ggerganov/llama.cpp

Pull Requests Commits

CUDA: vectorize same-type get_rows with int4 copy (#25929)

pwilkin committed 70 days ago

Verified 305ba519

hexagon: add CLAMP op (#25934)

tboinovski1 committed 70 days ago

Verified 76f46ad2

ui: Sidebar Conversations Bulk Action + Improved Settings logic/UI (#25815)

allozaur committed 70 days ago

Verified 2beefef6

llama_dsv4: write only used rows in state (#25325)

am17an committed 71 days ago

Verified 91d2fc38

ui: fix collapsed user bubble with markdown rendering (#25869)

ServeurpersoCom committed 71 days ago

Verified 4ee6a9af

UI: fix Settings/Display tool call content toggle (#25783)

ServeurpersoCom committed 71 days ago

Verified 43b5e635

ui: enable the agentic flow when only the JS sandbox is active (#25865)

ServeurpersoCom committed 71 days ago

Verified 1521a9ac

opencl: Support broadcast for Adreno MUL_MAT and honor `view_offs` for Adreno Q8_0 MUL_MAT for llama-server multi-stream (#25910)

wanghqc committed 71 days ago

Verified 178a6c44

model: rotate injected K/V cache for DFlash (#25823)

ruixiang63 committed 73 days ago

Verified 571d0d54

llama-quant : exclude i32 ffn_gate_tid2eid routing table from quantization (#25787)

devYRPauli committed 73 days ago

Verified 4937ca83

opencl: load and use `kernel_gemm_moe_q6_k_f32_ns` from bin kernel lib (#25797)

lhez committed 73 days ago

Verified 86a9c79f

opencl: read/write MoE dp4a activation tiles to local memory as 128-bit (vectorized LD/ST perf opt) for Adreno GPUs (#25810)

wanghqc committed 73 days ago

Verified 6bdd77f1

opencl: transpose q4_K noshuffle scales for coalesced reads (#25805)

wanghqc committed 74 days ago

Verified 86d86ed4

ggerganov committed 74 days ago

7d56da7e

ggml : bump version to 0.17.0 (ggml/1568)

ggerganov committed 74 days ago

37274040

tests : initialize all tensors in test_dsv4_hc to avoid NaNs in sentinel tensors (#25822)

fairydreaming committed 74 days ago

Verified 5d5306bf

common : auto-download dflash- and eagle3- HF sidecars (#25811)

ggerganov committed 74 days ago

Verified 635cdd5f

ggml-blas: default hadamard mul_mat to cpu routine (#25710)

taronaeo committed 74 days ago

Verified 11fd0a6f

vulkan: Support Q2_0 (#25430)

jeffbolznv committed 74 days ago

Verified 788e07dc

sycl: fix row calculation when K_QUANTS_PER_ITERATION is 1 (#25690)

malsbat committed 74 days ago

Verified 0bd0ec60

opencl: add ABS op (#25115)

Gezahegne committed 74 days ago

Verified b85833e9

opencl: loads quants as uint for q4_K and q5_K flat mv (optimization for Adreno A7x GPUs) (#25780)

wanghqc committed 74 days ago

Verified e8f19cc0

docs: added a note about using OpenCl with Adreno 810 (#25786)

akleine committed 74 days ago

Verified ac2557cb

DeepseekV4: Add fused hyper-connection ops (#25585)

am17an committed 74 days ago

Verified 0dc74e33

hexagon: L2 cache handling rework (dirty bit tracking with lazy flushing) and more MUL_MAT updates (#25762)

max-krasnyansky committed 74 days ago

Verified b2dd28a3

kleidiai: Add SME vs SME2 distinction in kernel dispatch (#25478)

matcraje committed 74 days ago

Verified f15bd609

vulkan: when using transfer queue for async copies, sync on event_wait to avoid race (#25229)

0cc4m committed 75 days ago

Verified b15ca938

conversion: accept BitNetForCausalLM architecture name (#25769)

khashayarghafouri committed 75 days ago

Verified 3278e921

TP: fix Phi3, Bert, Plamo2/3, ChatGLM (#25536)

JohannesGaessler committed 75 days ago

Verified 2e1fd764

vendor: update BoringSSL to 0.20260713.0 (#25624)

cabelo committed 75 days ago

Verified 86b719bf

Newer Older