Skip to content

Long story creation leads to single token responses when context grows #1944

Closed as not planned

Description

@RobinMeyer

I used a story generator with strata 0.1.42, which sends a request for every section with the content of all earlier request as context and there it happened, that after about 12000 tokens, the responses are cut in the middle and all following responses return only one token anymore.
I'm using RTX5070ti 16GB, 7950X3d and 64GB RAM. The configuration is:
"--expert-cache",
"auto",
"--prefill",
"auto",
"--spec",
"4",
"--mtp",
"D:\AI\Strata-data\mtp\rt",
"--max-context",
"131072",
"--kv",
"int8",
"--kv-resident",
"32768",
"--spec-min-p",
"0.5",
"--pool-workers",
"8",
"--adapt-every",
"1",
"--adapt-swaps",
"160",
"--adapt-decay",
"0.97"
With strata 0.1.39, the error seems not to happen. There, I just wrote a 90000 character story without any broken chapters with the same config.

Activity

  1. RobinMeyer commented on Oct 11, 2026

    @RobinMeyer
    Author

    Sorry, now the same error happened also with v0.1.39, when creating another story, while the same strata was still running.
    If you want to test the behavior: https://github.com/RobinMeyer/AI_Story_Generator

  2. RobinMeyer commented on Oct 11, 2026

    @RobinMeyer
    Author

    I made another test with 200k context and q4_0. There, a story with about 30k token was correctly finished. Another story creation without closing the program and strata ended after about 20k token. So it may help to increase the context.

  3. RobinMeyer commented on Oct 15, 2026

    @RobinMeyer
    Author

    Today, I made a story of about 45k token with these settings:

    "--expert-cache",
      "auto",
      "--prefill",
      "auto",
      "--spec",
      "4",
      "--mtp",
      "D:\\AI\\Strata-data\\mtp\\rt",
      "--max-context",
      "65536",
      "--kv",
      "k8v4",
      "--kv-resident",
      "32768",
      "--vram-reserve-mib",
      "1500",
      "--pcie-frac",
      "1.00",
      "--spec-min-p",
      "0.70",
      "--pool-workers",
      "10",
      "--adapt-every",
      "1",
      "--adapt-swaps",
      "160",
      "--adapt-decay",
      "0.97"
    

    with strata 0.1.44.1 and IQ3-XXS. It worked well and the story was consistent. The main change was doubling the VRAM reserve, but I don't know, if this is really the issue.

  4. Niko1221 commented on Oct 19, 2026

    @Niko1221
    Owner

    Thanks for taking the time to report this.

    We're resetting the tracker for 0.2.0-beta. There are more than 1,300 open issues and pull requests, too many for us to read properly, so we're closing everything that isn't being worked on right now. Closed here does not mean rejected.

    To reopen it takes about two minutes:

    1. Re-test on 0.2.0-beta. It may already be fixed. Write down what you see.
    2. Start the title with its area: [Engine], [GUI], [TUI], [Benchmark], [Docs] or [Other].
    3. Fill in the parts of the issue form: what happened, the Strata version, your graphics card, memory and system.
    4. Comment /reopen. A check runs and either reopens the issue or lists what is still missing.

    The rules are in CONTRIBUTING.md.

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