Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on Oct 27, 2023

• cuda : improve text-generation and batched decoding performance (#3776)

  Show description for 2f9ec7e
  ggerganov

  authored
  2f9ec7e

  Copy full SHA for 2f9ec7e

Commits on Oct 23, 2023

• llama : remove token functions with `context` args in favor of `model` (#3720)

  Show description for 5be6c80
  

  MarcusDunn

  and

  ggerganov

  authored
  5be6c80

  Copy full SHA for 5be6c80

Commits on Oct 22, 2023

• Add test for MPT tokenization (#3728)

  Show description for 9e70cc0
  goerch

  authored
  9e70cc0

  Copy full SHA for 9e70cc0

• llama : validate special token ids are in range when loading GGUF model (#3635)

  Show description for a5e7dbd
  

  KerfuffleV2

  and

  ggerganov

  authored
  a5e7dbd

  Copy full SHA for a5e7dbd

Commits on Oct 20, 2023

• sampling : refactor init to use llama_sampling_params (#3696)

  Show description for d1031cf
  ggerganov

  authored
  d1031cf

  Copy full SHA for d1031cf

• ggml : fix rope + llama minor optimizations (#3560)

  Show description for f439e50
  GermanAizek

  authored
  f439e50

  Copy full SHA for f439e50

Commits on Oct 18, 2023

• speculative : add tree-based sampling example (#3624)

  Show description for 0e89203
  ggerganov

  authored
  0e89203

  Copy full SHA for 0e89203

Commits on Oct 17, 2023

• fix embeddings when using CUDA (#3657)

  slaren

  authored
  cb33f43

  Copy full SHA for cb33f43

• llama : avoid fprintf in favor of LLAMA_LOG (#3538)

  ggerganov

  committed
  e1675d1

  Copy full SHA for e1675d1

• tokenizer : special token handling (#3538)

  Show description for 1a15955
  

  staviq

  and

  ggerganov

  authored
  1a15955

  Copy full SHA for 1a15955

Commits on Oct 15, 2023

• MPT : support GQA for replit-code-v1.5 (#3627)

  cebtenzzre

  authored
  11bff29

  Copy full SHA for 11bff29

Commits on Oct 13, 2023

• llama : remove n_threads from llama_decode_internal (#3614)

  Show description for 2a4bcba
  danbev

  authored
  2a4bcba

  Copy full SHA for 2a4bcba

Commits on Oct 10, 2023

• Minor improvements in GPT2 tokenizer (#3567)

  Show description for 233fc1c
  goerch

  authored
  233fc1c

  Copy full SHA for 233fc1c

• llm : add bloom models (#3553)

  Show description for 02d2875
  

  xingchensong

  and

  ggerganov

  authored
  02d2875

  Copy full SHA for 02d2875

• llm : add MPT support (#3417)

  Show description for f5f9121
  

  3 peopleauthored
  f5f9121

  Copy full SHA for f5f9121

Commits on Oct 9, 2023

• refact : fix convert script + zero out KV cache to avoid nans (#3523)

  Show description for fcca0a7
  ggerganov

  authored
  fcca0a7

  Copy full SHA for fcca0a7

Commits on Oct 8, 2023

• sync : ggml (ggml-backend) (#3548)

  Show description for db3abcc
  ggerganov

  authored
  db3abcc

  Copy full SHA for db3abcc

• llama : fix missing break in Persimmon arch case statements (#3535)

  KerfuffleV2

  authored
  63d3b06

  Copy full SHA for 63d3b06

Commits on Oct 7, 2023

• quantize : fail fast on write errors (#3521)

  cebtenzzre

  authored
  f1782c6

  Copy full SHA for f1782c6

• llm : support Adept Persimmon 8B (#3410)

  Show description for 0e797c2
  phillip-kravtsov

  authored
  0e797c2

  Copy full SHA for 0e797c2

• Fix for #3454 (#3455)

  Show description for 3a716b4
  goerch

  authored
  3a716b4

  Copy full SHA for 3a716b4

Commits on Oct 6, 2023

• kv cache slot search improvements (#3493)

  Show description for 9ca79d5
  KerfuffleV2

  authored
  9ca79d5

  Copy full SHA for 9ca79d5

• parallel : add option to load external prompt file (#3416)

  Show description for a8777ad
  

  pudepiedj

  and

  ggerganov

  authored
  a8777ad

  Copy full SHA for a8777ad

• llama : correct hparams comparison (#3446)

  Show description for 16820a5
  

  l3utterfly

  and

  cebtenzzre

  authored
  16820a5

  Copy full SHA for 16820a5

Commits on Oct 4, 2023

• llm : add Refact model (#3329)

  Show description for f8c90cd
  

  ds5t5

  and

  ggerganov

  authored
  f8c90cd

  Copy full SHA for f8c90cd

Commits on Oct 3, 2023

• llama : fix session saving/loading (#3400)

  Show description for ac2219f
  ggerganov

  authored
  ac2219f

  Copy full SHA for ac2219f

• llama : expose model's rope_freq_scale in the API (#3418)

  Show description for 48be797
  grencez

  authored
  48be797

  Copy full SHA for 48be797

• Work on the BPE tokenizer (#3252)

  Show description for ff5a3f0
  

  goerch

  and

  ggerganov

  authored
  ff5a3f0

  Copy full SHA for ff5a3f0

Commits on Oct 2, 2023

• metal : set log callback before initializing (#3427)

  phronmophobic

  authored
  a847676

  Copy full SHA for a847676

• infill : add new example + extend server API (#3296)

  Show description for c97f01c
  

  vvhg1

  and

  cebtenzzre

  authored
  c97f01c

  Copy full SHA for c97f01c

Commits on Sep 29, 2023

• llama : quantize up to 31% faster on Linux and Windows with mmap (#3206)

  Show description for 2777a84
  

  cebtenzzre

  and

  ggerganov

  authored
  2777a84

  Copy full SHA for 2777a84

Commits on Sep 28, 2023

• build : enable more non-default compiler warnings (#3200)

  cebtenzzre

  authored
  bc39553

  Copy full SHA for bc39553

• llama.cpp : split llama_context_params into model and context params (#3301)

  Show description for 16bc66d
  slaren

  authored
  16bc66d

  Copy full SHA for 16bc66d

• train : finetune LORA (#2632)

  Show description for 0e76a89
  

  3 peopleauthored
  0e76a89

  Copy full SHA for 0e76a89

• gguf : make token scores and types optional (#3347)

  cebtenzzre

  authored
  ecf90b1

  Copy full SHA for ecf90b1