How to play - VBPXKSMI/Wiki-Test GitHub Wiki
================================================================
This page will tell you where and how to start playing. It will not explain how to play in game nor cover any other unrelated topic.
================================================================
After reading that hefty introduction in the previous sections, you may be wondering where the hell the play button is located. Well, luckily for you, this page will spell out that, and clear a few things about the different versions of the game along the way.
Short summary (click on the links if you just want to go ahead and play):
-
Colab
- Latest up-to-date version
- Beta version
- Old "stable" version
-
Offline PC installations
- Base game and thadunge2 modded version
- Clover Edition
-
Official versions
- Website
- Android and IOS ports
================================================================
It's recommended to read the entire page, to have a clear understanding of the main core issue of any game, i.e. how to start playing it first.
================================================================
Fucking finally, I was wondering when would I be able to play the game. Lets start in the order you listed the summary, what is a "Colab"?
Colab is a short term for "Colaboratory", it is a plattform created and owned by Google to serve as a tool to further interests in researching, testing and developing machine learning. It's a Jupyter notebook environment where everything runs on Google's servers (AKA Google Cloud) through either GPUs or TPUs.
Thankfully, Colab its entirely free to use, as Google lends its spare servers to be used until they are needed elsewhere (like to suply the needs of a paying customer), although this last part will be explained later.
Cause Colab is free to use, it serves as the ideal frontend to play a game as resource intensive as AI Dungeon, reason why we use it.
So, we are using Google high performance hardware to play a game for free? Isn't that illegal/risky?
Morals aside (who cares about that?), there is no harm done as Google only lends you their spare hardware, so they wouldn't be using it anyway. Not to mention that there is nothing illegal about using it in the first place.
But more importantly, Colab has allowed thousands of people to play the games without needing very expensive hardware. In doing so, many got interested in something they may have previously never heard about: machine learning, the main intend of Colab, i.e. disseminating machine learning education and research.
Overall, using Colab to play the game has been more sucessful in bringing new minds into machine learning research than any ammount of propaganda Google may have used. So again, no harm done there either.
Ok, seems using Colab isn't that bad but why isn't the original author using it for his game then? It's free isn't it?
As a matter of fact, Colab was originally how you could play the game (the only one also), but due to the author deciding (quite quickly) to start a bussiness out of the game, Google stepped in and told him they wouldn't tolerate anyone from profitting for theirs hardware for free, so he (the original author) had to set up servers of his own (AKA rented from third parties) to host his game.
As this is an Open Source project, there is no profitting, so we are allowed to still use Colab to play the game, while the original author is not.
You mentioned it before, why is AI dungeon so slow/resource intensive when it's just a text based game?
You're not running a text game, you're running an AI brain with the RAM you use. You are using state-of-the-art natural language processing neural network, meaning that it works IRL time with no scripted actions. You shouldn't be surprised if it ddidn't run on your GTX 960. Think of it like playing in an "open world" game, just that everything in not scripted already, and instead of graphics there are just texts.
The base game requires at least 16GB of RAM or 12GB of VRAM at least just to get it running, so very few people have that kind of expensive hardware, unless you were a researcher or a developer yourself. Thankfully, this requirements were halved thanks to Anons works in Clover Edition, but it's still hardware that not everyone's have (specially if you are from a third world country).
In other words, the most easily available way to play the game is through Colab as the hardware needed is already provided by Google themselves to you the player.
So you are telling me that the most demanding game on the market right now is a text-based game? We have gone full-circle.
Effectively, think of this game in terms of what Zork was for his era and how it influenced the entire industry that came afterwards. This game is not different to that as it will probably mark the beggining of a new era for videogames, the next step would be transition to full graphics AI powered games but that will take many more years (probably decades), and ridiculy powerful hardware to run them.
Patience is a virtue that you don't seem to have but here they will be covered.
Take note that they are all hosted in this repository (if you didn't noticed it before, this repository contains more than just the Wiki):
Up to date Colab | Beta Colab | Old Colab |
---|---|---|
Latest | Beta | Colab v1.0 |
This is the Colab in the master branch, meaning is the one that is intended to be played by most, as it has gone out of Beta development, thus being polished and as bug free as possible. Is fairly easy to use as it is pretty straightfoward to get it running. Some Anons report having disconnects sometimes while using it, but otherwise it is pretty stable. | This is the Colab in the Beta branch, meaning bugs are to be expected as what is done here is mostly experimental and awaiting for feedback and tests before going into the master branch (which could take a while). This also means that, if you used it, you would have access to the new features before everyone else, while also providing the much needed feedback and helping the project in the process. So feel free to use it at your leisure and don't forget to give your feedback in the threads. | This is an old version of the Colab that is supposed to be used only when you encounter the disconnect bug too usually, as apparently it is more "stable" as reported by some Anons, but this might just be confirmation bias / placebo effect, so judge by yourself. Do note that it requires a few more steps to get it running (nothing too difficult really) but the main problem with it is that it's not getting updated actively anymore, so bugs may be encountered and new features may not be included on it. Otherwise it can be used with no problems. |
If you want to use different models other than the default one, go to the Models and Datasets page. |
---|
With this, you can proceed to play the game by just going there and reading the fairly short guides inside.
Now, for the next section...
The disconnect bug is (as you can infere) when your Colab gets disconnected suddenly by no apparent reason.
The main reason for why this happens is that Google kicks you out of their servers either as a preemption measure to avoid abuse of their servers (for more details see here), or to give their hardware to the actual paying users thus making that free hardware not available to you anymore.
A summary would be this:
- Colab is owned by Google, who lets you use some leftover TPUs and GPUs of them to run things on the Colab.
- But Google isn’t too stupid, so they limit your free usage by disconnecting you from their network at certain times of the day.
- Which times are those, nobody knows as they are not disclosed, but they seem to depend on which region the servers are located (most if not all are US based) and which time of the day it is, so your random encounters with the disconnect bug will depend entirely on your luck and which server you were put on and which time of the day you play in.
- This is also the reason why some never encounter the bug but some do, i.e. while some play and are being assigned to servers that never shutdown people, others aren’t so lucky. This is also correlated with the sudden increase of bug reports at certain times of the day.
So your available options are to either:
- Play at other times of the day.
- Use the local installs if you have the hardware.
- Use the supposedly "stable" Old Colab v1.0
- Pay Google for a private server.
- Play through the (inferior) official version website.
- Or pray to God to improve your luck.
Take note that other important influencing factors may be anything that could be interfering with your internet connection, even if for a fraction of a second, like by:
- Using a VPN.
- Using Wifi, instead of Ethernet.
Other minor influencing factors include (ranked from how likely they may be related to the bug):
- Switching between tabs too frequently (i.e. switching between the Colab tab in and out while waiting for an AI response) (Quite likely).
- Letting the Colab tab iddle for way too long (i.e. not inputing any new prompts or clicking on the page itself) (Quite likely).
- Which browser you use (Not very likely).
There's little play button in the top left corner of the little blocks, the blocks are the cells, you click that button to run them. Only run one at a time and wait for it to finish. They have a “play symbol” next to them, so you can easily see them. Also keep an eye on them, because some of them prompt you with questions and won't continue running until you respond.
This issue shouldn't be connected to the disconnect bug, and it's entirely possible to happen to people that use Colab for too long sessions.
To bypass it you can just make a new Google account and logging with it. Or make sure to completely exit out of Colab and wait for a while (like between 1-3 hours).
A bug this time, I’m using the Colab and my game cell keeps crashing when "ping" is active and the AI post a new response. What do?
Ping doesn't work in Colab (or any other sound system at the moment). So turn it off.
Or do this manually:
Copy the notebook to your Google Drive
On your desktop, reduce MaxHeight to ~"400" so the cell stays within the window (you can't do this on mobile because the code editor is fucking broken)
Load your custom notebook on your phone
Put phone in landscape
Run the modules
Tap the last cell so the options in the top right show up
Press the three dots
Press "view output fullscreen"
Yes, follow this.
Or you can try a new and better customization feature recently developed here. A better and easier way customizator is in the works at the moment, so make sure to pay attention to the Beta for when that feature comes out.
You can go to the Colab Troubleshooting page if you have more doubts/problems.
Now, if you finally don't have any more questions, lets proceed with the next section dealing with the local installs.
================================================================
So Colab was an online way to play the game, really useful if you have a potato PC, but what if I have a powerful gaming rig? Can I install the game on my computer without depending on Google hardware? Any benefits from doing so?
Yes you can, as long as you met the appropiate requisites there shouldn't be much differences between playing on Colab or locally.
The benefits:
- Not depending on Google hardware, which means you won't ever encounter the disconnection bug (good news if this happened to you too often).
- Not needing a Google Drive account, which is necessary for playing on Colab.
- Not needing internet connection at all to play the game.
- (Official versions only) Not having your sessions and stories being saved in a server, being it a publicly viewable one in Google Cloud, or in the original author personal servers.
- (Colab only) Not having your stories being saved inside Google Drive, which theorically Google could have access to, however that would be a tremendous breach of privacy, so you don't really have to worry about this unless you start writing about secret terrorist plots in Arabic.
First off, take note that different versions of the game do have different requisites. And those requisites are just the bare minimum to get the game running, lower specs may run it slower or don't get it running at all, while higher specs will run the game more smothly and faster.
Requisites:
- Base game and thadunge2 modded version:
- 64-bit Windows, either 16+ GB of RAM to run this on your CPU, or have a CUDA (NVidia cards only) enabled GPU with 12+ GB of VRAM
- Clover Edition:
- 64-bit Windows, either 8+ GB of RAM to run this on your CPU, or have a CUDA (NVidia cards only) enabled GPU with 6+ GB of VRAM
As you can see, the base game and thadunge2 modded version require hardware that very few people have, while Clover Edition requires half of that requisites and is thus more affordable to many.
Base game: is basically that, the original untouched game files, it doesn't have barely any extra commands or functionalities. Is good in and of itself, but is very barebones nonetheless.
To quote what Anons usually say about it:
- "If you've already tried thadunge2 or Clover Edition, going back to the mormon version is like drinking bland, boring, generic, store brand knockoff cola after tasting Classic Coke."
- "Once you get a taste of thadunge2 or Clover Edition you'll never accept an inferior product."
Thadunge2 modded version: a mod of the base game, adds plenty of commands and functionalities, thus making it a better product. Problem with it is that it requires the same requirements as the base game, thus is not very affordable for playing it locally, so most only use it on Colab.
Clover Edition: is another modded version, but instead of only being a mod this one is more of a complete rewrite of the base game, adding many (but currently not all) of thadunge2 commands and functionalities, extra features, and neat visuals. And its main selling point? Halving the requirements needed, thus making it very affordable to play locally if you have a mid-tier gaming PC.
There is no catch, to understand where the difference comes from you first need to understand where the main source of the requirements is: that is the AI's brain (AKA the model it uses).
Tensorflow: the base game, and thus also thadunge2, use models that need this installed, as the models need that to run as they were created using it. The main catch of this models created through this is that they are pretty big (5GB per 1558M model), and they require pretty good hardware to run (AKA the requirements from the base game). These models are also encoded using a 32-bits float precission, which all in all is unncessary as they bring very little gain for too much processing power.
PyTorch: this is what Clover Edition uses, models using and created through this are much more efficient in regards to processing power, with very little loss compared to tensorflow. Since creating the models through PyTorch is little difficult, what is usually done is creating the models using tensorflow, and then convert them to PyTorch. These models are encoded using 16-bits float precission, which means that the models require half of the usual requirements to run, while having very little loss in terms of model quality.
This is mainly where the differences come from. Other factors are that Clover Edition uses some word filtering while the models process your inputs that allows the AI to be less repetitive, and prevents the loopings it sometimes falls into.
thadunge2 accepts every pull request people send him, so it has more features but it breaks from time to time.
Clover has colors, more prompts (scrapped from the Custom Prompts Wiki pages) and can be run offline with only 8GB of VRAM (half of the one needed to run the base game).
Summary:
thadunge2 if on Colab or you want better commands.
CloverAnon if you run locally, want more compressed models and 1337 visuals.
Each version requires a different installation method, as they may use different kind of models (like Clover Edition does).
There are 2 ways in which it can be done, either installing the dependecies manually or doing so with authomatic installers, both ways work and are the same.
Base game + thadunge2 | Clover Edition |
---|---|
Wiki page | GitHub page |
How to install the base game, and how to apply any modded versions to it. Either manually or with authomatic installers. | How to install Clover Edition. Either manually or with authomatic installers. |
If you want to use different models other than the default one, go to the Models and Datasets page. |
---|
With that, the sections covering Anon's modded versions are concluded. Let's go to the final one, i.e. the one dealing with the official versions.
================================================================
Then you were wrong, what was discussed so far pertained only to 4chan Anons modded versions.
The official ones differ greatly, although not exactly in a possitive way.
Aside from what was discussed above (particularly here) there are other differences.
The negative ones:
- All official versions upload your stories to either a public bucket in Google Cloud (no joke), or go directly to the mormon servers. While all Anons versions have upload to the cloud off by default, nor recollect any kind of info.
- All official versions require you to provide an email to play (Colab too requires an email, in the form of a Google Drive account, but you are probably safer with Google than providing an email to a third party).
- All official versions are free to use, but have a separate patreon wall, and nobody knows if or when it will be mandatory to play. While all Anons versions are free to use and Open Source, and will continue to be so.
- All official versions are severily behind Anons versions, both contents, code and visual wise.
- All official versions are prone to be censored, the author confirmed he has plans to tackle "problematic words and content". While all Anons versions have "/censor" off by default, while there is no kind of restriction to the kind of content the models handle.
The possitive ones:
- As the author has more resources (thanks to his patreon bucks), he is able to maintain his own private server (rented to a third party). This means he doesn't have to rely on Colab and its Google hardware (no disconnect bug), nor in the user's hardware as he can provide the hardware himself. Even then, sometimes even his servers get overloaded so it isn't exactly fail proof.
- Availability, as you can play them with very few effort through the website or the mobile ports.
This all means that the official versions offer easiness of use for the players, although at the cost of game quality and very serious attempts to reduce freedom of the players(which includes both content, and free availability).
Nevertheless, we offer you the choice to play them if you wish so for whatever reason.
Seems that it has some flaws. By the way, I keep seeing references to a mormon or something, what's up with that?
Good catch, if you have been reading from the beggining of the table of contents, this is the first time a reference to ((HIM)) is made.
And all those references correspond to the original author's (Nick Walton) different nicknames.
The author earned himself several nicknames since the very first day the game was launched, mostly through how awful and idiotic his actions were. Thus, Anons made some nice nicknames to refer to him and his official versions, and have been calling out his bullshit since day 1.
Nobody hates him personally, just how he has managed everything since the beggining.
A logical/unbiased/chronological explanation of the events (as seen by Anons) goes as follows:
- As soon as the game got sort of popular (meaning since Day 1) he immediately launched a Patreon and rebuilt his website to beg for money under false pretenses ("it costs $62,000 a month for our servers to run this game!") and made promises he literally can't deliver on. It's obvious he's gone into full grifter mode to scam normalfags and retards into paying him an AI researcher's salary as a college undergrad.
It was a neat original idea, but he didn't do anything exceptional with it. He trained an already-made AI model on some data he downloaded from a website ("https://chooseyourstory.com", that is now pissed at him for doing so) using a FOS library called "gpt-2-simple", and then built a basic Python wrapper for formatting the input and trimming the output, and put that into a Jupyter notebook (something everyone learns how to do in their first machine learning class) on the Colab servers (using Google's processing power at no cost to him). Literally any undergrad CS student could have done what he did. We all taught ourselves how to do it once we got pissed off at how shitty his model was and we've been modding his game since Day 1.
Literally the only costs he incurred were hosting the 6GB model on his own servers like a retard, and aside from having a neat idea he's done nothing to deserve $140,000/year. His code was buggy as shit and crashed constantly until other people patched it up for him. He has no humility and went out of his way to grift people as soon as he got his fifteen seconds of fame.Even after all that, his versions are still provided here, even if there is no obligation to do so, and everyone reconigzes his due credit for being the first one to show the unexplored potential this kind of technology has for the videogame industry.
You may see him been refered by different nicknames, both in the Wiki pages as well as in the threads.
Most common nicknames are:
- The mormon (most common one, is due to him being a mormon, as well as attending a mormon university in Utah)
- ((HIM)) (common one, short version of the above)
- mormon-jew (self explanatory)
- The grifter
- Conman
- Scammer
- Todd v2.0
- Yanderdev v2.0
But enough talking about him, nobody cares what he does since a long time ago.
Here they are, do note that they all are fundamentally the same across platforms, and they all use the same default model as the Anons versions.
Official Website | Android + IOS ports |
---|---|
Website | Wiki page |
Direct and simple, just click and play. | The android and IOS ports of the game to play on mobile. |
With this the page is finished, now you know where and how to play the game.
Next page will detail how to play in game, i.e. what are and how the commands function.