I built an Excel Add-In that allows my girlfriend to quickly filter 7000 paper titles and abstracts for a review paper that she is writing [1]. It uses Gemma 2 2b which is a wonderful little model that can run on her laptop CPU. It works surprisingly well for this kind of binary classification task.
The nice thing is that she can copy/paste the titles and abstracts in to two columns and write e.g. "=PROMPT(A1:B1, "If the paper studies diabetic neuropathy and stroke, return 'Include', otherwise return 'Exclude'")" and then drag down the formula across 7000 rows to bulk process the data on her own because it's just Excel. There is a gif on the readme on the Github repo that shows it.
I don't know. This paper [1] reports accuracies in the 97-98% range on a similar task with more powerful models. With Gemma 2 2b the accuracy will certainly be lower.
Y'all definitely need to cross validate a small number of samples by hand. When I did this kind of research, I would hand validate to at least P < .01.
She and one other researcher has manually classified all 7000 papers as per standard protocol. Perhaps for the next article they will measure how this tool agreed with them against them and include it in the protocol if good enough.
Looks like I'm out...
Would be great if there was a google apps script alternative. My company gave all devs linux systems and the business team operates on windows. So I always use browser based tech like Gapps script for complex sheet manipulation
I've been using Llama models to identify cookie notices on websites, for the purpose of adding filter rules to block them in EasyList Cookie. Otherwise, this is normally done by, essentially, manual volunteer reporting.
Most cookie notices turn out to be pretty similar, HTML/CSS-wise, and then you can grab their `innerText` and filter out false positives with a small LLM. I've found the 3B models have decent performance on this task, given enough prompt engineering. They do fall apart slightly around edge cases like less common languages or combined cookie notice + age restriction banners. 7B has a negligible false-positive rate without much extra cost. Either way these things are really fast and it's amazing to see reports streaming in during a crawl with no human effort required.
This is so cool thanks for sharing. I can imagine it’s not technically possible (yet?) but it would be cool if this could simply be run as a browser extension rather than running a docker container
I did actually make a rough proof-of-concept of this! One of my long-term visions is to have it running natively in-browser, and able to automatically fix site issues caused by adblocking whenever they happen.
There are a couple of WebGPU LLM platforms available that form the building blocks to accomplish this right from the browser, especially since the models are so small.
It should be possible using native messaging [1] which can call out to an external binary. The 1password extensions use that to communicate with the password manager binary.
Ha, I'm not sure the EU is prepared to handle the deluge of petitions that would ensue.
On a more serious note, this must be the first time we can quantitatively measure the impact of cookie consent legislation across the web, so maybe there's something to be explored there.
why don't you spam the companies who want your data instead? The sites can simply stop gathering your data, then they will not require to ask for consent ...
It’s the same comments on HN as always. They think EU setting up rules is somehow worse than companies breaking them. We see how the US is turning out without pesky EU restrictions :)
It has higher salaries for privileged people like senior engineers. Try making ends meet in a lower class job.
And you have (almost) free and universal healthcare in Europa, good food available everywhere, drinking water that doesn't poison you, walkable cities, good public transport, somewhat decent police and a functioning legal system. The list goes on. Does this not impact your quality of life? Do you not care about these things?
How can you have a higher quality of life as a society with higher murders, much lower life-expectancy, so many people in jail, in debt, etc.
I think there is real potential here, for smart browsing. Have the llm get the page, replace all the ads with kittens, find non-paywall versions if possible and needed, spoof fingerprint data, detect and highlight AI generated drivel, etc. The site would have no way of knowing that it wasn’t touching eyeballs. We might be able to rake back a bit of the web this way.
You probably wouldn't want to run this in real-time on every site as it'll significantly increase the load on your browser, but as long as it's possible to generate adblock filter rules, the fixes can scale to a pretty large audience.
I was thinking running it in my home lab server as a proxy, but yeah, scaling it to the browser would require some pretty strong hardware. Still, maybe in a couple of years it could be mainstream.
To me this take is like smokers complaining that the evil government is forcing the good tobacco companies to degrade the experience by adding pictures of cancer patients on cigarette packs.
I have ollama responding to SMS spam texts. I told it to feign interest in whatever the spammer is selling/buying. Each number gets its own persona, like a millennial gymbro or 19th century British gentleman.
Given the source, I'm skeptical it's not just a troll, but found this explanation [0] plausible as to why those vague spam text exists. If true, this trolling helps the spammers warm those phone numbers up.
Carriers and SMS service providers (like Twillio) obey that, no matter what service is behind.
There are stories of people replying STOP to spam, then never getting a legit SMS because the number was re-used by another service. That's because it's being blocked between the spammer and the phone.
Calling Jessica an old chap is quite a giveaway that it's a bot xD
Nice idea indeed, but I have a feeling that it's just two LLMs now conversing with each other.
Yes it’s possible, but it’s not something you can easily scale.
I had a similar project a few years back that used OSX automations and Shortcuts and Python to send a message everyday to a friend. It required you to be signed in to iMessage on your MacBook.
Than was a send operation, the reading of replies is not something I implemented, but I know there is a file somewhere that holds a history of your recent iMessages. So you would have to parse it on file update and that should give you the read operation so you can have a conversation.
Very doable in a few hours unless something dramatic changed with how the messages apps works within the last few years.
If you are willing to use Apple Shortcuts on iOS it’s pretty easy to add something that will be trigged when a message is received and can call out to a service or even use SSH to do something with the contents, including replying
For something similar with FB chat, I use Selenium and run it on the same box that the llm is running on. Using multiple personalities is really cool though. I should update mine likewise!
You realize this is going to cause carriers to allow the number to send more spam, because it looks like engagement. The best thing to do is to report the offending message to 7726 (SPAM) so the carrier can take action. You can also file complaints at the FTC and FCC websites, but that takes a bit more effort.
I have a mini PC with an n100 CPU connected to a small 7" monitor sitting on my desk, under the regular PC. I have llama 3b (q4) generating endless stories in different genres and styles. It's fun to glance over at it and read whatever it's in the middle of making. I gave llama.cpp one CPU core and it generates slow enough to just read at a normal pace, and the CPU fans don't go nuts. Totally not productive or really useful but I like it.
FORTUNE=$(fortune) && echo $FORTUNE && echo "Convert the following output of the Unix `fortune` command into a small screenplay in the style of Shakespeare: \n\n $FORTUNE" | ollama run phi4
Do you find that it actually generates varied and diverse stories? Or does it just fall into the same 3 grooves?
Last week I tried to get an LLM (one of the recent Llama models running through Groq, it was 70B I believe) to produce randomly generated prompts in a variety of styles and it kept producing cyberpunk scifi stuff. When I told it to stop doing cyberpunk scifi stuff it went completely to wild west.
You should not ever expect an LLM to actually do what you want without handholding, and randomness in particular is one of the places it fails badly. This is probably fundamental.
That said, this is also not helped by the fact that all of the default interfaces lack many essential features, so you have to build the interface yourself. Neither "clear the context on every attempt" nor "reuse the context repeatedly" will give good results, but having one context producing just one-line summaries, then fresh contexts expanding each one will do slightly less badly.
(If you actually want the LLM to do something useful, there are many more things that need to be added beyond this)
Sounds to me like you might want to reduce the Top P - that will prevent the really unlikely next tokens from ever being selected, while still providing nice randomness in the remaining next tokens so you continue to get diverse stories.
They linked to an interactive explorer that nicely shows the diversity of the dataset, and the HF repo links to the GitHub repo that has the code that generated the stories: https://github.com/lennart-finke/simple_stories_generate
So, it seems there are ways to get varied stories.
It's a 3b model so the creativity is pretty limited. What helped for me was prompting for specific stories in specific styles. I have a python script that randomizes the prompt and the writing style, including asking for specific author styles.
> Do you find that it actually generates varied and diverse stories? Or does it just fall into the same 3 grooves?
> Last week I tried to get an LLM (one of the recent Llama models running through Groq, it was 70B I believe) to produce randomly generated prompts in a variety of styles and it kept producing cyberpunk scifi stuff.
oh wow that is actually such a brilliant little use case-- really cuts to the core of the real "magic" of ai: that it can just keep running continuously. it never gets tired, and never gets tired of thinking.
I have a small fish script I use to prompt a model to generate three commit messages based off of my current git diff. I'm still playing around with which model comes up with the best messages, but usually I only use it to give me some ideas when my brain isn't working. All the models accomplish that task pretty well.
Interesting idea. But those say what’s in the commit. The commit diff already tells you that. The best commit messages IMO tell you why you did it and what value was delivered. I think it’s gonna be hard for an LLM to do that since that context lives outside the code. But maybe it would, if you hook it to e.g. a ticketing system and include relevant tickets so it can grab context.
For instance, in your first example, why was that change needed? It was a fix, but for what issue?
In the second message: why was that a desirable change?
Typically I put the "why" of the commit in the body unless it's a super simple change, but that's a good point. Sometimes this function does generate a commit body to go with the summary, and sometimes it doesn't. It also has a habit of only looking at the first file in a diff and basing its messages off of that, instead of considering the whole patch.
I'll tweak the prompt when I have some time today and see if I can get some more consistency out of it.
I disagree. If you look back and all you see are commit messages summarizing the diff, you won't get any meaningful information.
Telling me `Changed timeout from 30s to 60s` means nothing, while `Increase timeout for slow <api name> requests` gives me an actual idea of why that was done.
Even better if you add meaningful messages to the commit body.
Take a look at commits from large repositories like the Linux kernel and we can see how good commit messages looks like.
Although there are better ways to test, I used a 3B model to speed up replies from my local AI server when testing out an application I was developing. Yes I could have mocked up HTTP replies etc., but in this case the small model let me just plug in and go.
I'm working on a plugin[1] that runs local LLMs from the Godot game engine. The optimal model sizes seem to be 2B-7B ish, since those will run fast enough on most computers. We recommend that people try it out with Gemma 2 2B (but it will work with any model that works with llama.cpp)
At those sizes, it's great for generating non-repetitive flavortext for NPCs. No more "I took an arrow to the knee".
Models at around the 2B size aren't really capable enough to act a competent adversary - but they are great for something like bargaining with a shopkeeper, or some other role where natural language can let players do a bit more immersive roleplay.
This reminds me of the antics of streamer DougDoug, who often uses LLM APIs to live-summarize, analyze, or interact with his (often multi-thousand-strong) Twitch chat. Most recently I saw him do a GeoGuessr stream where he had ChatGPT assume the role of a detective who must comb through the thousands of chat messages for clues about where the chat thinks the location is, then synthesizes the clamor into a final guess. Aside from constantly being trolled by people spamming nothing but "Kyoto, Japan" in chat, it occasionaly demonstrated a pretty effective incarnation of "the wisdom of the crowd" and was strikingly accurate at times.
What approach/stack would you recommend for listening to an ongoing conversation, transcribing it and passing through llm? I had some use cases in mind but I'm not very familiar with AI frameworks and tools
I'd love to hear more about the hardware behind this project. I've had concepts for tech requiring a mic on me at all times for various reasons. Always tricky to have enough power in a reasonable DIY form factor.
It's a lightweight tool that summarizes Hacker News articles. For example, here’s what it outputs for this very post, "Ask HN: Is anyone doing anything cool with tiny language models?":
"A user inquires about the use of tiny language models for interesting applications, such as spam filtering and cookie notice detection. A developer shares their experience with using Ollama to respond to SMS spam with unique personas, like a millennial gymbro or a 19th-century British gentleman. Another user highlights the effectiveness of 3B and 7B language models for cookie notice detection, with decent performance achieved through prompt engineering."
I originally used LLaMA 3:Instruct for the backend, which performs much better, but recently started experimenting with the smaller LLaMA 3.2:1B model.
It’s been cool seeing other people’s ideas too. Curious—does anyone have suggestions for small models that are good for summaries?
That's cool, I really like it. One piece of feedback: I am usually more interested in the HN comments than in the original article. If you'd include a link to the comments then I might switch to GopherSignal as a replacement for the HN frontpage.
My flow is generally: Look at the title and the amount of upvotes to decide if I'm interested in the article. Then view the comments to see if there's interesting discussion going on or if there's already someone adding essential context. Only then I'll decide if I want to read the article or not.
Of course no big deal if you're not interested in my patronage, just wanted to let you know your page already looks good enough for me to consider switching my most visited page to it if it weren't for this small detail. And maybe the upvote count.
Hey, thanks a ton for the feedback! That was super helpful to hear about your flow—makes a lot of sense and it's pretty similar to how I browse HN too. I usually only dive into the article after checking out the upvotes and seeing what context the comments add.
I'll definitely add a link to the comments and the upvote count—gotta keep my tiny but mighty userbase (my mom, me, and hopefully you soon) happy, right? lol
And if there's even a chance you'd use GopherSignal as your daily driver, that's a no-brainer for me. Really appreciate you taking the time to share your ideas and help me improve.
Before ollama and the others could do structured JSON output, I hacked together my own loop to correct the output. I used it that for dummy API endpoints to pretend to be online services but available locally, to pair with UI mockups. For my first test I made a recipe generator and then tried to see what it would take to "jailbreak" it. I also used uncensored models to allow it to generate all kinds of funny content.
I think the content you can get from the SLMs for fake data is a lot more engaging than say the ruby ffaker library.
Microsoft published a paper on their FLAME model (60M parameters) for Excel formula repair/completion which outperformed much larger models (>100B parameters).
That paper is from over a year ago, and it compared against codex-davinci... which was basically GPT-3, from what I understand. Saying >100B makes it sound a lot more impressive than it is in today's context... 100B models today are a lot more capable. The researchers also compared against a couple of other ancient(/irrelevant today), small models that don't give me much insight.
FLAME seems like a fun little model, and 60M is truly tiny compared to other LLMs, but I have no idea how good it is in today's context, and it doesn't seem like they ever released it.
This is wild. They claim it was trained exclusively on Excel formulas, but then they mention retrieval? Is it understanding the connection between English and formulas? Or am I misunderstanding retrieval in this context?
Edit: No, the retrieval is Formula-Formula, the model (nor I believe tokenizer) does not handle English.
But I feel we're going back full circle. These small models are not generalist, thus not really LLMs at least in terms of objective. Recently there has been a rise of "specialized" models that provide lots of values, but that's not why we were sold on LLMs.
Specialized models work much better still for most stuff. Really we need an LLM to understand the input and then hand it off to a specialized model that actually provides good results.
But that's the thing, I don't need my ML model to be able to write me a sonnet about the history of beets, especially if I want to run it at home for specific tasks like as a programming assistant.
I'm fine with and prefer specialist models in most cases.
I would love a model that knows SQL really well so I don't need to remember all the small details of the language. Beyond that, I don't see why the transformer architecture can't be applied to any problem that needs to predict sequences.
I think playing word games about what really counts as an LLM is a losing battle. It has become a marketing term, mostly. It’s better to have a functionalist point of view of “what can this thing do”.
Tiny language models can do a lot if they are fine tuned for a specific task, but IMO a few things are holding them back:
1. Getting the speed gains is hard unless you are able to pay for dedicated GPUs. Some services offer LoRA as serverless but you don't get the same performance for various technical reasons.
2. Lack of talent to actually do the finetuning. Regular engineers can do a lot of LLM implementation, but when it comes to actually performing training it is a scarcer skillset. Most small to medium orgs don't have people who can do it well.
3. Distribution. Sharing finetunes is hard. HuggingFace exists, but discoverability is an issue. It is flooded with random models with no documentation and it isn't easy to find a good oen for your task. Plus, with a good finetune you also need the prompt and possibly parsing code to make it work the way it is intended and the bundling hasn't been worked out well.
when you say fine-tuning skills or talent are scarce, do you have specific skills in mind? perhaps engineering for training models (eg making model parallelism work)? or the more ML type skills of designing experiments, choosing which methods to use, figuring out datasets for training, hyperparam tuning/evaluation, etc?
The technical parts are less common and specialized, like understanding the hyperparameters and all that, but I don't think that is the main problem. Most people don't understand how to build a good dataset or how to evaluate their finetune after training. Some parts of this are solid rules like always use a separate validation set, but the task dependent parts are harder to teach. It's a different problem every time.
We (avy.ai) are using models in that range to analyze computer activity on-device, in a privacy sensitive way, to help knowledge workers as they go about their day.
The local models do things ranging from cleaning up OCR, to summarizing meetings, to estimating the user's current goals and activity, to predicting search terms, to predicting queries and actions that, if run, would help the user accomplish their current task.
The capabilities of these tiny models have really surged recently. Even small vision models are becoming useful, especially if fine tuned.
So you are using a local small model to remove identifying information and make the question generic, which is then sent to a larger model? Is that understanding correct?
I think this would have some additional benefits of not confusing the larger model with facts it doesn't need to know about. My erasing information, you can allow its attention heads to focus on the pieces that matter.
> So you are using a local small model to remove identifying information and make the question generic, which is then sent to a larger model? Is that understanding correct?
You're using it to anonymize your code, not de-anonymize someone's code. I was confused by your comment until I read the replies and realized that's what you meant to say.
I read it the other way, their code contains eg fetch(url, pw:hunter123), and they're asking Claude anonymized questions like
"implement handler for fetch(url, {pw:mycleartrxtpw})"
And then claude replies
fetch(url, {pw:mycleartrxtpw}).then(writething)
And then the local llm converts the placeholder mycleartrxtpw into hunter123 using its access to the real code
> Put in all your work related questions in the plugin, an LLM will make it as an abstract question for you to preview and send it
So the LLM does both the anonymization into placeholders and then later the replacing of the placeholders too. Calling the latter step de-anonymization is confusing though, it's "de-anonymizing" yourself to yourself. And the overall purpose of the plugin is to anonymize OP to Claude, so to me at least that makes the whole thing clearer.
Llama 3.2 has about 3.2b parameters. I have to admit, I use bigger ones like phi-4 (14.7b) and Llama 3.3 (70.6b) but I think Llama 3.2 could do de-anonimization and anonimization of code
Llama 3.2 punches way above its weight. For general "language manipulation" tasks it's good enough - and it can be used on a CPU with acceptable speed.
Are you using the model to create a key-value pair to find/replace and then reverse to reanonymize, or are you using its outputs directly? If the latter, is it fast enough and reliable enough?
I've made a tiny ~1m parameter model that can generate random Magic the Gathering cards that is largely based on Karpathy's nanogpt with a few more features added on top.
I don't have a pre-trained model to share but you can make one yourself from the git repo, assuming you have an apple silicon mac.
It also does RAG on apps there, like the music player, contacts app and to-do app. I can ask it to recommend similar artists to listen to based on my music library for example or ask it to quiz me on my PDF papers.
Not sure it qualifies, but I've started building an Android app that wraps bergamot[0] (the firefox translation models) to have on-device translation without reliance on google.
Bergamot is already used inside firefox, but I wanted translation also outside the browser.
It's different. It doesn't always just give one translation but different options. I can do things like give it a phrase and then ask it to break it down. Or give it a word and if its translation doesn't make sense to me ask how it works in the context of a phrase.
llm -c, which continues the previous conversation, is specifically useful for that sort of manipulation.
It's also available from the command line, which I find convenient because I basically always have one open.
I wonder how big that model is in RAM/disk. I use LLMs for FFMPEG all the time, and I was thinking about training a model on just the FFMPEG CLI arguments. If it was small enough, it could be a package for FFMPEG. e.g. `ffmpeg llm "Convert this MP4 into the latest royalty-free codecs in an MKV."`
Please submit a blog post to HN when you're done. I'd be curious to know the most minimal LLM setup needed get consistently sane output for FFMPEG parameters.
You can train that size of a model on ~1 billion tokens in ~3 minutes on a rented 8xH100 80GB node (~$9/hr on Lambda Labs, RunPod io, etc.) using the NanoGPT speed run repo: https://github.com/KellerJordan/modded-nanogpt
For that short of a run, you'll spend more time waiting for the node to come up, downloading the dataset, and compiling the model, though.
If you have modern hardware, you can absolutely train that at home. Or very affordable on a cloud service.
I’ve seen a number of “DIY GPT-2” tutorials that target this sweet spot. You won’t get amazing results unless you want to leave a personal computer running for a number of hours/days and you have solid data to train on locally, but fine-tuning should be in the realm of normal hobbyists patience.
Not even on the edge. That's something you could train on a 2 GB GPU.
The general guidance I've used is that to train a model, you need an amount of RAM (or VRAM) equal to 8x the number of parameters, so a 0.125B model would need 1 GB of RAM to train.
Hm... I wonder what your use case it. I do the modern Enterprise Java and the tab completion is a major time saver.
While interactive AI is all about posing, meditating on the prompt, then trying to fix the outcome, IntelliJ tab completion... shows what it will complete as you type and you Tab when you are 100% OK with the completion, which surprisingly happens 90..99% of the time for me, depending on the project.
I've had good speed / reliability with TheBloke/rocket-3B-GGUF on Huggingface, the Q2_K model. I'm sure there are better models out there now, though.
It takes ~8-10 seconds to process an image on my M2 Macbook, so not quite quick enough to run on phones yet, but the accuracy of the output has been quite good.
what are you feed into the model? Image (like product packaging) or Image of Structured Table? I found out that model performs good in general with sturctured table, but fails a lot over images.
I am using smollm2 to extract some useful information (like remote, language, role, location, etc.) from "Who is hiring" monthly thread and create an RSS feed with specific filter.
Still not ready for Show HN, but working.
I am doing nothing, but I was wondering if it would make sense to combine a small LLM and SQLITE to parse date time human expressions. For example, given a human input like "last day of this month", the LLM will generate the following query `SELECT date('now','start of month','+1 month','-1 day');`
It is probably super overengineering, considering that pretty good libraries are already doing that on different languages, but it would be funny. I did some tests with chatGPT, and it worked sometimes. It would probably work with some fine-tuning, but I don't have the experience or the time right now.
LLMs tend to REALLY get this wrong. Ask it to generate a query to sum up likes on items uploaded in the last week, defined as the last monday-sunday week (not the last 7 days), and watch it get it subtly wrong almost every time.
I built https://ffprompt.ryanseddon.com using the chrome ai (Gemini nano). Allows you to do ffmpeg operations on videos using natural language all client side.
What are the prerequisites for this? I keep getting "Bummer, looks like your device doesn't support Chrome AI" on macOS 15.2 Chrome 132.0.6834.84 (Official Build) (arm64)
[Edit] Found it. I had to enable chrome://flags/#prompt-api-for-gemini-nano
We're using small language models to detect prompt injection. Not too cool, but at least we can publish some AI-related stuff on the internet without a huge bill.
My husband and me made a stock market analysis thing that gets it right about 55% of the time, so better than a coin toss. The problem is that it keeps making unethical suggestions, so we're not using it to trade stock. Does anyone have any idea what we can do with that?
Suggestion: calculate the out-of-sample Sharpe ratio[0] of the suggestions over a reasonable period to gauge how good the model would actually perform in terms of return compared to risks. It is better than vanilla accuracy or related metrics. Source: I'm a financial economist.
I think this is a good highlight of why context and reality checks are incredibly important when doing work like this. At first glance, it might look like 55% is a really good result, but in the previous year, a flat buy every day strategy would've been right 56.7% of the time.
You can literally flip coins and get better than 50% success in a bull market. Just buy index funds and spend your time on something that isn't trying to beat entropy. You won't be able to.
I built a platform to monitor LLMs that are given complete freedom in the form of a Docker container bash REPL. Currently the models have been offline for some time because I'm upgrading from a single DELL to a TinyMiniMicro Proxmox cluster to run multiple small LLMs locally.
The bots don't do a lot of interesting stuff though, I plan to add the following functionalities:
- Instead of just resetting every 100 messages, I'm going to provide them with a rolling window of context.
- Instead of only allowing BASH commands, they will be able to also respond with reasoning messages, hopefully to make them a bit smarter.
- Give them a better docker container with more CLI tools such as curl and a working package manager.
If you're interested in seeing the developments, you can subscribe on the platform!
We are building a framework to run this tiny language model in the web so anyone can access private LLMs in their browser: https://github.com/sauravpanda/BrowserAI.
With just three lines of code, you can run Small LLM models inside the browser. We feel this unlocks a ton of potential for businesses so that they can introduce AI without fear of cost and can personalize the experience using AI.
Would love your thoughts and what we can do more or better!
I think I know what he means. I use AI Chat. I load Qwen2.5-1.5B-Instruct with llama.cpp server, fully offloaded to the CPU, and then I config AI Chat to connect to the llama.cpp endpoint.
I think one major improvement for folks like me would be human->regex LLM translator, ideally also respecting different flavors/syntax for various languages and tools.
This has been a bane of me - I run into requirement to develop some complex regexes maybe every 2-3 years, so I dig deep into specs, work on it, deliver eventually if its even possible, and within few months almost completely forget all the details and start at almost same place next time. It gets better over time but clearly I will retire earlier than this skill settles in well.
What's your workflow like? I use AI Chat. I load Qwen2.5-1.5B-Instruct with llama.cpp server, fully offloaded to the CPU, and then I config AI Chat to connect to the llama.cpp endpoint.
I use a small model to rename my Linux ISOs. I gave it a custom prompt with examples of how I want the output filenames to be structured and then just feed it files to rename. The output only works 90ish percent of the time, so I wrote a little CLI to iterate through the files and accept / retry / edit the changes the LLM outputs.
I bought a tiny business in Brazil, the database (Excel) I inherited with previous customer data *do not include gender*. I need gender to start my marketing campaigns and learn more about my future customer. I used Gemma-2B and Python to determine gender based on the data and it worked perfect
I have several rhetoric and logic books of the sort you might use for training or whatever, and one of my best friends got a doctorate in a tangential field, and may have materials and insights.
We actually just threw a relationship curative app online in 17 hours around Thanksgiving., so they "owe" me, as it were.
I'm one of those people that can do anything practical with tech and the like, but I have no imagination for it - so when someone mentions something that I think would be beneficial for my fellow humans I get this immense desire to at least cheer on if not ask to help.
I'll be very positively impressed if you make this work; I spend all day every day for work trying to make more capable models perform basic reasoning, and often failing :-P
I have this idea that a tiny LM would be good at canonicalizing entered real estate addresses. We currently buy a data set and software from Experian, but it feels like something an LM might be very good at. There are lots of weirdnesses in address entry that regexes have a hard time with. We know the bulk of addresses a user might be entering, unless it's a totally new property, so we should be able to train it on that.
Has anyone ever tried to do some automatic email workflow autoresponder agents?
Lets say, I want some outcome and it will autonomousl handle the process prompt me and the other side for additional requirements if necessary and then based on that handle the process and reach the outcome?
Apple’s on device models are
around 3B if I’m nit mistaken, and they developed some nice tech around them that they published, if I’m not
mistaken - where they have just one model, but have switchable finetunings of that model so that it can perform different functionalities depending on context.
I've been working on a self-hosted, low-latency service for small LLM's. It's basically exactly what I would have wanted when I started my previous startup. The goal is for real time applications, where even the network time to access a fast LLM like groq is an issue.
I haven't benchmarked it yet but I'd be happy to hear opinions on it. It's written in C++ (specifically not python), and is designed to be a self-contained microservice based around llama.cpp.
Kinda? All local so very much personal, non-business use. I made Ollama talk in a specific persona styles with the idea of speaking like Spider Jerusalem, when I feel like retaining some level of privacy by avoiding phrases I would normally use. Uncensored llama just rewrites my post with a specific persona's 'voice'. Works amusingly well for that purpose.
when i feel like casually listening to something, instead of netflix/hulu/whatever, i'll run a ~3b model (qwen 2.5 or llama 3.2) and generate and audio stream of water cooler office gossip. (when it is up, it runs here: https://water-cooler.jothflee.com).
some of the situations get pretty wild, for the office :)
I'm interested in finding tiny models to create workflows stringing together several function/tools and running them on device using mcp.run servlets on Android (disclaimer: I work on that)
I’m tired of the bad playlists I get from algorithms, so I made a specific playlist with an Llama2 based on several songs I like. I started with 50, removed any I didn’t like, and added more to fill in the spaces. The small models were pretty good at this. Now I have a decent fixed playlist. It does get “tired” after a few weeks and I need to add more to it. I’ve never been able to do this myself with more than a dozen songs.
Interesting! I wrote a prompt for something similar[1], but I use Claude Sonnet for it. I wonder how a small model would handle it. Time to test, I guess.
I programmed my own version of Tic Tac Toe in Godot, using a Llama 3B as the AI opponent. Not for work flow, but figuring out how to beat it is entertaining during moments of boredom.
I don’t know if this counts as tiny but I use llama 3B in prod for summarization (kinda).
Its effective context window is pretty small but I have a much more robust statistical model that handles thematic extraction. The llm is essentially just rewriting ~5-10 sentences into a single paragraph.
I’ve found the less you need the language model to actually do, the less the size/quality of the model actually matters.
I'm using ollama, llama3.2 3b, and python to shorten news article titles to 10 words or less. I have a 3 column web site with a list of news articles in the middle column. Some of the titles are too long for this format, but the shorter titles appear OK.
Is there any experiments in a small models that does paraphrasing? I tried hsing some off-the-shelf models, but it didn't go well.
I was thinking of hooking them in RPGs with text-based dialogue, so that a character will say something slightly different every time you speak to them.
I put llama 3 on a RBPi 5 and have it running a small droid. I added a TTS engine so it can hear spoken prompts which it replies to in droid speak. It also has a small screen that translates the response to English. I gave it a backstory about being a astromech droid so it usually just talks about the hyperdrive but it's fun.
chatgpt did a stellar job parsing the "books on hard things" thread from a little while ago. my prompt was:
Can you identify all the books here, sorted by a weight which is determined based on a combo of the number of votes the comment has, the number of sub-comments, or the number of repeat mentions.
We're prototyping a text firewall (for Android) with Gemma2 2B (which limits us to English), though DeepSeek's R1 variants now look pretty promising [0]: Depending on the content, we rewrite the text or quarantine it from your view. Of course this is easy (for English) in the sense that the core logic is all LLMs [1], but the integration points (on Android) are not so straight forward for anything other than SMS. [2]
A more difficult problem we forsee is to turn it into a real-time (online) firewall (for calls, for example).
I built an Excel Add-In that allows my girlfriend to quickly filter 7000 paper titles and abstracts for a review paper that she is writing [1]. It uses Gemma 2 2b which is a wonderful little model that can run on her laptop CPU. It works surprisingly well for this kind of binary classification task.
The nice thing is that she can copy/paste the titles and abstracts in to two columns and write e.g. "=PROMPT(A1:B1, "If the paper studies diabetic neuropathy and stroke, return 'Include', otherwise return 'Exclude'")" and then drag down the formula across 7000 rows to bulk process the data on her own because it's just Excel. There is a gif on the readme on the Github repo that shows it.
[1] https://github.com/getcellm/cellm
https://x.com/Suhail/status/1882069209129340963
You could have called it CellMate
How accurate are the classifications?
I don't know. This paper [1] reports accuracies in the 97-98% range on a similar task with more powerful models. With Gemma 2 2b the accuracy will certainly be lower.
[1] https://www.medrxiv.org/content/10.1101/2024.10.01.24314702v...
Y'all definitely need to cross validate a small number of samples by hand. When I did this kind of research, I would hand validate to at least P < .01.
She and one other researcher has manually classified all 7000 papers as per standard protocol. Perhaps for the next article they will measure how this tool agreed with them against them and include it in the protocol if good enough.
> I don't know.
HN in a nutshell: I've built some cool tech but have no idea if it is helpful or even counter productive...
Real HN in a nutshell: People who don't build stuff telling people who do build stuff that the thing they built is useless :P
It's a hacker forum, let people hack!
If anything have a dig at OP for posting the thread too soon before the parent commenter has had the chance to gather any data, haha
Just because you can, doesn't mean you should
Sometimes people just like to build stuff for the sake of it.
Almost like hackers, doing shit just for the heck of it because they can (mostly)
You could have called it CellMate b
Very cool idea. I’ve used gemma2 2b for a few small things. Very good model for being so small.
Requirements: -Windows
Looks like I'm out... Would be great if there was a google apps script alternative. My company gave all devs linux systems and the business team operates on windows. So I always use browser based tech like Gapps script for complex sheet manipulation
I've been using Llama models to identify cookie notices on websites, for the purpose of adding filter rules to block them in EasyList Cookie. Otherwise, this is normally done by, essentially, manual volunteer reporting.
Most cookie notices turn out to be pretty similar, HTML/CSS-wise, and then you can grab their `innerText` and filter out false positives with a small LLM. I've found the 3B models have decent performance on this task, given enough prompt engineering. They do fall apart slightly around edge cases like less common languages or combined cookie notice + age restriction banners. 7B has a negligible false-positive rate without much extra cost. Either way these things are really fast and it's amazing to see reports streaming in during a crawl with no human effort required.
Code is at https://github.com/brave/cookiemonster. You can see the prompt at https://github.com/brave/cookiemonster/blob/main/src/text-cl....
It's funny that this is even necessary though - that great EU innovation at work.
Tracking, tracking cookies, banners etc. are a choice done by the website. There are browser addons for making it simpler, though.
The transparency requirements and consent for collecting all kinds of PII actually is a great innovation.
Tangentially related, I worked on something similar, using LLMs to find and skip sponsored content in YT videos:
https://butter.sonnet.io/
This is so cool thanks for sharing. I can imagine it’s not technically possible (yet?) but it would be cool if this could simply be run as a browser extension rather than running a docker container
I did actually make a rough proof-of-concept of this! One of my long-term visions is to have it running natively in-browser, and able to automatically fix site issues caused by adblocking whenever they happen.
The PoC is a bit outdated but it's here: https://github.com/brave/cookiemonster/tree/webext
There are a couple of WebGPU LLM platforms available that form the building blocks to accomplish this right from the browser, especially since the models are so small.
https://github.com/mlc-ai/web-llm
https://huggingface.co/docs/transformers.js/en/index
You do have to worry about WebGPU compatibility in browsers though.
https://caniuse.com/webgpu
It should be possible using native messaging [1] which can call out to an external binary. The 1password extensions use that to communicate with the password manager binary.
[1] https://developer.mozilla.org/en-US/docs/Mozilla/Add-ons/Web...
Maybe it could also send automated petitions to the EU to undo cookie consent legislation, and reverse some of the enshitification.
Ha, I'm not sure the EU is prepared to handle the deluge of petitions that would ensue.
On a more serious note, this must be the first time we can quantitatively measure the impact of cookie consent legislation across the web, so maybe there's something to be explored there.
why don't you spam the companies who want your data instead? The sites can simply stop gathering your data, then they will not require to ask for consent ...
It’s the same comments on HN as always. They think EU setting up rules is somehow worse than companies breaking them. We see how the US is turning out without pesky EU restrictions :)
The US has 3x higher salaries, larger houses and a much higher quality of life?
I work as a senior engineer in Europe and make barely $4k net per month... and that's considered a "good" salary!
It has higher salaries for privileged people like senior engineers. Try making ends meet in a lower class job.
And you have (almost) free and universal healthcare in Europa, good food available everywhere, drinking water that doesn't poison you, walkable cities, good public transport, somewhat decent police and a functioning legal system. The list goes on. Does this not impact your quality of life? Do you not care about these things?
How can you have a higher quality of life as a society with higher murders, much lower life-expectancy, so many people in jail, in debt, etc.
Touch grass. The US is a big place and is nothing like you seem to think it is.
Europe on the other hand can't even manage to defend itself and relies on the US for their sheer existence.
Because they have no reason to care about what you think or feel or they wouldn't be doing it in the first place.
Cookie notices just gave them another weapon in the end.
I think there is real potential here, for smart browsing. Have the llm get the page, replace all the ads with kittens, find non-paywall versions if possible and needed, spoof fingerprint data, detect and highlight AI generated drivel, etc. The site would have no way of knowing that it wasn’t touching eyeballs. We might be able to rake back a bit of the web this way.
You probably wouldn't want to run this in real-time on every site as it'll significantly increase the load on your browser, but as long as it's possible to generate adblock filter rules, the fixes can scale to a pretty large audience.
I was thinking running it in my home lab server as a proxy, but yeah, scaling it to the browser would require some pretty strong hardware. Still, maybe in a couple of years it could be mainstream.
To me this take is like smokers complaining that the evil government is forcing the good tobacco companies to degrade the experience by adding pictures of cancer patients on cigarette packs.
Those don’t really work: https://jamanetwork.com/journals/jamanetworkopen/fullarticle...
Do they help deter people from becoming smokers in the first place?
I have ollama responding to SMS spam texts. I told it to feign interest in whatever the spammer is selling/buying. Each number gets its own persona, like a millennial gymbro or 19th century British gentleman.
http://files.widloski.com/image10%20(1).png
http://files.widloski.com/image11.png
Given the source, I'm skeptical it's not just a troll, but found this explanation [0] plausible as to why those vague spam text exists. If true, this trolling helps the spammers warm those phone numbers up.
0 - https://x.com/nikitabier/status/1867029883387580571
Why does STOP work here?
Carriers and SMS service providers (like Twillio) obey that, no matter what service is behind.
There are stories of people replying STOP to spam, then never getting a legit SMS because the number was re-used by another service. That's because it's being blocked between the spammer and the phone.
STOP works thanks to the Telephone Consumer Protection Act (“TCPA”), which offers consumers spam protections and senders a framework on how to behave.
(Edit: It's relevant that STOP didn't come from the TCPA itself, but definitely has teeth due to it)
https://www.infobip.com/blog/a-guide-to-global-sms-complianc...
https://x.com/nikitabier/status/1867069169256308766
Again, no clue if this is true, but it seems plausible.
Calling Jessica an old chap is quite a giveaway that it's a bot xD Nice idea indeed, but I have a feeling that it's just two LLMs now conversing with each other.
This is fantastic. How have your hooked up a mobile number to the llm?
Android app that forwards to a Python service on remote workstation over MQTT. I can make a Show HN if people are interested.
I am so SO interested, please make a Show HN
Why MQTT over HTTP for a low volume, small scale integration?
I’m not OP, but I would hazard a guess that those are the tools that OP has at hand.
Interested for sure.
Yes, I'd be interested in that!
I’d love to see that. Could you simulate iMessage?
Yes it’s possible, but it’s not something you can easily scale.
I had a similar project a few years back that used OSX automations and Shortcuts and Python to send a message everyday to a friend. It required you to be signed in to iMessage on your MacBook.
Than was a send operation, the reading of replies is not something I implemented, but I know there is a file somewhere that holds a history of your recent iMessages. So you would have to parse it on file update and that should give you the read operation so you can have a conversation.
Very doable in a few hours unless something dramatic changed with how the messages apps works within the last few years.
They are all in a SQLite db on your disk.
If you mean hook this into iMessage, I don't know. I'm willing to bet it's way harder though because Apple
If you are willing to use Apple Shortcuts on iOS it’s pretty easy to add something that will be trigged when a message is received and can call out to a service or even use SSH to do something with the contents, including replying
For something similar with FB chat, I use Selenium and run it on the same box that the llm is running on. Using multiple personalities is really cool though. I should update mine likewise!
You realize this is going to cause carriers to allow the number to send more spam, because it looks like engagement. The best thing to do is to report the offending message to 7726 (SPAM) so the carrier can take action. You can also file complaints at the FTC and FCC websites, but that takes a bit more effort.
Yes, the very last thing to do is respond to spam (calls, email, text...) and inform that you are eligible to more solicitation.
Cool! Do you consider the risk of unintentional (and until some moment, an unknown) subscription to some paid SMS service and how do you mitigate it?
I have to whitelist a conversation before the LLM can respond.
I love this, more please!!!
Please tell me you have a blog/archive of these somewhere. This was such a joy to read!
I have a mini PC with an n100 CPU connected to a small 7" monitor sitting on my desk, under the regular PC. I have llama 3b (q4) generating endless stories in different genres and styles. It's fun to glance over at it and read whatever it's in the middle of making. I gave llama.cpp one CPU core and it generates slow enough to just read at a normal pace, and the CPU fans don't go nuts. Totally not productive or really useful but I like it.
That's neat. I just tried something similar:
Doesn't `fortune` inside double quotes execute the command in bash? You should use single quotes instead of backticks.
Do you find that it actually generates varied and diverse stories? Or does it just fall into the same 3 grooves?
Last week I tried to get an LLM (one of the recent Llama models running through Groq, it was 70B I believe) to produce randomly generated prompts in a variety of styles and it kept producing cyberpunk scifi stuff. When I told it to stop doing cyberpunk scifi stuff it went completely to wild west.
Set temperature to 1.0
You should not ever expect an LLM to actually do what you want without handholding, and randomness in particular is one of the places it fails badly. This is probably fundamental.
That said, this is also not helped by the fact that all of the default interfaces lack many essential features, so you have to build the interface yourself. Neither "clear the context on every attempt" nor "reuse the context repeatedly" will give good results, but having one context producing just one-line summaries, then fresh contexts expanding each one will do slightly less badly.
(If you actually want the LLM to do something useful, there are many more things that need to be added beyond this)
Sounds to me like you might want to reduce the Top P - that will prevent the really unlikely next tokens from ever being selected, while still providing nice randomness in the remaining next tokens so you continue to get diverse stories.
Someone mentioned generating millions of (very short) stories with an LLM a few weeks ago: https://news.ycombinator.com/item?id=42577644
They linked to an interactive explorer that nicely shows the diversity of the dataset, and the HF repo links to the GitHub repo that has the code that generated the stories: https://github.com/lennart-finke/simple_stories_generate
So, it seems there are ways to get varied stories.
It's a 3b model so the creativity is pretty limited. What helped for me was prompting for specific stories in specific styles. I have a python script that randomizes the prompt and the writing style, including asking for specific author styles.
Generate a list of 5000 possible topics you’d like it to talk about. Randomly pick one and inject that into your prompt.
> Do you find that it actually generates varied and diverse stories? Or does it just fall into the same 3 grooves?
> Last week I tried to get an LLM (one of the recent Llama models running through Groq, it was 70B I believe) to produce randomly generated prompts in a variety of styles and it kept producing cyberpunk scifi stuff.
100% relevant: "Someday" <https://en.wikipedia.org/wiki/Someday_(short_story)> by Isaac Asimov, 1956
this sounds pretty cool, do you have any video/media of it?
I don't have a video but here's a pic of the output: https://imgur.com/ip8GWIh
oh wow that is actually such a brilliant little use case-- really cuts to the core of the real "magic" of ai: that it can just keep running continuously. it never gets tired, and never gets tired of thinking.
this is so cool, any chance you post a video?
Just this pic: https://imgur.com/ip8GWIh
That's awesome!
I have a small fish script I use to prompt a model to generate three commit messages based off of my current git diff. I'm still playing around with which model comes up with the best messages, but usually I only use it to give me some ideas when my brain isn't working. All the models accomplish that task pretty well.
Here's the script: https://github.com/nozzlegear/dotfiles/blob/master/fish-func...
And for this change [1] it generated these messages:
[1] https://github.com/nozzlegear/dotfiles/commit/0db65054524d0d...Interesting idea. But those say what’s in the commit. The commit diff already tells you that. The best commit messages IMO tell you why you did it and what value was delivered. I think it’s gonna be hard for an LLM to do that since that context lives outside the code. But maybe it would, if you hook it to e.g. a ticketing system and include relevant tickets so it can grab context.
For instance, in your first example, why was that change needed? It was a fix, but for what issue?
In the second message: why was that a desirable change?
Typically I put the "why" of the commit in the body unless it's a super simple change, but that's a good point. Sometimes this function does generate a commit body to go with the summary, and sometimes it doesn't. It also has a habit of only looking at the first file in a diff and basing its messages off of that, instead of considering the whole patch.
I'll tweak the prompt when I have some time today and see if I can get some more consistency out of it.
Most of the time you are not able to fit the "Why?" in the summary.
That's what the body of the commit message is for.
I disagree. When you look at the git history in x months you're gonna have a hard time understanding what was done following your example.
I disagree. If you look back and all you see are commit messages summarizing the diff, you won't get any meaningful information.
Telling me `Changed timeout from 30s to 60s` means nothing, while `Increase timeout for slow <api name> requests` gives me an actual idea of why that was done.
Even better if you add meaningful messages to the commit body.
Take a look at commits from large repositories like the Linux kernel and we can see how good commit messages looks like.
By adding more context? I’m not sure who you’re replying to or what your objection is.
Those commit messages are pretty terrible, please try to come up with actual messages ;)
Awesome need to make one for naming variables too haha
Although there are better ways to test, I used a 3B model to speed up replies from my local AI server when testing out an application I was developing. Yes I could have mocked up HTTP replies etc., but in this case the small model let me just plug in and go.
I'm working on a plugin[1] that runs local LLMs from the Godot game engine. The optimal model sizes seem to be 2B-7B ish, since those will run fast enough on most computers. We recommend that people try it out with Gemma 2 2B (but it will work with any model that works with llama.cpp)
At those sizes, it's great for generating non-repetitive flavortext for NPCs. No more "I took an arrow to the knee".
Models at around the 2B size aren't really capable enough to act a competent adversary - but they are great for something like bargaining with a shopkeeper, or some other role where natural language can let players do a bit more immersive roleplay.
[1] https://github.com/nobodywho-ooo/nobodywho
We fine-tuned a Gemma 2B to identify urgent messages sent by new and expecting mothers on a government-run maternal health helpline.
https://idinsight.github.io/tech-blog/blog/enhancing_materna...
lovely application!
Genuine question: why not use (Modern)BERT instead for classification? (Is the json-output explanation so critical?)
Such a fun thread but this is the kind of applications that perk up my attention!
Very cool!
Oh that is a nice writeup. We have something similar in mind at work. Will forward it.
I have a tiny device that listens to conversations between two people or more and constantly tries to declare a "winner"
This reminds me of the antics of streamer DougDoug, who often uses LLM APIs to live-summarize, analyze, or interact with his (often multi-thousand-strong) Twitch chat. Most recently I saw him do a GeoGuessr stream where he had ChatGPT assume the role of a detective who must comb through the thousands of chat messages for clues about where the chat thinks the location is, then synthesizes the clamor into a final guess. Aside from constantly being trolled by people spamming nothing but "Kyoto, Japan" in chat, it occasionaly demonstrated a pretty effective incarnation of "the wisdom of the crowd" and was strikingly accurate at times.
I love that there's not even a vague idea of the winner "metric" in your explanation. Like it's just, _the_ winner.
Are you raising a funding round? I'm bought in. This is hilarious.
This made me actually laugh out loud. Can you share more details on hardware and models used?
This is a product I want
What approach/stack would you recommend for listening to an ongoing conversation, transcribing it and passing through llm? I had some use cases in mind but I'm not very familiar with AI frameworks and tools
I'd love to hear more about the hardware behind this project. I've had concepts for tech requiring a mic on me at all times for various reasons. Always tricky to have enough power in a reasonable DIY form factor.
wifey always wins. ;)
You can use the model to generate winning speeches also.
what model do you use for speech to text?
All computation on device?
Your SO must really love that lmao
https://gophersignal.com – I built GopherSignal!
It's a lightweight tool that summarizes Hacker News articles. For example, here’s what it outputs for this very post, "Ask HN: Is anyone doing anything cool with tiny language models?":
"A user inquires about the use of tiny language models for interesting applications, such as spam filtering and cookie notice detection. A developer shares their experience with using Ollama to respond to SMS spam with unique personas, like a millennial gymbro or a 19th-century British gentleman. Another user highlights the effectiveness of 3B and 7B language models for cookie notice detection, with decent performance achieved through prompt engineering."
I originally used LLaMA 3:Instruct for the backend, which performs much better, but recently started experimenting with the smaller LLaMA 3.2:1B model.
It’s been cool seeing other people’s ideas too. Curious—does anyone have suggestions for small models that are good for summaries?
Feel free to check it out or make changes: https://github.com/k-zehnder/gophersignal
That's cool, I really like it. One piece of feedback: I am usually more interested in the HN comments than in the original article. If you'd include a link to the comments then I might switch to GopherSignal as a replacement for the HN frontpage.
My flow is generally: Look at the title and the amount of upvotes to decide if I'm interested in the article. Then view the comments to see if there's interesting discussion going on or if there's already someone adding essential context. Only then I'll decide if I want to read the article or not.
Of course no big deal if you're not interested in my patronage, just wanted to let you know your page already looks good enough for me to consider switching my most visited page to it if it weren't for this small detail. And maybe the upvote count.
Hey, thanks a ton for the feedback! That was super helpful to hear about your flow—makes a lot of sense and it's pretty similar to how I browse HN too. I usually only dive into the article after checking out the upvotes and seeing what context the comments add.
I'll definitely add a link to the comments and the upvote count—gotta keep my tiny but mighty userbase (my mom, me, and hopefully you soon) happy, right? lol
And if there's even a chance you'd use GopherSignal as your daily driver, that's a no-brainer for me. Really appreciate you taking the time to share your ideas and help me improve.
May be even rate each post on the comments activity level.
Agreed..great suggestions. Id consider switching as well.
Micro Wake Word is a library and set of on device models for ESPs to wake on a spoken wake word. https://github.com/kahrendt/microWakeWord
Recently deployed in Home Assistants fully local capable Alexa replacement. https://www.home-assistant.io/voice_control/about_wake_word/
Nice idea.
Make sure your meeting participants know you’re transcribing them. Has similar notification requirements as recording state to state.
"Comedy Writing With Small Generative Models" by Jamie Brew (Strange Loop 2023)
https://m.youtube.com/watch?v=M2o4f_2L0No
Spend the 45 minutes watching this talk. It is a delight. If you are unsure, wait until the speaker picks up the guitar.
Seconded! This was my favorite talk at Strange Loop (including my own).
Before ollama and the others could do structured JSON output, I hacked together my own loop to correct the output. I used it that for dummy API endpoints to pretend to be online services but available locally, to pair with UI mockups. For my first test I made a recipe generator and then tried to see what it would take to "jailbreak" it. I also used uncensored models to allow it to generate all kinds of funny content.
I think the content you can get from the SLMs for fake data is a lot more engaging than say the ruby ffaker library.
Microsoft published a paper on their FLAME model (60M parameters) for Excel formula repair/completion which outperformed much larger models (>100B parameters).
https://arxiv.org/abs/2301.13779
That paper is from over a year ago, and it compared against codex-davinci... which was basically GPT-3, from what I understand. Saying >100B makes it sound a lot more impressive than it is in today's context... 100B models today are a lot more capable. The researchers also compared against a couple of other ancient(/irrelevant today), small models that don't give me much insight.
FLAME seems like a fun little model, and 60M is truly tiny compared to other LLMs, but I have no idea how good it is in today's context, and it doesn't seem like they ever released it.
This is wild. They claim it was trained exclusively on Excel formulas, but then they mention retrieval? Is it understanding the connection between English and formulas? Or am I misunderstanding retrieval in this context?
Edit: No, the retrieval is Formula-Formula, the model (nor I believe tokenizer) does not handle English.
But I feel we're going back full circle. These small models are not generalist, thus not really LLMs at least in terms of objective. Recently there has been a rise of "specialized" models that provide lots of values, but that's not why we were sold on LLMs.
Specialized models work much better still for most stuff. Really we need an LLM to understand the input and then hand it off to a specialized model that actually provides good results.
But that's the thing, I don't need my ML model to be able to write me a sonnet about the history of beets, especially if I want to run it at home for specific tasks like as a programming assistant.
I'm fine with and prefer specialist models in most cases.
I would love a model that knows SQL really well so I don't need to remember all the small details of the language. Beyond that, I don't see why the transformer architecture can't be applied to any problem that needs to predict sequences.
The trick is to find such problems with enough training data and some market potential. I am terrible at it.
I think playing word games about what really counts as an LLM is a losing battle. It has become a marketing term, mostly. It’s better to have a functionalist point of view of “what can this thing do”.
This is really cool. Is this already in Excel?
Tiny language models can do a lot if they are fine tuned for a specific task, but IMO a few things are holding them back:
1. Getting the speed gains is hard unless you are able to pay for dedicated GPUs. Some services offer LoRA as serverless but you don't get the same performance for various technical reasons.
2. Lack of talent to actually do the finetuning. Regular engineers can do a lot of LLM implementation, but when it comes to actually performing training it is a scarcer skillset. Most small to medium orgs don't have people who can do it well.
3. Distribution. Sharing finetunes is hard. HuggingFace exists, but discoverability is an issue. It is flooded with random models with no documentation and it isn't easy to find a good oen for your task. Plus, with a good finetune you also need the prompt and possibly parsing code to make it work the way it is intended and the bundling hasn't been worked out well.
when you say fine-tuning skills or talent are scarce, do you have specific skills in mind? perhaps engineering for training models (eg making model parallelism work)? or the more ML type skills of designing experiments, choosing which methods to use, figuring out datasets for training, hyperparam tuning/evaluation, etc?
The technical parts are less common and specialized, like understanding the hyperparameters and all that, but I don't think that is the main problem. Most people don't understand how to build a good dataset or how to evaluate their finetune after training. Some parts of this are solid rules like always use a separate validation set, but the task dependent parts are harder to teach. It's a different problem every time.
We (avy.ai) are using models in that range to analyze computer activity on-device, in a privacy sensitive way, to help knowledge workers as they go about their day.
The local models do things ranging from cleaning up OCR, to summarizing meetings, to estimating the user's current goals and activity, to predicting search terms, to predicting queries and actions that, if run, would help the user accomplish their current task.
The capabilities of these tiny models have really surged recently. Even small vision models are becoming useful, especially if fine tuned.
Is this along the lines of rewind.ai, MSCopilot, screenpipe, or something else entirely?
I simply use it to de-anonymize code that I typed in via Claude
Maybe should write a plugin for it (open source):
1. Put in all your work related questions in the plugin, an LLM will make it as an abstract question for you to preview and send it
2. And then get the answer with all the data back
E.g. df[“cookie_company_name”] becomes df[“a”] and back
So you are using a local small model to remove identifying information and make the question generic, which is then sent to a larger model? Is that understanding correct?
I think this would have some additional benefits of not confusing the larger model with facts it doesn't need to know about. My erasing information, you can allow its attention heads to focus on the pieces that matter.
Requires further study.
> So you are using a local small model to remove identifying information and make the question generic, which is then sent to a larger model? Is that understanding correct?
Yep that's it
You're using it to anonymize your code, not de-anonymize someone's code. I was confused by your comment until I read the replies and realized that's what you meant to say.
I read it the other way, their code contains eg fetch(url, pw:hunter123), and they're asking Claude anonymized questions like "implement handler for fetch(url, {pw:mycleartrxtpw})"
And then claude replies
fetch(url, {pw:mycleartrxtpw}).then(writething)
And then the local llm converts the placeholder mycleartrxtpw into hunter123 using its access to the real code
It's that yea
Flow would be:
1. Llama prompt: write a console log statement with my username and password: mettamage, superdupersecret
2. Claude prompt (edited by Llama): write a console log statement with my username and password: asdfhjk, sdjkfa
3. Claude replies: console.log('asdfhjk', 'sdjkfa')
4. Llama gets that input and replies to me: console.log('mettamage', 'superdupersecret')
> Put in all your work related questions in the plugin, an LLM will make it as an abstract question for you to preview and send it
So the LLM does both the anonymization into placeholders and then later the replacing of the placeholders too. Calling the latter step de-anonymization is confusing though, it's "de-anonymizing" yourself to yourself. And the overall purpose of the plugin is to anonymize OP to Claude, so to me at least that makes the whole thing clearer.
I could've been a bit more clear, sorry about that.
Could you recommend a tiny language model I could try out locally?
Llama 3.2 has about 3.2b parameters. I have to admit, I use bigger ones like phi-4 (14.7b) and Llama 3.3 (70.6b) but I think Llama 3.2 could do de-anonimization and anonimization of code
Llama 3.2 punches way above its weight. For general "language manipulation" tasks it's good enough - and it can be used on a CPU with acceptable speed.
How many tokens/s?
10-15t/s on 12400 with ddr5
+1 this idea. I do the same. Just do it locally using ollama, also using 3.2 3b
Are you using the model to create a key-value pair to find/replace and then reverse to reanonymize, or are you using its outputs directly? If the latter, is it fast enough and reliable enough?
I've made a tiny ~1m parameter model that can generate random Magic the Gathering cards that is largely based on Karpathy's nanogpt with a few more features added on top.
I don't have a pre-trained model to share but you can make one yourself from the git repo, assuming you have an apple silicon mac.
https://github.com/jlwitthuhn/TCGGPT
I have it running on a Raspberry Pi 5 for offline chat and RAG. I wrote this open-source code for it: https://github.com/persys-ai/persys
It also does RAG on apps there, like the music player, contacts app and to-do app. I can ask it to recommend similar artists to listen to based on my music library for example or ask it to quiz me on my PDF papers.
Does https://github.com/persys-ai/persys-server run on the rpi?
Is that design 3d printable? Or is that for paid users only.
I can publish it no problem. I’ll create a new repo with instructions for the hardware with CAD files.
Designing a new one for the NVIDIA Orin Nano Super so it might take a few days.
Up to you! Totally understand if you want to hold something back for a paid option!
Not sure it qualifies, but I've started building an Android app that wraps bergamot[0] (the firefox translation models) to have on-device translation without reliance on google.
Bergamot is already used inside firefox, but I wanted translation also outside the browser.
[0]: bergamot https://github.com/browsermt/bergamot-translator
I would be very interested if someone is aware of any small/tiny models to perform OCR, so the app can translate pictures as well
MiniCPM-V 2.6 isn't that small (8b) but it can do this.
Here is a demo.
* https://i.imgur.com/pAuTeAf.jpeg
Using this script:
* https://github.com/jabberjabberjabber/LLMOCR/
I made a shell alias to translate things from French to English, does that count?
Llama 3.2:3b is a fine French-English dictionary IMHO.Is it better than translatelocally? https://translatelocally.com/downloads/ (the same as used in firefox)
It's different. It doesn't always just give one translation but different options. I can do things like give it a phrase and then ask it to break it down. Or give it a word and if its translation doesn't make sense to me ask how it works in the context of a phrase.
llm -c, which continues the previous conversation, is specifically useful for that sort of manipulation.
It's also available from the command line, which I find convenient because I basically always have one open.
I used local LLMs via Ollama for generating H1's / marketing copy.
1. Create several different personas
2. Generate a ton of variation using a high temperature
3. Compare the variagtions head-to-head using the LLM to get a win / loss ratio
The best ones can be quite good.
0 - https://www.definite.app/blog/overkillm
What is an H1?
Main heading of an article
Not the OP, but they are "Headers". Probably coming from the <h1> tag in html. What outsiders probably call "Headlines".
clever name!
JetBrains' local single-line autocomplete model is 0.1B (w/ 1536-token context, ~170 lines of code): https://blog.jetbrains.com/blog/2024/04/04/full-line-code-co...
For context, GPT-2-small is 0.124B params (w/ 1024-token context).
I wonder how big that model is in RAM/disk. I use LLMs for FFMPEG all the time, and I was thinking about training a model on just the FFMPEG CLI arguments. If it was small enough, it could be a package for FFMPEG. e.g. `ffmpeg llm "Convert this MP4 into the latest royalty-free codecs in an MKV."`
Please submit a blog post to HN when you're done. I'd be curious to know the most minimal LLM setup needed get consistently sane output for FFMPEG parameters.
the jetbrains models are about 70MB zipped on disk (one model per language)
That’s a great idea, but I feel like it might be hard to get it to be correct enough
from a few days ago: https://news.ycombinator.com/item?id=42706637
You can train that size of a model on ~1 billion tokens in ~3 minutes on a rented 8xH100 80GB node (~$9/hr on Lambda Labs, RunPod io, etc.) using the NanoGPT speed run repo: https://github.com/KellerJordan/modded-nanogpt
For that short of a run, you'll spend more time waiting for the node to come up, downloading the dataset, and compiling the model, though.
That size is on the edge of something you can train at home
If you have modern hardware, you can absolutely train that at home. Or very affordable on a cloud service.
I’ve seen a number of “DIY GPT-2” tutorials that target this sweet spot. You won’t get amazing results unless you want to leave a personal computer running for a number of hours/days and you have solid data to train on locally, but fine-tuning should be in the realm of normal hobbyists patience.
Hmm is there anything reasonably ready made* for this spot? Training and querying a llm locally on an existing codebase?
* I don't mind compiling it myself but i'd rather not write it.
Not even on the edge. That's something you could train on a 2 GB GPU.
The general guidance I've used is that to train a model, you need an amount of RAM (or VRAM) equal to 8x the number of parameters, so a 0.125B model would need 1 GB of RAM to train.
Is that why their tab completion is so bad now?
Hm... I wonder what your use case it. I do the modern Enterprise Java and the tab completion is a major time saver.
While interactive AI is all about posing, meditating on the prompt, then trying to fix the outcome, IntelliJ tab completion... shows what it will complete as you type and you Tab when you are 100% OK with the completion, which surprisingly happens 90..99% of the time for me, depending on the project.
I used a small (3b, I think) model plus tesseract.js to perform OCR on an image of a nutritional facts table and output structured JSON.
What was the model? What kind of performance did you get out of it?
Could you share a link to your project, if it is public?
https://github.com/JLCarveth/nutrition-llama
I've had good speed / reliability with TheBloke/rocket-3B-GGUF on Huggingface, the Q2_K model. I'm sure there are better models out there now, though.
It takes ~8-10 seconds to process an image on my M2 Macbook, so not quite quick enough to run on phones yet, but the accuracy of the output has been quite good.
what are you feed into the model? Image (like product packaging) or Image of Structured Table? I found out that model performs good in general with sturctured table, but fails a lot over images.
All that server side or client?
I am using smollm2 to extract some useful information (like remote, language, role, location, etc.) from "Who is hiring" monthly thread and create an RSS feed with specific filter. Still not ready for Show HN, but working.
I am doing nothing, but I was wondering if it would make sense to combine a small LLM and SQLITE to parse date time human expressions. For example, given a human input like "last day of this month", the LLM will generate the following query `SELECT date('now','start of month','+1 month','-1 day');`
It is probably super overengineering, considering that pretty good libraries are already doing that on different languages, but it would be funny. I did some tests with chatGPT, and it worked sometimes. It would probably work with some fine-tuning, but I don't have the experience or the time right now.
LLMs tend to REALLY get this wrong. Ask it to generate a query to sum up likes on items uploaded in the last week, defined as the last monday-sunday week (not the last 7 days), and watch it get it subtly wrong almost every time.
What libraries have you seen that do this?
I built https://ffprompt.ryanseddon.com using the chrome ai (Gemini nano). Allows you to do ffmpeg operations on videos using natural language all client side.
What are the prerequisites for this? I keep getting "Bummer, looks like your device doesn't support Chrome AI" on macOS 15.2 Chrome 132.0.6834.84 (Official Build) (arm64)
[Edit] Found it. I had to enable chrome://flags/#prompt-api-for-gemini-nano
We're using small language models to detect prompt injection. Not too cool, but at least we can publish some AI-related stuff on the internet without a huge bill.
What kind of prompt injection attacks do you filter out? Have you tested with a prompt tuning framework?
My husband and me made a stock market analysis thing that gets it right about 55% of the time, so better than a coin toss. The problem is that it keeps making unethical suggestions, so we're not using it to trade stock. Does anyone have any idea what we can do with that?
Suggestion: calculate the out-of-sample Sharpe ratio[0] of the suggestions over a reasonable period to gauge how good the model would actually perform in terms of return compared to risks. It is better than vanilla accuracy or related metrics. Source: I'm a financial economist.
[0]: https://en.wikipedia.org/wiki/Sharpe_ratio
thank you! that's exactly the sort of thing I don't know.
Have you backtested this in times when markets were not constantly green? Nearly any strategy is good in the good times.
yep. the 55% is over a few years.
Right, but if 55% is avg over the last few years, “buy stock” is going to be correct more than not.
https://www.crestmontresearch.com/docs/Stock-Yo-Yo.pdf
I think this is a good highlight of why context and reality checks are incredibly important when doing work like this. At first glance, it might look like 55% is a really good result, but in the previous year, a flat buy every day strategy would've been right 56.7% of the time.
I'm curious what sort of unethical suggestions it's coming up with haha
so far, mostly buying companies owned/ran by horrible people.
Can't you adjust the prompt to filter out companies that fund genocide etc?
So if you filter out the Republican owned ones or whatever your bugbear is, does the 55% persist?
You can literally flip coins and get better than 50% success in a bull market. Just buy index funds and spend your time on something that isn't trying to beat entropy. You won't be able to.
INSUFFICIENT DATA FOR A MEANINGFUL ANSWER.
What data do you analyze?
I built a platform to monitor LLMs that are given complete freedom in the form of a Docker container bash REPL. Currently the models have been offline for some time because I'm upgrading from a single DELL to a TinyMiniMicro Proxmox cluster to run multiple small LLMs locally.
The bots don't do a lot of interesting stuff though, I plan to add the following functionalities:
- Instead of just resetting every 100 messages, I'm going to provide them with a rolling window of context.
- Instead of only allowing BASH commands, they will be able to also respond with reasoning messages, hopefully to make them a bit smarter.
- Give them a better docker container with more CLI tools such as curl and a working package manager.
If you're interested in seeing the developments, you can subscribe on the platform!
https://lama.garden
We are building a framework to run this tiny language model in the web so anyone can access private LLMs in their browser: https://github.com/sauravpanda/BrowserAI.
With just three lines of code, you can run Small LLM models inside the browser. We feel this unlocks a ton of potential for businesses so that they can introduce AI without fear of cost and can personalize the experience using AI.
Would love your thoughts and what we can do more or better!
Sounds cool. Anyway I can help.
No, but I use llama 3.2 1b and qwen2.5 1.5 as bash oneliner generator, always runnimg in console.
Could you elaborate?
I think I know what he means. I use AI Chat. I load Qwen2.5-1.5B-Instruct with llama.cpp server, fully offloaded to the CPU, and then I config AI Chat to connect to the llama.cpp endpoint.
Checkout the demo they have below
https://github.com/sigoden/aichat#shell-assistant
I just run llama-cli with the model. Every time I want some "awk" or "find" trickery, I just ask model. Good for throwaway python scripts too.
Can it do 'sed'?
I think one major improvement for folks like me would be human->regex LLM translator, ideally also respecting different flavors/syntax for various languages and tools.
This has been a bane of me - I run into requirement to develop some complex regexes maybe every 2-3 years, so I dig deep into specs, work on it, deliver eventually if its even possible, and within few months almost completely forget all the details and start at almost same place next time. It gets better over time but clearly I will retire earlier than this skill settles in well.
What's your workflow like? I use AI Chat. I load Qwen2.5-1.5B-Instruct with llama.cpp server, fully offloaded to the CPU, and then I config AI Chat to connect to the llama.cpp endpoint.
I use a small model to rename my Linux ISOs. I gave it a custom prompt with examples of how I want the output filenames to be structured and then just feed it files to rename. The output only works 90ish percent of the time, so I wrote a little CLI to iterate through the files and accept / retry / edit the changes the LLM outputs.
I bought a tiny business in Brazil, the database (Excel) I inherited with previous customer data *do not include gender*. I need gender to start my marketing campaigns and learn more about my future customer. I used Gemma-2B and Python to determine gender based on the data and it worked perfect
How did you verify it worked?
I'm playing with the idea of identifying logical fallacies stated by live broadcasters.
even better, podcasters probably easier to fetch the data as well
LLMs are notoriously unreliable with mathematics and logic. I wish you the best of luck, because this would nevertheless be an awesome tool to have.
I have several rhetoric and logic books of the sort you might use for training or whatever, and one of my best friends got a doctorate in a tangential field, and may have materials and insights.
We actually just threw a relationship curative app online in 17 hours around Thanksgiving., so they "owe" me, as it were.
I'm one of those people that can do anything practical with tech and the like, but I have no imagination for it - so when someone mentions something that I think would be beneficial for my fellow humans I get this immense desire to at least cheer on if not ask to help.
Automation to identify logical/rhetorical fallacies is a long held dream of mine, would love to follow along with this project if it picks up somehow
That's fantastic and I'd love to help
So far not much beyond this list of targets to identify https://en.wikipedia.org/wiki/List_of_fallacies
I'll be very positively impressed if you make this work; I spend all day every day for work trying to make more capable models perform basic reasoning, and often failing :-P
I think I am. At least I think I'm building things that will enable much smaller models: https://github.com/jmward01/lmplay/wiki/Sacrificial-Training
I'm making an agent that takes decompiled code and tries to understand the methods and replace variables and function names one at a time.
This sounds cool! Are you planningto opensource it?
I have this idea that a tiny LM would be good at canonicalizing entered real estate addresses. We currently buy a data set and software from Experian, but it feels like something an LM might be very good at. There are lots of weirdnesses in address entry that regexes have a hard time with. We know the bulk of addresses a user might be entering, unless it's a totally new property, so we should be able to train it on that.
Has anyone ever tried to do some automatic email workflow autoresponder agents?
Lets say, I want some outcome and it will autonomousl handle the process prompt me and the other side for additional requirements if necessary and then based on that handle the process and reach the outcome?
I am, in a way by using EHR/EMR data for fine tuning so agents can query each other for medical records in a HIPPA compliant manner.
Apple’s on device models are around 3B if I’m nit mistaken, and they developed some nice tech around them that they published, if I’m not mistaken - where they have just one model, but have switchable finetunings of that model so that it can perform different functionalities depending on context.
I've been working on a self-hosted, low-latency service for small LLM's. It's basically exactly what I would have wanted when I started my previous startup. The goal is for real time applications, where even the network time to access a fast LLM like groq is an issue.
I haven't benchmarked it yet but I'd be happy to hear opinions on it. It's written in C++ (specifically not python), and is designed to be a self-contained microservice based around llama.cpp.
https://github.com/thansen0/fastllm.cpp
Kinda? All local so very much personal, non-business use. I made Ollama talk in a specific persona styles with the idea of speaking like Spider Jerusalem, when I feel like retaining some level of privacy by avoiding phrases I would normally use. Uncensored llama just rewrites my post with a specific persona's 'voice'. Works amusingly well for that purpose.
when i feel like casually listening to something, instead of netflix/hulu/whatever, i'll run a ~3b model (qwen 2.5 or llama 3.2) and generate and audio stream of water cooler office gossip. (when it is up, it runs here: https://water-cooler.jothflee.com).
some of the situations get pretty wild, for the office :)
What prompt are you using for this?
Using llama 3.2 as an interface to a robot. If you can get the latency down, it works wonderfully
Would love to see this applied to a FPS bot in unreal engine.
I'm interested in finding tiny models to create workflows stringing together several function/tools and running them on device using mcp.run servlets on Android (disclaimer: I work on that)
I had an LLM create a playlist for me.
I’m tired of the bad playlists I get from algorithms, so I made a specific playlist with an Llama2 based on several songs I like. I started with 50, removed any I didn’t like, and added more to fill in the spaces. The small models were pretty good at this. Now I have a decent fixed playlist. It does get “tired” after a few weeks and I need to add more to it. I’ve never been able to do this myself with more than a dozen songs.
Interesting! I wrote a prompt for something similar[1], but I use Claude Sonnet for it. I wonder how a small model would handle it. Time to test, I guess.
[1]: https://git.sr.ht/~jamesponddotco/llm-prompts/tree/trunk/dat...
Huh, interesting. For me that often dreamed up artist and songs.
Interesting! I've sadly found more capable models to really fail on music recommendations for me.
I programmed my own version of Tic Tac Toe in Godot, using a Llama 3B as the AI opponent. Not for work flow, but figuring out how to beat it is entertaining during moments of boredom.
Number of players: zero
U.S. FIRST STRIKE WINNER: NONE
USSR FIRST STRIKE WINNER: NONE
NATO / WARSAW PACT WINNER: NONE
FAR EAST STRATEGY WINNER: NONE
US USSR ESCALATION WINNER: NONE
MIDDLE EAST WAR WINNER: NONE
USSR CHINA ATTACK WINNER: NONE
INDIA PAKISTAN WAR WINNER: NONE
MEDITERRANEAN WAR WINNER: NONE
HONGKONG VARIANT WINNER: NONE
Strange game. The only winning move is not to play
I don’t know if this counts as tiny but I use llama 3B in prod for summarization (kinda).
Its effective context window is pretty small but I have a much more robust statistical model that handles thematic extraction. The llm is essentially just rewriting ~5-10 sentences into a single paragraph.
I’ve found the less you need the language model to actually do, the less the size/quality of the model actually matters.
Many interesting projects, cool. I'm waiting to LLMs in games. That would make them much more fun. Any time now...
I'm using ollama for parsing and categorizing scraped jobs for a local job board dashboard I check everyday.
I'm using ollama, llama3.2 3b, and python to shorten news article titles to 10 words or less. I have a 3 column web site with a list of news articles in the middle column. Some of the titles are too long for this format, but the shorter titles appear OK.
Is there any experiments in a small models that does paraphrasing? I tried hsing some off-the-shelf models, but it didn't go well.
I was thinking of hooking them in RPGs with text-based dialogue, so that a character will say something slightly different every time you speak to them.
Intuitively this sounds like something that should be possible using almost any llm. This should be just a matter of prompting.
I put llama 3 on a RBPi 5 and have it running a small droid. I added a TTS engine so it can hear spoken prompts which it replies to in droid speak. It also has a small screen that translates the response to English. I gave it a backstory about being a astromech droid so it usually just talks about the hyperdrive but it's fun.
I copied all the text from this post and used an LLM to generate a list of all the ideas. I do the same for other similar HN post .
well, what are the ideas?
chatgpt did a stellar job parsing the "books on hard things" thread from a little while ago. my prompt was:
Can you identify all the books here, sorted by a weight which is determined based on a combo of the number of votes the comment has, the number of sub-comments, or the number of repeat mentions.
Ideally retain hyperlinks if possible.
I'm working on using them for agentic voice commands of a limited scope.
My needs are narrow and limited but I want a bit of flexibility.
We're prototyping a text firewall (for Android) with Gemma2 2B (which limits us to English), though DeepSeek's R1 variants now look pretty promising [0]: Depending on the content, we rewrite the text or quarantine it from your view. Of course this is easy (for English) in the sense that the core logic is all LLMs [1], but the integration points (on Android) are not so straight forward for anything other than SMS. [2]
A more difficult problem we forsee is to turn it into a real-time (online) firewall (for calls, for example).
[1] https://chat.deepseek.com/a/chat/s/d5aeeda1-fefe-4fc6-8c90-2...
[1] MediaPipe in particular makes it simple to prototype around Gemma2 on Android: https://ai.google.dev/edge/mediapipe/solutions/genai/llm_inf...
[2] Intend to open source it once we get it working for anything other than SMSes
Pretty sure they are mostly used as fine tuning targets, rather than as-is.
But for what purposes?
How accurate? are the classifications?