Code

Sgemm ppc mma#2

Code

Closed

amritahs-ibm wants to merge 183 commits intomasteramritahs-ibm/llama.cpp:masterfrom sgemm_ppc_mmaamritahs-ibm/llama.cpp:sgemm_ppc_mmaCopy head branch name to clipboard

+25,853-15,421

Commits

Commits on Sep 23, 2024

• metal : use F32 prec for K*Q in vec FA (ggerganov#9595)

  Show description for bf9c101
  ggerganov

  authored
  bf9c101

  Copy full SHA for bf9c101

• perplexity : remove extra new lines after chunks (ggerganov#9596)

  ggerganov

  authored
  37f8c7b

  Copy full SHA for 37f8c7b

• ggml : AVX512 gemm for Q4_0_8_8 (ggerganov#9532)

  Show description for 1e7b929
  

  Srihari-mcw

  and

  ggerganov

  authored
  1e7b929

  Copy full SHA for 1e7b929

• flake.lock: Update (ggerganov#9586)

  Show description for f3979df
  

  ggerganov

  and

  github-actions[bot]

  authored
  f3979df

  Copy full SHA for f3979df

• readme : add programmable prompt engine language CLI (ggerganov#9599)

  snowyu

  authored
  1d48e98

  Copy full SHA for 1d48e98

• threads: improve ggml_barrier scaling with large number of threads (ggerganov#9598)

  Show description for f0c7b5e
  

  max-krasnyansky

  and

  wtarreau

  authored
  f0c7b5e

  Copy full SHA for f0c7b5e

• server : add --no-context-shift option (ggerganov#9607)

  Show description for 0b3bf96
  

  ngxson

  and

  ggerganov

  authored
  0b3bf96

  Copy full SHA for 0b3bf96

Commits on Sep 24, 2024

• cuda: add q8_0->f32 cpy operation (ggerganov#9571)

  Show description for 116efee
  Nekotekina

  authored
  116efee

  Copy full SHA for 116efee

• threads: fix msvc build without openmp (ggerganov#9615)

  Show description for c087b6f
  max-krasnyansky

  authored
  c087b6f

  Copy full SHA for c087b6f

• sampling : avoid expensive softmax during greedy sampling (ggerganov#9605)

  Show description for b0f2736
  

  ggerganov

  and

  slaren

  authored
  b0f2736

  Copy full SHA for b0f2736

• server : add newline after chat example (ggerganov#9616)

  StrangeBytesDev

  authored
  0aa1501

  Copy full SHA for 0aa1501

• log : add CONT level for continuing previous log entry (ggerganov#9610)

  ggerganov

  authored
  cea1486

  Copy full SHA for cea1486

• llama : keep track of all EOG tokens in the vocab (ggerganov#9609)

  Show description for 31ac583
  ggerganov

  authored
  31ac583

  Copy full SHA for 31ac583

• examples : adapt to ggml.h changes (ggml/0)

  Show description for c038931
  ggerganov

  committed
  c038931

  Copy full SHA for c038931

• sync : ggml

  ggerganov

  committed
  bb5f819

  Copy full SHA for bb5f819

• ggml : add AVX512DQ requirement for AVX512 builds (ggerganov#9622)

  EZForever

  authored
  70392f1

  Copy full SHA for 70392f1

Commits on Sep 25, 2024

• cann: fix crash when llama-bench is running on multiple cann devices (ggerganov#9627)

  bachelor-dou

  authored
  904837e

  Copy full SHA for 904837e

• llama : add IBM Granite MoE architecture (ggerganov#9438)

  Show description for 3d6bf69
  

  gabe-l-hart

  and

  ggerganov

  authored
  3d6bf69

  Copy full SHA for 3d6bf69

• server : add more env vars, improve gen-docs (ggerganov#9635)

  Show description for afbbfaa
  ngxson

  authored
  afbbfaa

  Copy full SHA for afbbfaa

• ggml : remove assert for AArch64 GEMV and GEMM Q4 kernels (ggerganov#9217)

  Show description for 1e43630
  chaxu01

  authored
  1e43630

  Copy full SHA for 1e43630

• ci : fix docker build number and tag name (ggerganov#9638)

  Show description for ea9c32b
  ngxson

  authored
  ea9c32b

  Copy full SHA for ea9c32b

Commits on Sep 26, 2024

• mtgpu: enable VMM (ggerganov#9597)

  Show description for 7691654
  yeahdongcn

  authored
  7691654

  Copy full SHA for 7691654

• [SYCL] add missed dll file in package (ggerganov#9577)

  Show description for 95bc82f
  

  NeoZhangJianyu

  and

  arthw

  authored
  95bc82f

  Copy full SHA for 95bc82f

Commits on Sep 27, 2024

• cmake : add option for common library (ggerganov#9661)

  iboB

  authored
  44f59b4

  Copy full SHA for 44f59b4

• readme : update hot topics

  ggerganov

  authored
  b5de3b7

  Copy full SHA for b5de3b7

Commits on Sep 28, 2024

• Enable use to the rebar feature to upload buffers to the device. (ggerganov#9251)

  mtavenrath

  authored
  89f9944

  Copy full SHA for 89f9944

• ggml : add run-time detection of neon, i8mm and sve (ggerganov#9331)

  Show description for 6a0f779
  eddnjjn

  authored
  6a0f779

  Copy full SHA for 6a0f779

• readme : add tool (ggerganov#9655)

  akx

  authored
  43bcdd9

  Copy full SHA for 43bcdd9

• llama : add support for Chameleon (ggerganov#8543)

  Show description for 9a91311
  

  nopperl

  and

  compilade

  authored
  9a91311

  Copy full SHA for 9a91311

• vocab : refactor tokenizer to reduce init overhead (ggerganov#9449)

  Show description for 6102037
  

  kylo5aby

  and

  ggerganov

  authored
  6102037

  Copy full SHA for 6102037

• llama : add comment about thread-safety [no ci] (ggerganov#9449)

  ggerganov

  committed
  7398427

  Copy full SHA for 7398427

• test-backend-ops : use flops for some performance tests (ggerganov#9657)

  Show description for 1b2f992
  slaren

  authored
  1b2f992

  Copy full SHA for 1b2f992

• llama : add reranking support (ggerganov#9510)

  Show description for f4d2b88
  

  ggerganov

  and

  ngxson

  authored
  f4d2b88

  Copy full SHA for f4d2b88

Commits on Sep 29, 2024

• contrib : add Resources section (ggerganov#9675)

  ggerganov

  authored
  589b48d

  Copy full SHA for 589b48d

• py : add model class for Chameleon conversion (ggerganov#9683)

  nopperl

  authored
  f99d3f8

  Copy full SHA for f99d3f8

• common : ensure llama_batch size does not exceed max size (ggerganov#9668)

  Show description for faac0ba
  matiaslin

  authored
  faac0ba

  Copy full SHA for faac0ba

• ggml : fix GGML_MAX_N_THREADS + improve formatting (ggml/969)

  ggerganov

  committed
  6084bfb

  Copy full SHA for 6084bfb

• vulkan : argsort barriers must be under uniform control flow (ggml/951)

  Show description for 544f409
  

  smeso

  authored and

  ggerganov

  committed
  544f409

  Copy full SHA for 544f409

• vulkan : fix build for GGML_VULKAN_RUN_TESTS, add TFLOPS to log (ggml/961)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  0de8b20

  Copy full SHA for 0de8b20

• vulkan : multithread pipeline creation (ggml/963)

  

  jeffbolznv

  authored and

  ggerganov

  committed
  641002f

  Copy full SHA for 641002f

• CUDA: remove bad assert (ggml/972)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  aaa4099

  Copy full SHA for aaa4099

• sync : ggml

  ggerganov

  committed
  d0b1d66

  Copy full SHA for d0b1d66

• ggml : define missing HWCAP flags (ggerganov#9684)

  Show description for c919d5d
  

  ggerganov

  and

  wtarreau

  authored
  c919d5d

  Copy full SHA for c919d5d

Commits on Sep 30, 2024

• console : utf-8 fix for windows stdin (ggerganov#9690)

  Show description for 8277a81
  

  hasaranga

  and

  ggerganov

  authored
  8277a81

  Copy full SHA for 8277a81

• flake.lock: Update (ggerganov#9680)

  Show description for ace4f4b
  

  ggerganov

  and

  github-actions[bot]

  authored
  ace4f4b

  Copy full SHA for ace4f4b

• py : update transfomers version (ggerganov#9694)

  Show description for 08a43d0
  Vaibhavs10

  authored
  08a43d0

  Copy full SHA for 08a43d0

• ci : reduce severity of unused Pyright ignore comments (ggerganov#9697)

  compilade

  authored
  511636d

  Copy full SHA for 511636d

• Fix Docker ROCM builds, use AMDGPU_TARGETS instead of GPU_TARGETS (ggerganov#9641)

  Show description for 6f1d9d7
  serhii-nakon

  authored
  6f1d9d7

  Copy full SHA for 6f1d9d7

Commits on Oct 1, 2024

• convert : refactor rope_freqs generation (ggerganov#9396)

  Show description for 1927378
  compilade

  authored
  1927378

  Copy full SHA for 1927378

• llama : print correct model type for Llama 3.2 1B and 3B

  ggerganov

  committed
  a90484c

  Copy full SHA for a90484c

• metal : reduce command encoding overhead (ggerganov#9698)

  Show description for cad341d
  ggerganov

  authored
  cad341d

  Copy full SHA for cad341d

• ggml: fix gradient allocation logic (ggml/966)

  Show description for 7254cdf
  

  JohannesGaessler

  authored and

  ggerganov

  committed
  7254cdf

  Copy full SHA for 7254cdf

• ggml : fix ggml_cast (ggml/973)

  

  iboB

  authored and

  ggerganov

  committed
  6c53224

  Copy full SHA for 6c53224

• vulkan : mul_mat: fix UB with small warps (ggml/952)

  Show description for cb00020
  

  smeso

  authored and

  ggerganov

  committed
  cb00020

  Copy full SHA for cb00020

• test: fix OPT_STEP_ADAMW for test-backend-ops (ggml/974)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  e98c1c1

  Copy full SHA for e98c1c1

• sync : ggml

  ggerganov

  committed
  f1b8c42

  Copy full SHA for f1b8c42

• Update README.md (ggerganov#9591)

  Show description for 3f1ae2e
  32bitmicro

  authored
  3f1ae2e

  Copy full SHA for 3f1ae2e

Commits on Oct 2, 2024

• examples : remove benchmark (ggerganov#9704)

  Show description for 148844f
  ggerganov

  authored
  148844f

  Copy full SHA for 148844f

• gguf-split : improve --split and --merge logic (ggerganov#9619)

  Show description for 76b37d1
  

  3 peopleauthored
  76b37d1

  Copy full SHA for 76b37d1

• vulkan : do not use tensor->extra (ggerganov#9407)

  Show description for 00b7317
  

  rgerganov

  and

  0cc4m

  authored
  00b7317

  Copy full SHA for 00b7317

• [SYCL] Initial cmake support of SYCL for AMD GPUs (ggerganov#9658)

  Show description for f536f4c
  Alcpz

  authored
  f536f4c

  Copy full SHA for f536f4c

• llama : reduce compile time and binary size (ggerganov#9712)

  Show description for a39ab21
  ngxson

  authored
  a39ab21

  Copy full SHA for a39ab21

• ggml-backend : add device and backend reg interfaces (ggerganov#9707)

  Show description for c83ad6d
  

  slaren

  and

  JohannesGaessler

  authored
  c83ad6d

  Copy full SHA for c83ad6d

Commits on Oct 3, 2024

• Fixed dequant precision issues in Q4_1 and Q5_1 (ggerganov#9711)

  OuadiElfarouki

  authored
  5639971

  Copy full SHA for 5639971

• rpc : enable vulkan (ggerganov#9714)

  Show description for 841713e
  rgerganov

  authored
  841713e

  Copy full SHA for 841713e

• convert : handle tokenizer merges format from transformers 4.45 (ggerganov#9696)

  compilade

  authored
  e3c355b

  Copy full SHA for e3c355b

• ggml: unify backend logging mechanism (ggerganov#9709)

  Show description for d6fe7ab
  bandoti

  authored
  d6fe7ab

  Copy full SHA for d6fe7ab

• ggml-backend : add device description to CPU backend (ggerganov#9720)

  slaren

  authored
  a7ad553

  Copy full SHA for a7ad553

• metal : fix compute pass descriptor autorelease crash (ggerganov#9718)

  jmousseau

  authored
  5d5ab1e

  Copy full SHA for 5d5ab1e

• ggml: refactor cross entropy loss CPU impl. (ggml/976)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  eee39bd

  Copy full SHA for eee39bd

• ggml/ex: calculate accuracy in graph, adapt MNIST (ggml/980)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  fabdc3b

  Copy full SHA for fabdc3b

• sync : ggml

  ggerganov

  committed
  1bb8a64

  Copy full SHA for 1bb8a64

• metal : remove abort (skip) (ggml/0)

  ggerganov

  committed
  d5ed2b9

  Copy full SHA for d5ed2b9

Commits on Oct 4, 2024

• Fixed RNG seed docs (ggerganov#9723)

  Show description for 133c7b4
  d-kleine

  authored
  133c7b4

  Copy full SHA for 133c7b4

• ci : fine-grant permission (ggerganov#9710)

  ngxson

  authored
  f3fdcfa

  Copy full SHA for f3fdcfa

• ggml : fixes after sync (ggml/983)

  Show description for ff56576
  

  slaren

  authored and

  ggerganov

  committed
  ff56576

  Copy full SHA for ff56576

• ggml : fix typo in example usage ggml_gallocr_new (ggml/984)

  

  danbev

  authored and

  ggerganov

  committed
  55951c0

  Copy full SHA for 55951c0

• sync : ggml

  ggerganov

  committed
  1788077

  Copy full SHA for 1788077

• Add Llama Assistant (ggerganov#9744)

  vietanhdev

  authored
  71967c2

  Copy full SHA for 71967c2

Commits on Oct 5, 2024

• metal : zero-init buffer contexts (whisper/0)

  ggerganov

  committed
  905f548

  Copy full SHA for 905f548

• sync : ggml

  ggerganov

  committed
  58b1669

  Copy full SHA for 58b1669

• rerank : use [SEP] token instead of [BOS] (ggerganov#9737)

  Show description for 8c475b9
  ggerganov

  authored
  8c475b9

  Copy full SHA for 8c475b9

Commits on Oct 6, 2024

• vulkan : retry allocation with fallback flags (whisper/2451)

  Show description for b0915d5
  

  2 peopleauthored and

  ggerganov

  committed
  b0915d5

  Copy full SHA for b0915d5

• sync : llama.cpp

  ggerganov

  committed
  b6d6c52

  Copy full SHA for b6d6c52

• readme : fix typo [no ci]

  ggerganov

  authored
  f4b2dcd

  Copy full SHA for f4b2dcd

• contrib : simplify + minor edits [no ci]

  ggerganov

  authored
  d5cb868

  Copy full SHA for d5cb868

Commits on Oct 7, 2024

• metal : single allocation of encode_async block (ggerganov#9747)

  Show description for 96b6912
  

  ptsochantaris

  and

  ggerganov

  authored
  96b6912

  Copy full SHA for 96b6912

• ggml : add metal backend registry / device (ggerganov#9713)

  Show description for d5ac8cf
  

  ggerganov

  and

  slaren

  authored
  d5ac8cf

  Copy full SHA for d5ac8cf

• flake.lock: Update (ggerganov#9753)

  Show description for 6279dac
  

  ggerganov

  and

  github-actions[bot]

  authored
  6279dac

  Copy full SHA for 6279dac

• Update building for Android (ggerganov#9672)

  Show description for f1af42f
  amqdn

  authored
  f1af42f

  Copy full SHA for f1af42f

• ggml : add backend registry / device interfaces to BLAS backend (ggerganov#9752)

  Show description for 6374743
  slaren

  authored
  6374743

  Copy full SHA for 6374743

Commits on Oct 8, 2024

• scripts : fix spelling typo in messages and comments (ggerganov#9782)

  Show description for fa42aa6
  standby24x7

  authored
  fa42aa6

  Copy full SHA for fa42aa6

• server : better security control for public deployments (ggerganov#9776)

  Show description for 458367a
  ngxson

  authored
  458367a

  Copy full SHA for 458367a

• ggml : fix BLAS with unsupported types (ggerganov#9775)

  Show description for dca1d4b
  slaren

  authored
  dca1d4b

  Copy full SHA for dca1d4b

Commits on Oct 9, 2024

• examples : remove llama.vim

  Show description for 3dc48fe
  ggerganov

  committed
  3dc48fe

  Copy full SHA for 3dc48fe

• perplexity : fix integer overflow (ggerganov#9783)

  Show description for e702206
  ggerganov

  authored
  e702206

  Copy full SHA for e702206

• cmake : do not build common library by default when standalone (ggerganov#9804)

  slaren

  authored
  c81f3bb

  Copy full SHA for c81f3bb

Commits on Oct 10, 2024

• examples : do not use common library in simple example (ggerganov#9803)

  Show description for c7499c5
  slaren

  authored
  c7499c5

  Copy full SHA for c7499c5

• musa: add docker image support (ggerganov#9685)

  Show description for cf8e0a3
  yeahdongcn

  authored
  cf8e0a3

  Copy full SHA for cf8e0a3

• rpc : add backend registry / device interfaces (ggerganov#9812)

  Show description for 0e9f760
  slaren

  authored
  0e9f760

  Copy full SHA for 0e9f760

• common : use common_ prefix for common library functions (ggerganov#9805)

  Show description for 7eee341
  

  slaren

  and

  ggerganov

  authored
  7eee341

  Copy full SHA for 7eee341

Commits on Oct 11, 2024

• ggml : move more prints to the ggml log system (ggerganov#9839)

  Show description for 9677640
  slaren

  authored
  9677640

  Copy full SHA for 9677640

Commits on Oct 12, 2024

• musa : update doc (ggerganov#9856)

  Show description for 943d20b
  yeahdongcn

  authored
  943d20b

  Copy full SHA for 943d20b

• llama : improve infill support and special token detection (ggerganov#9798)

  Show description for 11ac980
  ggerganov

  authored
  11ac980

  Copy full SHA for 11ac980

• server : remove legacy system_prompt feature (ggerganov#9857)

  Show description for 95c76e8
  ggerganov

  authored
  95c76e8

  Copy full SHA for 95c76e8

• server : remove self-extend features (ggerganov#9860)

  Show description for 1bde94d
  ggerganov

  authored
  1bde94d

  Copy full SHA for 1bde94d

• server : add option to time limit the generation phase (ggerganov#9865)

  Show description for edc2656
  ggerganov

  authored
  edc2656

  Copy full SHA for edc2656

Commits on Oct 13, 2024

• flake.lock: Update (ggerganov#9870)

  Show description for 92be9f1
  

  ggerganov

  and

  github-actions[bot]

  authored
  92be9f1

  Copy full SHA for 92be9f1

• server : reuse cached context chunks (ggerganov#9866)

  Show description for c7181bd
  ggerganov

  authored
  c7181bd

  Copy full SHA for c7181bd

• server : accept extra_context for the infill endpoint (ggerganov#9874)

  Show description for d4c19c0
  ggerganov

  authored
  d4c19c0

  Copy full SHA for d4c19c0

Commits on Oct 14, 2024

• Vectorize load instructions in dmmv f16 CUDA kernel (ggerganov#9816)

  Show description for 13dca2a
  

  agray3

  and

  JohannesGaessler

  authored
  13dca2a

  Copy full SHA for 13dca2a

• server : handle "logprobs" field with false value (ggerganov#9871)

  Show description for a89f75e
  

  VoidIsVoid

  and

  Gimling

  authored
  a89f75e

  Copy full SHA for a89f75e

Commits on Oct 15, 2024

• readme : update bindings list (ggerganov#9889)

  srgtuszy

  authored
  4c42f93

  Copy full SHA for 4c42f93

• server : update preact (ggerganov#9895)

  ggerganov

  authored
  dcdd535

  Copy full SHA for dcdd535

• sampling : add XTC sampler (ggerganov#9742)

  Show description for fbc98b7
  MaggotHATE

  authored
  fbc98b7

  Copy full SHA for fbc98b7

• server : improve infill context reuse (ggerganov#9894)

  Show description for 223c25a
  ggerganov

  authored
  223c25a

  Copy full SHA for 223c25a

• llama : add infill sampler (ggerganov#9896)

  Show description for 755a9b2
  ggerganov

  authored
  755a9b2

  Copy full SHA for 755a9b2

Commits on Oct 16, 2024

• [CANN] Fix cann compilation error (ggerganov#9891)

  Show description for becfd38
  leo-pony

  authored
  becfd38

  Copy full SHA for becfd38

• ggml-alloc : remove buffer_id from leaf_alloc (ggml/987)

  Show description for cd60b88
  

  danbev

  authored and

  ggerganov

  committed
  cd60b88

  Copy full SHA for cd60b88

• sync : ggml

  ggerganov

  committed
  0e41b30

  Copy full SHA for 0e41b30

• server : fix the disappearance of the end of the text (ggerganov#9867)

  Show description for 1f66b69
  z80maniac

  authored
  1f66b69

  Copy full SHA for 1f66b69

• llama : add tensor name for "result_norm" (ggerganov#9907)

  Show description for 10433e8
  MollySophia

  authored
  10433e8

  Copy full SHA for 10433e8

• grammar : fix JSON Schema for string regex with top-level alt. (ggerganov#9903)

  Show description for 66c2c93
  jemc

  authored
  66c2c93

  Copy full SHA for 66c2c93

• llava : fix typo in error message [no ci] (ggerganov#9884)

  danbev

  authored
  dbf18e4

  Copy full SHA for dbf18e4

• llama : suppress conversion from 'size_t' to 'int' (ggerganov#9046)

  Show description for 9e04102
  danbev

  authored
  9e04102

  Copy full SHA for 9e04102

• fix: use vm_allocate to allocate CPU backend buffer on macOS (ggerganov#9875)

  Show description for 73afe68
  giladgd

  authored
  73afe68

  Copy full SHA for 73afe68

• fix: allocating CPU buffer with size 0 (ggerganov#9917)

  giladgd

  authored
  2194200

  Copy full SHA for 2194200

Commits on Oct 17, 2024

• vulkan : add backend registry / device interfaces (ggerganov#9721)

  Show description for f010b77
  slaren

  authored
  f010b77

  Copy full SHA for f010b77

• readme : update bindings list (ggerganov#9918)

  Show description for 3752217
  

  ShenghaiWang

  and

  Tim Wang

  authored
  3752217

  Copy full SHA for 3752217

• llama : infill sampling handle very long tokens (ggerganov#9924)

  Show description for 99bd4ac
  ggerganov

  authored
  99bd4ac

  Copy full SHA for 99bd4ac

• llama : change warning to debug log

  ggerganov

  committed
  9f45fc1

  Copy full SHA for 9f45fc1

• readme : remove --memory-f32 references (ggerganov#9925)

  ggerganov

  committed
  17bb928

  Copy full SHA for 17bb928

• llama : rename batch_all to batch (ggerganov#8881)

  Show description for 6f55bcc
  danbev

  authored
  6f55bcc

  Copy full SHA for 6f55bcc

Commits on Oct 18, 2024

• server : add n_indent parameter for line indentation requirement (ggerganov#9929)

  Show description for 8901755
  ggerganov

  authored
  8901755

  Copy full SHA for 8901755

• add amx kernel for gemm (ggerganov#8998)

  Show description for 60ce97c
  mingfeima

  authored
  60ce97c

  Copy full SHA for 60ce97c

• [SYCL] Add SYCL Backend registry, device and Event Interfaces (ggerganov#9705)

  Show description for 87421a2
  OuadiElfarouki

  authored
  87421a2

  Copy full SHA for 87421a2

• rpc : backend refactoring (ggerganov#9912)

  Show description for afd9909
  rgerganov

  authored
  afd9909

  Copy full SHA for afd9909

• llama : remove all_pos_0, all_pos_1, all_seq_id from llama_batch (ggerganov#9745)

  Show description for cda0e4b
  ngxson

  authored
  cda0e4b

  Copy full SHA for cda0e4b

Commits on Oct 20, 2024

• readme : update infra list (ggerganov#9942)

  Show description for 7cab208
  icppWorld

  authored
  7cab208

  Copy full SHA for 7cab208

• readme : update bindings list (ggerganov#9951)

  Show description for 45f0976
  lcarrere

  authored
  45f0976

  Copy full SHA for 45f0976

Commits on Oct 21, 2024

• fix mul_mat_vec_q and *_vec_q error (ggerganov#9939)

  Show description for 1db8c84
  

  NeoZhangJianyu

  and

  arthw

  authored
  1db8c84

  Copy full SHA for 1db8c84

• speculative : fix handling of some input params (ggerganov#9963)

  Show description for bc21975
  ggerganov

  authored
  bc21975

  Copy full SHA for bc21975

• llama : default sampling changes + greedy update (ggerganov#9897)

  Show description for 55e4778
  ggerganov

  authored
  55e4778

  Copy full SHA for 55e4778

• rpc : pack only RPC structs (ggerganov#9959)

  rgerganov

  authored
  d5ebd79

  Copy full SHA for d5ebd79

• ggml : add asserts for type conversion in fattn kernels (ggerganov#9971)

  Show description for f594bc8
  ggerganov

  authored
  f594bc8

  Copy full SHA for f594bc8

• llama.vim : plugin for Neovim (ggerganov#9787)

  ggerganov

  authored
  dbd5f2f

  Copy full SHA for dbd5f2f

• arg : fix attention non-causal arg value hint (ggerganov#9985)

  Show description for 94008cc
  danbev

  authored
  94008cc

  Copy full SHA for 94008cc

• readme : update UI list (ggerganov#9972)

  Show description for 994cfb1
  a-ghorbani

  authored
  994cfb1

  Copy full SHA for 994cfb1

• llama.vim : move info to the right of screen [no ci] (ggerganov#9787)

  Show description for e01c67a
  ggerganov

  committed
  e01c67a

  Copy full SHA for e01c67a

• llama.vim : fix info text display [no ci] (ggerganov#9787)

  ggerganov

  committed
  e94a138

  Copy full SHA for e94a138

Commits on Oct 22, 2024

• arg : fix typo in embeddings argument help [no ci] (ggerganov#9994)

  Show description for 674804a
  danbev

  authored
  674804a

  Copy full SHA for 674804a

• [CANN] Adapt to dynamically loadable backends mechanism (ggerganov#9970)

  Show description for 6b84473
  leo-pony

  authored
  6b84473

  Copy full SHA for 6b84473

• llama : add chat template for RWKV-World + fix EOT (ggerganov#9968)

  Show description for 4ff7fe1
  MollySophia

  authored
  4ff7fe1

  Copy full SHA for 4ff7fe1

• lora : warn user if new token is added in the adapter (ggerganov#9948)

  ngxson

  authored
  c421ac0

  Copy full SHA for c421ac0

• Rwkv chat template fix (ggerganov#10001)

  Show description for 11d4705
  

  MollySophia

  and

  ngxson

  authored
  11d4705

  Copy full SHA for 11d4705

• llama : rename batch to ubatch (ggerganov#9950)

  Show description for 19d900a
  danbev

  authored
  19d900a

  Copy full SHA for 19d900a

• llama : fix empty batch causing llama_batch_allocr to crash (ggerganov#9966)

  Show description for c8c07d6
  

  ngxson

  and

  ggerganov

  authored
  c8c07d6

  Copy full SHA for c8c07d6

Commits on Oct 23, 2024

• flake.lock: Update

  Show description for 873279b
  

  github-actions[bot]

  authored and

  SomeoneSerge

  committed
  873279b

  Copy full SHA for 873279b

• metal : add POOL2D and fix IM2COL (ggerganov#9943)

  Show description for 4c9388f
  junhee-yoo

  authored
  4c9388f

  Copy full SHA for 4c9388f

• llama.vim : add classic vim support (ggerganov#9995)

  Show description for ac113a0
  

  m18coppola

  and

  Michael Coppola

  authored
  ac113a0

  Copy full SHA for ac113a0

• ggml : remove redundant set of contexts used field (ggml/978)

  Show description for c19af0a
  

  danbev

  authored and

  ggerganov

  committed
  c19af0a

  Copy full SHA for c19af0a

• CUDA: fix 1D im2col, add tests (ggml/993)

  

  JohannesGaessler

  authored and

  ggerganov

  committed
  80273a3

  Copy full SHA for 80273a3

• llama.vim : bump generation time limit to 3s [no ci]

  ggerganov

  committed
  2d3aba9

  Copy full SHA for 2d3aba9

• sync : ggml

  ggerganov

  committed
  190a37d

  Copy full SHA for 190a37d

• server : samplers accept the prompt correctly (ggerganov#10019)

  wwoodsTM

  authored
  0a1c750

  Copy full SHA for 0a1c750

Commits on Oct 24, 2024

• CUDA: fix MMQ for non-contiguous src0, add tests (ggerganov#10021)

  Show description for c39665f
  JohannesGaessler

  authored
  c39665f

  Copy full SHA for c39665f

• CUDA: fix insufficient buffer clearing for MMQ (ggerganov#10032)

  JohannesGaessler

  authored
  167a515

  Copy full SHA for 167a515

• ci : fix cmake flags for SYCL

  ggerganov

  committed
  40f2555

  Copy full SHA for 40f2555

• server : refactor slot input data, move tokenizer to HTTP thread (ggerganov#10023)

  Show description for 958367b
  ngxson

  authored
  958367b

  Copy full SHA for 958367b

Commits on Oct 25, 2024

• server : check that the prompt fits in the slot's context (ggerganov#10030)

  Show description for bc5ba00
  ggerganov

  authored
  bc5ba00

  Copy full SHA for bc5ba00

• llamafile : extend sgemm.cpp support for Q5_0 models (ggerganov#10010)

  Srihari-mcw

  authored
  2f8bd2b

  Copy full SHA for 2f8bd2b

• llama: string_split fix (ggerganov#10022)

  Show description for d80fb71
  Xarbirus

  authored
  d80fb71

  Copy full SHA for d80fb71

• llama : add DRY sampler (ggerganov#9702)

  Show description for ff252ea
  

  3 peopleauthored
  ff252ea

  Copy full SHA for ff252ea

• metal : support permuted matrix multiplicaions (ggerganov#10033)

  Show description for 6687503
  ggerganov

  authored
  6687503

  Copy full SHA for 6687503

Commits on Oct 26, 2024

• scripts : fix amx sync [no ci]

  ggerganov

  committed
  9e4a256

  Copy full SHA for 9e4a256

• increase cuda_cpy block size (ggml/996)

  Show description for 8c60a8a
  

  2 peopleauthored and

  ggerganov

  committed
  8c60a8a

  Copy full SHA for 8c60a8a

• sync : ggml

  ggerganov

  committed
  cc2983d

  Copy full SHA for cc2983d

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

• PPC MMA implementation

  Show description for 46dcd2b
  amritahs-ibm

  committed
  46dcd2b

  Copy full SHA for 46dcd2b

• Merge branch 'ggerganov:master' into sgemm_ppc_mma

  amritahs-ibm

  authored
  4a9a3c8

  Copy full SHA for 4a9a3c8