人生苦短，学不完啊 2025-06-11T08:34:52.000Z https://puzhiyuan.github.io/ puzhiyuan Hexo https://puzhiyuan.github.io/llm/rlhf/verl_multinode/ 2025-06-11T08:34:52.000Z 2025-06-11T08:34:52.000Z

<h1 id="verl-multi-node-train"><a href="#verl-multi-node-train" class="headerlink" title="verl multi-node train"></a>verl multi-node

https://puzhiyuan.github.io/llm/rlhf/grpo/ 2025-06-03T09:36:20.000Z 2025-06-03T09:36:20.000Z

<h1 id="GRPO"><a href="#GRPO" class="headerlink" title="GRPO"></a>GRPO</h1><h2 id="核心流程概述"><a href="#核心流程概述" class="headerlink"

https://puzhiyuan.github.io/tools/docker/nvidia_container_toolkit/ 2025-05-11T05:41:52.000Z 2025-05-11T05:41:52.000Z

<h2 id="🧰-nvidia-container-toolkit"><a href="#🧰-nvidia-container-toolkit" class="headerlink" title="🧰 nvidia-container-toolkit"></a>🧰

https://puzhiyuan.github.io/tools/docker/docker_mirrors/ 2025-05-11T05:41:23.000Z 2025-05-11T05:41:23.000Z

<h2 id="Docker-mirrors"><a href="#Docker-mirrors" class="headerlink" title="Docker mirrors"></a>Docker mirrors</h2><h3

https://puzhiyuan.github.io/vlm/qwen2_5_vl/ 2025-05-07T07:43:05.000Z 2025-05-07T07:43:05.000Z

<h1 id="Qwen2-5-VL"><a href="#Qwen2-5-VL" class="headerlink" title="Qwen2.5-VL"></a>Qwen2.5-VL</h1><p>年初，Qwen团队推出了最新的视觉-语言（Vision-Language,

https://puzhiyuan.github.io/vlm/clip/ 2025-04-27T05:27:44.000Z 2025-04-27T05:27:44.000Z

<h1 id="CLIP"><a href="#CLIP" class="headerlink" title="CLIP"></a>CLIP</h1><h2 id="Abstract"><a href="#Abstract" class="headerlink"

https://puzhiyuan.github.io/deeplearning/distribute/zero/ 2025-02-17T11:28:28.000Z 2025-02-17T11:28:28.000Z

<h1 id="ZeRO"><a href="#ZeRO" class="headerlink" title="ZeRO:"></a>ZeRO:</h1><h2 id="0x00-Abstract"><a href="#0x00-Abstract"

https://puzhiyuan.github.io/deeplearning/distribute/modelparallel/ 2025-02-15T14:08:19.000Z 2025-02-15T14:08:19.000Z

<h2 id="模型并行"><a href="#模型并行" class="headerlink"

https://puzhiyuan.github.io/deeplearning/distribute/dataparallel/ 2025-02-15T09:04:17.000Z 2025-02-15T09:04:17.000Z

<p>深度学习模型通常包含大量的参数，随着模型规模和数据量的增长，单一设备往往无法承载训练过程中的计算与存储需求。为了解决这个问题，<strong>分布式训练</strong>技术应运而生，它能够有效地利用多台设备（如 GPU/TPU

https://puzhiyuan.github.io/llm/sft/lora/ 2025-01-06T13:01:39.000Z 2025-01-06T13:01:39.000Z

<h2 id="LoRA"><a href="#LoRA" class="headerlink" title="LoRA"></a>LoRA</h2><p><a href="https://arxiv.org/abs/2106.09685">原文链接：LoRA:

https://puzhiyuan.github.io/deploy/tensorrt_install/ 2024-12-28T09:19:49.000Z 2024-12-28T09:19:49.000Z

<h2 id="TensorRT"><a href="#TensorRT" class="headerlink" title="TensorRT"></a>TensorRT</h2><p>NVIDIA TensorRT 的<strong>核心是一个

https://puzhiyuan.github.io/autonomousdriving/sensorscalibration/lidar2camera/ 2024-12-08T14:44:07.000Z 2024-12-08T14:44:07.000Z

<h2 id="相机和激光雷达联合标定"><a href="#相机和激光雷达联合标定" class="headerlink"

https://puzhiyuan.github.io/deeplearning/fundamentals/fundamentals_of_deep_learning/ 2024-12-07T15:12:01.000Z 2024-12-07T15:12:01.000Z

<h2 id="LOSS"><a href="#LOSS" class="headerlink" title="LOSS"></a>LOSS</h2><p><a

https://puzhiyuan.github.io/python/python_tips/ 2024-12-07T05:38:22.000Z 2024-12-07T05:38:22.000Z

<h2 id="Q：python-函数调用是值传递还是引用传递"><a href="#Q：python-函数调用是值传递还是引用传递" class="headerlink" title="Q：python 函数调用是值传递还是引用传递"></a>Q：python

https://puzhiyuan.github.io/tools/terminal/tmux_screen/ 2024-11-22T14:18:35.000Z 2024-11-22T14:18:35.000Z

<h2 id="1-背景与简介"><a href="#1-背景与简介" class="headerlink" title="1. 背景与简介"></a>1. 背景与简介</h2><h3 id="1-1-tmux"><a href="#1-1-tmux"

https://puzhiyuan.github.io/tools/ssh/ssh_port_forwarding/ 2024-11-21T12:27:31.000Z 2024-11-21T12:27:31.000Z

<h3 id="SSH实现端口映射"><a href="#SSH实现端口映射" class="headerlink" title="SSH实现端口映射"></a>SSH实现端口映射</h3><p>SSH 端口映射（Port Forwarding）是一种通过 SSH

https://puzhiyuan.github.io/autonomousdriving/object_detection/3d/camera/sparse4dv2/ 2024-11-20T13:29:33.000Z 2024-11-20T13:29:33.000Z

<p><a href="https://arxiv.org/abs/2305.14018">原文链接：Sparse4D v2: Recurrent Temporal Fusion with Sparse Model</a></p> <p><a

https://puzhiyuan.github.io/tools/docker/docker_use/ 2024-10-26T13:32:04.000Z 2024-10-26T13:32:04.000Z

<h2 id="Docker-使用教程"><a href="#Docker-使用教程" class="headerlink" title="Docker 使用教程"></a>Docker 使用教程</h2><h3 id="1-什么是-Docker？"><a

https://puzhiyuan.github.io/autonomousdriving/object_detection/3d/camera/sparse4dv3/ 2024-10-24T13:34:25.000Z 2024-10-24T13:34:25.000Z

<p><a href="https://github.com/linxuewu/Sparse4D">Github：Sparse4Dv3</a></p> <h3 id="0-Abstract"><a href="#0-Abstract" class="headerlink"

https://puzhiyuan.github.io/tools/ssh/ssh_key/ 2024-10-23T13:37:16.000Z 2024-10-23T13:37:16.000Z

<p>最近在实习的时候用的 vscode