All issues

Search results

gemma4 GGUF: embedding quant type hardcoded to Q6_K, so a Q4_0-embedding GGUF fails to load with a bare AssertionError

Status: Open.

#188 In FlashML-org/FreeToken;

· salekseev opened on Aug 25, 2026

Use the main app folder's drive for venv install

Status: Open.

#187 In FlashML-org/FreeToken;

· VictorZakharov opened on Aug 25, 2026

GGUF MoE GEMV crashes with cudaErrorInvalidValue once tokens*top_k exceeds 65535 (grid.z limit)

Status: Open.

#186 In FlashML-org/FreeToken;

· erichstuntebeck opened on Aug 25, 2026

Apodex-1.1-mini-NVFP4 fails: dense_quant hardcoded to NVFP4 when experts are NVFP4, but shared_expert is FP8 (modelopt MIXED_PRECISION)

Status: Open.

#183 In FlashML-org/FreeToken;

· zangetsu787 opened on Aug 25, 2026

FP8 row-wise _scaled_mm hangs on SM89 with PyTorch 2.11

Status: Open.

#182 In FlashML-org/FreeToken;

· endenis opened on Aug 25, 2026

Try to load antirez/deepseek-v4-gguf (q2 87GB) but I get the error: architecture 'deepseek4' is not supported

Status: Open.

#181 In FlashML-org/FreeToken;

· dan64 opened on Aug 25, 2026

K8S helm chart

Status: Open.

#180 In FlashML-org/FreeToken;

· jomcgi opened on Aug 25, 2026

can I run Qwen 3.8 27b on RTX 2060 8GB + 48GB RAM?

Status: Open.

#179 In FlashML-org/FreeToken;

· krisheeetu opened on Aug 25, 2026

Technical Question About Model Size Limits and MoE Execution

Status: Open.

#178 In FlashML-org/FreeToken;

· Hosseinghorbani0 opened on Aug 25, 2026

Feature: Predictive Expert Offloading

Status: Open.

#176 In FlashML-org/FreeToken;

· Chavell3 opened on Aug 25, 2026

Loading files into the model.

Status: Open.

#175 In FlashML-org/FreeToken;

· MrHousek opened on Aug 25, 2026

MoE offload cache: LRU realizes ~8-18% decode hit rate on DSV4 fine-grained MoE, frequency-pinning oracle shows 46% at the same size

Status: Open.

#174 In FlashML-org/FreeToken;

· angelbaba01L opened on Aug 25, 2026