Datasets:

Languages:
code
ArXiv:
License:

The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Octopack

Dataset Card for CommitPackFT

Dataset Summary

CommitPackFT is a 2GB filtered version of CommitPack to contain only high-quality commit messages that resemble natural language instructions.

  • Creation: The dataset can be recreated using instructions available here.
  • Languages: 277
  • OctoPack🐙🎒:
Data CommitPack 4TB of GitHub commits across 350 programming languages
CommitPackFT Filtered version of CommitPack for high-quality commit messages that resemble instructions
Model OctoCoder StarCoder (16B parameters) instruction tuned on CommitPackFT + OASST
OctoGeeX CodeGeeX2 (6B parameters) instruction tuned on CommitPackFT + OASST
Evaluation   HumanEvalPack Extension of OpenAI's HumanEval to cover 3 scenarios across 6 languages

Dataset Structure

Data Instances

An example looks as follows:

{
  'commit': '0c17311f7fd511f5dae8f8e4acc2dce1a2de3cf5',
  'old_file': 'main.py',
  'new_file': 'main.py',
  'old_contents': "import numpy as np\nimport matplotlib.pyplot as plt\n\n# generate sample data\nx_data = np.linspace(-5, 5, 20)\ny_data = np.random.normal(0.0, 1.0, x_data.size)\n\nplt.plot(x_data, y_data, 'o')\nplt.show()\n",
  'new_contents': "import math\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# generate sample data\nx_data = np.linspace(-math.pi, math.pi, 30)\ny_data = np.sin(x_data) + np.random.normal(0.0, 0.1, x_data.size)\n\nplt.plot(x_data, y_data, 'o')\nplt.show()\n\n",
  'subject': 'Change to sin() function with noise',
  'message': 'Change to sin() function with noise\n',
  'lang': 'Python',
  'license': 'mit',
  'repos': 'MorganR/basic-gaussian-process'
}

Data Fields

The data fields are the same among all splits:

  • commit: unique commit id
  • old_file: name of the file before the commit
  • new_file: name of the file after the commit
  • old_contents: contents of the file before the commit
  • new_contents: contents of the file after the commit
  • subject: subject of the commit (this is used for all experiments in the paper)
  • message: message of the commit (commonly the same as the subject)
  • lang: programming language
  • license: license of the repository the code stems from, one of ['mit', 'artistic-2.0', 'isc', 'cc0-1.0', 'epl-1.0', 'mpl-2.0', 'unlicense', 'unknown', 'apache-2.0', 'bsd-3-clause', 'agpl-3.0', 'lgpl-2.1', 'bsd-2-clause']
  • repos: name of the the repository the code stems from (if multiple, they are comma separated)

Data Splits

Name Megabytes % of total Samples % of total
total 1545.02 100.0% 702062 100.0%
ruby 195.292 12.6401% 69413 9.887%
yaml 190.876 12.3543% 114320 16.2835%
python 132.68 8.5876% 56025 7.9801%
markdown 131.152 8.4887% 62518 8.9049%
javascript 125.008 8.091% 52989 7.5476%
json 86.744 5.6144% 39777 5.6657%
shell 66.864 4.3277% 31217 4.4465%
text 66.664 4.3148% 46588 6.6359%
php 60.22 3.8977% 24791 3.5312%
java 56.284 3.6429% 20635 2.9392%
html 48.42 3.1339% 20214 2.8792%
c# 26.84 1.7372% 9346 1.3312%
xml 23.676 1.5324% 9337 1.3299%
html+erb 23.104 1.4954% 10910 1.554%
c 21.08 1.3644% 8506 1.2116%
ini 21.04 1.3618% 11360 1.6181%
coffeescript 16.96 1.0977% 5513 0.7853%
swift 16.272 1.0532% 4849 0.6907%
restructuredtext 15.728 1.018% 6560 0.9344%
typescript 14.284 0.9245% 5868 0.8358%
c++ 14.136 0.9149% 4992 0.711%
scss 13.208 0.8549% 6829 0.9727%
go 12.132 0.7852% 5004 0.7128%
scala 11.184 0.7239% 5040 0.7179%
haml 10.74 0.6951% 4415 0.6289%
css 9.364 0.6061% 5049 0.7192%
rust 7.244 0.4689% 2996 0.4267%
toml 5.584 0.3614% 3424 0.4877%
jsx 5.5 0.356% 2199 0.3132%
kotlin 5.368 0.3474% 2214 0.3154%
clojure 5.068 0.328% 2403 0.3423%
perl 4.988 0.3228% 2288 0.3259%
bitbake 4.464 0.2889% 1308 0.1863%
groovy 4.168 0.2698% 1486 0.2117%
twig 3.956 0.256% 1610 0.2293%
nix 3.84 0.2485% 1593 0.2269%
sql 3.74 0.2421% 2069 0.2947%
less 3.724 0.241% 1360 0.1937%
haskell 3.308 0.2141% 1389 0.1978%
handlebars 3.292 0.2131% 1429 0.2035%
unknown 3.048 0.1973% 1597 0.2275%
batchfile 2.984 0.1931% 1466 0.2088%
cucumber 2.588 0.1675% 976 0.139%
makefile 2.528 0.1636% 960 0.1367%
elixir 2.348 0.152% 1150 0.1638%
jade 2.348 0.152% 1119 0.1594%
cmake 2.268 0.1468% 981 0.1397%
powershell 2.064 0.1336% 991 0.1412%
slim 2.056 0.1331% 1052 0.1498%
emacs-lisp 1.972 0.1276% 1015 0.1446%
dart 1.96 0.1269% 765 0.109%
viml 1.956 0.1266% 1063 0.1514%
asciidoc 1.864 0.1206% 523 0.0745%
lua 1.852 0.1199% 920 0.131%
llvm 1.6 0.1036% 780 0.1111%
smarty 1.588 0.1028% 737 0.105%
diff 1.48 0.0958% 680 0.0969%
common-lisp 1.448 0.0937% 778 0.1108%
saltstack 1.412 0.0914% 617 0.0879%
vue 1.384 0.0896% 587 0.0836%
sass 1.364 0.0883% 705 0.1004%
fish 1.328 0.086% 813 0.1158%
erlang 1.192 0.0772% 480 0.0684%
freemarker 1.028 0.0665% 510 0.0726%
stylus 0.948 0.0614% 480 0.0684%
qml 0.936 0.0606% 368 0.0524%
hcl 0.912 0.059% 421 0.06%
html+django 0.848 0.0549% 399 0.0568%
mako 0.756 0.0489% 170 0.0242%
ada 0.728 0.0471% 265 0.0377%
ocaml 0.704 0.0456% 333 0.0474%
f# 0.656 0.0425% 254 0.0362%
elm 0.62 0.0401% 265 0.0377%
tex 0.564 0.0365% 307 0.0437%
rdoc 0.552 0.0357% 270 0.0385%
csv 0.532 0.0344% 375 0.0534%
protocol-buffer 0.524 0.0339% 181 0.0258%
smalltalk 0.46 0.0298% 284 0.0405%
arduino 0.456 0.0295% 225 0.032%
java-server-pages 0.452 0.0293% 173 0.0246%
scheme 0.42 0.0272% 213 0.0303%
groff 0.396 0.0256% 192 0.0273%
objective-c++ 0.376 0.0243% 86 0.0122%
desktop 0.364 0.0236% 186 0.0265%
factor 0.356 0.023% 113 0.0161%
crystal 0.348 0.0225% 182 0.0259%
rhtml 0.348 0.0225% 135 0.0192%
haxe 0.344 0.0223% 174 0.0248%
glsl 0.34 0.022% 164 0.0234%
gas 0.336 0.0217% 193 0.0275%
html+php 0.332 0.0215% 150 0.0214%
qmake 0.32 0.0207% 140 0.0199%
julia 0.312 0.0202% 180 0.0256%
cython 0.308 0.0199% 123 0.0175%
html+eex 0.292 0.0189% 135 0.0192%
tcl 0.292 0.0189% 103 0.0147%
org 0.272 0.0176% 136 0.0194%
perl6 0.268 0.0173% 122 0.0174%
m4 0.264 0.0171% 101 0.0144%
xslt 0.256 0.0166% 99 0.0141%
svg 0.252 0.0163% 169 0.0241%
nimrod 0.236 0.0153% 67 0.0095%
r 0.228 0.0148% 121 0.0172%
robotframework 0.212 0.0137% 85 0.0121%
racket 0.196 0.0127% 117 0.0167%
textile 0.184 0.0119% 61 0.0087%
assembly 0.172 0.0111% 105 0.015%
purescript 0.172 0.0111% 80 0.0114%
unity3d-asset 0.156 0.0101% 101 0.0144%
visual-basic 0.152 0.0098% 48 0.0068%
dm 0.148 0.0096% 16 0.0023%
pod 0.148 0.0096% 54 0.0077%
standard-ml 0.148 0.0096% 72 0.0103%
fortran 0.144 0.0093% 70 0.01%
gettext-catalog 0.132 0.0085% 72 0.0103%
idris 0.132 0.0085% 38 0.0054%
livescript 0.128 0.0083% 63 0.009%
xtend 0.128 0.0083% 55 0.0078%
actionscript 0.12 0.0078% 49 0.007%
vala 0.116 0.0075% 50 0.0071%
awk 0.104 0.0067% 52 0.0074%
ceylon 0.1 0.0065% 49 0.007%
jupyter-notebook 0.1 0.0065% 48 0.0068%
dockerfile 0.096 0.0062% 39 0.0056%
rouge 0.096 0.0062% 41 0.0058%
asp 0.092 0.006% 22 0.0031%
sqf 0.092 0.006% 45 0.0064%
edn 0.088 0.0057% 48 0.0068%
liquid 0.088 0.0057% 30 0.0043%
xquery 0.084 0.0054% 39 0.0056%
linker-script 0.08 0.0052% 37 0.0053%
mediawiki 0.08 0.0052% 33 0.0047%
parrot-internal-representation 0.08 0.0052% 23 0.0033%
solidity 0.08 0.0052% 37 0.0053%
json5 0.076 0.0049% 33 0.0047%
systemverilog 0.076 0.0049% 35 0.005%
thrift 0.076 0.0049% 28 0.004%
groovy-server-pages 0.072 0.0047% 25 0.0036%
processing 0.072 0.0047% 35 0.005%
cuda 0.068 0.0044% 25 0.0036%
graphviz-dot 0.068 0.0044% 35 0.005%
inno-setup 0.064 0.0041% 16 0.0023%
api-blueprint 0.06 0.0039% 23 0.0033%
nsis 0.06 0.0039% 15 0.0021%
gentoo-ebuild 0.056 0.0036% 16 0.0023%
logtalk 0.056 0.0036% 21 0.003%
jasmin 0.052 0.0034% 9 0.0013%
literate-coffeescript 0.052 0.0034% 19 0.0027%
webidl 0.052 0.0034% 6 0.0009%
coldfusion-cfc 0.048 0.0031% 20 0.0028%
opencl 0.048 0.0031% 23 0.0033%
openscad 0.048 0.0031% 21 0.003%
pan 0.048 0.0031% 23 0.0033%
pascal 0.048 0.0031% 25 0.0036%
pony 0.048 0.0031% 16 0.0023%
turtle 0.048 0.0031% 21 0.003%
chapel 0.044 0.0028% 20 0.0028%
ioke 0.044 0.0028% 25 0.0036%
ooc 0.044 0.0028% 15 0.0021%
sparql 0.044 0.0028% 23 0.0033%
applescript 0.04 0.0026% 19 0.0027%
augeas 0.04 0.0026% 13 0.0019%
g-code 0.04 0.0026% 7 0.001%
mirah 0.04 0.0026% 16 0.0023%
capn-proto 0.036 0.0023% 12 0.0017%
digital-command-language 0.036 0.0023% 19 0.0027%
hy 0.036 0.0023% 12 0.0017%
logos 0.036 0.0023% 19 0.0027%
modelica 0.036 0.0023% 15 0.0021%
vcl 0.036 0.0023% 18 0.0026%
antlr 0.032 0.0021% 15 0.0021%
gdscript 0.032 0.0021% 9 0.0013%
graphql 0.032 0.0021% 17 0.0024%
hlsl 0.032 0.0021% 11 0.0016%
gnuplot 0.028 0.0018% 17 0.0024%
http 0.028 0.0018% 19 0.0027%
ninja 0.028 0.0018% 14 0.002%
oz 0.028 0.0018% 8 0.0011%
raml 0.028 0.0018% 9 0.0013%
aspectj 0.024 0.0016% 8 0.0011%
autohotkey 0.024 0.0016% 15 0.0021%
fancy 0.024 0.0016% 8 0.0011%
moonscript 0.024 0.0016% 10 0.0014%
piglatin 0.024 0.0016% 11 0.0016%
stata 0.024 0.0016% 10 0.0014%
urweb 0.024 0.0016% 6 0.0009%
xs 0.024 0.0016% 7 0.001%
yang 0.024 0.0016% 6 0.0009%
agda 0.02 0.0013% 10 0.0014%
coldfusion 0.02 0.0013% 9 0.0013%
emberscript 0.02 0.0013% 7 0.001%
latte 0.02 0.0013% 7 0.001%
literate-haskell 0.02 0.0013% 7 0.001%
postscript 0.02 0.0013% 9 0.0013%
scilab 0.02 0.0013% 10 0.0014%
tcsh 0.02 0.0013% 10 0.0014%
volt 0.02 0.0013% 9 0.0013%
apl 0.016 0.001% 7 0.001%
genshi 0.016 0.001% 3 0.0004%
jsonld 0.016 0.001% 6 0.0009%
krl 0.016 0.001% 4 0.0006%
lean 0.016 0.001% 3 0.0004%
lfe 0.016 0.001% 6 0.0009%
metal 0.016 0.001% 4 0.0006%
monkey 0.016 0.001% 4 0.0006%
mupad 0.016 0.001% 4 0.0006%
nesc 0.016 0.001% 7 0.001%
nit 0.016 0.001% 3 0.0004%
pike 0.016 0.001% 6 0.0009%
purebasic 0.016 0.001% 5 0.0007%
renpy 0.016 0.001% 3 0.0004%
vhdl 0.016 0.001% 5 0.0007%
xproc 0.016 0.001% 3 0.0004%
zephir 0.016 0.001% 4 0.0006%
apacheconf 0.012 0.0008% 2 0.0003%
boo 0.012 0.0008% 2 0.0003%
brainfuck 0.012 0.0008% 2 0.0003%
bro 0.012 0.0008% 3 0.0004%
cartocss 0.012 0.0008% 3 0.0004%
creole 0.012 0.0008% 2 0.0003%
csound 0.012 0.0008% 4 0.0006%
dylan 0.012 0.0008% 2 0.0003%
eagle 0.012 0.0008% 4 0.0006%
ecl 0.012 0.0008% 4 0.0006%
eiffel 0.012 0.0008% 2 0.0003%
flux 0.012 0.0008% 3 0.0004%
io 0.012 0.0008% 4 0.0006%
jsoniq 0.012 0.0008% 6 0.0009%
lilypond 0.012 0.0008% 6 0.0009%
lsl 0.012 0.0008% 3 0.0004%
mask 0.012 0.0008% 4 0.0006%
nginx 0.012 0.0008% 2 0.0003%
nu 0.012 0.0008% 2 0.0003%
pov-ray-sdl 0.012 0.0008% 5 0.0007%
ragel-in-ruby-host 0.012 0.0008% 4 0.0006%
slash 0.012 0.0008% 4 0.0006%
sourcepawn 0.012 0.0008% 3 0.0004%
squirrel 0.012 0.0008% 4 0.0006%
ston 0.012 0.0008% 6 0.0009%
uno 0.012 0.0008% 2 0.0003%
wisp 0.012 0.0008% 3 0.0004%
xbase 0.012 0.0008% 3 0.0004%
yacc 0.012 0.0008% 3 0.0004%
zig 0.012 0.0008% 4 0.0006%
abap 0.008 0.0005% 1 0.0001%
arc 0.008 0.0005% 2 0.0003%
ats 0.008 0.0005% 3 0.0004%
blitzmax 0.008 0.0005% 1 0.0001%
bluespec 0.008 0.0005% 2 0.0003%
c2hs-haskell 0.008 0.0005% 2 0.0003%
clean 0.008 0.0005% 1 0.0001%
dns-zone 0.008 0.0005% 2 0.0003%
forth 0.008 0.0005% 2 0.0003%
harbour 0.008 0.0005% 1 0.0001%
igor-pro 0.008 0.0005% 1 0.0001%
inform-7 0.008 0.0005% 2 0.0003%
isabelle 0.008 0.0005% 2 0.0003%
jflex 0.008 0.0005% 1 0.0001%
literate-agda 0.008 0.0005% 1 0.0001%
maple 0.008 0.0005% 2 0.0003%
mathematica 0.008 0.0005% 1 0.0001%
module-management-system 0.008 0.0005% 1 0.0001%
mtml 0.008 0.0005% 2 0.0003%
netlinx 0.008 0.0005% 1 0.0001%
parrot-assembly 0.008 0.0005% 2 0.0003%
pawn 0.008 0.0005% 3 0.0004%
propeller-spin 0.008 0.0005% 1 0.0001%
pure-data 0.008 0.0005% 1 0.0001%
rebol 0.008 0.0005% 3 0.0004%
red 0.008 0.0005% 1 0.0001%
sage 0.008 0.0005% 1 0.0001%
sas 0.008 0.0005% 1 0.0001%
scaml 0.008 0.0005% 1 0.0001%
smt 0.008 0.0005% 3 0.0004%
supercollider 0.008 0.0005% 2 0.0003%
unrealscript 0.008 0.0005% 1 0.0001%
xpages 0.008 0.0005% 1 0.0001%

Additional Information

Licensing Information

Each sample comes from a code repository with a permissive license. The license is provided by the license field for each sample.

Citation Information

@article{muennighoff2023octopack,
      title={OctoPack: Instruction Tuning Code Large Language Models}, 
      author={Niklas Muennighoff and Qian Liu and Armel Zebaze and Qinkai Zheng and Binyuan Hui and Terry Yue Zhuo and Swayam Singh and Xiangru Tang and Leandro von Werra and Shayne Longpre},
      journal={arXiv preprint arXiv:2308.07124},
      year={2023}
}
Downloads last month
64