[Optimization] The pre- and post-processing pipeline do not perform dict conversion (#5494)

* to_request_for_infer initial commit * refact to from_chat_completion_request * preprocess use request initial commit * bugfix * processors refact to using request * bug fix * refact Request from_generic_request * post process initial commit * bugfix * postprocess second commit * bugfix * serving_embedding initial commit * serving_reward initial commit * bugfix * replace function name * async_llm initial commit * offline initial commit and fix bug * bugfix * fix async_llm * remove add speculate_metrics into data * fix logprobs bug * fix echo bug * fix bug * fix reasoning_max_tokens * bugfix * bugfix and modify unittest * bugfix and modify unit test * bugfix * bugfix * bugfix * modify unittest * fix error when reasong_content is none for text_processor * remove some unnessary logic * revert removed logic * implement add and set method for RequestOutput and refact code * modify unit test * modify unit test * union process_request and process_request_obj * remove a unit test * union process_response and process_response_obj * support qwen3_vl_processor * modify unittest and remove comments * fix prompt_logprobs * fix codestyle * add v1 * v1 * fix unit test * fix unit test * fix pre-commit * fix * add process request * add process request * fix * fix * fix unit test * fix unit test * fix unit test * fix unit test * fix unit test * remove file * add unit test * add unit test * add unit test * fix unit test * fix unit test * fix * fix --------- Co-authored-by: Jiaxin Sui <95567040+plusNew001@users.noreply.github.com> Co-authored-by: luukunn <981429396@qq.com> Co-authored-by: luukunn <83932082+luukunn@users.noreply.github.com> Co-authored-by: Zhang Yulong <35552275+ZhangYulongg@users.noreply.github.com>
2026-04-25 09:57:51 +08:00 · 2026-01-22 00:50:52 +08:00
parent fe5ba4b509
commit 6e416c62dd
66 changed files with 16614 additions and 739 deletions
@@ -0,0 +1,15 @@
+"""
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+"""
@@ -0,0 +1,109 @@
+"""
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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 base64
+import datetime
+import hashlib
+import os
+import threading
+import uuid
+from pathlib import Path
+
+import requests
+from PIL import Image
+
+RAW_VIDEO_DIR = "./download_tmp/raw_video/"
+RAW_IMAGE_DIR = "./download_tmp/raw_images/"
+EXTRACTED_FRAME_DIR = "./download_tmp/extracted_frames/"
+TMP_DIR = "./download_tmp/upload_tmp/"
+
+
+def file_download(url, download_dir, save_to_disk=False, retry=0, retry_interval=3):
+    """
+    Description: 下载url，如果url是PIL直接返回
+    Args:
+        url(str, PIL): http/本地路径/io.Bytes，注意io.Bytes是图片字节流
+        download_path: 在save_to_disk=True的情况下生效，返回保存地址
+        save_to_disk: 是否保存在本地路径
+
+    """
+    from .video_utils import VideoReaderWrapper
+
+    if isinstance(url, Image.Image):
+        return url
+    elif isinstance(url, VideoReaderWrapper):
+        return url
+    elif url.startswith("http"):
+        response = requests.get(url)
+        bytes_data = response.content
+    elif os.path.isfile(url):
+        if save_to_disk:
+            return url
+        bytes_data = open(url, "rb").read()
+    else:
+        bytes_data = base64.b64decode(url)
+    if not save_to_disk:
+        return bytes_data
+
+    download_path = os.path.join(download_dir, get_filename(url))
+    Path(download_path).parent.mkdir(parents=True, exist_ok=True)
+    with open(download_path, "wb") as f:
+        f.write(bytes_data)
+    return download_path
+
+
+def get_filename(url=None):
+    """
+    Get Filename
+    """
+    if url is None:
+        return str(uuid.uuid4()).replace("-", "")
+    t = datetime.datetime.now()
+    if not isinstance(url, bytes):
+        url = url.encode("utf-8")
+
+    md5_hash = hashlib.md5(url).hexdigest()
+    pid = os.getpid()
+    tid = threading.get_ident()
+
+    # 去掉后缀，防止save-jpg报错
+    image_filname = f"{t.year}-{t.month:02d}-{t.day:02d}-{pid}-{tid}-{md5_hash}"
+    return image_filname
+
+
+def get_downloadable(
+    url,
+    download_dir=RAW_VIDEO_DIR,
+    save_to_disk=False,
+    retry=0,
+    retry_interval=3,
+):
+    """download video and store it in the disk
+
+    return downloaded **path** if save_to_disk is set to true
+    return downloaded **bytes** if save_to_disk is set to false
+    """
+
+    if not os.path.exists(download_dir):
+        os.makedirs(download_dir)
+    downloaded_path = file_download(
+        url,
+        download_dir,
+        save_to_disk=save_to_disk,
+        retry=retry,
+        retry_interval=retry_interval,
+    )
+    return downloaded_path
@@ -0,0 +1,103 @@
+"""
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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 os
+from pathlib import Path
+
+from PIL import Image, ImageDraw, ImageFont
+
+cur_directory = Path(__file__).parent.absolute()
+FONT_PATH = os.path.join(cur_directory, "Roboto-Regular.ttf")
+
+
+def render_single_image_with_timestamp(image: Image, number: str, rate: float, font_path: str = FONT_PATH):
+    """
+    函数功能: 给pil.image的图片渲染时间戳
+    时间戳的大小为 min(width, height)的rate
+    字体的颜色为黑色, 轮廓是白色, 轮廓的大小是字体的10%
+    返回一个 Image 对象
+    """
+    draw = ImageDraw.Draw(image)  # 创建一个可绘制对象
+    width, height = image.size  # 获取图片大小
+    font_size = int(min(width, height) * rate)  # 设置字体大小
+    outline_size = int(font_size * 0.1)  # 设置轮廓大小
+    font = ImageFont.truetype(font_path, font_size)  # 加载字体文件, 设置字体大小
+    x = 0
+    y = 0  # 文本的x坐标, y坐标
+
+    # 绘制黑色的时间戳，白色的边框
+    draw.text(
+        (x, y),
+        number,
+        font=font,
+        fill=(0, 0, 0),
+        stroke_width=outline_size,
+        stroke_fill=(255, 255, 255),
+    )
+
+    return image
+
+
+def timestamp_converting(time_stamp_in_seconds):
+    """
+    convert timestamp format from seconds to hr:min:sec
+    """
+    # get hours
+    hours = 0
+    while time_stamp_in_seconds >= 3600:
+        hours += 1
+        time_stamp_in_seconds -= 3600
+    # get minutes
+    mins = 0
+    while time_stamp_in_seconds >= 60:
+        mins += 1
+        time_stamp_in_seconds -= 60
+    time_hours = f"{int(hours):02d}"
+    time_mins = f"{int(mins):02d}"
+    time_secs = f"{time_stamp_in_seconds:05.02f}"
+    fi_time_stamp = time_hours + ":" + time_mins + ":" + time_secs
+
+    return fi_time_stamp
+
+
+def get_timestamp_for_uniform_frame_extraction(num_frames, frame_id, duration):
+    """
+    function: get the timestamp of a frame, 在均匀抽帧时用。
+
+    num_frames: 总帧数
+    frameid_list: 被抽帧的帧的索引
+    duration: 视频的总时长
+    return: timestamp; xx:xx:xx (str)
+    """
+    time_stamp = duration * 1.0 * frame_id / num_frames
+
+    return time_stamp
+
+
+def render_frame_timestamp(frame, timestamp, font_rate=0.1):
+    """
+    函数功能, 给frame, 按照顺序将 index 渲染上去
+    逻辑思路: 把index渲染到图片的左上方
+
+    frame: 帧，PIL.Image object
+    timestamp: 时间戳，单位是秒
+    font_rate: 字体大小占 min(wi, hei)的比率
+    """
+
+    time_stamp = "time: " + timestamp_converting(timestamp)
+    new_frame = render_single_image_with_timestamp(frame, time_stamp, font_rate)
+
+    return new_frame
@@ -0,0 +1,83 @@
+"""
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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 io
+import os
+from tempfile import NamedTemporaryFile as ntf
+
+import decord
+
+try:
+    # moviepy 1.0
+    import moviepy.editor as mp
+except:
+    # moviepy 2.0
+    import moviepy as mp
+
+
+def is_gif(data: bytes) -> bool:
+    """
+    check if a bytes is a gif based on the magic head
+    """
+    return data[:6] in (b"GIF87a", b"GIF89a")
+
+
+class VideoReaderWrapper(decord.VideoReader):
+    """
+    Solving memory leak bug
+
+    https://github.com/dmlc/decord/issues/208
+    """
+
+    def __init__(self, video_path, *args, **kwargs):
+        with ntf(delete=True, suffix=".gif") as gif_file:
+            gif_input = None
+            self.original_file = None
+            if isinstance(video_path, str):
+                self.original_file = video_path
+                if video_path.lower().endswith(".gif"):
+                    gif_input = video_path
+            elif isinstance(video_path, bytes):
+                if is_gif(video_path):
+                    gif_file.write(video_path)
+                    gif_input = gif_file.name
+            elif isinstance(video_path, io.BytesIO):
+                video_path.seek(0)
+                tmp_bytes = video_path.read()
+                video_path.seek(0)
+                if is_gif(tmp_bytes):
+                    gif_file.write(tmp_bytes)
+                    gif_input = gif_file.name
+
+            if gif_input is not None:
+                clip = mp.VideoFileClip(gif_input)
+                mp4_file = ntf(delete=False, suffix=".mp4")
+                clip.write_videofile(mp4_file.name, verbose=False, logger=None)
+                clip.close()
+                video_path = mp4_file.name
+                self.original_file = video_path
+
+            super().__init__(video_path, *args, **kwargs)
+            self.seek(0)
+
+    def __getitem__(self, key):
+        frames = super().__getitem__(key)
+        self.seek(0)
+        return frames
+
+    def __del__(self):
+        if self.original_file and os.path.exists(self.original_file):
+            os.remove(self.original_file)