Hermes Agent自动部署DeepSeek v4 Flash，双DGX Sparks单流82 tok/s
Hermes agent看了条帖子就自动部署好DeepSeek v4 Flash，双DGX Sparks跑出82 tok/s，省人力还快。
MiaAI实验室用Hermes agent根据一条X帖子，在一夜之间自动完成了DeepSeek v4 Flash GA (0731)在2块NVIDIA DGX Sparks上的部署。
部署使用官方FP8权重，单流推理速度达82 tok/s，峰值约95 tok/s，三会话并发时达到135 tok/s。该方案已发布在GitHub上，并感谢了0xSero提交的PR，后续还会改进部署配方。
Just gave Hermes agent this post and it set it up overnight with no issues: Mia @MiaAI_lab Run DeepSeek v4 Flash GA (0731) on 2x @NVIDIAAI DGX Sparks with easy ✨ 82 tok/s single stream, peak ~95 tok/s 135 tok/s at 3 sessions Uses the official FP8 weights. Special thanks to @0xSero for a PR on this. Further improvements to the recipe are planned! Get it here: github.com/MiaAI-Lab/Deep… 🔗 View Quoted Tweet📊 2 ⚡
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include DeepSeek v4 Flash, Hermes, NVIDIA DGX Sparks, FP8.
Source attribution: curated from andrew chen (@andrewchen). Cross-source corroboration: 11 independent sources have covered this development (including IT之家, 深度求索 DeepSeek, 歸藏(guizang.ai)), confirming industry-wide signal strength.
