Sabitlenmiş Tweet
Adhyaay Karnwal
1.8K posts

Adhyaay Karnwal
@AdhyaayK
15 and working on https://t.co/ktKproOFh2
Katılım Nisan 2025
237 Takip Edilen118 Takipçiler

This is the most hilarious thing I saw and did today
Ran gemma-4-12B-coder-fable5-composer2.5-v1-GGUF locally with 8 GB VRAM at 20+ tok/sec
Anthropic's Claude Fable 5 launched June 9.
By June 12 it was banned. I can't access it. You can't either.
But here's the twist: I'm running a model trained on its chain of thought at 20 tok/s on my RTX 4060 8GB.
Locally. Offline. No cloud. No export control.
Enter: Gemma4-12B-Coder GGUF (Q4_K_M)
Base: Google's gemma-4-12B-it
Fine-tuned on verifiable Python CoT data:
- Primary: Composer 2.5 real reasoning traces (only passing solutions kept)
- Auxiliary: Fable 5 used to redo the hard cases Composer missed.
Every training example's reasoning led to code that actually ran. No hallucinated logic.
Llama.cpp flags:
-m gemma4-coding-Q4_K_M.gguf -cnv -ngl 44 -c 64000 -v
(huggingface model link in comments)
Flag breakdown:
-ngl 44 → offload 44 layers to GPU (tune this for your VRAM)
-c 64000 → 64K context window
-cnv → conversation/chat mode
-v → verbose output
The irony writes itself.
Anthropic spent weeks telling the world Fable 5 (mythos) is too powerful to release. Then released it. Then got banned from serving it, including their own researchers.
Meanwhile: a Gemma 4 12B fine tune, trained on Fable 5's reasoning, runs fully offline on my mid range consumer GPU
No API. No cloud. Just me and llama.cpp.
This is why local AI matters.
Check out the model's link in the comments. How's your experience been with this model?
Hugging Models@HuggingModels
Gemma 4 12B Coder is here and it's a game changer for local code generation. This GGUF model packs Google's latest gemma-4 architecture into a compact 12B size, perfect for running on consumer hardware. It's optimized for reasoning and thinking, making it ideal for developers who want fast, private coding assistance without the cloud.
English

@calcsam Wait damn what happens to Mastra
The last time I used it was like early 2025 or smth that’s crazy
English

introducing t-rex
with t-rex enabled, greptile doesn't just review your PR, it runs your branch in a sandbox to find bugs.
it mocks api calls, clicks around the UI, and writes + runs unit tests
in our benchmarks, it caught ~20% more bugs than base greptile. most of the new bugs caught could not have been caught with more inference, they required code execution.
t-rex is available in beta to all greptile users
English

yo @DevinAI @cognition my birthday's around the corner(5 days)
could use some devin credits I just used mine up tryna make smth👀
special gift or smth yk yk
English

@leodev @stolinski @Coobyk_ nah but it worked for me tho
cuz i was on the waitlist before and i used it and it just randomly unlocked it
English

Day 20of posting content till I join @10x_apps
team. This is the last video with this format( still just 3 days) .
Hours of editing on this one, give it a watch.
English

@KrzysztofStaron true you gotta filter through it but might be easier
English

Making more training data.
Got bored, outsourced it to my sister.
I got her bubble tea for it.
Krzysztof Staroń@KrzysztofStaron
Built a model that writes with virtual pen and paper. Thinking about connecting it to a robotic arm next to see how well it transfers to the real world. Live demo + repo below:
English

@whitekr_ On everything this gave me access I promise
You can see that before recorded I tried something because I didn't believe it worked
I was on the waitlist and my phone already had Apple Intelligence so maybe it was already available just needed to be unlocked to put me off the wait
English

@dunk3ey But I never got access I was on the waitlist (and doing this gave me access)
I promise that i was on the waitlist before trying this and it worked
U gotta be on iOS 27 and on the waitlist I think
And already have Apple Intelligence/your phone being compatible
English

@whitekr_ it literally worked tho (i recorded it after i already tried the hack bc i realized it actually worked)
English

after hitting $3M ARR, 2.5B+ views, 37K+ App Store reviews, and millions of app downloads
i wrote an X article on how i launch and scale UGC programs
best outcome of my UGC programs? a comment from giannis

Ben Aratame@benaratame
English















