增加环绕侦察场景适配

2026-01-08 15:44:38 +08:00
parent 3eba1f962b
commit 10c5bb5a8a
5441 changed files with 40219 additions and 379695 deletions
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_client.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_client.py
@@ -135,7 +135,7 @@ class InferenceClient:
            Note: for better compatibility with OpenAI's client, `model` has been aliased as `base_url`. Those 2
            arguments are mutually exclusive. If a URL is passed as `model` or `base_url` for chat completion, the `(/v1)/chat/completions` suffix path will be appended to the URL.
        provider (`str`, *optional*):
-            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"clarifai"`, `"cohere"`, `"fal-ai"`, `"featherless-ai"`, `"fireworks-ai"`, `"groq"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"nscale"`, `"openai"`, `"publicai"`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"`, `"wavespeed"` or `"zai-org"`.
+            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"clarifai"`, `"cohere"`, `"fal-ai"`, `"featherless-ai"`, `"fireworks-ai"`, `"groq"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"nscale"`, `"openai"`, `"ovhcloud"`, `"publicai"`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"`, `"wavespeed"` or `"zai-org"`.
            Defaults to "auto" i.e. the first of the providers available for the model, sorted by the user's order in https://hf.co/settings/inference-providers.
            If model is a URL or `base_url` is passed, then `provider` is not used.
        token (`str`, *optional*):
@@ -452,6 +452,7 @@ class InferenceClient:
            api_key=self.token,
        )
        response = self._inner_post(request_parameters)
+        response = provider_helper.get_response(response, request_params=request_parameters)
        return AutomaticSpeechRecognitionOutput.parse_obj_as_instance(response)

    @overload
@@ -1028,7 +1029,7 @@ class InferenceClient:
        normalize: Optional[bool] = None,
        prompt_name: Optional[str] = None,
        truncate: Optional[bool] = None,
-        truncation_direction: Optional[Literal["Left", "Right"]] = None,
+        truncation_direction: Optional[Literal["left", "right"]] = None,
        model: Optional[str] = None,
    ) -> "np.ndarray":
        """
@@ -1053,7 +1054,7 @@ class InferenceClient:
            truncate (`bool`, *optional*):
                Whether to truncate the embeddings or not.
                Only available on server powered by Text-Embedding-Inference.
-            truncation_direction (`Literal["Left", "Right"]`, *optional*):
+            truncation_direction (`Literal["left", "right"]`, *optional*):
                Which side of the input should be truncated when `truncate=True` is passed.

        Returns:
@@ -3195,10 +3196,7 @@ class InferenceClient:
        )
        response = self._inner_post(request_parameters)
        output = _bytes_to_dict(response)
-        return [
-            ZeroShotClassificationOutputElement.parse_obj_as_instance({"label": label, "score": score})
-            for label, score in zip(output["labels"], output["scores"])
-        ]
+        return ZeroShotClassificationOutputElement.parse_obj_as_list(output)

    def zero_shot_image_classification(
        self,
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_common.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_common.py
@@ -144,7 +144,7 @@ def _open_as_mime_bytes(content: Optional[ContentT]) -> Optional[MimeBytes]:
    if hasattr(content, "read"):  # duck-typing instead of isinstance(content, BinaryIO)
        logger.debug("Reading content from BinaryIO")
        data = content.read()
-        mime_type = mimetypes.guess_type(content.name)[0] if hasattr(content, "name") else None
+        mime_type = mimetypes.guess_type(str(content.name))[0] if hasattr(content, "name") else None
        if isinstance(data, str):
            raise TypeError("Expected binary stream (bytes), but got text stream")
        return MimeBytes(data, mime_type=mime_type)
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_generated/_async_client.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_generated/_async_client.py
@@ -126,7 +126,7 @@ class AsyncInferenceClient:
            Note: for better compatibility with OpenAI's client, `model` has been aliased as `base_url`. Those 2
            arguments are mutually exclusive. If a URL is passed as `model` or `base_url` for chat completion, the `(/v1)/chat/completions` suffix path will be appended to the URL.
        provider (`str`, *optional*):
-            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"clarifai"`, `"cohere"`, `"fal-ai"`, `"featherless-ai"`, `"fireworks-ai"`, `"groq"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"nscale"`, `"openai"`, `"publicai"`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"`, `"wavespeed"` or `"zai-org"`.
+            Name of the provider to use for inference. Can be `"black-forest-labs"`, `"cerebras"`, `"clarifai"`, `"cohere"`, `"fal-ai"`, `"featherless-ai"`, `"fireworks-ai"`, `"groq"`, `"hf-inference"`, `"hyperbolic"`, `"nebius"`, `"novita"`, `"nscale"`, `"openai"`, `"ovhcloud"`, `"publicai"`, `"replicate"`, `"sambanova"`, `"scaleway"`, `"together"`, `"wavespeed"` or `"zai-org"`.
            Defaults to "auto" i.e. the first of the providers available for the model, sorted by the user's order in https://hf.co/settings/inference-providers.
            If model is a URL or `base_url` is passed, then `provider` is not used.
        token (`str`, *optional*):
@@ -472,6 +472,7 @@ class AsyncInferenceClient:
            api_key=self.token,
        )
        response = await self._inner_post(request_parameters)
+        response = provider_helper.get_response(response, request_params=request_parameters)
        return AutomaticSpeechRecognitionOutput.parse_obj_as_instance(response)

    @overload
@@ -1055,7 +1056,7 @@ class AsyncInferenceClient:
        normalize: Optional[bool] = None,
        prompt_name: Optional[str] = None,
        truncate: Optional[bool] = None,
-        truncation_direction: Optional[Literal["Left", "Right"]] = None,
+        truncation_direction: Optional[Literal["left", "right"]] = None,
        model: Optional[str] = None,
    ) -> "np.ndarray":
        """
@@ -1080,7 +1081,7 @@ class AsyncInferenceClient:
            truncate (`bool`, *optional*):
                Whether to truncate the embeddings or not.
                Only available on server powered by Text-Embedding-Inference.
-            truncation_direction (`Literal["Left", "Right"]`, *optional*):
+            truncation_direction (`Literal["left", "right"]`, *optional*):
                Which side of the input should be truncated when `truncate=True` is passed.

        Returns:
@@ -3245,10 +3246,7 @@ class AsyncInferenceClient:
        )
        response = await self._inner_post(request_parameters)
        output = _bytes_to_dict(response)
-        return [
-            ZeroShotClassificationOutputElement.parse_obj_as_instance({"label": label, "score": score})
-            for label, score in zip(output["labels"], output["scores"])
-        ]
+        return ZeroShotClassificationOutputElement.parse_obj_as_list(output)

    async def zero_shot_image_classification(
        self,
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_generated/types/base.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_generated/types/base.py
@@ -19,6 +19,8 @@ import types
 from dataclasses import asdict, dataclass
 from typing import Any, TypeVar, Union, get_args

+from typing_extensions import dataclass_transform
+

 T = TypeVar("T", bound="BaseInferenceType")

@@ -29,6 +31,7 @@ def _repr_with_extra(self):
    return f"{self.__class__.__name__}({', '.join(f'{k}={self.__dict__[k]!r}' for k in fields + other_fields)})"


+@dataclass_transform()
 def dataclass_with_extra(cls: type[T]) -> type[T]:
    """Decorator to add a custom __repr__ method to a dataclass, showing all fields, including extra ones.

--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_generated/types/feature_extraction.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_generated/types/feature_extraction.py
@@ -8,7 +8,7 @@ from typing import Literal, Optional, Union
 from .base import BaseInferenceType, dataclass_with_extra


-FeatureExtractionInputTruncationDirection = Literal["Left", "Right"]
+FeatureExtractionInputTruncationDirection = Literal["left", "right"]


@dataclass_with_extra
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_mcp/cli.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_mcp/cli.py
@@ -5,8 +5,8 @@ import traceback
 from typing import Optional

 import typer
-from rich import print

+from ...utils import ANSI
 from ._cli_hacks import _async_prompt, _patch_anyio_open_process
 from .agent import Agent
 from .utils import _load_agent_config
@@ -55,10 +55,10 @@ async def run_agent(
        if first_sigint:
            first_sigint = False
            abort_event.set()
-            print("\n[red]Interrupted. Press Ctrl+C again to quit.[/red]", flush=True)
+            print(ANSI.red("\nInterrupted. Press Ctrl+C again to quit."), flush=True)
            return

-        print("\n[red]Exiting...[/red]", flush=True)
+        print(ANSI.red("\nExiting..."), flush=True)
        exit_event.set()

    try:
@@ -75,8 +75,12 @@ async def run_agent(

        if len(inputs) > 0:
            print(
-                "[bold blue]Some initial inputs are required by the agent. "
-                "Please provide a value or leave empty to load from env.[/bold blue]"
+                ANSI.bold(
+                    ANSI.blue(
+                        "Some initial inputs are required by the agent. "
+                        "Please provide a value or leave empty to load from env."
+                    )
+                )
            )
            for input_item in inputs:
                input_id = input_item["id"]
@@ -98,15 +102,17 @@ async def run_agent(

                if not input_usages:
                    print(
-                        f"[yellow]Input '{input_id}' defined in config but not used by any server or as an API key."
-                        " Skipping.[/yellow]"
+                        ANSI.yellow(
+                            f"Input '{input_id}' defined in config but not used by any server or as an API key."
+                            " Skipping."
+                        )
                    )
                    continue

                # Prompt user for input
                env_variable_key = input_id.replace("-", "_").upper()
                print(
-                    f"[blue] • {input_id}[/blue]: {description}. (default: load from {env_variable_key}).",
+                    ANSI.blue(f" • {input_id}") + f": {description}. (default: load from {env_variable_key}).",
                    end=" ",
                )
                user_input = (await _async_prompt(exit_event=exit_event)).strip()
@@ -118,10 +124,12 @@ async def run_agent(
                if not final_value:
                    final_value = os.getenv(env_variable_key, "")
                    if final_value:
-                        print(f"[green]Value successfully loaded from '{env_variable_key}'[/green]")
+                        print(ANSI.green(f"Value successfully loaded from '{env_variable_key}'"))
                    else:
                        print(
-                            f"[yellow]No value found for '{env_variable_key}' in environment variables. Continuing.[/yellow]"
+                            ANSI.yellow(
+                                f"No value found for '{env_variable_key}' in environment variables. Continuing."
+                            )
                        )
                resolved_inputs[input_id] = final_value

@@ -150,9 +158,9 @@ async def run_agent(
            prompt=prompt,
        ) as agent:
            await agent.load_tools()
-            print(f"[bold blue]Agent loaded with {len(agent.available_tools)} tools:[/bold blue]")
+            print(ANSI.bold(ANSI.blue("Agent loaded with {} tools:".format(len(agent.available_tools)))))
            for t in agent.available_tools:
-                print(f"[blue] • {t.function.name}[/blue]")
+                print(ANSI.blue(f" • {t.function.name}"))

            while True:
                abort_event.clear()
@@ -165,13 +173,13 @@ async def run_agent(
                    user_input = await _async_prompt(exit_event=exit_event)
                    first_sigint = True
                except EOFError:
-                    print("\n[red]EOF received, exiting.[/red]", flush=True)
+                    print(ANSI.red("\nEOF received, exiting."), flush=True)
                    break
                except KeyboardInterrupt:
                    if not first_sigint and abort_event.is_set():
                        continue
                    else:
-                        print("\n[red]Keyboard interrupt during input processing.[/red]", flush=True)
+                        print(ANSI.red("\nKeyboard interrupt during input processing."), flush=True)
                        break

                try:
@@ -195,7 +203,7 @@ async def run_agent(
                                        print(f"{call.function.arguments}", end="")
                        else:
                            print(
-                                f"\n\n[green]Tool[{chunk.name}] {chunk.tool_call_id}\n{chunk.content}[/green]\n",
+                                ANSI.green(f"\n\nTool[{chunk.name}] {chunk.tool_call_id}\n{chunk.content}\n"),
                                flush=True,
                            )

@@ -203,12 +211,12 @@ async def run_agent(

                except Exception as e:
                    tb_str = traceback.format_exc()
-                    print(f"\n[bold red]Error during agent run: {e}\n{tb_str}[/bold red]", flush=True)
+                    print(ANSI.red(f"\nError during agent run: {e}\n{tb_str}"), flush=True)
                    first_sigint = True  # Allow graceful interrupt for the next command

    except Exception as e:
        tb_str = traceback.format_exc()
-        print(f"\n[bold red]An unexpected error occurred: {e}\n{tb_str}[/bold red]", flush=True)
+        print(ANSI.red(f"\nAn unexpected error occurred: {e}\n{tb_str}"), flush=True)
        raise e

    finally:
@@ -236,10 +244,10 @@ def run(
    try:
        asyncio.run(run_agent(path))
    except KeyboardInterrupt:
-        print("\n[red]Application terminated by KeyboardInterrupt.[/red]", flush=True)
+        print(ANSI.red("\nApplication terminated by KeyboardInterrupt."), flush=True)
        raise typer.Exit(code=130)
    except Exception as e:
-        print(f"\n[bold red]An unexpected error occurred: {e}[/bold red]", flush=True)
+        print(ANSI.red(f"\nAn unexpected error occurred: {e}"), flush=True)
        raise e


--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_mcp/utils.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_mcp/utils.py
@@ -57,10 +57,10 @@ def format_result(result: "mcp_types.CallToolResult") -> str:
        elif item.type == "resource":
            resource = item.resource

-            if hasattr(resource, "text"):
+            if hasattr(resource, "text") and isinstance(resource.text, str):
                formatted_parts.append(resource.text)

-            elif hasattr(resource, "blob"):
+            elif hasattr(resource, "blob") and isinstance(resource.blob, str):
                formatted_parts.append(
                    f"[Binary Content ({resource.uri}): {resource.mimeType}, {_get_base64_size(resource.blob)} bytes]\n"
                    f"The task is complete and the content accessible to the User"
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/init.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/init.py
@@ -38,8 +38,15 @@ from .nebius import (
 from .novita import NovitaConversationalTask, NovitaTextGenerationTask, NovitaTextToVideoTask
 from .nscale import NscaleConversationalTask, NscaleTextToImageTask
 from .openai import OpenAIConversationalTask
+from .ovhcloud import OVHcloudConversationalTask
 from .publicai import PublicAIConversationalTask
-from .replicate import ReplicateImageToImageTask, ReplicateTask, ReplicateTextToImageTask, ReplicateTextToSpeechTask
+from .replicate import (
+    ReplicateAutomaticSpeechRecognitionTask,
+    ReplicateImageToImageTask,
+    ReplicateTask,
+    ReplicateTextToImageTask,
+    ReplicateTextToSpeechTask,
+)
 from .sambanova import SambanovaConversationalTask, SambanovaFeatureExtractionTask
 from .scaleway import ScalewayConversationalTask, ScalewayFeatureExtractionTask
 from .together import TogetherConversationalTask, TogetherTextGenerationTask, TogetherTextToImageTask
@@ -70,6 +77,7 @@ PROVIDER_T = Literal[
    "novita",
    "nscale",
    "openai",
+    "ovhcloud",
    "publicai",
    "replicate",
    "sambanova",
@@ -166,10 +174,14 @@ PROVIDERS: dict[PROVIDER_T, dict[str, TaskProviderHelper]] = {
    "openai": {
        "conversational": OpenAIConversationalTask(),
    },
+    "ovhcloud": {
+        "conversational": OVHcloudConversationalTask(),
+    },
    "publicai": {
        "conversational": PublicAIConversationalTask(),
    },
    "replicate": {
+        "automatic-speech-recognition": ReplicateAutomaticSpeechRecognitionTask(),
        "image-to-image": ReplicateImageToImageTask(),
        "text-to-image": ReplicateTextToImageTask(),
        "text-to-speech": ReplicateTextToSpeechTask(),
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/_common.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/_common.py
@@ -32,6 +32,7 @@ HARDCODED_MODEL_INFERENCE_MAPPING: dict[str, dict[str, InferenceProviderMapping]
    "hyperbolic": {},
    "nebius": {},
    "nscale": {},
+    "ovhcloud": {},
    "replicate": {},
    "sambanova": {},
    "scaleway": {},
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/fal_ai.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/fal_ai.py
@@ -112,7 +112,7 @@ class FalAIAutomaticSpeechRecognitionTask(FalAITask):
        text = _as_dict(response)["text"]
        if not isinstance(text, str):
            raise ValueError(f"Unexpected output format from FalAI API. Expected string, got {type(text)}.")
-        return text
+        return {"text": text}


 class FalAITextToImageTask(FalAITask):
--- a/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/replicate.py
+++ b/backend_service/venv/lib/python3.13/site-packages/huggingface_hub/inference/_providers/replicate.py
@@ -72,6 +72,67 @@ class ReplicateTextToSpeechTask(ReplicateTask):
        return payload


+class ReplicateAutomaticSpeechRecognitionTask(ReplicateTask):
+    def __init__(self) -> None:
+        super().__init__("automatic-speech-recognition")
+
+    def _prepare_payload_as_dict(
+        self,
+        inputs: Any,
+        parameters: dict,
+        provider_mapping_info: InferenceProviderMapping,
+    ) -> Optional[dict]:
+        mapped_model = provider_mapping_info.provider_id
+        audio_url = _as_url(inputs, default_mime_type="audio/wav")
+
+        payload: dict[str, Any] = {
+            "input": {
+                **{"audio": audio_url},
+                **filter_none(parameters),
+            }
+        }
+
+        if ":" in mapped_model:
+            payload["version"] = mapped_model.split(":", 1)[1]
+
+        return payload
+
+    def get_response(self, response: Union[bytes, dict], request_params: Optional[RequestParameters] = None) -> Any:
+        response_dict = _as_dict(response)
+        output = response_dict.get("output")
+
+        if isinstance(output, str):
+            return {"text": output}
+
+        if isinstance(output, list) and output:
+            first_item = output[0]
+            if isinstance(first_item, str):
+                return {"text": first_item}
+            if isinstance(first_item, dict):
+                output = first_item
+
+        text: Optional[str] = None
+        if isinstance(output, dict):
+            transcription = output.get("transcription")
+            if isinstance(transcription, str):
+                text = transcription
+
+            translation = output.get("translation")
+            if isinstance(translation, str):
+                text = translation
+
+            txt_file = output.get("txt_file")
+            if isinstance(txt_file, str):
+                text_response = get_session().get(txt_file)
+                text_response.raise_for_status()
+                text = text_response.text
+
+        if text is not None:
+            return {"text": text}
+
+        raise ValueError("Received malformed response from Replicate automatic-speech-recognition API")
+
+
 class ReplicateImageToImageTask(ReplicateTask):
    def __init__(self):
        super().__init__("image-to-image")