Code

master b1794#70

Code

Merged

Nexesenex merged 64 commits intoNexesenex:masterNexesenex/croco.cpp:masterfrom ggerganov:masterggerganov/llama.cpp:masterCopy head branch name to clipboardJan 8, 2024

+7,697-5,360

Commits

Commits on Dec 30, 2023

• clip : refactor + bug fixes (#4696)

  Show description for 9fbda71
  ggerganov

  authored
  9fbda71

  Copy full SHA for 9fbda71

Commits on Dec 31, 2023

• ggml : add ggml_vdotq_s32 alias (#4715)

  Show description for e39106c
  ggerganov

  authored
  e39106c

  Copy full SHA for e39106c

• flake.nix: expose full scope in legacyPackages

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  1e3900e

  Copy full SHA for 1e3900e

• flake.nix: rocm not yet supported on aarch64, so hide the output

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  a5c088d

  Copy full SHA for a5c088d

• flake.nix: expose checks

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  356ea17

  Copy full SHA for 356ea17

• workflows: nix-ci: init; build flake outputs

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  7adedec

  Copy full SHA for 7adedec

• workflows: nix-ci: add a job for eval

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  1e9ae54

  Copy full SHA for 1e9ae54

• workflows: weekly nix flake update

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  c523994

  Copy full SHA for c523994

• workflows: nix-flakestry: drop tag filters

  Show description for 06f2a5d
  

  SomeoneSerge

  authored and

  philiptaron

  committed
  06f2a5d

  Copy full SHA for 06f2a5d

• workflows: nix-ci: add a qemu job for jetsons

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  d836174

  Copy full SHA for d836174

• flake.nix: suggest the binary caches

  

  SomeoneSerge

  authored and

  philiptaron

  committed
  198ed7e

  Copy full SHA for 198ed7e

• flake.lock: update

  Show description for edd1ab7
  

  SomeoneSerge

  authored and

  philiptaron

  committed
  edd1ab7

  Copy full SHA for edd1ab7

Commits on Jan 2, 2024

• metal : enable shader debugging (cmake option) (#4705)

  Show description for 58ba655
  ggerganov

  authored
  58ba655

  Copy full SHA for 58ba655

• finetune: fix typo in README.md (#4733)

  Show description for 775ac87
  danbev

  authored
  775ac87

  Copy full SHA for 775ac87

• py : re-enable mmap in convert hf (#4732)

  Show description for 26f3071
  

  4 peopleauthored
  26f3071

  Copy full SHA for 26f3071

• server : add --override-kv parameter (#4710)

  Show description for 5d7002d
  

  3 peopleauthored
  5d7002d

  Copy full SHA for 5d7002d

• editorconfig : fix whitespace and indentation #4710

  ggerganov

  committed
  32866c5

  Copy full SHA for 32866c5

• llama : differentiate the KV dims in the attention (#4657)

  Show description for 83e633c
  

  postmasters

  and

  ggerganov

  authored
  83e633c

  Copy full SHA for 83e633c

• llama : replace all API facing int's with int32_t (#4577)

  Show description for 0040d42
  MarcusDunn

  authored
  0040d42

  Copy full SHA for 0040d42

• llama : llama_model_desc print number of experts

  ggerganov

  committed
  540938f

  Copy full SHA for 540938f

• server : add token counts to html footer (#4738)

  Show description for 0ef3ca2
  phiharri

  authored
  0ef3ca2

  Copy full SHA for 0ef3ca2

• metal : optimize ggml_mul_mat_id (faster Mixtral PP) (#4725)

  Show description for f3f62f0
  ggerganov

  authored
  f3f62f0

  Copy full SHA for f3f62f0

Commits on Jan 3, 2024

• server : throw an error when slot unavailable (#4741)

  jparkerweb

  authored
  f2eb19b

  Copy full SHA for f2eb19b

• ggml : extend ggml_get_rows, ggml_repeat, ggml_concat (ggml/639)

  Show description for 5f66ebc
  

  gwenzek

  and

  ggerganov

  committed
  5f66ebc

  Copy full SHA for 5f66ebc

• scripts : fix sync order + metal sed

  ggerganov

  committed
  ab62fc3

  Copy full SHA for ab62fc3

• metal : add kernel_get_rows_i32

  Show description for 2893137
  ggerganov

  committed
  2893137

  Copy full SHA for 2893137

• sync : ggml

  Show description for 75e3fd8
  ggerganov

  committed
  75e3fd8

  Copy full SHA for 75e3fd8

• cuda : mark I16 and I32 ops as unsupported

  Show description for d55356d
  ggerganov

  committed
  d55356d

  Copy full SHA for d55356d

• cuda : simplify expression

  Show description for 7bed7eb
  

  ggerganov

  and

  slaren

  committed
  7bed7eb

  Copy full SHA for 7bed7eb

• swift : update Package.swift to use ggml as dependency (#4691)

  Show description for ece9a45
  1-ashraful-islam

  authored
  ece9a45

  Copy full SHA for ece9a45

• train : fix typo in overlapping-samples help msg (#4758)

  Show description for cb1e281
  danbev

  authored
  cb1e281

  Copy full SHA for cb1e281

Commits on Jan 4, 2024

• llama.swiftui : fix build of ggml.metallib (#4754)

  Show description for 46cea79
  

  singularity-s0

  and

  ggerganov

  authored
  46cea79

  Copy full SHA for 46cea79

• ggml : include stdlib.h before intrin.h (#4736)

  ggerganov

  authored
  dc891b7

  Copy full SHA for dc891b7

• server : fix options in README.md (#4765)

  Show description for e580431
  

  m18coppola

  and

  ggerganov

  authored
  e580431

  Copy full SHA for e580431

• llama.swiftui : support loading custom model from file picker (#4767)

  Show description for 3c0b585
  singularity-s0

  authored
  3c0b585

  Copy full SHA for 3c0b585

• Print backend name on test-backend-ops failure (#4751)

  JohannesGaessler

  authored
  a919280

  Copy full SHA for a919280

• server : send token probs for "stream == false" (#4714)

  ggerganov

  authored
  012cf34

  Copy full SHA for 012cf34

• finetune : remove unused includes (#4756)

  Show description for b3a7c20
  danbev

  authored
  b3a7c20

  Copy full SHA for b3a7c20

Commits on Jan 5, 2024

• examples : add few-shot translation example (#4783)

  ggerganov

  authored
  3681f22

  Copy full SHA for 3681f22

• ggml : do not sched_yield when calling BLAS (#4761)

  Show description for c1d7cb2
  ggerganov

  authored
  c1d7cb2

  Copy full SHA for c1d7cb2

• ggml : add error handling to graph_compute (whisper/1714)

  

  finnvoor

  authored and

  ggerganov

  committed
  1bf681f

  Copy full SHA for 1bf681f

• ggml : fix q2_k bpw in comments (ggml/680)

  ggerganov

  committed
  d061bf9

  Copy full SHA for d061bf9

• metal : switch back to default.metallib (ggml/681)

  Show description for 91d3887
  ggerganov

  committed
  91d3887

  Copy full SHA for 91d3887

• flake.nix : fix typo (#4700)

  Show description for be36bb9
  eltociear

  authored
  be36bb9

  Copy full SHA for be36bb9

• cmake : check for openblas64 (#4134)

  Show description for eec22a1
  

  a-n-n-a-l-e-e

  authored
  eec22a1

  Copy full SHA for eec22a1

Commits on Jan 6, 2024

• examples : improve base-translate.sh script (#4783)

  ggerganov

  committed
  96e80da

  Copy full SHA for 96e80da

• llama.swiftui : use correct pointer for llama_token_eos (#4797)

  illescasDaniel

  authored
  c75ca5d

  Copy full SHA for c75ca5d

Commits on Jan 7, 2024

• server : fix n_predict check (#4798)

  ggerganov

  authored
  6798492

  Copy full SHA for 6798492

• ggml : use __builtin_amdgcn_sudot4 in __dp4a for gfx11 (#4787)

  kzhuravl

  authored
  63ee677

  Copy full SHA for 63ee677

• llama.swiftui : add visionOS target (#4805)

  azarovalex

  authored
  3418c03

  Copy full SHA for 3418c03

• llama : print tensor meta for debugging

  ggerganov

  committed
  d117d4d

  Copy full SHA for d117d4d

• llama.swiftui : use llama.cpp as SPM package (#4804)

  azarovalex

  authored
  72d8407

  Copy full SHA for 72d8407

• llama : remove redundant GQA check (#4796)

  ggerganov

  authored
  3c36213

  Copy full SHA for 3c36213

• llama : remove unused vars (#4796)

  ggerganov

  committed
  9dede37

  Copy full SHA for 9dede37

• CUDA: fixed redundant value dequantization (#4809)

  JohannesGaessler

  authored
  d5a410e

  Copy full SHA for d5a410e

• llama-bench : add no-kv-offload parameter (#4812)

  slaren

  authored
  226460c

  Copy full SHA for 226460c

• readme : add lgrammel/modelfusion JS/TS client for llama.cpp (#4814)

  lgrammel

  authored
  b7e7982

  Copy full SHA for b7e7982

Commits on Jan 8, 2024

• examples : add passkey test (#3856)

  Show description for b0034d9
  ggerganov

  authored
  b0034d9

  Copy full SHA for b0034d9

• main : add self-extend support (#4815)

  Show description for 52531fd
  ggerganov

  authored
  52531fd

  Copy full SHA for 52531fd

• llama.swiftui : update readme

  ggerganov

  authored
  42ea63c

  Copy full SHA for 42ea63c

• swift : exclude ggml-metal.metal from the package (#4822)

  ggerganov

  authored
  668b31f

  Copy full SHA for 668b31f

• SOTA 2-bit quants (#4773)

  Show description for dd5ae06
  

  ikawrakow

  and

  Kawrakow

  authored
  dd5ae06

  Copy full SHA for dd5ae06

• readme : add link to SOTA models

  ggerganov

  authored
  a9a8c5d

  Copy full SHA for a9a8c5d

• common : fix the short form of --grp-attn-w, not -gat (#4825)

  Show description for 1fc2f26
  howlger

  authored
  1fc2f26

  Copy full SHA for 1fc2f26