Repository navigation

Issues

New issue

changelog : libllama API
#9289 · ggerganov opened on Sep 3, 2024
15
changelog : llama-server REST API
#9291 · ggerganov opened on Sep 3, 2024
21
tutorials : list for llama.cpp
#13523 · ggerganov opened on May 14, 2025
24

Feature Request: restrict qwen4exp QSA to a sequence's own cells with a unified KV cache
ggml-org/llama.cpp#32181 · opened on Nov 13, 2026
Misc. bug: HIP/gfx1200 FLASH_ATTN_EXT produces varying results and NaNs
ggml-org/llama.cpp#32178 · opened on Nov 13, 2026
Feature Request: ternary weights with a 1024-wide Hadamard rotation (one plane and multi-plane) - which way upstream?
ggml-org/llama.cpp#32160 · opened on Nov 12, 2026
1
Misc. bug: tools/ui npm ci fails on npm 12 with EALLOWREMOTE after #32112 (llama-brand tarball dependency)
ggml-org/llama.cpp#32158 · opened on Nov 12, 2026
Misc. bug: pplx-decider detection matches non-Qwen3.5 models by file names only, blocks conversion
ggml-org/llama.cpp#32152 · opened on Nov 12, 2026
Misc. bug: backend sampling ignores min_keep and drops all tokens with min_p = 1
ggml-org/llama.cpp#32141 · opened on Nov 12, 2026
Eval bug: CUDA MoE MUL_MAT_ID crash on large ubatch
ggml-org/llama.cpp#32139 · opened on Nov 12, 2026
Feature Request: Support Qwen3-ForcedAligner-0.6B
ggml-org/llama.cpp#32127 · opened on Nov 12, 2026
Misc. bug: Vulkan on Windows + AMD Strix Halo (APU): idle weights evicted from device-local memory within ~1.5 s; depends on commit headroom; ping keep-alive ineffective
ggml-org/llama.cpp#32124 · opened on Nov 12, 2026
1
Misc. bug: server slot discards the restored prompt if the LoRA is different for another call/model
ggml-org/llama.cpp#32123 · opened on Nov 12, 2026
1
RPC: GGML_OP_LIGHTNING_INDEXER scheduled on remote Metal device aborts when indexer head_count != 64 (GLM-5.3)
ggml-org/llama.cpp#32122 · opened on Nov 12, 2026
Misc. bug: HIP: CU_CHECK(err) in the VMM pool lets cuMemAddressReserve spin forever
ggml-org/llama.cpp#32105 · opened on Nov 11, 2026
Misc. bug: hf downloader mode not router aware
ggml-org/llama.cpp#32104 · opened on Nov 11, 2026
2
Eval bug: qwen2vl graph ignores NVFP4 FFN scales (NULL ffn_*_s), NVIDIA Qwen2.5-VL-7B-Instruct-NVFP4 gives NaN / PPL ~1e6
ggml-org/llama.cpp#32101 · opened on Nov 11, 2026
Misc. bug: llama-server stops after first token when a stop string is empty
ggml-org/llama.cpp#32092 · opened on Nov 11, 2026
Misc. bug: Tool call allow/deny buttons disappear on reload
ggml-org/llama.cpp#32075 · opened on Nov 11, 2026
1
Misc. bug: Skip reasoning button disappears if the page is reloaded
ggml-org/llama.cpp#32074 · opened on Nov 11, 2026
1
Misc. bug: Skip reasoning doesn't work after tool calls
ggml-org/llama.cpp#32073 · opened on Nov 11, 2026
Eval bug: LFM2-VL-1.6B: Slow image processing and image context retention on Android (Galaxy S24)
ggml-org/llama.cpp#32070 · opened on Nov 11, 2026
Feature Request: Support Zaya1 -8B Model
ggml-org/llama.cpp#32069 · opened on Nov 11, 2026
CUDA backend: fatal CUDA error + NVIDIA driver reset (nvlddmkm Event 153) during sustained batched prompt processing while an NVENC encoder shares the GPU — Vulkan build is stable at the same VRAM footprint (depth-dependent)
ggml-org/llama.cpp#32062 · opened on Nov 11, 2026
1
Eval bug: Qwen3-TTS repeated requests produce different audio with Vulkan cooperative matrices on Radeon 8060S
ggml-org/llama.cpp#32050 · opened on Nov 10, 2026
1
Misc. bug: server: slot erase does not clear the --cache-ram prompt cache, so an earlier prompt is restored after erase
ggml-org/llama.cpp#32048 · opened on Nov 10, 2026
3
Misc. bug: scripts/moe-cache-tune.py crashes on Windows with "ValueError: Unsupported signal: 2" and leaves llama-server running
ggml-org/llama.cpp#32041 · opened on Nov 10, 2026
build-sanitize.yml: ctest is non-blocking for all three sanitizers (job-level continue-on-error)
ggml-org/llama.cpp#32037 · opened on Nov 10, 2026
1