forked from ggml-org/llama.cpp
-
Notifications
You must be signed in to change notification settings - Fork 89
Pull requests: PrismML-Eng/llama.cpp
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
cuda: query the device type with cudaDeviceGetAttribute, not cudaGetDeviceProperties (+8% decode)
CUDA
ggml
#163
opened Sep 7, 2026 by
bri-prism
Loading…
cuda: native PTQ1_0 MMQ tile loader (closes the dequantize-to-cuBLAS prefill fallback)
CUDA
ggml
#160
opened Sep 6, 2026 by
bri-prism
Loading…
metal: TQ1_0 kernel performance (dense decode +39-55%, prefill +78-90%)
Apple Metal
ggml
#156
opened Sep 5, 2026 by
bri-prism
Loading…
dflash: AngelSpec DFly drafter support
conversion
documentation
Improvements or additions to documentation
model
testing
#146
opened Sep 3, 2026 by
bri-prism
Loading…
cuda: optimize Bonsai low-bit kernels for GB10
CUDA
ggml
testing
#135
opened Aug 30, 2026 by
pooyakhosravi
Loading…
cuda: AMD RDNA4 Q1_0/Q2_0 — HIP-path vec_dots (+37%/+16% decode), opt-in quant dedup, opt-in hipBLASLt prefill routes
CUDA
ggml
#116
opened Aug 11, 2026 by
The-Monk
Loading…
cuda: MoE prefill — fused SwiGLU epilogue and fused weighted reduction
CUDA
ggml
#107
opened Aug 3, 2026 by
bri-prism
Loading…
5 tasks done
cuda: enable the Hopper wgmma Q1_0/Q2_0 prefill path by default
CUDA
ggml
#106
opened Aug 2, 2026 by
bri-prism
Loading…
4 of 5 tasks
Refactor CUDA workflow and remove ROCm steps
devops
documentation
Improvements or additions to documentation
#105
opened Aug 1, 2026 by
MasterCool4389
Loading…
POWER10/11 MMA acceleration for all quantized formats (silicon-validated)
ggml
#100
opened Jul 23, 2026 by
mavin2009
Loading…
speculative: window dspark drafter staging to its trained position range
documentation
Improvements or additions to documentation
server
testing
#98
opened Jul 22, 2026 by
thadreber-web
Loading…
ggml-cpu: enable Q2_0 VNNI kernel on AVX-VNNI-only CPUs
ggml
#76
opened Jul 15, 2026 by
gondoi
Loading…
ProTip!
Type g i on any issue or pull request to go back to the issue listing page.