WHAT NEW IN X10.28 WED

AI collections from X · 22 collections · 58,924 posts

jevIMAGE

LLM2Jev: open-source local multimodal decision engine

LLM2Jev turns a local LLM into a structured decision engine that accepts text and images, needs only prefill, and ships with three runnable demos.

SuSu_酥酥👅
@NFT_ChenArchived 2026.09.24Originally posted 2026.09.23
LLM2Jev: open-source local multimodal decision engine (image 1 of 3)
LLM2Jev: open-source local multimodal decision engine (image 2 of 3)
LLM2Jev: open-source local multimodal decision engine (image 3 of 3)
SuSu_酥酥👅 @NFT_ChenOriginally posted 2026.09.23
Open original on X

💥逆天了!支持多模态的本地 Jev 模型来了!图文皆可喂! LLM2Jev 把本地大模型变成结构化决策引擎,图文都能一起喂进去,模型可以看着图做判断,开源、数据不出本地、模型随便换。 核心亮点 🔹仅需 prefill,不做自回归解码
🔹图文多模态正式支持(SGLang / Transformers) 🔹Choice / Score / Noul 三类结构化输出,选项顺序不影响分数
🔹兼容 System One API,后续还有 MLX
🔹三个新 Demo:网页秒懂玩法、AI 自己玩贪吃蛇、视觉判断驱动机械臂 上手 4 步 1️⃣git clone https://github.com/Yinsongxu/LLM2Jev.git 2️⃣cd LLM2Jev && uv sync --extra sglang 3️⃣source .venv/bin/activate 4️⃣python examples/sglang_inference.py --model-path /你的模型路径 文档、原理、测试结果都齐了,新朋友也能从零上手。 
GitHub:https://github.com/Yinsongxu/LLM2Jev #LLM2Jev #Jev #多模态 #开源AI #本地大模型 #SystemOne #SGLang

This post quotes another post: “DiffusionGemma deploys as a Jev-compatible decision endpoint in one command”

x.com/nft_chen/status/2102643562928308671Archived 2026.09.24 10:17