T5-nl2cmd

T5-nl2cmd is a model that generates bash commands from natural language descriptions.

This repository contains the trained model, fine-tuned from Flan-T5 base.

Training data

The model is fine-tuned from the Flan-T5 base model.

It has been trained on a single Nvidia L4 from Scaleway, for 4 hours and 4 epochs.

The model aims to generate bash commands from natural language descriptions.

The results are evaluated on an 500 eval subset of the data cited above.

Task	Command
Count time, calls, and errors for each system call and report a summary on program exit	`slurmctl -c path/to/file`
Search for a word in all the normal/regular files in the current folder	`find . -type f -exec grep -H`
Start the daemon with verbose output coming from specific files in the bluetoothd or plugins source	`bluetoothd-start --verbose path/to/file`
Find all directories under current directory and make them read, write, and executable for owner & group and remove read-write-execute permission for others	`find . -type d -exec chmod ug=`
Set the verbosity level (error, warn, info, debug, or trace)	`svn -v verbosity`
Show the configuration of a virtual machine, specifying its ID	`qm show vm_id`
Import VPN configuration	`ipconfig import config`
Find all the files ending with ".coffee" in the current folder and search for the words "re" in each line	`find . -name "*.coffee" -exec grep`
Run a program from a file	`pstool --run path/to/file`
Display details and metadata about a specified portable service image	`portablectl show-image path/to/image`
Display all the files in the current folder excluding those which are in the path of ".git"	`find . -path '*.git' -prune -`
Find all *.ogg files on the system ignoring the case	`find / -iname "*.ogg"`
Print and recursively remove the alphabetically last directory in the current directory	`rmdir -rf`