ngxson/llama.cpp

Pull Requests Commits
hexagon: tiling, tracing and optimizations for unary ops (#25474)
aparmp-quic committed 2 days ago
Verified fb30ba9a
server : move chat-template thinking probe inside the init try/catch (#24093)
palios-taey committed 2 days ago
Verified 82fce65d
ggml : fix conv 2d dw (#25490)
ggerganov committed 2 days ago
Verified 5c3a5868
meta: add hard emphasis on agents not writing descriptions/comments (#25480)
pwilkin committed 2 days ago
Verified c15c5c77
Refactor: Consistently use smart pointers in `test-backend-ops` (#25440)
ORippler committed 2 days ago
Verified f84a5194
Only index by compile times + always multiply/add (#25445)
ORippler committed 2 days ago
Verified 683f0c72
llama-bench : init params.offline (#25476)
angt committed 2 days ago
Verified 259f2e2a
metal : add CONV_2D_DW (depthwise convolution) support (#21565)
Sou-ly committed 2 days ago
Verified 92b187c9
ggml-hip: enable -funsafe-math-optimizations (#24668)
RapidMark committed 2 days ago
Verified ccb0c342
cuda: align snake fusion matcher with the other backends (#25460)
ServeurpersoCom committed 2 days ago
Verified 2021515a
server : respect min-step when splitting prompt batches (#25420)
aldehir committed 2 days ago
Verified 64c8b7db
hexagon: add VISION RoPE support (#25216)
aparmp-quic committed 2 days ago
Verified f2d1c2f3
ggml-webgpu: tune subgroup split (d_split) in flash_attn_vec (#25418)
yomaytk committed 2 days ago
Verified 32e41fa5
opencl: Q6_K GEMM/GEMV fix for ne01 of weights that are not multiples of 128. (#25464)
wanghqc committed 3 days ago
Verified 92366df3
vulkan: disable FA mask_opt on GCN to improve performance (#24362)
0cc4m committed 3 days ago
Verified a646006f
opencl: ragged-tile MoE prefill FP16 GEMM optimization (skip padded expert tiles) (#25433)
wanghqc committed 3 days ago
Verified 167d0576
llama-batch: fix allowed decreasing pos in a seq (#25449)
am17an committed 3 days ago
Verified 1ee09393
vulkan: for small AMD GPUs, reduce submission threshold based on CU count (#25240)
0cc4m committed 3 days ago
Verified 0bbc87b1
hexagon: new vtcm layouts and improved pipelines for MUL_MAT, MUL_MAT_ID and FLASH_ATTN_EXT (#25425)
max-krasnyansky committed 3 days ago
Verified 81ff7abe
cli : move to HTTP-based implementation (#24948)
ngxson committed 3 days ago
Verified c264f65f
Make hip quality check run on all changes (#25403)
ORippler committed 3 days ago
Verified 07e012af
cuda : add support for f16->f16 GGML_OP_SET_ROWS (#25367)
fairydreaming committed 3 days ago
Verified ed8c2615
llama: refactor fused ops (#24646)
am17an committed 3 days ago
Verified 90e0f5cf
server-stream: follow-up on SSE Replay Buffer (#23226) (#25047)
ServeurpersoCom committed 3 days ago
Verified bbebeec4
llama-batch: add n_keep_tail in split_equal for recurrent models (#25278)
am17an committed 3 days ago
Verified 230ea9d2
common: auto-create prompts-log-dir at argument parsing, so all tools using the flag benefit (#25322)
rankaiyx committed 3 days ago
Verified f296fdfb
ui: Context usage gauge and panel (#25340)
allozaur committed 3 days ago
Verified f1161b15
llama-eval : fix crash when answer is None in HTML dump (#25435)
ggerganov committed 3 days ago
Verified da46e59c
metal : add set_rows with src0 f16 (#25434)
fairydreaming committed 3 days ago
Verified 0512ef1e
fix: OOB reads in UGM tokenizer (precompiled_charsmap handling) (#18750)
hourhl committed 3 days ago
Verified 4a7ee312
Terms Of Use Privacy Policy Refund Policy Impressum