I’ve recovered, remade a dataset, and re-fine-tuned. By doing that I’ve made the best performing model so far.
The model now generates actual English sentences instead of gluing stuff together with commas and the word ‘and’. I’m now working on making it hallucinating less.
Log in to leave a comment
I hate my life.
I’ve tried a bunch of things in the past 2 hours, but I still can’t stop if from hallucinating random things not present in the input.
I’ve removed the required output JSON structure so the model only needs to output final entry (and uses two-passes generation for things like title and mood). I’ve programmatically converted input into actual sentences in hope that it’ll somehow makes it perform better. While it did, it’s minimal and the issue still persists.
I gave up and just used Qwen3-0.6B without any fine tune. It easily outperform my 14 hours effort. It even cutted my parameters count (and thus app size) by a whopping 600MB, wtf?
bruh, I’m overcomplicating things. I always am.
Log in to leave a comment
Hi! I’m sorry for logging 15 hours of work. Putting this into flavortown was really a last-minute decision. I’ll try my best to make up for it by writing the whole devlog here.
MomentSLM is an on device Small Language Model that powers my ‘write the bare minimum’ journal app, “Moment”. Under the hood, it’s a fine tuned version of LFM-1.2B-instruct with my own custom dataset. It’s really simple, but it took a lot of time because there’s simply too many things I can change. For example:
Qwen3-0.6B, Qwen3-1.7B, LFM2.5-1.2B-Base, LFM-2.5-1.2B-Instruct
It’s not an easy task. Being a small language model, MomentSLM has some trouble understanding the context of my input, and thus generated some quite interesting outputs. For instance:
While it’s not perfect, I’m quite happy with it’s current status, and will write the devlog here once I’ve managed to get it working better.
Log in to leave a comment
I’ve made a CLI tool that helps me manage devlogs and blogs.
It can:
I’ll also mirror the devlog on flavortown to my website’s project devlog from now on. You can check it here: https://mebaddev.net/project/personal-website
Log in to leave a comment
Heavily inspired by Flavortown and Summer of making lol
Check it out for yourself:
https://mebaddev.net/project/personal-website
Log in to leave a comment
This is my personal website. It has:
Here’s the change log I’ve made:
Log in to leave a comment
I’ve refactored the unholy spaghetti that is my codebase. That’s all I have to update.
I’ve attempted to include 2 videos in the last devlog, but for some reason it only shows 1 (and it’s dulplicated, i am so confused), so I’ve squeezed out an extra devlog just to upload that video.
Log in to leave a comment
I’ve implemented a simple server! It’s using WebSocket because I’d like to provide a web demo in the future, and WebSocket is perfect for that. Also, I’ll be able to host dedicated servers in addition to LAN P2P with the same codebase, which is a nice bonus.
I’ve also (embarrassingly) attached my tetr.io gameplay clip instead of my project in the last devlog (when can we get the ability to edit devlogs btw), so I’ve attached it to this one. Sorry! :P
Log in to leave a comment
Very cool animations!!
bruh it only uploaded one video twice just like every single one of my devlog :\
i guess i’ll work on refactoring my codebase and show the server in the next devlog
I’ve implemented a garbage system (hopefully) identical to guideline Tetris! I’ve also implemented a “Backfire” mode where you basically attack yourself with a 0.5x multiplier. That way I can show this demo.
I think I’ll now begin working on multiplayer :D
Log in to leave a comment
WAIT THATS THE WRONG VIDEO
I just realize it’s impossible to delete devlogs, so sorry for that. I’ll include the intended video on the next devlog
meanwhile hope you enjoy the tetris lol
good tetrio gameplay 👍
There’s now SFX for placing pieces, spins, comboing and perfect clear. I’ve definitely spent way more time then a hour but hackatime is not avaliable for audacity and musescore :(
I’m think the sound of spin is kind of weird and does not fit, though. I might change it later.
Anyways, here’s the video!
Log in to leave a comment
I looked for feedbacks in slack and some gamedev discord server and altered a few things:
I’ll now begin to work on adding sound effects.
Log in to leave a comment
nice kde theme
Thanks, I use arch btw :3
Now when you perfect clear, it does a cool spin!
You might think that this update is not as ‘packed’ as my previous devlogs, but it’s because while I’ve done quite a lot, it’s all just refactoring my spaghetti, and I don’t know how to present it.
I’ve also poked around the repo of Flavortown. In theory, if I include this link: https://hackclub.slack.com/archives/C01504DCLVD/p1766762314668419, I’ll get an achievement, let’s try it!
We’ve also officially hitted the 10 hour mark, so that’s pretty cool.
Log in to leave a comment
That didn’t work :(
Well, I don’t need 20 cookies that much currently anyways, so it’s fine.
Well, that’s a really general description, and I don’t even know how to convert it into English sentence, so here’s a changelog!
Rebuilt everything using Godot’s Control node for easier handling of different screen sizes and stuff
Every display (the board, next and hold piece) now uses the same skin system, so I don’t have to apply skin 3 times. If you look closely at my last devlog you’ll see some discrimination, which is now fixed :D
Reworked the spin animation so it looks more smooth
Adjusted Next piece display’s height so it doesn’t have an awkward blank space
Most importantly, I’ve visualized what clears have you made and the back to back chain, and I’m quite happy with the animation!
(In case this devlog looks weirdly spaced, sorry. Flavortown’s markdown is in a weird state rn. The subtitle renders correctly but not the bullet points and they just disappeared. I’m adding some space between those bullet points to try to make it easier to read, and I have no idea how this will turn out.)
Log in to leave a comment
For the past 2 hours or so I’ve been working on juicing things up:
I also feel like now’s a good time to yap a bit about what I want this game to be.
Basically, I love tetris, but now there’s a skill gap between me and my friends so they no longer want to play with me :(. I want to create a Tetris party game that will be fun and balanced even if players have skill gap, and make it enjoyable for everyone.
I’ll talk more about it in later devlogs. For now, cya!
Log in to leave a comment
WOW, now that looks good. You made really good job with the user feedback animations.
Yep, that’s right. I spent 5.2 hours making a Tetris engine. Please trust me that it’s worth it, though, and no, this is not my first time using Godot.
You see, while Tetris is a popular simple beginner project with tons of tutorials, most of them implements it ‘incorrectly’. There’s actually standards on what Tetris is, including rotation system, kick table, bag system, piece spawn position/orientation etc, which allows for crazily fast Tetris. Most tutorials just go as far as ‘blocks fall and when line full clear line’, but that’s not enough for me.
Here’s what I’ve done:
The engine I’ve created is exteremly module, you can swap rotation system or throw some pentrominos at the player and it won’t complain.
Here’s a video of it! (I’m an S+ player on tetr.io, btw :D)
Log in to leave a comment