Skip to content

Massive performance drop from 0.1.44 moving onwards #4482

Open

Description

@tmielsch

After the update from 0.1.44 to the next version I went from 80-90 to 40-60 decode. I reverted to 0.1.44 and stayed there so far. I retried with 0.2.4 and the new dsh based launcher, and its still the same

Ryzen 3950x, 4070 Ti super 16gb, 128gb DDR4

{
"exe": "Strata\engine\strata.exe",
"args": [
"--pack",
"Strata-data\packs\swift-iq2_xs",
"--native",
"Strata-data\models\swift-IQ2_XS\Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf",
"--ple-gguf",
"Strata-data\models\swift-IQ2_XS\Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf",
"--expert-profile",
"Strata\data\expert-profile.bin",
"--expert-cache",
"auto",
"--prefill",
"auto",
"--spec",
"4",
"--spec-min-p",
"0.5",
"--mtp",
"E:\_AI\Strata-data\mtp\rt",
"--max-context",
"524288",
"--rope-scaling",
"yarn",
"--rope-scale",
"2",
"--kv",
"int8",
"--kv-resident",
"32768",
"--vision",
"--vram-reserve-mib",
"300",
"--prompt-cache",
"12",
"--prompt-cache-root",
"2048",
"--conversation-cache-mib",
"37000",
"--conversation-cache-slots",
"12"
],
"cwd": "Strata",
"tokenizer": Strata-data\packs\swift-iq2_xs\tokenizer",
"model_name": "swift-1.5-iq2_xs",
"log": "Strata\strata-swift-iq2_xs.log",
"lib_dirs": [
"Strata\.venv\Lib\site-packages\nvidia\cu13\bin\x86_64"
],
"port": 8080,
"gpu": 0,
"gpus_asked": true,
"host": "0.0.0.0",
"api_key": "123",
"vision": {
"exe": "Strata\engine\strata-vision.exe",
"mmproj": "Strata-data\models\mmproj-Swift-Qwen3.8-Flash-Next-BF16.gguf",
"model": "Strata-data\models\swift-IQ2_XS\Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf",
"gpu": false,
"max_tokens": 8192
},
"allowed_hosts": [
"xxxxxx"
],
"effort_position": "end"
}

Activity

  1. github-actions commented on Oct 28, 2026

    @github-actions

    Thanks for the issue. To keep the tracker readable, every item needs an area prefix in the title and the parts of the issue form filled in. This one is missing:

    • Start the title with its area: [Engine], [GUI], [TUI], [Benchmark], [Docs] or [Other], then a short description. Example: "[Engine] decode stops after 2 minutes".

    Edit the issue (title and description) and this note goes away on its own. If it is still missing parts in 5 days it will be closed, which is not a rejection: fix it and comment /reopen.

    Details: CONTRIBUTING.md

  2. Niko1221 commented on Oct 29, 2026

    @Niko1221
    Owner

    Please post the engine start log of a slow start (the lines with expert cache, the PCIe probe and the VRAM free figures) and say whether STRATA_DEFAULTS=026 or dropping --vision or --kv-resident brings back 80 tok/s.

  3. stratbasher commented on Oct 31, 2026

    @stratbasher

    I'm having the exact same experience - 4070 Super and 64GB of DDR5 system memory, i7 14700k.

    Goes from 60t/s decode on 0.1.44 with Swift 1.5 IQ3_XXS INT8 down to like 10-30t/s decode on the latest 0.2 release.

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

    needs-formatTitle or template does not follow CONTRIBUTING.md yet

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions