Skip to content

Spill kv-cache to nvme #47

Closed
Closed

Description

@maedoc

(hi first thx for this project, i'd not dreamed I could run such a model on a 2yo GPU!)

I've had some success with the ninfer 4090 project tiering kv cache to nvme, and on my system, there's no enough main RAM for keeping full kv cache, so I'm wondering if it would make sense to enable spilling kv cache out to an nvme ssd, so that across many different long agent sessions, refills stay rare. I'm thinking of taking a stab at it and sending a PR, but if you have initial thoughts I'd be happy to have them.

Activity

  1. Niko1221 commented on Sep 28, 2026

    @Niko1221
    Owner

    I will leave this opened. First I need to do some kernel magic and after that, I'll try this as well.

  2. Niko1221 commented on Sep 28, 2026

    @Niko1221
    Owner

    Keeping this open to track the NVMe tier: it will build on the shared snapshot core from #57 once that's merged (see #52).

  3. Niko1221 commented on Sep 30, 2026

    @Niko1221
    Owner

    The NVMe tier is part of the unified plan in #57 now (spill on eviction on the shared snapshot core). Closing in favour of #57.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions