ngxson/llama.cpp

Pull Requests Commits

ggml-webgpu: tune subgroup split (d_split) in flash_attn_vec (#25418)

yomaytk committed 5 days ago

Verified 32e41fa5

opencl: Q6_K GEMM/GEMV fix for ne01 of weights that are not multiples of 128. (#25464)

wanghqc committed 5 days ago

Verified 92366df3

vulkan: disable FA mask_opt on GCN to improve performance (#24362)

0cc4m committed 5 days ago

Verified a646006f

opencl: ragged-tile MoE prefill FP16 GEMM optimization (skip padded expert tiles) (#25433)

wanghqc committed 5 days ago

Verified 167d0576

llama-batch: fix allowed decreasing pos in a seq (#25449)

am17an committed 5 days ago

Verified 1ee09393

vulkan: for small AMD GPUs, reduce submission threshold based on CU count (#25240)

0cc4m committed 5 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 5 days ago

Verified 81ff7abe

cli : move to HTTP-based implementation (#24948)

ngxson committed 5 days ago

Verified c264f65f

Make hip quality check run on all changes (#25403)

ORippler committed 5 days ago

Verified 07e012af

cuda : add support for f16->f16 GGML_OP_SET_ROWS (#25367)

fairydreaming committed 5 days ago

Verified ed8c2615

llama: refactor fused ops (#24646)

am17an committed 5 days ago

Verified 90e0f5cf

server-stream: follow-up on SSE Replay Buffer (#23226) (#25047)

ServeurpersoCom committed 5 days ago

Verified bbebeec4

llama-batch: add n_keep_tail in split_equal for recurrent models (#25278)

am17an committed 5 days ago

Verified 230ea9d2

common: auto-create prompts-log-dir at argument parsing, so all tools using the flag benefit (#25322)

rankaiyx committed 5 days ago

Verified f296fdfb

ui: Context usage gauge and panel (#25340)

allozaur committed 5 days ago

Verified f1161b15

llama-eval : fix crash when answer is None in HTML dump (#25435)

ggerganov committed 5 days ago

Verified da46e59c

metal : add set_rows with src0 f16 (#25434)

fairydreaming committed 5 days ago

Verified 0512ef1e

fix: OOB reads in UGM tokenizer (precompiled_charsmap handling) (#18750)

hourhl committed 6 days ago

Verified 4a7ee312

ggml : fix A indexing in simd_gemm scalar tail-column path (#25390)

tyronecai committed 6 days ago

Verified 57b50e1f

ggml : add support for CPU f16->f16 GGML_OP_SET_ROWS (#25344)

fairydreaming committed 6 days ago

Verified 68a521b5

opencl: fix potential crash in aos reconstruct (#25383)

lhez committed 6 days ago

Verified 931ca30b

Add Q2_0 quantization: type definition and CPU backend (#24448)

khosravipasha committed 6 days ago

Verified bec4772f

spec : fix naming, spacing (#25410)

ggerganov committed 6 days ago

Verified c198af4d

CUDA: Fuse MMVQ post-scale for NVFP4 (#24481)

ORippler committed 6 days ago

Verified 3899b39c

server : fix draft model fit vs load inconsistency (#25056)

wadealexc committed 6 days ago

Verified f5525f7e

server : add timings and progress to /responses API stream (#25348)

thom-dev-fr committed 6 days ago

Verified 5eca4e3c

server: enforce prompt cache RAM limit (#25070)

tarruda committed 6 days ago

Verified 6c487e2f

common : add missing <fstream> include in common.h (#25220)

zhangrunda committed 6 days ago

Verified c1a411fb

ggml-hip : add -fno-finite-math-only alongside -ffast-math (#25373)

asf0 committed 6 days ago

Verified 33ca0dcb

llama: fix quantized kv-cache for dsv4 (#25202)

am17an committed 6 days ago

Verified 024c46ae

Newer Older