veadk-python 0.2.11__tar.gz → 0.2.12__tar.gz

{veadk_python-0.2.11 → veadk_python-0.2.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: veadk-python
-Version: 0.2.11
+Version: 0.2.12
 Summary: Volcengine agent development kit, integrations with Volcengine cloud services.
 Author-email: Yaozheng Fang <fangyozheng@gmail.com>, Guodong Li <cu.eric.lee@gmail.com>, Zhi Han <sliverydayday@gmail.com>, Meng Wang <mengwangwm@gmail.com>
 License:                                  Apache License

{veadk_python-0.2.11 → veadk_python-0.2.12}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "veadk-python"
-version = "0.2.11"
+version = "0.2.12"
 description = "Volcengine agent development kit, integrations with Volcengine cloud services."
 readme = "README.md"
 requires-python = ">=3.10"

{veadk_python-0.2.11 → veadk_python-0.2.12}/veadk/cli/cli.py

@@ -22,6 +22,7 @@ from veadk.cli.cli_kb import kb
 from veadk.cli.cli_pipeline import pipeline
 from veadk.cli.cli_prompt import prompt
 from veadk.cli.cli_web import web
+from veadk.cli.cli_uploadevalset import uploadevalset
 from veadk.version import VERSION
 
 
@@ -41,6 +42,7 @@ veadk.add_command(web)
 veadk.add_command(pipeline)
 veadk.add_command(eval)
 veadk.add_command(kb)
+veadk.add_command(uploadevalset)
 
 if __name__ == "__main__":
     veadk()

veadk_python-0.2.12/veadk/cli/cli_uploadevalset.py

@@ -0,0 +1,125 @@
+# Copyright (c) 2025 Beijing Volcano Engine Technology Co., Ltd. and/or its affiliates.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import click
+import json
+import requests
+from veadk.utils.logger import get_logger
+from veadk.config import getenv
+from pathlib import Path
+
+logger = get_logger(__name__)
+
+
+@click.command()
+@click.option("--file", required=True, help="JSON file path containing dataset items")
+@click.option("--cozeloop-workspace-id", default=None, help="CozeLoop workspace ID")
+@click.option("--cozeloop-evalset-id", default=None, help="CozeLoop evaluation set ID")
+@click.option(
+    "--cozeloop-api-key",
+    default=None,
+    help="CozeLoop API key (or set COZELOOP_API_KEY env var)",
+)
+def uploadevalset(
+    file: str,
+    cozeloop_workspace_id: str,
+    cozeloop_evalset_id: str,
+    cozeloop_api_key: str,
+) -> None:
+    """Upload dataset items to CozeLoop evaluation set."""
+
+    if not cozeloop_workspace_id:
+        cozeloop_workspace_id = getenv(
+            "OBSERVABILITY_OPENTELEMETRY_COZELOOP_SERVICE_NAME"
+        )
+    if not cozeloop_evalset_id:
+        cozeloop_evalset_id = getenv("OBSERVABILITY_OPENTELEMETRY_COZELOOP_EVALSET_ID")
+    if not cozeloop_api_key:
+        cozeloop_api_key = getenv("OBSERVABILITY_OPENTELEMETRY_COZELOOP_API_KEY")
+
+    # Read JSON file
+    file_path = Path(file)
+    if not file_path.exists():
+        logger.error(f"File not found: {file}")
+        return
+
+    logger.info(f"Reading dataset from {file}")
+    with open(file_path, "r", encoding="utf-8") as f:
+        data = json.load(f)
+
+    # Prepare items
+    items = []
+    for case in data.get("eval_cases", []):
+        conversation = case.get("conversation", [])
+        for turn in conversation:
+            user_text = (
+                turn.get("user_content", {}).get("parts", [{}])[0].get("text", "")
+            )
+            output_text = (
+                turn.get("final_response", {}).get("parts", [{}])[0].get("text", "")
+            )
+
+            items.append(
+                {
+                    "turns": [
+                        {
+                            "field_datas": [
+                                {
+                                    "name": "input",
+                                    "content": {
+                                        "content_type": "Text",
+                                        "text": user_text,
+                                    },
+                                },
+                                {
+                                    "name": "output",
+                                    "content": {
+                                        "content_type": "Text",
+                                        "text": output_text,
+                                    },
+                                },
+                            ]
+                        }
+                    ]
+                }
+            )
+
+    # Upload to CozeLoop
+    url = f"https://api.coze.cn/v1/loop/evaluation/evaluation_sets/{cozeloop_evalset_id}/items"
+    logger.info(
+        f"Uploading {len(items)} items to workspace_id={cozeloop_workspace_id} evalset_id={cozeloop_evalset_id}"
+    )
+
+    response = requests.post(
+        url=url,
+        headers={
+            "Authorization": f"Bearer {cozeloop_api_key}",
+            "Content-Type": "application/json",
+            "X-TT-ENV": "ppe_eval_openapi",
+            "x-use-ppe": "1",
+        },
+        json={
+            "workspace_id": cozeloop_workspace_id,
+            "is_allow_partial_add": True,
+            "is_skip_invalid_items": True,
+            "items": items,
+        },
+    )
+
+    if response.status_code == 200:
+        logger.info(
+            f"Successfully uploaded dataset to CozeLoop evalset {cozeloop_evalset_id}"
+        )
+    else:
+        logger.error(f"Failed to upload dataset: {response.text}")

{veadk_python-0.2.11 → veadk_python-0.2.12}/veadk/cli/cli_web.py

@@ -56,10 +56,14 @@ def _get_ltm_from_env() -> LongTermMemory | None:
     logger = get_logger(__name__)
 
     long_term_memory_backend = os.getenv("LONG_TERM_MEMORY_BACKEND")
+    app_name = os.getenv("VEADK_WEB_APP_NAME", "")
+    user_id = os.getenv("VEADK_WEB_USER_ID", "")
 
     if long_term_memory_backend:
         logger.info(f"Long term memory: backend={long_term_memory_backend}")
-        return LongTermMemory(backend=long_term_memory_backend)  # type: ignore
+        return LongTermMemory(
+            backend=long_term_memory_backend, app_name=app_name, user_id=user_id
+        )  # type: ignore
     else:
         logger.warning("No long term memory backend settings detected.")
         return None
@@ -131,7 +135,13 @@ def patch_adkwebserver_disable_openapi():
 
 @click.command()
 @click.option("--host", default="127.0.0.1", help="Host to run the web server on")
-def web(host: str) -> None:
+@click.option(
+    "--app_name", default="", help="The `app_name` for initializing long term memory"
+)
+@click.option(
+    "--user_id", default="", help="The `user_id` for initializing long term memory"
+)
+def web(host: str, app_name: str, user_id: str) -> None:
     """Launch web with long term and short term memory."""
     import os
     from typing import Any
@@ -175,6 +185,9 @@ def web(host: str) -> None:
         self.session_service = short_term_memory.session_service
         self.memory_service = long_term_memory
 
+    os.environ["VEADK_WEB_APP_NAME"] = app_name
+    os.environ["VEADK_WEB_USER_ID"] = user_id
+
     import google.adk.cli.adk_web_server
 
     google.adk.cli.adk_web_server.AdkWebServer.__init__ = init_for_veadk

{veadk_python-0.2.11 → veadk_python-0.2.12}/veadk/memory/long_term_memory_backends/vikingdb_memory_backend.py

@@ -137,7 +137,7 @@ class VikingDBLTMBackend(BaseLongTermMemoryBackend):
     ) -> list[str]:
         filter = {
             "user_id": user_id,
-            "memory_type": ["sys_event_v1"],
+            "memory_type": ["sys_event_v1", "event_v1"],
         }
 
         logger.debug(

{veadk_python-0.2.11 → veadk_python-0.2.12}/veadk/tools/builtin_tools/generate_image.py

@@ -33,6 +33,10 @@ from veadk.consts import (
 from veadk.utils.logger import get_logger
 from veadk.utils.misc import formatted_timestamp, read_file_to_bytes
 from veadk.version import VERSION
+import asyncio
+import concurrent.futures
+import contextvars
+
 
 logger = get_logger(__name__)
 
@@ -43,11 +47,164 @@ client = Ark(
     base_url=getenv("MODEL_IMAGE_API_BASE", DEFAULT_IMAGE_GENERATE_MODEL_API_BASE),
 )
 
+executor = concurrent.futures.ThreadPoolExecutor(max_workers=8)
+tracer = trace.get_tracer("gcp.vertex.agent")
 
-async def image_generate(
-    tasks: list,
-    tool_context: ToolContext,
-) -> Dict:
+
+def _build_input_parts(item: dict, task_type: str, image_field):
+    input_part = {"role": "user"}
+    input_part["parts.0.type"] = "text"
+    input_part["parts.0.text"] = json.dumps(item, ensure_ascii=False)
+
+    if image_field:
+        if task_type.startswith("single"):
+            assert isinstance(image_field, str), (
+                f"single_* task_type image must be str, got {type(image_field)}"
+            )
+            input_part["parts.1.type"] = "image_url"
+            input_part["parts.1.image_url.name"] = "origin_image"
+            input_part["parts.1.image_url.url"] = image_field
+        elif task_type.startswith("multi"):
+            assert isinstance(image_field, list), (
+                f"multi_* task_type image must be list, got {type(image_field)}"
+            )
+            assert len(image_field) <= 10, (
+                f"multi_* task_type image list length must be <= 10, got {len(image_field)}"
+            )
+            for i, image_url in enumerate(image_field):
+                idx = i + 1
+                input_part[f"parts.{idx}.type"] = "image_url"
+                input_part[f"parts.{idx}.image_url.name"] = f"origin_image_{i}"
+                input_part[f"parts.{idx}.image_url.url"] = image_url
+
+    return input_part
+
+
+def handle_single_task_sync(
+    idx: int, item: dict, tool_context
+) -> tuple[list[dict], list[str]]:
+    logger.debug(f"handle_single_task_sync item {idx}: {item}")
+    success_list: list[dict] = []
+    error_list: list[str] = []
+    total_tokens = 0
+    output_tokens = 0
+    output_part = {"message.role": "model"}
+
+    task_type = item.get("task_type", "text_to_single")
+    prompt = item.get("prompt", "")
+    response_format = item.get("response_format", None)
+    size = item.get("size", None)
+    watermark = item.get("watermark", None)
+    image_field = item.get("image", None)
+    sequential_image_generation = item.get("sequential_image_generation", None)
+    max_images = item.get("max_images", None)
+
+    input_part = _build_input_parts(item, task_type, image_field)
+
+    inputs = {"prompt": prompt}
+    if size:
+        inputs["size"] = size
+    if response_format:
+        inputs["response_format"] = response_format
+    if watermark is not None:
+        inputs["watermark"] = watermark
+    if sequential_image_generation:
+        inputs["sequential_image_generation"] = sequential_image_generation
+
+    with tracer.start_as_current_span(f"call_llm_task_{idx}") as span:
+        try:
+            if (
+                sequential_image_generation
+                and sequential_image_generation == "auto"
+                and max_images
+            ):
+                response = client.images.generate(
+                    model=getenv("MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME),
+                    **inputs,
+                    sequential_image_generation_options=SequentialImageGenerationOptions(
+                        max_images=max_images
+                    ),
+                )
+            else:
+                response = client.images.generate(
+                    model=getenv("MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME),
+                    **inputs,
+                )
+
+            if not response.error:
+                logger.debug(f"task {idx} Image generate response: {response}")
+
+                total_tokens += getattr(response.usage, "total_tokens", 0) or 0
+                output_tokens += getattr(response.usage, "output_tokens", 0) or 0
+
+                for i, image_data in enumerate(response.data):
+                    image_name = f"task_{idx}_image_{i}"
+                    if "error" in image_data:
+                        logger.error(f"Image {image_name} error: {image_data.error}")
+                        error_list.append(image_name)
+                        continue
+
+                    if getattr(image_data, "url", None):
+                        image_url = image_data.url
+                    else:
+                        b64 = getattr(image_data, "b64_json", None)
+                        if not b64:
+                            logger.error(
+                                f"Image {image_name} missing data (no url/b64)"
+                            )
+                            error_list.append(image_name)
+                            continue
+                        image_bytes = base64.b64decode(b64)
+                        image_url = _upload_image_to_tos(
+                            image_bytes=image_bytes, object_key=f"{image_name}.png"
+                        )
+                        if not image_url:
+                            logger.error(f"Upload image to TOS failed: {image_name}")
+                            error_list.append(image_name)
+                            continue
+                        logger.debug(f"Image saved as ADK artifact: {image_name}")
+
+                    tool_context.state[f"{image_name}_url"] = image_url
+                    output_part[f"message.parts.{i}.type"] = "image_url"
+                    output_part[f"message.parts.{i}.image_url.name"] = image_name
+                    output_part[f"message.parts.{i}.image_url.url"] = image_url
+                    logger.debug(
+                        f"Image {image_name} generated successfully: {image_url}"
+                    )
+                    success_list.append({image_name: image_url})
+            else:
+                logger.error(
+                    f"Task {idx} No images returned by model: {response.error}"
+                )
+                error_list.append(f"task_{idx}")
+
+        except Exception as e:
+            logger.error(f"Error in task {idx}: {e}")
+            traceback.print_exc()
+            error_list.append(f"task_{idx}")
+
+        finally:
+            add_span_attributes(
+                span,
+                tool_context,
+                input_part=input_part,
+                output_part=output_part,
+                output_tokens=output_tokens,
+                total_tokens=total_tokens,
+                request_model=getenv(
+                    "MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME
+                ),
+                response_model=getenv(
+                    "MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME
+                ),
+            )
+    logger.debug(
+        f"task {idx} Image generate success_list: {success_list}\nerror_list: {error_list}"
+    )
+    return success_list, error_list
+
+
+async def image_generate(tasks: list[dict], tool_context) -> Dict:
     """
     Seedream 4.0: batch image generation via tasks.
     Args:
@@ -127,193 +284,72 @@ async def image_generate(
     - 如果想要指定生成组图的数量，请在prompt里添加数量说明，例如："生成3张图片"。
     - size 推荐使用 2048x2048 或表格里的标准比例，确保生成质量。
     """
-
+    logger.debug(
+        f"Using model: {getenv('MODEL_IMAGE_NAME', DEFAULT_IMAGE_GENERATE_MODEL_NAME)}"
+    )
     success_list: list[dict] = []
-    error_list = []
-
-    for idx, item in enumerate(tasks):
-        input_part = {"role": "user"}
-        output_part = {"message.role": "model"}
-        total_tokens = 0
-        output_tokens = 0
-        tracer = trace.get_tracer("gcp.vertex.agent")
-        with tracer.start_as_current_span("call_llm") as span:
-            task_type = item.get("task_type", "text_to_single")
-            prompt = item.get("prompt", "")
-            response_format = item.get("response_format", None)
-            size = item.get("size", None)
-            watermark = item.get("watermark", None)
-            image = item.get("image", None)
-            sequential_image_generation = item.get("sequential_image_generation", None)
-            max_images = item.get("max_images", None)
-
-            input_part["parts.0.type"] = "text"
-            input_part["parts.0.text"] = json.dumps(item, ensure_ascii=False)
-            inputs = {
-                "prompt": prompt,
-            }
-
-            if size:
-                inputs["size"] = size
-            if response_format:
-                inputs["response_format"] = response_format
-            if watermark:
-                inputs["watermark"] = watermark
-            if image:
-                if task_type.startswith("single"):
-                    assert isinstance(image, str), (
-                        f"single_* task_type image must be str, got {type(image)}"
-                    )
-                    input_part["parts.1.type"] = "image_url"
-                    input_part["parts.1.image_url.name"] = "origin_image"
-                    input_part["parts.1.image_url.url"] = image
-                elif task_type.startswith("multi"):
-                    assert isinstance(image, list), (
-                        f"multi_* task_type image must be list, got {type(image)}"
-                    )
-                    assert len(image) <= 10, (
-                        f"multi_* task_type image list length must be <= 10, got {len(image)}"
-                    )
-                    for i, image_url in enumerate(image):
-                        input_part[f"parts.{i + 1}.type"] = "image_url"
-                        input_part[f"parts.{i + 1}.image_url.name"] = (
-                            f"origin_image_{i}"
-                        )
-                        input_part[f"parts.{i + 1}.image_url.url"] = image_url
-
-            if sequential_image_generation:
-                inputs["sequential_image_generation"] = sequential_image_generation
-
-            try:
-                if (
-                    sequential_image_generation
-                    and sequential_image_generation == "auto"
-                    and max_images
-                ):
-                    response = client.images.generate(
-                        model=getenv(
-                            "MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME
-                        ),
-                        **inputs,
-                        sequential_image_generation_options=SequentialImageGenerationOptions(
-                            max_images=max_images
-                        ),
-                    )
-                else:
-                    response = client.images.generate(
-                        model=getenv(
-                            "MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME
-                        ),
-                        **inputs,
-                    )
-                if not response.error:
-                    for i, image_data in enumerate(response.data):
-                        image_name = f"task_{idx}_image_{i}"
-                        if "error" in image_data:
-                            error_details = (
-                                f"Image {image_name} error: {image_data.error}"
-                            )
-                            logger.error(error_details)
-                            error_list.append(image_name)
-                            continue
-                        if image_data.url:
-                            image = image_data.url
-                            tool_context.state[f"{image_name}_url"] = image
-
-                            output_part[f"message.parts.{i}.type"] = "image_url"
-                            output_part[f"message.parts.{i}.image_url.name"] = (
-                                image_name
-                            )
-                            output_part[f"message.parts.{i}.image_url.url"] = image
-
-                        else:
-                            image = image_data.b64_json
-                            image_bytes = base64.b64decode(image)
-
-                            tos_url = _upload_image_to_tos(
-                                image_bytes=image_bytes, object_key=f"{image_name}.png"
-                            )
-                            if tos_url:
-                                tool_context.state[f"{image_name}_url"] = tos_url
-                                image = tos_url
-                                output_part[f"message.parts.{i}.type"] = "image_url"
-                                output_part[f"message.parts.{i}.image_url.name"] = (
-                                    image_name
-                                )
-                                output_part[f"message.parts.{i}.image_url.url"] = image
-                            else:
-                                logger.error(
-                                    f"Upload image to TOS failed: {image_name}"
-                                )
-                                error_list.append(image_name)
-                                continue
-
-                            logger.debug(f"Image saved as ADK artifact: {image_name}")
-
-                        total_tokens += response.usage.total_tokens
-                        output_tokens += response.usage.output_tokens
-                        success_list.append({image_name: image})
-                else:
-                    error_details = (
-                        f"No images returned by Doubao model: {response.error}"
-                    )
-                    logger.error(error_details)
-                    error_list.append(f"task_{idx}")
-
-            except Exception as e:
-                error_details = f"Error: {e}"
-                logger.error(error_details)
-                traceback.print_exc()
-                error_list.append(f"task_{idx}")
-
-            add_span_attributes(
-                span,
-                tool_context,
-                input_part=input_part,
-                output_part=output_part,
-                output_tokens=output_tokens,
-                total_tokens=total_tokens,
-                request_model=getenv(
-                    "MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME
-                ),
-                response_model=getenv(
-                    "MODEL_IMAGE_NAME", DEFAULT_IMAGE_GENERATE_MODEL_NAME
-                ),
-            )
-    if len(success_list) == 0:
+    error_list: list[str] = []
+    logger.debug(f"image_generate tasks: {tasks}")
+    with tracer.start_as_current_span("image_generate"):
+        base_ctx = contextvars.copy_context()
+
+        def make_task(idx, item):
+            ctx = base_ctx.copy()
+            return lambda: ctx.run(handle_single_task_sync, idx, item, tool_context)
+
+        loop = asyncio.get_event_loop()
+        futures = [
+            loop.run_in_executor(executor, make_task(idx, item))
+            for idx, item in enumerate(tasks)
+        ]
+
+        results = await asyncio.gather(*futures, return_exceptions=True)
+
+        for res in results:
+            if isinstance(res, Exception):
+                logger.error(f"Task raised exception: {res}")
+                error_list.append("unknown_task_exception")
+                continue
+            s, e = res
+            success_list.extend(s)
+            error_list.extend(e)
+
+    if not success_list:
+        logger.debug(
+            f"image_generate success_list: {success_list}\nerror_list: {error_list}"
+        )
         return {
             "status": "error",
             "success_list": success_list,
             "error_list": error_list,
         }
-    else:
-        app_name = tool_context._invocation_context.app_name
-        user_id = tool_context._invocation_context.user_id
-        session_id = tool_context._invocation_context.session.id
+    app_name = tool_context._invocation_context.app_name
+    user_id = tool_context._invocation_context.user_id
+    session_id = tool_context._invocation_context.session.id
+    artifact_service = tool_context._invocation_context.artifact_service
+
+    if artifact_service:
+        for image in success_list:
+            for _, image_tos_url in image.items():
+                filename = f"artifact_{formatted_timestamp()}"
+                await artifact_service.save_artifact(
+                    app_name=app_name,
+                    user_id=user_id,
+                    session_id=session_id,
+                    filename=filename,
+                    artifact=Part(
+                        inline_data=Blob(
+                            display_name=filename,
+                            data=read_file_to_bytes(image_tos_url),
+                            mime_type=mimetypes.guess_type(image_tos_url)[0],
+                        )
+                    ),
+                )
 
-        artifact_service = tool_context._invocation_context.artifact_service
-        if artifact_service:
-            for image in success_list:
-                for _, image_tos_url in image.items():
-                    filename = f"artifact_{formatted_timestamp()}"
-                    await artifact_service.save_artifact(
-                        app_name=app_name,
-                        user_id=user_id,
-                        session_id=session_id,
-                        filename=filename,
-                        artifact=Part(
-                            inline_data=Blob(
-                                display_name=filename,
-                                data=read_file_to_bytes(image_tos_url),
-                                mime_type=mimetypes.guess_type(image_tos_url)[0],
-                            )
-                        ),
-                    )
-        return {
-            "status": "success",
-            "success_list": success_list,
-            "error_list": error_list,
-        }
+    logger.debug(
+        f"image_generate success_list: {success_list}\nerror_list: {error_list}"
+    )
+    return {"status": "success", "success_list": success_list, "error_list": error_list}
 
 
 def add_span_attributes(