AtAndDev (alkinun)

reacted to nroggendorff's post with 🚀 about 1 hour ago

Post

2615

I'm not really doing much on HuggingFace right now due to their new Docker space policies, so if you want to keep up with most of what I'm up to, follow my [instagram](https://sly.sh/ig)

11 replies

·

reacted to andito's post with 🔥 1 day ago

Post

3613

🧠👁️ Can AI visualize solutions?

Humans often solve visual problems by sketching ideas in our minds. What if Vision-Language Models (VLMs) could do something similar, not by generating full images, but by using internal “mental sketches”?

That’s the idea behind Mirage, a new framework that empowers VLMs to reason using latent visual tokens. Instead of just thinking in words, Mirage mixes in abstract visual representations that help the model solve complex tasks.

These aren't photorealistic images. They're compact, internal representations optimized purely to support reasoning.

🔧 Mirage is trained in two phases:

1) Grounding: It learns to produce latent tokens anchored in real images.
2) Refinement: The model drops the images and learns to generate visual tokens on its own.

📈 And yes, it works!
On challenging benchmarks like Visual Spatial Planning, Jigsaw puzzles, and Spatial Attention Tasks, Mirage clearly outperforms GPT-4o and other strong baselines.
Smart sketches > empty words.

By mimicking the way humans visualize solutions, Mirage gives AI a new kind of imagination, one that’s faster, more efficient, and more human-like.
Kudos to the teams at UMass Amherst and MIT behind this exciting work.
Check the paper: Machine Mental Imagery: Empower Multimodal Reasoning with Latent Visual Tokens (2506.17218)

1 reply

·

reacted to YerbaPage's post with 🔥👀😎 1 day ago

Post

2562

Is 100% Pass Rate on HumanEval possible? Yes! ✅

Meet MGDebugger if you are tired of LLMs failing on complex bugs 🤔 Our MGDebugger, just hit 100% accuracy on HumanEval using the DeepSeek-R1 model. 🚀

✨ Demo: learnmlf/MGDebugger
📝 Paper: From Code to Correctness: Closing the Last Mile of Code Generation with Hierarchical Debugging (2410.01215)
💻 Code: https://github.com/YerbaPage/MGDebugger

HumanEval may be retired, we're ready for the next challenge In more complex scenarios! You may also take look at this repo for a collection of awesome repo-level coding tasks!

🖥️ https://github.com/YerbaPage/Awesome-Repo-Level-Code-Generation

reacted to Jaward's post with 👍 4 days ago

Post

1969

I played around with the new RXTX paper (XX^T) and was able to train nanogpt with 4x4 RXTX matmuls in both attention layer and optimizer🤕
It just works (well I had to add some guardrails) but still saves 5% of memory usage:
The Patch:
- Computes attention scores with a 4x4 blockwise RXTX matmuls (no pytorch dot prod)
- Handles arbitrary sequence lengths by padding to the nearest multiple of 4.
- An RXTX variant of shampoo with params reshaped into 4x4 blocks during each optimizer step.
- Uses 5% less ops
Code: https://github.com/Jaykef/ai-algorithms/blob/main/nanogpt-rxtx.ipynb
Paper: https://arxiv.org/pdf/2505.09814

replied to danielhanchen's post 5 days ago

unsloth ftw

reacted to danielhanchen's post with ❤️🚀🔥 5 days ago

Post

2209

Gemma 3n finetuning is now 1.5x faster and uses 50% less VRAM in Unsloth!

Click "Use this model" and click "Google Colab"!

unsloth/gemma-3n-E4B-it

unsloth/gemma-3n-E2B-it

https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Gemma3N_(4B)-Conversational.ipynb

2 replies

·

reacted to FlameF0X's post with 👍 9 days ago

Post

1913

SnowflakeCore-G1 development update: We're building a 24-layer transformer with 32K context and 1024 embedding dimensions - pretty ambitious! Even running at batch_size=1 with heavy gradient accumulation, we're hitting memory walls at 300GB RAM. Scaling up to ~1TB will take some time, but the architecture is looking promising. Thanks for following along with the journey! 😅

1 reply

·

reacted to samihalawa's post with 🚀 9 days ago

Post

2460

🔥BEST DEBUG PROMPT FOR CLAUDE_CODE
😲FIXES ANY REPO

How can I prompt you so there will not be any More bug or thing? So list the 20 kind of errors and bugs more common in a codebase like this one and create a comprehensive table list of each of them and all the main files and fucnionality and fix until cross all out. Don't add complexity. Ignore security. Only fatal problems
Proceed in the most exhaustive and comprehensive way possible. Don't miss a line of code. SYSTEMATICALLY FIX EVERYTHING

reacted to jsulz's post with 🚀 9 days ago

Post

4169

It's been a bit since I took a step back and looked at

xet-team progress to migrate Hugging Face from Git LFS to Xet, but every time I do it boggles the mind.

A month ago there were 5,500 users/orgs on Xet with 150K repos and 4PB. Today?
🤗 700,000 users/orgs
📈 350,000 repos
🚀 15PB

Meanwhile, our migrations have pushed throughput to numbers that are bonkers. In June, we hit upload speeds of 577Gb/s (crossing 500Gb/s for the first time).

These are hard numbers to put into context, but let's try:

The latest run of the Common Crawl from

commoncrawl was 471 TB.

We now have ~32 crawls stored in Xet. At peak upload speed we could move the latest crawl into Xet in about two hours.

We're moving to a new phase in the process, so stay tuned.

This shift in gears means it's also time to roll up our sleeves and look at all the bytes we have and the value we're adding to the community.

I already have some homework from @RichardErkhov to look at the dedupe across their uploads, and I'll be doing the same for other early adopters, big models/datasets, and frequent uploaders (looking at you @bartowski 👀)

Let me know if there's anything you're interested in; happy to dig in!

4 replies

·

replied to jsulz's post 9 days ago

Great job fellas keep it up!

reacted to yeonseok-zeticai's post with 👀 9 days ago

Post

3117

🚀 Real-Time On-Device AI Agent with Polaris-4B — Run It Yourself, No Cloud, No Cost

We just deployed a real-time on-device AI agent using the Polaris-4B-Preview model — one of the top-performing <6B open LLMs on Hugging Face.

📱 What’s remarkable?
This model runs entirely on a mobile device, without cloud, and without any manual optimization. It was built using ZETIC.MLange, and the best part?

➡️ It’s totally automated, free to use, and anyone can do it.
You don’t need to write deployment code, tweak backends, or touch device-specific SDKs. Just upload your model — and ZETIC.MLange handles the rest.

🧠 About the Model
- Model: Polaris-4B-Preview
- Size: ~4B parameters
- Ranking: Top 3 on Hugging Face LLM Leaderboard (<6B)
- Tokenizer: Token-incremental inference supported
- Modifications: None — stock weights, just optimized for mobile

⚙️ What ZETIC.MLange Does
ZETIC.MLange is a fully automated deployment framework for On-Device AI, built for AI engineers who want to focus on models — not infrastructure.

Here’s what it does in minutes:
- 📊 Analyzes model structure
- ⚙️ Converts to mobile-optimized format (e.g., GGUF, ONNX)
- 📦 Generates a runnable runtime environment with pre/post-processing
- 📱 Targets real mobile hardware (CPU, GPU, NPU — including Qualcomm, MediaTek, Apple)
- 🎯 Gives you a downloadable SDK or mobile app component — ready to run
And yes — this is available now, for free, at https://mlange.zetic.ai

🧪 For AI Engineers Like You, If you want to:
- Test LLMs directly on-device
- Run models offline with no latency
- Avoid cloud GPU costs
- Deploy to mobile without writing app-side inference code

Then this is your moment. You can do exactly what we did, using your own models — all in a few clicks.

🎯 Start here → https://mlange.zetic.ai

📬 Want to try Polaris-4B on your own app? [email protected], or just visit https://mlange.zetic.ai , it is opened as free!

Great work @Chancy , @Zhihui , @tobiaslee !

reacted to BFFree's post with 👍 15 days ago

Post

711

Love the subtle retro style of this sculptural piece! Took a bit of fine tuning the style but one of the more satisfying things I have created recently.

reacted to brainhome's post with 👍👀 15 days ago

Post

1737

Trinity-Synthesis: A Multi-Agent Architecture for AI Agents That Think Before They Speak
Ever felt your AI agent is "shooting from the hip"? It latches onto a single line of thought and fails to produce a robust, well-rounded plan. This is a common struggle I've called the "AI Reasoning Paradox."

To tackle this, I developed Trinity-Synthesis, a multi-agent architecture designed to force reflection and synthesis before delivering a final answer. The philosophy is simple: constructive conflict between different perspectives leads to better solutions.

Here’s the core idea:

Instead of one agent, it uses four agents running on the same base model but with different "personalities" defined by their system prompts and temperature settings:

🧠 The Visionary: Thinks outside the box (high temp: 1.0).
📊 The Analyst: Focuses on logic, data, and structure (low temp: 0.3).
🛠️ The Pragmatist: Evaluates feasibility, costs, and risks (mid temp: 0.5).
These three "thinkers" work in parallel on the same problem. Then, a final Synthesizer agent critically analyzes their outputs, rejects flawed arguments, and integrates the best points into a single, coherent, and often superior strategy.

The result is a more robust reasoning process that balances creativity with analytical rigor, making it ideal for solving complex, strategic problems where answer quality is critical.

I've written a deep dive on how it works, including a detailed case study ("The Helios Initiative") and the Python source code for you to experiment with.

Read the full article on Medium:
https://medium.com/@brainhome9/trinity-synthesis-how-i-built-an-ai-agent-that-thinks-before-it-speaks-d45d45c2827c

I'd love to hear your feedback and see what you build with it!

#AI #AIAgents #LLM #Reasoning #MultiAgent

4 replies

·

reacted to yeonseok-zeticai's post with 🔥 15 days ago

Post

1389

🚀 Try deep-research without Network Connection and Data Leak, Jan-Nano production-ready on-device AI in 6 hours!

Jan-Nano has been making waves as one of HuggingFace's most trending 4B parameter models, outperforming even 671B models on SimpleQA benchmarks. But here's what changes everything: ZETIC.MLange just transformed Jan-Nano into a blazing-fast on-device AI solution.

✨ 6-hour deployment from huggingface to production-ready library!

Zero cloud dependency - complete privacy and offline capability
While others struggle with complex on-device deployments taking weeks or months, ZETIC.MLange's automated pipeline makes it effortless. No manual optimization, no vendor-specific coding, no compromise on performance.

📱 Ready to transform your AI models? Try ZETIC.MLange, it is totally free now!

The future of AI is on-device. Make it happen in hours, not months.

#OnDeviceAI #EdgeAI #MLOptimization #NPU #PrivacyFirst

4 replies

·

reacted to AdinaY's post with 🔥 17 days ago

Post

1281

SongGeneration 🎵 A model by Tencent for vocal + accompaniment music generation.

Model:
tencent/SongGeneration
Demo:
https://huggingface.co/spaces/waytan22/SongGeneration-LeVo

✨ Mixed & dual-track token modeling
✨ Hi-fi music with custom codec
✨ Currently supports Chinese, English version coming soon!

alkinun

AI & ML interests

Recent Activity

Organizations

alkinun

AI & ML interests

Recent Activity

Organizations

AtAndDev's activity