Skip to content

Latest commit

 

History

History

llama.cpp

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
imatrix
imatrix
 
 
llama-bench
llama-bench
 
 
llava
llava
 
 
main
main
 
 
perplexity
perplexity
 
 
quantize
quantize
 
 
server
server
 
 
BUILD.mk
BUILD.mk
 
 
LICENSE
LICENSE
 
 
README.llamafile
README.llamafile
 
 
base64.h
base64.h
 
 
build-info.cpp
build-info.cpp
 
 
common.cpp
common.cpp
 
 
common.h
common.h
 
 
console.cpp
console.cpp
 
 
console.h
console.h
 
 
cores.cpp
cores.cpp
 
 
cores.h
cores.h
 
 
ggml-aarch64.c
ggml-aarch64.c
 
 
ggml-aarch64.h
ggml-aarch64.h
 
 
ggml-alloc.c
ggml-alloc.c
 
 
ggml-alloc.h
ggml-alloc.h
 
 
ggml-backend-impl.h
ggml-backend-impl.h
 
 
ggml-backend.c
ggml-backend.c
 
 
ggml-backend.h
ggml-backend.h
 
 
ggml-common.h
ggml-common.h
 
 
ggml-cuda.cu
ggml-cuda.cu
 
 
ggml-cuda.h
ggml-cuda.h
 
 
ggml-impl.h
ggml-impl.h
 
 
ggml-metal.h
ggml-metal.h
 
 
ggml-metal.m
ggml-metal.m
 
 
ggml-metal.metal
ggml-metal.metal
 
 
ggml-quants-amd-avx.c
ggml-quants-amd-avx.c
 
 
ggml-quants-amd-avx2.c
ggml-quants-amd-avx2.c
 
 
ggml-quants-amd-avx512.c
ggml-quants-amd-avx512.c
 
 
ggml-quants-amd-avx512vl.c
ggml-quants-amd-avx512vl.c
 
 
ggml-quants-amd-k8.c
ggml-quants-amd-k8.c
 
 
ggml-quants-amd-ssse3.c
ggml-quants-amd-ssse3.c
 
 
ggml-quants-arm80.c
ggml-quants-arm80.c
 
 
ggml-quants.cpp
ggml-quants.cpp
 
 
ggml-quants.h
ggml-quants.h
 
 
ggml-quants.inc
ggml-quants.inc
 
 
ggml-quants.py
ggml-quants.py
 
 
ggml-vector-amd-avx.c
ggml-vector-amd-avx.c
 
 
ggml-vector-amd-avx2.c
ggml-vector-amd-avx2.c
 
 
ggml-vector-amd-avx512.c
ggml-vector-amd-avx512.c
 
 
ggml-vector-amd-avx512bf16.c
ggml-vector-amd-avx512bf16.c
 
 
ggml-vector-amd-avx512vl.c
ggml-vector-amd-avx512vl.c
 
 
ggml-vector-amd-f16c.c
ggml-vector-amd-f16c.c
 
 
ggml-vector-amd-fma.c
ggml-vector-amd-fma.c
 
 
ggml-vector-amd-k8.c
ggml-vector-amd-k8.c
 
 
ggml-vector-amd-ssse3.c
ggml-vector-amd-ssse3.c
 
 
ggml-vector-arm80.c
ggml-vector-arm80.c
 
 
ggml-vector-arm82.c
ggml-vector-arm82.c
 
 
ggml-vector.cpp
ggml-vector.cpp
 
 
ggml-vector.h
ggml-vector.h
 
 
ggml-vector.inc
ggml-vector.inc
 
 
ggml-vector.py
ggml-vector.py
 
 
ggml.c
ggml.c
 
 
ggml.h
ggml.h
 
 
grammar-parser.cpp
grammar-parser.cpp
 
 
grammar-parser.h
grammar-parser.h
 
 
json-schema-to-grammar.cpp
json-schema-to-grammar.cpp
 
 
json-schema-to-grammar.h
json-schema-to-grammar.h
 
 
json.h
json.h
 
 
llama-grammar.cpp
llama-grammar.cpp
 
 
llama-grammar.h
llama-grammar.h
 
 
llama-impl.h
llama-impl.h
 
 
llama-sampling.cpp
llama-sampling.cpp
 
 
llama-sampling.h
llama-sampling.h
 
 
llama-vocab.cpp
llama-vocab.cpp
 
 
llama-vocab.h
llama-vocab.h
 
 
llama.cpp
llama.cpp
 
 
llama.h
llama.h
 
 
llamafile.h
llamafile.h
 
 
log.h
log.h
 
 
ngram-cache.cpp
ngram-cache.cpp
 
 
ngram-cache.h
ngram-cache.h
 
 
sampling.cpp
sampling.cpp
 
 
sampling.h
sampling.h
 
 
string.h
string.h
 
 
unicode-data.cpp
unicode-data.cpp
 
 
unicode-data.h
unicode-data.h
 
 
unicode.cpp
unicode.cpp
 
 
unicode.h
unicode.h
 
 

README.llamafile

DESCRIPTION

  llama.cpp is a machine learning library for large language models

LICENSE

  MIT

ORIGIN

  https://github.com/ggerganov/llama.cpp/
  8b3befc0e2ed8fb18b903735831496b8b0c80949
  2024-08-16

LOCAL MODIFICATIONS

  - See [jart] and [kawrakow] annotations
  - Remove MAP_POPULATE because it makes mmap(tinyllama) block for 100ms
  - Refactor ggml.c, llama.cpp, and llava to use llamafile_open() APIs
  - Unify main, server, and llava-cli into single llamafile program
  - Make cuBLAS / hipBLAS optional by introducing tinyBLAS library
  - Add support to main() programs for Cosmo /zip/.args files
  - Introduce pledge() SECCOMP sandboxing to improve security
  - Call exit() rather than abort() when GGML_ASSERT() fails
  - Clamp bf16/f32 values before passing to K quantizers
  - Make GPU logger callback API safer and less generic
  - Write log to /dev/null when main.log fails to open
  - Make main and llava-cli print timings on ctrl-c
  - Make emebeddings CLI program shell scriptable
  - Avoid bind() conflicts on port 8080 w/ server
  - Use runtime dispatching for matmul quants
  - Remove operating system #ifdef statements
  - Remove stdout logging from LLaVA