RCreddit.com·
暂不在当前实时榜单
Getting stupidly good results on my 4x3060ti setup.
A user achieved impressive results with their 4x3060ti setup by investing in a new motherboard and CPU to utilize existing graphics cards from an old crypto miner. They then employed a Vllm setup with the syv-ai/qwen38-27b-rtx3090 fast-variant serving shape model, which is a reduced reasoning finetune of Qwen3.8-27B built from Swift's own weights and outputs, leading to their success.
This report details a specific model, syv-ai/qwen38-27b-rtx3090, and its performance on a 4x3060ti setup, unlike general discussions of hardware optimization.
时间与来源
时间显示为 UTC
显示时区:UTC
本地时区尚不可用,暂时显示 UTC。
收录当时偏移:UTC+02026年9月26日 20:00 UTC
- 收录
- 2026年9月26日 20:00
- 来源类型
- 开发者社区
本站未收录正文。
前往源站阅读 →