ggml-org/llama.cpp

Pull Requests Commits

Merge remote-tracking branch 'origin/master' into qwen4exp/qwen3.8-flash-next

ngxson committed 2 days ago

3d7baaeb

quantize: cap working memory size to avoid loading big tensors onto RAM (#27795)

ngxson committed 2 days ago

Verified 732707df

LazyChunkedTensor

ngxson committed 2 days ago

a510c82e

improve the m-rope image for get_prev_tokens

ngxson committed 2 days ago

fbc5135a

Feature: Added LIGHTNING_INDEXER support for Deepseek V4 ops on Vulkan Backend (#27453)

shenron0101 committed 2 days ago

Verified cb300598

replace some changes with #27795

ngxson committed 2 days ago

250b6144

pr2wt : use ssh/https remote in worktree depending on base (#27800)

CISC committed 2 days ago

Verified 1a946ec7

ngxson committed 2 days ago

af1ffaf3

Merge remote-tracking branch 'origin/master' into qwen4exp/qwen3.8-flash-next

ngxson committed 2 days ago

539c35e9

llama: model_loader: add TENSOR_READ_LAZY (#27794)

ngxson committed 2 days ago

Verified fac889fb

revert LLAMA_MMAP_RANDOM

ngxson committed 2 days ago

41b4b707

ui: Improve Chat Form Actions UI/UX (models selector, add panel) (#27746)

allozaur committed 2 days ago

Verified cae63579

clean up new comments

ngxson committed 2 days ago

24391470

Merge commit 'refs/pull/27742/head' of github.com:ggerganov/llama.cpp into qwen4exp/qwen3.8-flash-next

ngxson committed 2 days ago

56672bf6

convert : fix Nemotron-H LoRA GGUF conversion (#27356)

frozenblade1224 committed 2 days ago

Verified bcb6084a

clean up code comments

ngxson committed 2 days ago

5b34ce29

qwen4exp: bias the QSA selection per block, not per cell

danielhanchen committed 2 days ago

ef687669

FACP (Fewer Acronym Classes Please)

CISC committed 2 days ago

Verified e5227ab2

ui: Replace per-conversation MCP overrides with per-conversation tool policy (#27745)

allozaur committed 2 days ago

Verified fe235f43

spec: Add benchmark-only synthetic speculative acceptance options (#27711)

gaugarg-nv committed 2 days ago

Verified 2bb9bdda

model : simplify MiniMax-01 graph (#27790)

fairydreaming committed 2 days ago

Verified deae5ee1

Merge commit 'refs/pull/27742/head' of github.com:ggerganov/llama.cpp into qwen4exp/qwen3.8-flash-next

ngxson committed 2 days ago

69a1269f

args: add --video-* CLI arguments (#24318)

ngxson committed 2 days ago

Verified f2955121

llama: fold the random-access prefetch into its own feature flag

danielhanchen committed 2 days ago

b8bdf73b

llama: narrow the random-access mmap advice to the gather table

danielhanchen committed 2 days ago

58325573

llama: opt-in random-access mmap advice for host-resident gather tables

danielhanchen committed 2 days ago

95da4ba8

qwen4exp: shrink the PLE hparams storage

danielhanchen committed 2 days ago

c3259772

llama: give the qwen4exp full memory context its indexer cache

danielhanchen committed 2 days ago

1caf6c57

no more ple_hist (use master version)

ngxson committed 2 days ago

5eeb97d1

Merge remote-tracking branch 'origin/master' into qwen4exp/qwen3.8-flash-next

ngxson committed 2 days ago

b5b5fc30

Older