Search results

Feature Request: server: expose per-device memory usage (weights / context / compute)

Status: Open.

#26129 In ggml-org/llama.cpp;

· tobocop2 opened on Jul 26, 2026

Misc. bug: server prompt cache is incompatible with the RPC backend — GGML_ASSERT(tensor->data != NULL) in prompt_save() → state_seq_get_data(), triggered as soon as -np > 1

Status: Open.

#26128 In ggml-org/llama.cpp;

· doornail opened on Jul 26, 2026

Windows HIP release omits gfx1152; Radeon 860M fails with invalid device function

Status: Open.

#26127 In ggml-org/llama.cpp;

· AIwork4me opened on Jul 26, 2026

Feature Request: Support Motif-3-Beta

enhancementNew feature or requestNew feature or request

Status: Open.

#26125 In ggml-org/llama.cpp;

· EugeoSynthesisThirtyTwo opened on Jul 26, 2026

Hexagon: dspqueue_read failed: 0x0000002e during graph compute on Snapdragon 8 Gen 2 (v73), reproducible at 4B+ params, works fine at 1B

Status: Open.

#26123 In ggml-org/llama.cpp;

· secretlytravis opened on Jul 25, 2026

Eval bug: --swa-full + -sm tensor crashes on load for Gemma4-31B (ISWA KV cache / ggml_backend_meta_alloc_ctx_tensors_from_buft)

bug-unconfirmed

Status: Open.

#26122 In ggml-org/llama.cpp;

· piet8stevens opened on Jul 25, 2026

Security Issue: Detected: Trojan:Win32/Wacatac.H!ml - Build b10107

Status: Open.

#26120 In ggml-org/llama.cpp;

· njardim opened on Jul 25, 2026

CUDA cublasSgemm_v2 unsupported parameter during speculative decoding on V100

Status: Open.

#26119 In ggml-org/llama.cpp;

· GrassfedMod opened on Jul 25, 2026

allow llama serve -hf to use llama-server in router mode

Status: Open.

#26116 In ggml-org/llama.cpp;

· ngxson opened on Jul 25, 2026

Feature Request: support Upstage's Solar-Open2

enhancementNew feature or requestNew feature or request

Status: Open.

#26115 In ggml-org/llama.cpp;

· eleius opened on Jul 25, 2026

--load-mode refactor (#20834) removed the only safe combo for --n-cpu-moe CPU-offload: "no mmap, but pin the CPU-resident tensors"

bug-unconfirmed

Status: Open.

#26110 In ggml-org/llama.cpp;

· fillika opened on Jul 25, 2026

Eval bug: Failed to read image when using DFlash (Qwen3.6 27B)

bug-unconfirmed

Status: Open.

#26108 In ggml-org/llama.cpp;

· fantasyz opened on Jul 25, 2026