Search results

CUDA illegal memory access with fixed Turbo4 on mixed RTX 5090 + 4090 layer split

Status: Open.

#81 In spiritbuun/buun-llama-cpp;

· hughmadden opened on Jul 12, 2026

Feature Request: pull support for TQ3_4S from https://github.com/turbo-tan/llama.cpp-tq3

enhancementNew feature or requestNew feature or request

Status: Open.

#51 In spiritbuun/buun-llama-cpp;

· mahald opened on May 10, 2026

Misc. bug: [CUDA error] In a dual RTX 4090 environment.

Status: Open.

#47 In spiritbuun/buun-llama-cpp;

· LateLinux opened on May 8, 2026

Eval bug: buun 分支的 DFlash 实现在你的异构双卡（16G + 6G、两个不同代 GPU）上存在 Bug

Status: Open.

#34 In spiritbuun/buun-llama-cpp;

· andytsao2023 opened on Apr 30, 2026

Eval bug: CUDA illegal memory access with dual GPU + DFlash speculative decoding + turbo4 TCQ

Status: Open.

#28 In spiritbuun/buun-llama-cpp;

· Nullicc opened on Apr 27, 2026

[Bug] Crash with GGML_ASSERT(ggml_can_repeat) when using DFlash + TurboQuant together

Status: Open.

#22 In spiritbuun/buun-llama-cpp;

· hamesgit opened on Apr 25, 2026

Eval bug: Performance issue in Prompt Processing with Gemma 4 26B-A4B Model

Status: Open.

#16 In spiritbuun/buun-llama-cpp;

· adrianhoehne opened on Apr 11, 2026