Skip to content

Instantly share code, notes, and snippets.

View seouri's full-sized avatar

Joon Lee seouri

  • Activate Care
  • Boston, MA
View GitHub Profile
@seouri
seouri / grpo_demo.py
Created January 31, 2025 19:30 — forked from willccbb/grpo_demo.py
GRPO Llama-1B
# train_grpo.py
import re
import torch
from datasets import load_dataset, Dataset
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import LoraConfig
from trl import GRPOConfig, GRPOTrainer
# Load and prep dataset
#!/bin/sh
git filter-branch --env-filter '
OLD_EMAIL="your-old-email@example.com"
CORRECT_NAME="Your Correct Name"
CORRECT_EMAIL="your-correct-email@example.com"
if [ "$GIT_COMMITTER_EMAIL" = "$OLD_EMAIL" ]
then

Simplecov aggregated coverage report from CircleCI 2.0 parallel builds

Problem Statement

We have Rails application which is running tests on circleCI 2.0, we have simplecov configured to track the coverage of our test suite. Now the problem is with parallelism enabled, we have partial coverage reports in all different containers according to the tests those containers ran.

We obviously want to have consolidated simplecov coverage report which actually shows us overall coverage report.

M[16],X=16,W,k;main(){T(system("stty cbreak")
);puts(W&1?"WIN":"LOSE");}K[]={2,3,1};s(f,d,i
,j,l,P){for(i=4;i--;)for(j=k=l=0;k<4;)j<4?P=M
[w(d,i,j++)],W|=P>>11,l*P&&(f?M[w(d,i,k)]=l<<
(l==P):0,k++),l=l?P?l-P?P:0:l:P:(f?M[w(d,i,k)
]=l:0,++k,W|=2*!l,l=0);}w(d,i,j){return d?w(d
-1,j,3-i):4*i+j;}T(i){for(i=X+rand()%X;M[i%X]
*i;i--);i?M[i%X]=2<<rand()%2:0;for(W=i=0;i<4;
)s(0,i++);for(i=X,puts("\e[2J\e[H");i--;i%4||
puts(""))printf(M[i]?"%4d|":" |",M[i]);W-2
namespace :heroku do
def app_name
@app_name ||= Heroku::Command::BaseWithApp.new([]).app
end
def latest_bundle(timeout = 30)
puts "Attempting to get latest bundle..."
get_bundle = lambda do
bundles = Heroku::Command.run('bundles', {})
bundles.sort { |b1, b2| b1[:created_at] <=> b2[:created_at] }

Rails 2.3.5 on App Engine

We assumed Rails 2 would never work without rubygems, and we committed to gem bunlder for JRuby on App Engine, so we were waiting for Rails 3. Fortunately, Takeru Sasaki was able to patch the Rails 2.3.5 calls to rubygems, and now we have it working. Rails 2.3.5 currently spins up several seconds faster than Rails 3, and just a few seconds behind Sinatra.

Install the Development Environment

The gems for the development environment include a pre-release appengine-tools gem that provides a pre-release version of jruby-rack.

sudo gem install google-appengine
sudo gem install appengine-tools --pre
development: &global_settings
database: textual_development
host: 127.0.0.1
port: 27017
test:
database: textual_test
<<: *global_settings
production:
@seouri
seouri / Gemfile .rb
Created December 14, 2009 19:32 — forked from woodie/Gemfile .rb
# Critical default settings:
disable_system_gems
disable_rubygems
bundle_path ".gems/bundler_gems"
# List gems to bundle here:
gem "rails", "3.0.pre", :git => "git://github.com/rails/rails.git"
gem "arel", :git => "git://github.com/rails/arel.git"
gem "i18n"
gem "dm-appengine"
@seouri
seouri / url_dsl.rb
Created December 14, 2009 19:18 — forked from defunkt/url_dsl.rb
require 'open-uri'
# url dsl -- the ultimate url dsl!
#
# You just can't beat this:
#
# $ irb -r url_dsl
# >> include URLDSL
# => Object
# >> http://github.com/defunkt.json
@seouri
seouri / gist:71852
Created February 28, 2009 03:16 — forked from ryanb/gist:71618
class Tag
has_many :taggings
named_scope :important, :include => { :taggings => :important }
end
class Tagging
belongs_to :tag
belongs_to :posts
# uses a string 'flag' column but could be changed to a boolean important