llamp 0.0.4

LLAMP - Large Language Model for Planning and Acting

LLamp - Large Languge Models for Planning

This is a package that uses LLMs (closed and open-source) for planning.

Pre-requisites:

1. Python3.9
2. (recommended) virtualenv

Installation:

1. pip install -r requirements.txt
2. alfworld-download
3. pip install -e . (this installs llamp)

Exporting API Keys:

1. export OPENAI_API_KEY=""
2. export CEREBRAS_API_KEY=""
3. ... (and so on for all the providers you want to use, e.g. Anthropic, Nvidia, Cohere)

Testing everything works:

1. Basic test:

cd test
./test.sh

2. More advanced test:

cd root_folder
./playgrounds/run_alfworld_eval.sh test_ours

Running Evaluation

1. Running Eval:

cd root_folder
./playgrounds/run_alfworld_eval.sh cerebras_main

Creating Conda Env: (Note this might be work in progress)

conda env create -f environment_stateact.yml -n env_name

Webshop, Run from Docker:

docker container run -p 3000:3000 ainikolai/webshop:latest

---

Previous README (currently being archived and refactored.)

WARNING PACKAGE IS STILL UNDER DEVELOPMENT and requirements needs cleaning up.

Installation:

1. Textworld Game (pip install textworld)
2. Textworld Visualisation (pip install -r requirements_textworld_visualisation.txt)
3. (install chromedriver or firefox driver)

Playgame:

The following is accepted:

python3 playgrounds/playground_tw_gym.py {human/openai/...} --custom/--simple {PARAMS}

e.g.:

python3 playgrounds/playground_tw_gym.py human --custom 1 2 2

Or:

1. (In terminal with browser visualiser) tw-play tw_games/first_game.z8 --viewer
2. (as Gym environement in terminal) python3 playgrounds/playground_tw_gym.py

Generate New Textworld games using helper script

python3 generate_games.py --simple/--custom {PARAMS}

e.g.

python3 generate_games.py --custom 2 2 2 1234

Generate New Textworld games using TW

1. tw-make custom --world-size 2 --nb-objects 10 --quest-length 5 --seed 1234 --output games/tw_games/w2_o10_l5_game.z8

2. tw-make tw-simple --rewards dense --goal detailed --seed 1234 --output games/tw_games/simple/r_dense__g_detailed__seed_1234.z8

Rewards: (dense, balanced, sparse) Goal: (detailed, brief, none)

Reference: [https://textworld.readthedocs.io/en/stable/tw-make.html#types-of-game-to-create]

Available Commands to agent:

Available commands:
  look:                describe the current room
  goal:                print the goal of this game
  inventory:           print player's inventory
  go <dir>:            move the player north, east, south or west
  examine ...:         examine something more closely
  eat ...:             eat edible food
  open ...:            open a door or a container
  close ...:           close a door or a container
  drop ...:            drop an object on the floor
  take ...:            take an object that is on the floor
  put ... on ...:      place an object on a supporter
  take ... from ...:   take an object from a container or a supporter
  insert ... into ...: place an object into a container
  lock ... with ...:   lock a door or a container with a key
  unlock ... with ...: unlock a door or a container with a key

Running jupyter notebooks in your own environment:

1. [https://medium.com/@WamiqRaza/how-to-create-virtual-environment-jupyter-kernel-python-6836b50f4bf4]

pip install ipython
pip install ipykernel

ipython kernel install --user --name=myenv

python -m ipykernel install --user --name=myenv