github-featured / ollama--ollama

发布

0 发布 616 Git 标签

RSS 订阅源
  • v0.11.4 f2e9c9aff5
    比较

    server: Reduce gpt-oss context length for small VRAM GPUs

    frostbyte_neo 发布于 2025-08-07 21:23:55 +00:00

    gpt-oss works best with a context length of at least 8k. However,
    for GPUs with limited amount of VRAM, there is a significant
    performance hit to this increased context. In these cases, we
    switch to the Ollama default of 4k

    下载附件