HF Viewer device farm!
Summary
HF Viewer 创作者宣布推出 embedl.com，一个收录 1000+ 针对边缘设备优化模型的平台，用户可通过免费的真实边缘设备农场（如 NVIDIA Orin、AGX Thor、RTX GPU）直接运行模型并获得详细性能分析。
Similar Articles
@haoge6699: Back in the day, fine-tuning large models was a pipe dream—no GPUs, no cash, no setup, and regular folks couldn't even …
Unsloth Studio 现已接入免费的 Google Colab，支持超过 500 个开源模型（如 Qwen、DeepSeek、Gemma），让普通用户无需 GPU 或付费硬件即可进行大模型微调。
@hank_aibtc: Amazing! Running Gemma 4 in the browser, on par with ChatGPT?! Completely zero server, zero data upload, offline, pure WebGPU local inference! Xenova has open-sourced all 27 custom WebGPU kernels written by Fable 5: - Gemma 4 E2B (2.3B parameters...)
The article introduces Xenova's open-sourcing of 27 custom WebGPU kernels, enabling Gemma 4 to run fully offline and locally in the browser at 255 tok/s, and discusses advantages like privacy and offline use. It also mentions FLUX.2's 3D generation capability.
@eternityspring: https://x.com/eternityspring/status/2084666970558074971
Detailed explanation of how to run the MiniMax H3 video generation model locally on an RTX 4080 16GB using ComfyUI's native workflow, including model download, directory configuration, parameter tuning, and common pitfalls.
An 80B MoE (Qwen3-Next, 48 GB GGUF) answering in a Chrome tab: llama.cpp's WebGPU backend, experts paged in from disk
作者将上周的 MoE 专家按需从磁盘流式加载的方案接入 llama.cpp 的 WebGPU 后端，使 80B MoE 模型（Qwen3-Next，48GB GGUF）也能在 Chrome 标签页中运行，无需为每个模型手写引擎。该方案新增了 ggml-webgpu 的缓冲类型、slot 映射的 MUL_MAT_ID 着色器和驻留逻辑，实测在 M4 Pro MacBook 上以 2.9 tok/s 解码 80B 模型，较小模型可达 22 tok/s。
@MinLiBuilds: https://x.com/MinLiBuilds/status/2089338660386992295
This article compares the performance of NVIDIA DGX Spark and a modified RTX 4090 in locally deploying the Qwen3.8-27B and Ling-3.0-flash models, providing benchmark data and purchase recommendations.