😳震惊！vLLM直接砍掉Kimi K3.1 Pro 75%专家，体积暴降到476GB！
This is a dev post classified by AGK Intelligence as Hosting & infra (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
😳震惊！vLLM直接砍掉Kimi K3.1 Pro 75%专家，体积暴降到476GB！ 2.8万亿参数的超大MoE模型，被粗暴剪到只剩约700-750B参数，模型文件直接压到476GB，原始每层896个专家只保留224个，砍掉75%，属于校准无关的实验性剪枝，专门用来做vLLM服务和性能测试。 核心亮点： 🔹参数从2.8T暴降到约700-750B，体积只剩476GB 🔹每层专家：896 → 224（保留25%） 🔹定位是烟雾测试/性能调优，不是能正常用的模型 🔹GSM8K准确率从原版约95.5%直接掉到34.4%，质量回退非常明显 🔹官方明确说输出基本不可用，只适合跑服务和基准 使用步骤： 1️⃣安装支持该模型的vLLM（需trust-remote-code） 2️⃣直接运行： vllm serve mgoin/Kimi-K3-pruned75 \ --served-model-name kimi-k3-pruned-75pct \ --tensor-parallel-size 4 \ --trust-remote-code \ --gpu-memory-utilization 0.95 \ --max-model-len 32768 \ --kv-cache-dtype fp8 3️⃣服务起来后用OpenAI兼容接口调用测试 模型地址：http
Posted by SuSu_酥酥👅 (26.8k followers) 1 h ago · 14 likes · 855 views · view the original post on X. Kept by the Dev Radar as Hosting & infra. Tools mentioned: kimi-k3-pruned75.
More dev work like this
- A Hybrid #MachineLearning with #AWS Application. #BigData #Analytics #DataScience #AI… — @gp_pulipaka
- Grafana, Jenkins and Coder all send telemetry data to their makers by default, even when… — @IntCyberDigest
- Another great startup program is AWS Activate. — @tarasshyn
- Incredible work by Ashley and team! — @Jilles
- Connect @Cloudflare to Superconductor, and your agents can deploy Workers, set up DNS… — @Superconductor
- What does it really take to scale AI infrastructure? — @CoreWeave
- Managing Cross Connect services just got simpler. Updated Cross Connect APIs and a… — @Equinix
- portless now supports Docker Compose — @RaillyHugo
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.7k posts from 5k X accounts over the last 21 days, 9.4k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-05 22:53 UTC. Full method.