All issues

Search results

Desktop engine installer ignores existing FreeToken engine path and always installs a new venv in AppData

bugSomething isn't workingSomething isn't working

Desktopproblem related to FreeToken Desktopproblem related to FreeToken Desktop

windows

Status: Open.

#387 In FlashML-org/FreeToken;

· MSM-App opened on Sep 4, 2026

Build on Windows?

bugSomething isn't workingSomething isn't working

windows

Status: Open.

#384 In FlashML-org/FreeToken;

· XeonG opened on Sep 4, 2026

--moe-cache-auto ignores --num-tokens / --num-pages: expert fill overruns the budget and the KV pool OOMs at boot

linux

Status: Open.

#383 In FlashML-org/FreeToken;

· gdevenyi opened on Sep 4, 2026

Move the The Cache config element in the console window for QOL improvement.

Desktopproblem related to FreeToken Desktopproblem related to FreeToken Desktop

featureNew feature or requestNew feature or request

Status: Open.

#382 In FlashML-org/FreeToken;

· bmgjet opened on Sep 4, 2026

[Bug]: qwen3_5_moe fails to convert/load compressed-tensors NVFP4 checkpoints with model.language_model prefix

Status: Open.

#381 In FlashML-org/FreeToken;

· silvertakana opened on Sep 4, 2026

ci(container): publish official Docker images to GHCR

featureNew feature or requestNew feature or request

Status: Open.

#379 In FlashML-org/FreeToken;

· Fgeeha opened on Sep 4, 2026

Congratulations for the project, finally able to run something useful locally using this project

Status: Open.

#373 In FlashML-org/FreeToken;

· rodjjo opened on Sep 3, 2026

TP > 1: every rank keeps torch's default intra-op thread count, so a multi-rank weight load busy-waits itself to a standstill on a small-core host (19 min -> 61 s)

Status: Open.

#371 In FlashML-org/FreeToken;

· lukascechovic opened on Sep 3, 2026

fix(checkpoint): FTW output drops the PLE n-gram shards, so qwen4_exp checkpoints cannot be served from it

Status: Open.

#370 In FlashML-org/FreeToken;

· sgzxs opened on Sep 3, 2026

Unified-memory platforms (DGX Spark, sm_121): moe-backend auto resolves to offload, adding ~130s/request of pure overhead

Status: Open.

#369 In FlashML-org/FreeToken;

· iamanishx opened on Sep 3, 2026

TP>1: the first request can be lost in the rank-0→rank-N ZeroMQ PUB/SUB relay if it arrives before rank N's subscription is registered — both ranks then block forever while /v1/models keeps answering 200

Status: Open.

#364 In FlashML-org/FreeToken;

· lukascechovic opened on Sep 3, 2026

Feature Request: Community‑driven hardware benchmark sharing database

Status: Open.

#363 In FlashML-org/FreeToken;

· mayipiaopiao opened on Sep 3, 2026