Top
Best
New

Posted by xenova 4 hours ago

Bonsai 27B: A 27B-Class Model that runs on a phone(prismml.com)
185 points | 54 commentspage 2
liuliu 3 hours ago|
The problem, of course, is if you run the UD_Q2 variant (Unsloth) which does only post-training, the number is pretty close to 1-bit model here and the 5% drop in tool-call is significant than it suggests in real-life use cases.
liuliu 3 hours ago|
You also need to pay close attention to BFCLv3 multi-turn result, that helps you to get a sense how frequently these quants will be in a doom loop.
h14h 1 hour ago||
I'm curious what kind of results one could get from combining the clever quantization PrismML is doing here with something like LiquidAI's antidoom:

https://github.com/Liquid4All/antidoom

luckystarr 2 hours ago||
Tried it on Android and got "!!!!!!!!!!!!!" for answers.
gunalx 1 hour ago||
The qwen models really seem to have this as a failure mode, its so annoying having a proper trace ending up in !!!!!! Garbage.
amelius 35 minutes ago||
Wait in a regular sentence, what is the probability of "!!!" being followed by "!"?

Sounds like the model is not following a proper probabilistic choice here, so maybe more a programming error than a model training error.

jldugger 9 minutes ago||
After the third !, the probability of a fourth probably skyrockets =)
verdverm 1 hour ago||
That's what happens when you quant too hard. I'm working on quant strats and evals for the same underlying qwen 27b models.

When I saw 27b on a phone, I thought not fitting, big phone, or aggressive quant. NVFP4 still takes 27G before KV cache.

syntaxing 2 hours ago||
For those curious about their demo, I’m pretty sure it’s using Locally AI (iOS only) that lmstudio acquired/aquihired a couple months ago.
thomasjb 2 hours ago||
I've been watching and waiting for this, interested to see how smart it is, as it fits with my interest of getting the smartest possible model running in 10GB of VRAM (RTX3060 that has to drive 2 monitors and run an llm)
dakolli 1 hour ago|
start saving your money.
syntaxing 2 hours ago||
I don’t know if the llama cpp implementation is wonky (and only supports the binary version) but it’s a lot slower than 35B-A3B @ Q4_KM + MTP with CPU offloading.
pulse7 2 hours ago|
Most probably not optimized yet for this model...
wy35 1 hour ago||
Entire blog post seems to be AI-generated :/
wmf 1 hour ago|
Do you think people who work on AI for a living are not going to use it?
wy35 1 hour ago||
Of course not, personally almost all of my code these days is generated.

The LLM style of writing is just very distracting to read. “It unlocks X”, “Y changes the equation”, and why is there always something shifting? Makes my eyes glaze over in an otherwise interesting post.

arjie 1 hour ago||
The text is mostly content-free. Headline + charts are enough for most HN stories.
0xbadcafebee 1 hour ago||
27B is way more than you need for a phone. Doesn't matter how much you try to compress it, it's the wrong application of the wrong tool. There are already useful tiny models that fit on phones and do basic things really well. Dumb down a big model too much and it becomes worse than a small fine-tuned model.
xyzsparetimexyz 3 hours ago||
That's awesome. What's the largest model that could fit onto a single 16gb gpu at 1.125 effects bits per weight?
Catloafdev 2 hours ago||
Doing some naive math, the F16 filesize is ~53.8gb, the 1-bit version is ~3.8gb, about 7% of the original size. The F16 size is roughly 2x param count, so that gives a rough ballpark of ~110B.
kroaton 10 minutes ago||
Which would be very interesting to test, as larger models (such as Deepseek V4 Flash or Qwen 397B) seem to compress better. Their Q2 quants are usable as is, even without the ternary compression.
drob518 49 minutes ago||
Yep, that’s the question. I asked just that when Bonsai’s first models got released. Super interesting if we can push the parameter count over 100B with 1.125 bit quantization and still keep pretty good performance versus 16-bit 100B models. That’s a definite sweet spot.
erelong 2 hours ago||
I was trying Ornith 9B locally (it's up on Ollama) which claims:

> Ornith-1.0-9B, which can be easily deployed on edge devices, matches or exceeds the performance of much larger models such as Gemma 4-31B and Qwen 3.6 35B.

https://deep-reinforce.com/ornith_1_0.html

Only tried it so much so far; it did a little better than Qwen 9B

liuliu 2 hours ago||
Note that 3.5 9B cannot do thinking (while 3.6 27B can, pretty effectively, quite verbosely).
gunalx 1 hour ago||
3.5 9B can do thinking. Its just disabled by default in its gguf chat template.
liuliu 1 hour ago||
It is disabled because it doesn't work :) Try it and see the doom loop it gets itself in.
janalsncm 2 hours ago|||
Is that a 1-bit LLM? I don’t understand the connection with this article.
erelong 2 hours ago||
Oh, I don't actually know the difference if you want to explain it

The title says it's 27B grade running on a phone and what I was comparing it to in my mind was a model that runs at 35B grade that could presumably run on a phone "better"?

edit: I asked AI for the difference and understand a little better, thanks for the heads up to learn the difference between models... I think the thing was, although ornith was created for a specific agentic purpose, it was still outperforming a previous generalist model I had running locally (so in my mind I thought it was still a better local model) - I'd like to try bonsai out if I can figure out how to run it lol

verdverm 1 hour ago||
Orinth was not impressive in my vibes testing, I just completed my first grid analysis with real evals on qwen 27b. I can now scale that grid analysis and intend to include the qwen 9b ftunes I've seen going around. They were actually a main motivation because so many claim this or that one is better, but very little in the way of evals
drob518 48 minutes ago||
I tried it, too, and it got stuck in some loops where it couldn’t recover. Shame, it was promising for the same reason as Bonsai’s models.
verdverm 40 minutes ago||
check out geyron-9b, I've only used it a bit, but seems better than orinth on vibe evals

huggingface.co/Tivaphraen/Geryon-9B-v1

drob518 11 minutes ago||
Interesting, thanks. Looking at the model card on Huggingface, it’s combining the Qwythos and Qwable fine tunes from Empero.
verdverm 4 minutes ago||
yea, it's an experiment in merging multiple fine-tuned models
Havoc 3 hours ago|
This must be some sort of unpublished app?

I can just see their image tool on the app store

smallerize 5 minutes ago||
One of the links on the sidebar goes to "Locally AI" https://apps.apple.com/us/app/locally-ai-by-lm-studio/id6741... it requires an iPhone 17 Pro or Pro Max to run the 27B model though.
Catloafdev 2 hours ago||
It's a LLM model, not a phone app.

Available on HuggingFace: https://huggingface.co/collections/prism-ml/bonsai-27b

Havoc 41 minutes ago||
Indeed.

The article is about running it on a phone though, and shows an app with their branding running this in text mode on a phone. I'm asking where can I find this app to try what is being demonstrated in this article & video? Appstore only has an image gen app by them and other MLX apps I've tried don't seem to support this model

More comments...