🚨 AI News | TestingCatalog· @testingcatalog · X·· 20 天前AI 评分55
AI 导读
nace-ai 在 Hugging Face 发布 drex-max-v1 的 27 GB Q8_0 量化权重,可通过 Nace 版本的 llama.cpp 和 Ollama 在 Mac 上运行。在 M5 Pro、48 GB 内存的测试中,该量化模型对样例请求的答案概率与运行在 NVIDIA H100 上的完整模型相差在 0.02 以内。单个问题最多可给出 255 个答案,权重与代码已在该 Hugging Face 页面开放。
正文 · 原文
The weights are available on Hugging Face as a 27 GB Q8_0 file that can run on a Mac through Nace's versions of llama.cpp and Ollama.
In Nace's test performed on an M5 Pro with 48 GB of memory, its answer probabilities for a sample request were within 0.02 of the full model running on an NVIDIA H100.
A single question can offer up to 255 answers.
Weights and code 👇
https://huggingface.co/nace-ai/drex-max-v1
来源:🚨 AI News | TestingCatalog · x.com