stableAPY.hl

Building HyperFolio - HyperEVM Portfolio Tracker

[stableAPY.hl @stableAPY]

I just released a new feature for Hyperfolio and the Hyperfolio API: Yield.

You can now browse thousands of yield opportunities on HyperEVM across nearly 30 protocols

[stableAPY.hl @stableAPY]

https://t.co/u3awiuaCNb

[stableAPY.hl @stableAPY]

I've ran some tests on Gemma 4 12b on my RTX 3060 12gb

it seems to handle context pretty good and I have around 30tok/s almost all the way to 130k context

I asked GPT to test it out a bit for coding and comparing to Qwen 3.6 35B, and it seems quite close

I have tried the Q4_K_M from Unsloth, I want to see if the Q6 does any better quality wise

[stableAPY.hl @stableAPY]

wow this looks cool for my 3060 12gb gonna try it

[Google Gemma @googlegemma]

Meet Gemma 4 12B!

A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license.

Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇

[stableAPY.hl @stableAPY]

@cheeez42 I'm downloading the Q6 rn to see, what context did you use?

[stableAPY.hl @stableAPY]

@RW9cgvdlo Ho I'm gonna try those different levels or reasoning then

[stableAPY.hl @stableAPY]

I've ran some tests on Gemma 4 12b on my RTX 3060 12gb

it seems to handle context pretty good and I have around 30tok/s almost all the way to 130k context

I asked GPT to test it out a bit for coding and comparing to Qwen 3.6 35B, and it seems quite close

I have tried the Q4_K_M from Unsloth, I want to see if the Q6 does any better quality wise

[stableAPY.hl @stableAPY]

wow this looks cool for my 3060 12gb gonna try it

[stableAPY.hl @stableAPY]

@sami3dat that is pretty decent actually! have you used it? is the model good? based on benchmarks it looks a bit better than qwen 3.5 9B on coding

[stableAPY.hl @stableAPY]

@0xSero https://t.co/BPVUkNEgHf

Does this kind of videos gives some hints?

[stableAPY.hl @stableAPY]

@runbloc This is what I needed

[Bloc @runbloc]

We're building the Docker for Local AI.

A simple way to package and share complete AI environments including models, runtimes, dependencies, and configuration.

One command. Same setup. Run anywhere.
Free and Open-source

we’re live on Product Hunt today.

[stableAPY.hl @stableAPY]

@g0t4 Feels like most harness now are "bloated lists of tools and skills" except for pi

[stableAPY.hl @stableAPY]

27B running on my 3090 that delegate 2 code reviews to two 35B subagents running on my 3060 and 3080

pi is such a perfect harness for local AI

is this AGI?

[stableAPY.hl @stableAPY]

@Sn4eh5hWGJ @neverselling777 It does https://t.co/ZsKAv9wjcM

[maxSuffering @neverselling777]

@stableAPY yo this was a big help. i got qwen3.6 35b a3b running at around 59 t/s after consulting with grok throughout the afternoon. big big help and a huge upgrade i can tell already from qwen 3.5 9b

[stableAPY.hl @stableAPY]

@neverselling777 my maaaaan happy that it helped 🫡

[stableAPY.hl @stableAPY]

@Sn4eh5hWGJ @neverselling777 I have a Ryzen 5900x too, 64gb DDR4 and it runs at 40tok/s on my 3060 12gb so that seems feasible on a 4070

[stableAPY.hl @stableAPY]

@gtrakGT I could

[stableAPY.hl @stableAPY]

On what I read it's more about having some other model with fresh context than a really better model for reviews, but yeah I'd love to have enough compute to be able to run 3 27Bs lol

Then 35B gives back some findings and it's up to 27B to give them credit or not, it fixes some findings because they are legitimate, some it doesn't

Tbh I'm trying a bunch of setups to see if I can achieve the 3 agents I feel comfortable with for coding

[stableAPY.hl @stableAPY]

@sami3dat @softwarevlogger @summitainotes Did you ? It felt less capable than 35B but I haven't tested it properly tho

[stableAPY.hl @stableAPY]

@Dev_Jodd idk, I'd say you have to test it out and see on your specific use case

For coding 160k seems fine on the 27B and 131k for sub agents too for what I do

but if you have ultra specific needs and need that much context that won't fit and I think Qwen is max 250k something no

[stableAPY.hl @stableAPY]

@brah_ddah https://t.co/PZNAxoymQg

[stableAPY.hl @stableAPY]

@Alharari01 100% for me

pi is just so light by default, and everything is plugin you can get online or build yourself

it feels way less bloated, and I also prefer it's UI/UX

[stableAPY.hl @stableAPY]

@softwarevlogger @summitainotes Seems to do quite the job actually, it's not a big production project it's just a nice side project and this setup is doing good

27B need less Vram but can't really be offloaded like MoE it depends on your hardware tbh

[stableAPY.hl @stableAPY]

@softwarevlogger Qwen 3.6 35 A3B, unsloth version Q4_K_M quant