dashscope 1.27.2__tar.gz → 1.27.3__tar.gz

{dashscope-1.27.2 → dashscope-1.27.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dashscope
-Version: 1.27.2
+Version: 1.27.3
 Summary: dashscope client sdk library
 Home-page: https://dashscope.aliyun.com/
 Author: Alibaba Cloud
@@ -28,16 +28,16 @@ Requires-Dist: cryptography
 Requires-Dist: certifi
 Requires-Dist: typer>=0.9.0
 Requires-Dist: rich>=13.0.0
-Requires-Dist: httpx>=0.27.0
+Requires-Dist: httpx>=0.25.0
 Requires-Dist: httpx-sse>=0.4.0
-Requires-Dist: prompt-toolkit>=3.0
-Requires-Dist: pydantic>=2.0
-Requires-Dist: tenacity
-Requires-Dist: tomli>=2.0; python_version < "3.11"
-Requires-Dist: PyYAML>=6.0
-Requires-Dist: textual>=0.50
+Requires-Dist: typing_extensions>=4.0
 Provides-Extra: tokenizer
 Requires-Dist: tiktoken; extra == "tokenizer"
+Provides-Extra: acli
+Requires-Dist: prompt-toolkit>=3.0; extra == "acli"
+Requires-Dist: textual>=0.50; extra == "acli"
+Requires-Dist: PyYAML>=6.0; extra == "acli"
+Requires-Dist: tomli>=2.0; python_version < "3.11" and extra == "acli"
 Provides-Extra: acli-anthropic
 Requires-Dist: anthropic>=0.40; extra == "acli-anthropic"
 Provides-Extra: acli-openai
@@ -48,11 +48,22 @@ Requires-Dist: numpy>=1.20; extra == "acli-voice"
 Provides-Extra: acli-camera
 Requires-Dist: opencv-python>=4.5; extra == "acli-camera"
 Provides-Extra: acli-all
+Requires-Dist: prompt-toolkit>=3.0; extra == "acli-all"
+Requires-Dist: textual>=0.50; extra == "acli-all"
+Requires-Dist: PyYAML>=6.0; extra == "acli-all"
+Requires-Dist: tomli>=2.0; python_version < "3.11" and extra == "acli-all"
 Requires-Dist: anthropic>=0.40; extra == "acli-all"
 Requires-Dist: openai>=1.30; extra == "acli-all"
 Requires-Dist: sounddevice>=0.4; extra == "acli-all"
 Requires-Dist: numpy>=1.20; extra == "acli-all"
 Requires-Dist: opencv-python>=4.5; extra == "acli-all"
+Provides-Extra: rl
+Requires-Dist: pydantic>=2.0; extra == "rl"
+Requires-Dist: tenacity; extra == "rl"
+Requires-Dist: PyYAML>=6.0; extra == "rl"
+Requires-Dist: fastapi>=0.100; extra == "rl"
+Requires-Dist: uvicorn>=0.20; extra == "rl"
+Requires-Dist: opentelemetry-sdk>=1.20; extra == "rl"
 Dynamic: author
 Dynamic: author-email
 Dynamic: classifier
@@ -69,6 +80,8 @@ Dynamic: summary
 
 # DashScope Python SDK
 
+> **English** | [中文](README_zh.md)
+
 The DashScope Python SDK provides a comprehensive interface to [Alibaba Cloud Model Studio (Bailian)](https://www.alibabacloud.com/help/en/model-studio/) APIs, covering text generation, multi-modal understanding, embeddings, reranking, image/video generation, speech synthesis & recognition, and more.
 
 ## What is New
@@ -81,16 +94,20 @@ To install the DashScope Python SDK, simply run:
 pip install dashscope
 ```
 
+The base install covers SDK API calls and the `dashscope` CLI command.
+Optional feature groups are available as extras:
+
+| Extra | Provides | Install |
+|-------|----------|---------|
+| `acli` | Interactive AI assistant (DashScope SDK Expert) | `pip install "dashscope[acli]"` |
+| `rl` | Agentic RL fine-tuning | `pip install "dashscope[rl]"` |
+| `tokenizer` | Local tokenizer without downloads | `pip install "dashscope[tokenizer]"` |
+
 If you clone the code from github, you can install from  source by running:
 ```shell
 pip install -e .
 ```
 
-To use tokenizer in local mode without downloading any files, run:
-```shell
-pip install dashscope[tokenizer]
-```
-
 
 ## Quick Start
 

{dashscope-1.27.2 → dashscope-1.27.3}/README.md

@@ -1,5 +1,7 @@
 # DashScope Python SDK
 
+> **English** | [中文](README_zh.md)
+
 The DashScope Python SDK provides a comprehensive interface to [Alibaba Cloud Model Studio (Bailian)](https://www.alibabacloud.com/help/en/model-studio/) APIs, covering text generation, multi-modal understanding, embeddings, reranking, image/video generation, speech synthesis & recognition, and more.
 
 ## What is New
@@ -12,16 +14,20 @@ To install the DashScope Python SDK, simply run:
 pip install dashscope
 ```
 
+The base install covers SDK API calls and the `dashscope` CLI command.
+Optional feature groups are available as extras:
+
+| Extra | Provides | Install |
+|-------|----------|---------|
+| `acli` | Interactive AI assistant (DashScope SDK Expert) | `pip install "dashscope[acli]"` |
+| `rl` | Agentic RL fine-tuning | `pip install "dashscope[rl]"` |
+| `tokenizer` | Local tokenizer without downloads | `pip install "dashscope[tokenizer]"` |
+
 If you clone the code from github, you can install from  source by running:
 ```shell
 pip install -e .
 ```
 
-To use tokenizer in local mode without downloading any files, run:
-```shell
-pip install dashscope[tokenizer]
-```
-
 
 ## Quick Start
 

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/__init__.py

@@ -1,8 +1,14 @@
 # -*- coding: utf-8 -*-
 from __future__ import annotations
 
+import uuid
+
 __version__ = "0.6.2"
 
+# Per-process identifier sent as x-dashscope-sdk-session-id so the
+# backend can group multi-turn requests from one CLI run.
+SDK_SESSION_ID = uuid.uuid4().hex
+
 # Expose the lightweight programmatic SDK at the package root.
 try:
     from dashscope.acli.sdk import (

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/cli/repl.py

@@ -195,6 +195,10 @@ async def _run_loop(config: Config):
         get_mcp_clients_fn=lambda: _mcp_clients,
     )
 
+    from dashscope.acli.tools.vision import register_vision_tools
+
+    register_vision_tools(config, get_provider_fn=get_provider)
+
     from dashscope.acli.tools.evolution import register_evolution_tools
 
     register_evolution_tools(get_agent=lambda: agent)

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/cli/runners.py

@@ -358,6 +358,10 @@ def _run_tui_mode(config: Config):
         get_mcp_clients_fn=lambda: _mcp_clients,
     )
 
+    from dashscope.acli.tools.vision import register_vision_tools
+
+    register_vision_tools(config, get_provider_fn=get_provider)
+
     from dashscope.acli.tools.evolution import register_evolution_tools
 
     register_evolution_tools(get_agent=lambda: agent)

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/providers/tongyi.py

@@ -3,10 +3,12 @@
 from __future__ import annotations
 
 import json
+import os
 from typing import AsyncIterator
 
 import httpx
 
+from dashscope.acli import SDK_SESSION_ID, __version__
 from dashscope.acli.providers.base import LLMChunk, LLMResponse, ToolCall
 
 DASHSCOPE_BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
@@ -123,6 +125,10 @@ class TongyiProvider:
         headers = {"Content-Type": "application/json"}
         if self.api_key:
             headers["Authorization"] = f"Bearer {self.api_key}"
+        if not os.environ.get("DASHSCOPE_DISABLE_SDK_HEADERS"):
+            headers["x-dashscope-sdk-client"] = "acli"
+            headers["x-dashscope-sdk-version"] = __version__
+            headers["x-dashscope-sdk-session-id"] = SDK_SESSION_ID
         return headers
 
     async def chat(

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/tools/browser.py

@@ -488,6 +488,7 @@ async def scrape_web_screenshot(
         return f"Error: screenshot failed — {e}"
 
     return (
-        f"Screenshot saved to {output_path}. Reference it with @path "
-        f"to send the image to a vision model for analysis."
+        f"Screenshot saved to {output_path}. Call the view_image tool "
+        f"with this path to analyze it (the user can also attach it "
+        f"via @path)."
     )

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/tools/camera.py

@@ -11,8 +11,8 @@ from dashscope.acli.tools.registry import PermissionLevel, tool
     name="camera_capture",
     description=(
         "Take a photo with the webcam and save it to the given path. "
-        "The user can then reference it via @path to send it to a "
-        "vision model for analysis."
+        "The photo can then be analyzed with the view_image tool (the "
+        "user can also attach it via @path)."
     ),
     permission=PermissionLevel.CONFIRM,
 )

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/tools/registry.py

@@ -368,6 +368,7 @@ class ToolRegistry:
                 "switch_provider",
                 "capability_enable",
                 "capability_disable",
+                "view_image",
             ],
         )
         include_tools.update(["mcp_connect", "mcp_disconnect"])

dashscope-1.27.3/dashscope/acli/tools/vision.py

@@ -0,0 +1,263 @@
+# -*- coding: utf-8 -*-
+"""Built-in vision tool — let the agent analyze local images itself.
+
+Previously the only multimodal entry point was user-input ``@path``
+expansion, so after ``scrape_web_screenshot`` / ``camera_capture`` the
+model could not see the saved image and sometimes echoed the
+user-facing ``@path`` hint verbatim, ending the turn prematurely.
+``view_image`` closes that gap by routing the image to a
+vision-capable model on the active provider.
+
+Very tall/wide images (e.g. a long announcement table) get downscaled
+or truncated by vision-model backends until text is illegible, so
+``view_image`` auto-slices them with Pillow (when installed) and runs
+one vision call per slice. Without Pillow it falls back to sending the
+whole image as-is.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import base64
+import io
+import math
+from typing import Callable
+
+from dashscope.acli.config import Config, is_vision_model
+from dashscope.acli.tools.registry import (
+    PermissionLevel,
+    ToolDefinition,
+    registry,
+)
+
+_FALLBACK_VISION_MODELS = {
+    "tongyi": "qwen-vl-max",
+    "openai": "gpt-4o",
+    "anthropic": "claude-sonnet-4-20250514",
+}
+
+# Slicing triggers: beyond these the model backend downscales or
+# truncates the image and fine text becomes unreadable.
+_SLICE_MAX_LONG_SIDE = 8000
+_SLICE_MAX_RATIO = 4.0
+_SLICE_STEP = 2000
+_SLICE_OVERLAP = 100
+_SLICE_MAX_COUNT = 32
+_SLICE_JPEG_QUALITY = 88
+
+
+def pick_vision_model(config: Config) -> str:
+    """Choose a vision-capable model on the active provider.
+
+    Preference: the active model when it accepts images, then the first
+    vision model declared on the provider's extension block, then a
+    per-provider built-in fallback.
+    """
+    if is_vision_model(config.model):
+        return config.model
+
+    from dashscope.acli.extensions import find_provider
+
+    ext = find_provider(config.provider)
+    if ext and ext.vision_models:
+        return ext.vision_models[0]
+    return _FALLBACK_VISION_MODELS.get(config.provider, "qwen-vl-max")
+
+
+def plan_slices(width: int, height: int):
+    """Return crop boxes for an oversized image, or None if it fits.
+
+    Slices run along the long axis with a small overlap so rows cut at
+    a box boundary still appear complete in one of the two neighbors.
+    """
+    long_side = max(width, height)
+    short_side = max(min(width, height), 1)
+    if (
+        long_side <= _SLICE_MAX_LONG_SIDE
+        and long_side / short_side <= _SLICE_MAX_RATIO
+    ):
+        return None
+
+    step = _SLICE_STEP
+    overlap = _SLICE_OVERLAP
+    count = math.ceil((long_side - overlap) / (step - overlap))
+    if count > _SLICE_MAX_COUNT:
+        # Re-derive a coarser step; drop the overlap so the cap holds.
+        step = math.ceil(long_side / _SLICE_MAX_COUNT)
+        overlap = 0
+
+    boxes = []
+    if height >= width:
+        y = 0
+        while y < height:
+            y2 = min(y + step, height)
+            boxes.append((0, y, width, y2))
+            if y2 == height:
+                break
+            y = y2 - overlap
+    else:
+        x = 0
+        while x < width:
+            x2 = min(x + step, width)
+            boxes.append((x, 0, x2, height))
+            if x2 == width:
+                break
+            x = x2 - overlap
+    return boxes
+
+
+def _slice_data_urls(image, boxes) -> list:
+    """Crop ``boxes`` out of a PIL image, JPEG-encode as data URLs."""
+    urls = []
+    for box in boxes:
+        crop = image.crop(box)
+        if crop.mode not in ("RGB", "L"):
+            crop = crop.convert("RGB")
+        buf = io.BytesIO()
+        crop.save(buf, format="JPEG", quality=_SLICE_JPEG_QUALITY)
+        b64 = base64.b64encode(buf.getvalue()).decode()
+        urls.append(f"data:image/jpeg;base64,{b64}")
+    return urls
+
+
+async def _ask_vision(provider, data_url: str, text: str) -> str:
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": text},
+                {"type": "image_url", "image_url": {"url": data_url}},
+            ],
+        },
+    ]
+    response = await provider.chat(messages)
+    return response.content or "(vision model returned no content)"
+
+
+def register_vision_tools(config: Config, *, get_provider_fn: Callable):
+    """Register ``view_image``; called once after agent construction."""
+
+    async def view_image(
+        image_path: str,
+        question: str = (
+            "Describe this image in detail. If it contains text or "
+            "tables, transcribe them verbatim."
+        ),
+        slice_mode: str = "auto",
+    ) -> str:
+        from dashscope.acli.utils.paths import validate_path
+
+        try:
+            safe_path = validate_path(image_path)
+        except (ValueError, OSError) as e:
+            return f"Error: failed to read image: {e}"
+
+        model = pick_vision_model(config)
+        provider = get_provider_fn(
+            config.provider,
+            model,
+            config.api_key,
+            base_url=config.base_url or None,
+        )
+
+        boxes = None
+        image = None
+        if slice_mode != "never":
+            try:
+                from PIL import Image
+
+                image = Image.open(safe_path)
+                image.load()
+                boxes = plan_slices(*image.size)
+            except Exception:  # pylint: disable=broad-except
+                # No Pillow, or not a decodable image: fall back to the
+                # whole-image path and let it report the real error.
+                boxes = None
+                image = None
+
+        if boxes:
+            axis = (
+                "top-to-bottom"
+                if image.height >= image.width
+                else ("left-to-right")
+            )
+            urls = _slice_data_urls(image, boxes)
+            image.close()
+            total = len(urls)
+            tasks = [
+                _ask_vision(
+                    provider,
+                    url,
+                    (
+                        f"Slice {i + 1}/{total} ({axis}) of one large "
+                        "image. " + question
+                    ),
+                )
+                for i, url in enumerate(urls)
+            ]
+            results = await asyncio.gather(*tasks, return_exceptions=True)
+            parts = [
+                f"(Image auto-split into {total} slices {axis}; "
+                "results per slice below.)",
+            ]
+            for i, r in enumerate(results, start=1):
+                if isinstance(r, BaseException):
+                    parts.append(f"--- slice {i}/{total} ---\nError: {r}")
+                else:
+                    parts.append(f"--- slice {i}/{total} ---\n{r}")
+            return "\n\n".join(parts)
+
+        try:
+            from dashscope.acli.utils.images import image_to_data_url
+
+            data_url = image_to_data_url(safe_path)
+        except (ValueError, OSError) as e:
+            return f"Error: failed to read image: {e}"
+        try:
+            return await _ask_vision(provider, data_url, question)
+        except Exception as e:  # pylint: disable=broad-except
+            return f"Error: vision call failed: {e}"
+
+    registry.register(
+        ToolDefinition(
+            name="view_image",
+            description=(
+                "Analyze a local image file (screenshot, photo, chart) "
+                "with a vision model: OCR/transcribe text, describe "
+                "layout or content. Very tall/wide images are split "
+                "automatically. Use after scrape_web_screenshot or "
+                "camera_capture, or whenever you need to read an image."
+            ),
+            permission=PermissionLevel.AUTO,
+            func=view_image,
+            parameters={
+                "type": "object",
+                "properties": {
+                    "image_path": {
+                        "type": "string",
+                        "description": (
+                            "Path to the image file (relative to the "
+                            "workspace or absolute)"
+                        ),
+                    },
+                    "question": {
+                        "type": "string",
+                        "description": (
+                            "What to extract from the image; defaults "
+                            "to a full description + verbatim OCR"
+                        ),
+                    },
+                    "slice_mode": {
+                        "type": "string",
+                        "enum": ["auto", "never"],
+                        "description": (
+                            "auto (default): split oversized images "
+                            "into slices; never: always send the whole "
+                            "image in one call"
+                        ),
+                    },
+                },
+                "required": ["image_path"],
+            },
+        ),
+    )

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/acli/ui/tui.py

@@ -2651,6 +2651,11 @@ class AgenticCLIApp(App):
             except Exception as e:
                 output.write(Text(f"Error: {e}", style="red"))
 
+        # Commands like /model switch the model inside the executor
+        # thread, after the post-submission hint refresh already ran;
+        # re-read agent.model_name so the status line is current.
+        self._refresh_hint_label()
+
         try:
             input_widget = self.query_one("#command-input", CommandInput)
             input_widget.focus()

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/agentstudio/transport.py

@@ -32,6 +32,7 @@ from dashscope.agentstudio.constants import (
     AGENTSTUDIO_DEFAULT_TIMEOUT,
     AGENTSTUDIO_MAX_RETRIES,
 )
+from dashscope.common.utils import get_sdk_headers
 
 logger = logging.getLogger("dashscope.agentstudio")
 
@@ -229,6 +230,7 @@ def build_headers(
         HEADER_AUTHORIZATION: f"Bearer {api_key}",
         HEADER_USER_AGENT: user_agent,
     }
+    headers.update(get_sdk_headers())
     if json_body:
         headers["Content-Type"] = "application/json"
         headers["Accept"] = "application/json"

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/aigc/video_synthesis.py

@@ -47,9 +47,12 @@ class VideoSynthesis(BaseAsyncApi):
         REFERENCE_IMAGE = "reference_image"
         REFERENCE_VIDEO = "reference_video"
         REFERENCE_VOICE = "reference_voice"
+        REFERENCE_AUDIO = "reference_audio"
         VIDEO = "video"
         FIRST_CLIP = "first_clip"
         DRIVING_AUDIO = "driving_audio"
+        FILE = "file"
+        LINK = "link"
 
     @classmethod
     def call(  # type: ignore[override]
@@ -84,6 +87,7 @@ class VideoSynthesis(BaseAsyncApi):
         ratio: Optional[str] = None,
         shot_type: Optional[str] = None,
         audio_setting: Optional[str] = None,
+        enable_thinking: Optional[bool] = None,
         **kwargs,
     ) -> VideoSynthesisResponse:
         """Call video synthesis service and get result.
@@ -126,6 +130,7 @@ class VideoSynthesis(BaseAsyncApi):
             shot_type (str, optional): Shot type for video generation.
             audio_setting (str, optional): Audio setting for video,
                 "auto" or "origin".
+            enable_thinking (bool, optional): Whether to use Agentic PE.
             **kwargs: Additional parameters passed to the API.
 
         Raises:
@@ -152,6 +157,8 @@ class VideoSynthesis(BaseAsyncApi):
             kwargs["shot_type"] = shot_type
         if audio_setting is not None:
             kwargs["audio_setting"] = audio_setting
+        if enable_thinking is not None:
+            kwargs["enable_thinking"] = enable_thinking
         return super().call(  # type: ignore[return-value]
             model,
             prompt,
@@ -263,7 +270,8 @@ class VideoSynthesis(BaseAsyncApi):
         if media:
             for i, m_file in enumerate(media):
                 if isinstance(m_file, dict):
-                    if m_file.get("url"):
+                    # exclude when media is a link
+                    if m_file.get("url") and m_file.get("type") != "link":
                         tasks.append(
                             {
                                 "type": "media",
@@ -397,6 +405,7 @@ class VideoSynthesis(BaseAsyncApi):
         ratio: Optional[str] = None,
         shot_type: Optional[str] = None,
         audio_setting: Optional[str] = None,
+        enable_thinking: Optional[bool] = None,
         **kwargs,
     ) -> VideoSynthesisResponse:
         """Create a video synthesis task, and return task information.
@@ -435,6 +444,7 @@ class VideoSynthesis(BaseAsyncApi):
             shot_type (str, optional): Shot type for video generation.
             audio_setting (str, optional): Audio setting for video,
                 "auto" or "origin".
+            enable_thinking (bool, optional): Whether to use Agentic PE.
             **kwargs: Additional parameters passed to the API.
 
         Raises:
@@ -462,6 +472,8 @@ class VideoSynthesis(BaseAsyncApi):
             kwargs["shot_type"] = shot_type
         if audio_setting is not None:
             kwargs["audio_setting"] = audio_setting
+        if enable_thinking is not None:
+            kwargs["enable_thinking"] = enable_thinking
         task_group, function = _get_task_group_and_task(__name__)
 
         inputs, kwargs, task = cls._get_input(
@@ -656,6 +668,7 @@ class AioVideoSynthesis(BaseAsyncAioApi):
         ratio: str = None,
         shot_type: str = None,
         audio_setting: str = None,
+        enable_thinking: bool = None,
         **kwargs,
     ) -> VideoSynthesisResponse:
         """Call video synthesis service and get result.
@@ -694,6 +707,7 @@ class AioVideoSynthesis(BaseAsyncAioApi):
             shot_type (str, optional): Shot type for video generation.
             audio_setting (str, optional): Audio setting for video,
                 "auto" or "origin".
+            enable_thinking (bool, optional): Whether to use Agentic PE.
             **kwargs: Additional parameters passed to the API.
 
         Raises:
@@ -720,6 +734,8 @@ class AioVideoSynthesis(BaseAsyncAioApi):
             kwargs["shot_type"] = shot_type
         if audio_setting is not None:
             kwargs["audio_setting"] = audio_setting
+        if enable_thinking is not None:
+            kwargs["enable_thinking"] = enable_thinking
         task_group, f = _get_task_group_and_task(__name__)
         # pylint: disable=protected-access
         inputs, kwargs, task = VideoSynthesis._get_input(
@@ -792,6 +808,8 @@ class AioVideoSynthesis(BaseAsyncAioApi):
         resolution: str = None,
         ratio: str = None,
         shot_type: str = None,
+        audio_setting: str = None,
+        enable_thinking: bool = None,
         **kwargs,
     ) -> VideoSynthesisResponse:
         """Create a video synthesis task, and return task information.
@@ -828,6 +846,8 @@ class AioVideoSynthesis(BaseAsyncAioApi):
             resolution (str, optional): Output resolution.
             ratio (str, optional): Aspect ratio, e.g. "16:9".
             shot_type (str, optional): Shot type for video generation.
+            enable_thinking (bool, optional): Whether to use Agentic PE.
+            audio_setting (str, optional): Audio setting for video,
             **kwargs: Additional parameters passed to the API.
 
         Raises:
@@ -853,6 +873,10 @@ class AioVideoSynthesis(BaseAsyncAioApi):
             kwargs["ratio"] = ratio
         if shot_type is not None:
             kwargs["shot_type"] = shot_type
+        if audio_setting is not None:
+            kwargs["audio_setting"] = audio_setting
+        if enable_thinking is not None:
+            kwargs["enable_thinking"] = enable_thinking
         task_group, function = _get_task_group_and_task(__name__)
 
         # pylint: disable=protected-access

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/api_entities/api_request_factory.py

@@ -18,6 +18,7 @@ from dashscope.common.constants import (
 )
 from dashscope.common.error import InputDataRequired, UnsupportedApiProtocol
 from dashscope.common.logging import logger
+from dashscope.common.utils import get_sdk_headers
 from dashscope.protocol.websocket import WebsocketStreamingMode
 
 
@@ -191,8 +192,11 @@ def _build_api_request(  # pylint: disable=too-many-branches
             "websocket]",
         )
 
+    merged_headers = dict(get_sdk_headers())
     if headers is not None:
-        request.add_headers(headers=headers)
+        merged_headers.update(headers)
+    if merged_headers:
+        request.add_headers(headers=merged_headers)
 
     if input is None and form is None:
         raise InputDataRequired("There is no input data and form data")

{dashscope-1.27.2 → dashscope-1.27.3}/dashscope/api_entities/base_request.py

@@ -5,7 +5,7 @@ import os
 from abc import ABC, abstractmethod
 
 from dashscope.common.constants import DASHSCOPE_DISABLE_DATA_INSPECTION_ENV
-from dashscope.common.utils import get_user_agent
+from dashscope.common.utils import get_sdk_headers, get_user_agent
 
 
 class BaseRequest(ABC):
@@ -17,6 +17,7 @@ class BaseRequest(ABC):
             ua += "; " + user_agent
 
         self.headers = {"user-agent": ua}
+        self.headers.update(get_sdk_headers())
         disable_data_inspection = os.environ.get(
             DASHSCOPE_DISABLE_DATA_INSPECTION_ENV,
             "true",