Pull Requests ggml-org/llama.cpp

ggml-hexagon: add HTP unary ops for ABS and LOG ggml Hexagon

#27786 opened 2026-08-27 05:56 by cqderek

hexagon: support for device discovery and create sessions on demand ggml Hexagon

#27785 opened 2026-08-27 04:38 by max-krasnyansky

tests: add ABS and LOG coverage to test-backend-ops testing

#27782 opened 2026-08-27 03:13 by cqderek

speculative : fix shared-KV detection for draft-mtp contexts

#27781 opened 2026-08-27 02:29 by Gwonk1

ggml-cpu: add F16 input to the FWHT testing ggml Apple Metal

#27779 opened 2026-08-27 01:24 by bri-prism

Revise function calling documentation documentation

#27778 opened 2026-08-27 01:08 by apollo-2006

[Metal/Apple Silicon Performance] Disable K-quant mul_mv_ext for M3 Pro in lieu of regular mul_mv ggml Apple Metal

#27776 opened 2026-08-27 00:49 by DavidLandup0

opencl: perf optimization for mamba2 ssm_scan by folding 4 dim rows into one workgroup and add more coverage ggml OpenCL

#27775 opened 2026-08-26 23:48 by wanghqc

add GLM-5.3-Flash (GLM5-Next) support model mtmd conversion

#27773 opened 2026-08-26 23:27 by timkhronos

llama-quantize: Add MoE chunk queue for faster multi-threaded quant creation

#27770 opened 2026-08-26 22:37 by bartowski1182

opencl: add bin kernels `kernel_gemm_moe_q4_0_q8_1_dp4a_bin`, `kernel_gemm_moe_mxfp4_q8_1_dp4a_bin` ggml OpenCL

#27768 opened 2026-08-26 22:04 by shawngu-quic

vulkan: add TQ1_0 support (mm, mat-vec, mat-vec-id, dequant, get_rows) testing Vulkan ggml

#27765 opened 2026-08-26 20:56 by Anjielon

chat : split specialized parsers into common/parsers

#27764 opened 2026-08-26 20:17 by pwilkin

server : report live generation throughput server

#27760 opened 2026-08-26 19:08 by xternet

metal : fix memory leaks due to missing autoreleasepools ggml merge ready Apple Metal

#27758 opened 2026-08-26 18:32 by nikwen

tests : run test-save-load-state across all architectures model testing

#27755 opened 2026-08-26 17:37 by ggerganov

model: add GLM-5-Next (GLM-5.3-Flash) model testing mtmd conversion

#27754 opened 2026-08-26 16:53 by danielhanchen

ci : build only the ggml-hip backend for windows-rocm release devops

#27753 opened 2026-08-26 16:12 by harkgill-amd

model : add GLM-5.3-Flash (glm5next) model testing conversion

#27752 opened 2026-08-26 16:07 by eauchs

server : fix Responses API item parsing for reasoning and Codex compatibility draft

#27751 opened 2026-08-26 15:04 by shengbox

ui: Improve Chat Form Actions UI/UX (models selector, add panel) server/ui

#27746 opened 2026-08-26 13:51 by allozaur

ui: Replace per-conversation MCP overrides with per-conversation tool policy server/ui

#27745 opened 2026-08-26 13:48 by allozaur

model: add Qwen3.8-Flash-Next (qwen4exp) model testing conversion

#27742 opened 2026-08-26 13:43 by danielhanchen

fix: model loader tensor count mismatch when MTP block contains NVFP4 weights with scales / input scales.

#27736 opened 2026-08-26 12:18 by stevelikesrhino

server : accept data: URLs for input_video and input_audio server

#27735 opened 2026-08-26 11:48 by geckguy

ggml-vulkan: allow fp32-only devices (Haswell hasvk) Vulkan ggml

#27723 opened 2026-08-25 23:59 by scottjoyner

CUDA graphs on Pascal (sm_61): +40% on MoE, +7% on dense, no regression ggml CUDA

#27721 opened 2026-08-25 22:49 by ArtyomITA

server : make /v1/models created timestamp stable server

#27719 opened 2026-08-25 20:16 by theycallmeloki

fix: Prevent Qwen3.6 reranker from allocating exponential memory as physical batch size increases model

#27715 opened 2026-08-25 18:21 by sredman

spec: Add benchmark-only synthetic speculative acceptance options documentation testing server

#27711 opened 2026-08-25 16:57 by gaugarg-nv