Search results

Support slot save/restore for slots containing images

enhancementNew feature or requestNew feature or request

Status: Open.

#25854 In ggml-org/llama.cpp;

· CHIPMUNK-T0T opened on Jul 18, 2026

Eval bug: CUDA GGML_ASSERT(tensor->nb[0] == ggml_element_size(tensor)) fitting Deepseek V4 Flash, but manual tensor split works

bug-unconfirmed

Status: Open.

#25851 In ggml-org/llama.cpp;

· jettoblack opened on Jul 18, 2026

Can't disable thinking after upgrade, Qwen_Qwen3.6 webui

Status: Open.

#25849 In ggml-org/llama.cpp;

· bitraft opened on Jul 18, 2026

Misc. bug: llama_bench error: failed to run depth (deepseek v4 + metal)

bug-unconfirmed

Status: Open.

#25843 In ggml-org/llama.cpp;

· tarruda opened on Jul 17, 2026

Eval bug: server -np 3 decode drops ~4.4x (166 -> 38 t/s) whenever the two busy slots are the non-contiguous pair {0,2} — hybrid/recurrent model (Qwen3.6-35B-A3B), deterministic, b9966

bug-unconfirmed

Status: Open.

#25840 In ggml-org/llama.cpp;

· mike22beatty opened on Jul 17, 2026

Eval bug: deepseek4 (DeepSeek-V4-Flash) garbled output on Apple Silicon (ARM/NEON) CPU — repack of attn_output_a/_b

Status: Open.

#25837 In ggml-org/llama.cpp;

· KikoCis opened on Jul 17, 2026

Eval bug: BF16/F16 models crash with "ROCm error: invalid argument" (MUL_MAT) on gfx1201/Windows - second rocBLAS GEMM with different lda fails

Status: Open.

#25836 In ggml-org/llama.cpp;

· badbat75 opened on Jul 17, 2026

Eval bug: VRAM leak with CUDA Graphs(Enable CUDA graphs on Volta+Turing #25749) on V100

bug-unconfirmed

Status: Open.

#25835 In ggml-org/llama.cpp;

· QiDian2025 opened on Jul 17, 2026

Eval bug: b10063cannot run model https://huggingface.co/prism-ml/Ternary-Bonsai-27B-gguf

bug-unconfirmed

Status: Open.

#25833 In ggml-org/llama.cpp;

· cnzyan opened on Jul 17, 2026

Eval bug: split-mode = tensor is broken since b10054

bug-unconfirmed

Status: Open.

#25829 In ggml-org/llama.cpp;

· wOvAN opened on Jul 17, 2026

Eval bug: mtmd llava models run one ViT layer short (subtracted at conversion AND graph build)

bug-unconfirmed

Status: Open.

#25817 In ggml-org/llama.cpp;

· waleedabujaish opened on Jul 17, 2026

Eval bug: mtmd llava graph feeds projected [CLS] to the LLM and shifts patch position embeddings

bug-unconfirmed

Status: Open.

#25814 In ggml-org/llama.cpp;

· waleedabujaish opened on Jul 17, 2026