

src/speaches/routers/chat.py

@@ -73,10 +73,10 @@ class CompletionCreateParamsBase(OpenAICompletionCreateParamsBase):
 
     @model_validator(mode="after")
     def validate_audio_format_when_stream(self) -> Self:
-        # NOTE: OpenAI only supports pcm format for streaming. We can support any format but keeping this hardcoded for consistency  # noqa: E501
+        # NOTE: OpenAI only supports pcm format for streaming. We can support any format but keeping this hardcoded for consistency
         if self.stream and self.audio is not None and self.audio.format != "pcm16":
             raise ValueError(
-                f"Unsupported value: 'audio.format' does not support '{self.audio.format}' when stream=true. Supported values are: 'pcm16'."  # noqa: E501
+                f"Unsupported value: 'audio.format' does not support '{self.audio.format}' when stream=true. Supported values are: 'pcm16'."
             )
         return self
 
@@ -220,7 +220,7 @@ async def handle_completions(  # noqa: C901
     for i, message in enumerate(body.messages):
         if message.role == "user":
             content = message.content
-            # per https://platform.openai.com/docs/guides/audio?audio-generation-quickstart-example=audio-in#quickstart, input audio should be within the `message.content` list  # noqa: E501
+            # per https://platform.openai.com/docs/guides/audio?audio-generation-quickstart-example=audio-in#quickstart, input audio should be within the `message.content` list
             if not isinstance(content, list):
                 continue
 
@@ -250,7 +250,7 @@ async def handle_completions(  # noqa: C901
                 function_call=message.function_call,
             )
 
-    # NOTE: rather than doing a `model_copy` it might be better to override the fields when doing the `model_dump` and destructuring  # noqa: E501
+    # NOTE: rather than doing a `model_copy` it might be better to override the fields when doing the `model_dump` and destructuring
     proxied_body = body.model_copy(deep=True)
     proxied_body.modalities = ["text"]
     proxied_body.audio = None