forked from ggml-org/llama.cpp
-
Notifications
You must be signed in to change notification settings - Fork 41
Pull requests: unslothai/llama.cpp
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
qwen4exp: gather-based sparse attention for QSA decode (+ cuda top-k batching fix)
#165
opened Aug 31, 2026 by
abdel-darwish-27
Loading…
ggml-cuda: avoid direct ROCm_Host compute on HIP integrated GPUs (port of ggml-org#25863)
bug
Something isn't working
#158
opened Aug 31, 2026 by
danielhanchen
Member
Loading…
ggml-cuda: use cudaMemcpyDefault in the ggml_cuda_cpy 2D fast path
bug
Something isn't working
#157
opened Aug 31, 2026 by
danielhanchen
Member
Loading…
kv-cells: hoist the sequence list out of the for_each_token_in cell loop
#154
opened Aug 31, 2026 by
danielhanchen
Member
Loading…
llama: map each contiguous run of a context's tensors, not one span over all of them
#152
opened Aug 31, 2026 by
danielhanchen
Member
Loading…
qwen4exp: compute QSA cache-layout inputs once per ubatch, not per layer
#150
opened Aug 30, 2026 by
talexeyft
Loading…
ggml-cuda: make GGML_CUDA_ENABLE_UNIFIED_MEMORY=0 actually disable it, and say so on HIP
bug
Something isn't working
#149
opened Aug 30, 2026 by
danielhanchen
Member
Loading…
ci: ship test-backend-ops in the Vulkan and ROCm prebuilts
#146
opened Aug 30, 2026 by
danielhanchen
Member
Loading…
llama: let an MTP draft borrow the target's embeddings and lm head
#142
opened Aug 28, 2026 by
danielhanchen
Member
Loading…
llama: batched readahead for lazily read gather tables
#137
opened Aug 28, 2026 by
danielhanchen
Member
Loading…
hip: enable GGML_CUDA_USE_CUB via hipCUB (fixes TOP_K abort for >1024 rows)
#136
opened Aug 27, 2026 by
davidcanar
Loading…
qwen4exp: keep the QSA indexer cache in lockstep with the attention cache
#135
opened Aug 27, 2026 by
vladimir-voinea
Loading…
conversion : fix Qwen3.5 MTP config lookup from text_config (fixes unslothai/unsloth#8443)
#102
opened Aug 17, 2026 by
chakshu-dhannawat
Loading…
fix(qwen): resolve GGUF export failure when MTP config is nested under text_config
#101
opened Aug 12, 2026 by
IshantDere
Loading…
sampling: index penalties by token id instead of scanning every candidate
#95
opened Aug 11, 2026 by
danielhanchen
Member
Loading…
Prototype: pin the CPU legs' glibc floor with a container, not the runner
#79
opened Aug 7, 2026 by
danielhanchen
Member
•
Draft
kimi-k3 : the MoonViT-3d vision tower and full-size loading fixes
#70
opened Aug 6, 2026 by
danielhanchen
Member
Loading…
IQ1_XS, IQ1_XXS, IQ1_XXXS: three quant types below IQ1_S
#61
opened Aug 3, 2026 by
danielhanchen
Member
Loading…
ProTip!
Follow long discussions with comments:>50.