All issues

Search results

fix(engine): ft ctl cache live resize: OOM during rebuild breaks rollback and wedges the server

Status: Open.

#526 In FlashML-org/FreeToken;

· alvarorsouza-arch opened on Sep 19, 2026

Failed to load kernel error on install

bugSomething isn't workingSomething isn't working

Desktopproblem related to FreeToken Desktopproblem related to FreeToken Desktop

linux

Status: Open.

#523 In FlashML-org/FreeToken;

· zerochance1958 opened on Sep 18, 2026

Add freetoken CLI command to avoid PowerShell’s built-in ft alias

Desktopproblem related to FreeToken Desktopproblem related to FreeToken Desktop

featureNew feature or requestNew feature or request

Status: Open.

#522 In FlashML-org/FreeToken;

· miseqs opened on Sep 18, 2026

[TP=2] Quantized MoE expert parallelism on heterogeneous consumer CUDA GPUs — a working NVFP4 port, bit-exact vs TP=1

Status: Open.

#520 In FlashML-org/FreeToken;

· Yuepixel opened on Sep 18, 2026

[Bug] Desktop installer fails / breaks when changing installation directory to a custom path (e.g., D: drive)

bugSomething isn't workingSomething isn't working

Desktopproblem related to FreeToken Desktopproblem related to FreeToken Desktop

windows

Status: Open.

#519 In FlashML-org/FreeToken;

· Novicracker opened on Sep 18, 2026

Feature request / guidance: tensor-parallel across heterogeneous and idle consumer GPUs (quantized MoE experts) - a hobbyist request, with reference to #478

Status: Open.

#510 In FlashML-org/FreeToken;

· Yuepixel opened on Sep 18, 2026

Yarn support for Qwen 3.8 Next Flash

featureNew feature or requestNew feature or request

Status: Open.

#508 In FlashML-org/FreeToken;

· manisabri opened on Sep 17, 2026

FrozenInstanceError: cannot assign to field 'audio_config'

bugSomething isn't workingSomething isn't working

linux

Status: Open.

#506 In FlashML-org/FreeToken;

· TheFausap opened on Sep 17, 2026

[Feature] Expose per-request inference performance metrics in /v1/chat/completions

featureNew feature or requestNew feature or request

Status: Open.

#503 In FlashML-org/FreeToken;

· FantasticCode2019 opened on Sep 17, 2026

Cached-prefix prefill runs 10-40x slower than a cold prefill on a hybrid MoE model (scheduler spends 90% in _invalidate_prefill_buffer + stream sync)

Status: Open.

#501 In FlashML-org/FreeToken;

· alvarorsouza-arch opened on Sep 17, 2026

[BUG] --reasoning-parser auto gives non-thinking Instruct checkpoints a thinking parser, so message.content comes back empty

bugSomething isn't workingSomething isn't working

linux

Status: Open.

#497 In FlashML-org/FreeToken;

· ljrirene opened on Sep 17, 2026

[BUG] shard_tensor() chunks 0-dim ModelOpt scales; kv-cache scales break the qwen3_moe qkv merge

bugSomething isn't workingSomething isn't working

linux

Status: Open.

#496 In FlashML-org/FreeToken;

· ljrirene opened on Sep 17, 2026