Post
Conversation
![David Watson 🥑](data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="96" height="96"><rect fill-opacity="0"/></svg>)
Post your reply
Google released Gemma 3 270M, a new model for hyper-efficient local AI!
We'll fine-tune this model and make it very smart at playing chess and predict the next move.
Tech stack:
- for efficient fine-tuning.
- transformers to run it locally.
Let's go! 
0:01 / 0:12
Load the model
We start by loading the Gemma 3 270M and its tokenizer using Unsloth.
Check this 
Define LoRA config
We'll use LoRA for efficient fine-tuning.
To do this, we use Unsloth's PEFT and specify:
- The model
- LoRA low-rank (r)
- Layers for fine-tuning (target_modules)
Check this code 
Load dataset
We'll fine-tune Gemma 3 to make it extremly smart at playing chess.
Given a set of previous move (one move missing) & the final result it has to predict the missing move.
In order to do this we're using the ChessInstruct dataset from HuggingFace.
Check this
Prepare dataset
Next, we use a conversation style dataset to fine-tune Gemma 3.
The standardize_data_formats method converts the dataset to the correct format for finetuning purposes!
Define Trainer
Here, we create a Trainer object by specifying the training config, like learning rate, model, tokenizer, and more.
Check this out 
Train
With that done, we initiate training.
The loss is generally decreasing with steps, which means the model is being fine-tuned correctly.
Check this code and training logs 
Finally, the video shows prompting the LLM before and after fine-tuning.
After fine-tuning, the model is able to find the exact missing chess move instead of randomly generating some moves.
Check this
0:06
If you found it insightful, reshare with your network.
Find me → 
For more insights and tutorials on LLMs, AI Agents, and Machine Learning!
For more insights and tutorials on LLMs, AI Agents, and Machine Learning!Quote
![](data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg" width="48" height="48"><rect fill-opacity="0"/></svg>)
Akshay
@akshay_pachaar
Google just dropped a new LLM!
You can run it locally on just 0.5 GB RAM.
Let's fine-tune this on our own data (100% locally):
Cool, now get it to run on a pixel watch without a data plan or phoning home - create your own, ultra-personal small-scale AI that doesn't tattle on you but can help you monitor and decode yourself.
Serious question from someone without deep understanding of this stuff: could that run in a browser window as long as GPU support (webasm?) is provided?
Wow that’s interesting. What are some typical use cases for this type of local LLM?
Low memory footprint makes AI accessible to everyone - democratizing local ML development!
Brilliant guide to fine-tuning, Akshay. I misinterpreted 270M as 270B when I heard about the release, since hardly any powerful LLM comes with parameters in "M" these days :D. The level of performance compared to its scale is great.
Thanks for the demo / PoC.
What kind of setup/bootstrap/data would I need to fine tune this for spam/cold outreach email detection?
Impressive efficiency. Could this democratize customized AI models? Localized fine-tuning raises interesting privacy questions.
Wow! 
Running a full LLM locally on just 0.5 GB RAM is amazing.
Can’t wait to see what fine-tuning on custom data can do—fast, private, and powerful AI experiments ahead!
Running a full LLM locally on just 0.5 GB RAM is amazing.
Can’t wait to see what fine-tuning on custom data can do—fast, private, and powerful AI experiments ahead!
I enjoyed reading this. Would you recommend a similar approach for fine tuning a model for classifying text (spam/ham) or would something like Facebook’s fastText still be the recommended approach?
That sounds like a very interesting LLM. What can one do with it?
What does it take (hardware) to train a model like this locally?
Nice
How long would it take to fine-tune that on 8-12 GB of VRAM?
Don't wanna ask grok as he doesn't have a clue 
Tuning an LLM to play chess makes absolutely no sense! Why don't you just download the chess engine Stockfish? Its evaluation function uses an embedded neural network. It is the strongest chess engine in the world.
Since this model is so small, is it possible to fine tune it for classification tasks like sentiment analysis and NER?
Is this really powerful for its size? And which use cases that'll be suitable for this?
Founder of Reddit ($25B)
Founder of Loom ($1B)
Execs at Uber ($200B), Tinder ($8B), and more.
You can get personalized advice from them on Intro.
No pitch decks. No cold emails.
Just real advice. 1:1. Live.
→
Blow to high value nvidia chips and large jumbo data centers ? Remember when deepseek was released on February 202 - What happened ??
using these kind of small models ( Gemma 3 270M) on a well know problem ( like chess) is a great way to learn the fine tuning of LLMs.
Quote
Akshay
@akshay_pachaar
Replying to @akshay_pachaar
Google released Gemma 3 270M, a new model for hyper-efficient local AI!
We'll fine-tune this model and make it very smart at playing chess and predict the next move.
Tech stack:
- @UnslothAI for efficient fine-tuning.
- @huggingface transformers to run it locally.
Let's go!
why the fuck are you doing parameter efficient finetuning? it's already 270m parameters smh
Impressive, but will this localized approach limit its learning potential long-term?
