Skip to content

bigger context window, less weight = good local model

Notifications You must be signed in to change notification settings

samsja/reason_net

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ReasonNet

generate data

cd data_gen
cargo run --release -- --min 3 --max 5 --size 20000000 --seed 32 --save-path data-20m-add --chunk-size=500000 --operators=+

mv data-20m-add ../datasets/.


cargo run --release -- --min 3 --max 5 --size 20000000 --seed 32 --save-path data-20m-all --chunk-size=500000 
mv data-20m-all ../datasets/.

run the training

normal

python reason_net/run.py --config-path configs --config-name all-14m.yaml

reason middle

python reason_net/run.py --config-path configs --config-name all-14m.yaml reason_mode=true 

reason left

python reason_net/run.py --config-path configs --config-name all-14m.yaml reason_mode=true +data.reason.reason_token_pos="left" 

With small model for test

python reason_net/run.py --config-path configs --config-name all-14m.yaml module/model=910K

About

bigger context window, less weight = good local model

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published