Seqmod is a library to run sequence modelling on texts built on top of Pytorch. Seqmod is mostly the result of a personal attempt at having a tested and structured framework for:
- (i) Quick model and experiment development
- (ii) Built-in tooling for running experiments
Accordingly, the core of seqmod consists of (i) pluggable Pytorch modules (inside /seqmod/modules/)
with a thin abstraction layer and (ii) tools for training and testing models (inside /seqmod/misc/).
Currently, the easiest way to install seqmod is to grab the package (clone or just download) and put make
it discoverable by Python by either adding the path to seqmod to your PYTHONPATH:
# ~/.(bashrc|profile|bash_profile)
export PYTHONPATH=$PYTHONPATH:/path/to/seqmod/or (less comfy) by adding the following to your script:
# my_python_script.py
import sys
sys.append('/path/to/seqmod/')Additionally, seqmod comes with a number of scripts to train Language Models and Encoder-Decoder models
without having to write a single line of code. See inside scripts.
A final warning. Seqmod evolves rapidly according to my research needs, so API changes are common
(although I mostly work on the dev branch and keep master untouched/stable for longer periods of time.).