Code

merge upstream#44

Code

Merged

l3utterfly merged 51 commits intolayla-buildl3utterfly/llama.cpp:layla-buildfrom masterl3utterfly/llama.cpp:masterCopy head branch name to clipboardNov 5, 2024

+24,515-28,325

Commits

Commits on Oct 27, 2024

• llama : switch KQ multiplication to F32 precision by default (ggerganov#10015)

  Show description for 8841ce3
  ggerganov

  authored
  8841ce3

  Copy full SHA for 8841ce3

Commits on Oct 28, 2024

• server : don't overfill the batch during infill (ggerganov#10018)

  Show description for 8125e6c
  ggerganov

  authored
  8125e6c

  Copy full SHA for 8125e6c

• musa: workaround for Guilty Lockup in cleaning src0 (ggerganov#10042)

  Show description for 524afee
  yeahdongcn

  authored
  524afee

  Copy full SHA for 524afee

• flake.lock: Update (ggerganov#10063)

  Show description for 07028f9
  

  ggerganov

  and

  github-actions[bot]

  authored
  07028f9

  Copy full SHA for 07028f9

• llama : Add IBM granite template (ggerganov#10013)

  Show description for 61715d5
  

  arch-btw

  and

  ngxson

  authored
  61715d5

  Copy full SHA for 61715d5

Commits on Oct 29, 2024

• llama : remove Tail-Free sampling (ggerganov#10071)

  Show description for 8d8ff71
  ggerganov

  authored
  8d8ff71

  Copy full SHA for 8d8ff71

• ggml: Add POOL2D OP for GPU acceleration to the Vulkan backend in the MobileVLM model. (ggerganov#9763)

  Show description for 8f275a7
  cyzero-kim

  authored
  8f275a7

  Copy full SHA for 8f275a7

Commits on Oct 30, 2024

• llama : refactor model loader with backend registry (ggerganov#10026)

  slaren

  authored
  c5b0f4b

  Copy full SHA for c5b0f4b

• ggml : add Q4_0_8_8 RISC-V GEMV and GEMM kernels (ggerganov#10029)

  Show description for fc83a9e
  xctan

  authored
  fc83a9e

  Copy full SHA for fc83a9e

• convert : more detailed convert lora usage docs (ggerganov#10065)

  richdougherty

  authored
  79a2bc0

  Copy full SHA for 79a2bc0

• readme : more lora detail in main example readme (ggerganov#10064)

  richdougherty

  authored
  6763f71

  Copy full SHA for 6763f71

• ggml : fix memory leaks when loading invalid gguf files (ggerganov#10094)

  Show description for b9e02e8
  slaren

  authored
  b9e02e8

  Copy full SHA for b9e02e8

• kompute: add backend registry / device interfaces (ggerganov#10045)

  Show description for 61408e7
  slp

  authored
  61408e7

  Copy full SHA for 61408e7

Commits on Oct 31, 2024

• kompute: add mul_mat_q4_k shader (ggerganov#10097)

  Show description for 1329c0a
  slp

  authored
  1329c0a

  Copy full SHA for 1329c0a

• ggml : check tensor name lengths in gguf files (ggerganov#10100)

  slaren

  authored
  dea5e86

  Copy full SHA for dea5e86

• server : include scheme when printing URL (ggerganov#10106)

  bakkot

  authored
  0a683e8

  Copy full SHA for 0a683e8

• loader: refactor tensor weights storage (ggerganov#9935)

  Show description for ab3d71f
  

  kylo5aby

  and

  slaren

  authored
  ab3d71f

  Copy full SHA for ab3d71f

• llama : fix buffer checks for mamba and rwk (ggerganov#10111)

  Show description for c02e5ab
  slaren

  authored
  c02e5ab

  Copy full SHA for c02e5ab

• quantize : fix --keep-split (ggerganov#10114)

  slaren

  authored
  1e9f949

  Copy full SHA for 1e9f949

• llama : improve output buffer type selection (ggerganov#10098)

  slaren

  authored
  85679d3

  Copy full SHA for 85679d3

Commits on Nov 1, 2024

• build: fix build error in Windows env with OneAPI setup (ggerganov#10107)

  kylo5aby

  authored
  e597e50

  Copy full SHA for e597e50

• ggml : alloc ggml_contexts on the heap (whisper/2525)

  ggerganov

  committed
  f221d56

  Copy full SHA for f221d56

• sync : ggml

  ggerganov

  committed
  815fe72

  Copy full SHA for 815fe72

• ggml : remove ggml_scratch (ggerganov#10121)

  Show description for 1804adb
  ggerganov

  authored
  1804adb

  Copy full SHA for 1804adb

• server : fix smart selection of available slot (ggerganov#10120)

  Show description for d865d14
  sasha0552

  authored
  d865d14

  Copy full SHA for d865d14

• readme : update hot topics

  ggerganov

  authored
  ba6f62e

  Copy full SHA for ba6f62e

• vulkan : improve ggml_vk_create_buffer error handling (ggerganov#9898)

  FanShupei

  authored
  418f5ee

  Copy full SHA for 418f5ee

• llama : use smart pointers for ggml resources (ggerganov#10117)

  slaren

  authored
  e991e31

  Copy full SHA for e991e31

• llama : add simple-chat example (ggerganov#10124)

  Show description for a6744e4
  

  slaren

  and

  ngxson

  authored
  a6744e4

  Copy full SHA for a6744e4

Commits on Nov 2, 2024

• convert-lora : make --base optional (ggerganov#10110)

  Show description for 7554aa4
  ngxson

  authored
  7554aa4

  Copy full SHA for 7554aa4

• simple-chat : only add bos on first prompt (ggerganov#10129)

  slaren

  authored
  b634f8a

  Copy full SHA for b634f8a

• llama : adjust default context size + print warnings (ggerganov#10136)

  Show description for 1926d6e
  ggerganov

  authored
  1926d6e

  Copy full SHA for 1926d6e

• server : fix endpoint checks (ggerganov#10135)

  Show description for 4595041
  ggerganov

  authored
  4595041

  Copy full SHA for 4595041

• server : fix slot selection by lru (ggerganov#10126)

  Show description for 42cadc7
  sasha0552

  authored
  42cadc7

  Copy full SHA for 42cadc7

• Add apple arm to presets (ggerganov#10134)

  Show description for 9830b69
  kohnech

  authored
  9830b69

  Copy full SHA for 9830b69

Commits on Nov 3, 2024

• flake.lock: Update (ggerganov#10146)

  ggerganov

  authored
  1839f69

  Copy full SHA for 1839f69

• metal : minor fixup in FA kernel (ggerganov#10143)

  Show description for 08828a6
  ggerganov

  authored
  08828a6

  Copy full SHA for 08828a6

• ggml : move CPU backend to a separate file (ggerganov#10144)

  slaren

  authored
  9f40989

  Copy full SHA for 9f40989

Commits on Nov 4, 2024

• metal : fix minor string leaks (ggml/1004)

  

  pminev

  authored and

  ggerganov

  committed
  e2292aa

  Copy full SHA for e2292aa

• cmake : make it possible linking ggml as external lib (ggml/1003)

  

  ykhrustalev

  authored and

  ggerganov

  committed
  284e5b0

  Copy full SHA for 284e5b0

• sync : ggml

  ggerganov

  committed
  ce027ad

  Copy full SHA for ce027ad

• CANN: adjust backend registry refactor. (ggerganov#10158)

  Show description for 329ed91
  leo-pony

  authored
  329ed91

  Copy full SHA for 329ed91

• metal : move dequantize templates to beginning of MSL source (#0)

  ggerganov

  committed
  f8e5813

  Copy full SHA for f8e5813

• metal : simplify f16 and f32 dequant kernels (#0)

  ggerganov

  committed
  05697f6

  Copy full SHA for 05697f6

• cuda : clear error after changing peer access (ggerganov#10153)

  slaren

  authored
  ea02c75

  Copy full SHA for ea02c75

• fix build break on arm64 linux (ggerganov#10166)

  Show description for 6a066b9
  snadampal

  authored
  6a066b9

  Copy full SHA for 6a066b9

• server : clarify /slots endpoint, add is_processing (ggerganov#10162)

  Show description for 9e0ecfb
  ngxson

  authored
  9e0ecfb

  Copy full SHA for 9e0ecfb

• ggml : fix q4xx mat mul, increase ggml_aligned_malloc alignment (ggerganov#10167)

  slaren

  authored
  401558b

  Copy full SHA for 401558b

• ggml : fix gelu tables initialization (ggerganov#10172)

  slaren

  authored
  d5a409e

  Copy full SHA for d5a409e

• Q6_K AVX improvements (ggerganov#10118)

  Show description for 3407364
  netrunnereve

  authored
  3407364

  Copy full SHA for 3407364

• ggml : fix arch check in bf16_to_fp32 (ggerganov#10164)

  slaren

  authored
  a9e8a9a

  Copy full SHA for a9e8a9a