Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up

All HF Hub posts

SeaWolf-AIΒ 
posted an update 2 days ago
view post
Post
3958
πŸ”₯ 128 Blackwell GPUs β€” Thank You, Hugging Face

I've been awarded 128 NVIDIA Blackwell GPUs through NIPA (Korea's National IT Industry Promotion Agency). Sharing this here first β€” because Hugging Face is where it all started.

I design LLM architectures from scratch. HF was my lab β€” dissecting Transformers internals, analyzing thousands of checkpoints, iterating on Spaces with global feedback.

Our FINAL Bench reached #5 globally in HF dataset popularity, and this research is exactly what earned the GPU grant.
πŸ‘‰ FINAL-Bench/Leaderboard

These 128 Blackwells will scale AETHER-Net β€” our Proto-AGI architecture (Emergence Engine Β· Meta-Cognition Β· SLAI Β· Multi-Intelligence Β· Synergy & Critique) β€” validated at 0.8B with MoE expansion to 2.1B params. Next stop: 166B.

People I must thank:

@John6666 β€” Guardian of this ecosystem. Never misses a forum question, interested in every project, active 24/7. I've genuinely wondered if you're a machine. Remarkable.

@bartowski β€” Master of quantization. The hidden infrastructure of open-source LLM. Countless experiments possible thanks to you.

@SaylorTwift β€” You see what others miss. Insight that cuts to the essence. Deep respect.

My promise: AETHER-Net design docs, training recipes, checkpoints, and failure logs β€” all shared here openly.

πŸ€— Thank you, Hugging Face. Let's turn the next page together. πŸš€

vidraft Β· VIDRAFT
#OpenScience #HuggingFace #ProtoAGI #AETHER #LLMArchitecture #Blackwell #NIPA
  • 7 replies
Β·
alibidaranΒ 
posted an update 2 days ago
view post
Post
3836
With the release of Gemma 4, I launched a new Space called MEDPAI β€” a medical imaging analysis tool that combines object detection with multimodal AI.
Here's how it works:

Upload a CT scan or X-ray
Computer vision models detect and annotate findings
Gemma 4 33B generates a report or answers your questions about the image

Currently available detectors: dental analysis and bone fracture detection.
More models are in the pipeline β€” follow the Space to stay updated!
alibidaran/MEDPAI
  • 1 reply
Β·
ArtelTalebΒ 
posted an update 1 day ago
view post
Post
1273
🎡 MP3 Player - Drop your music, hit play. No install

MP3 Player - brings that energy back - straight in your browser.

- Drop your files - MP3, WAV, FLAC, AAC, OGG, AIFF, WMA β€” it reads them all
- Build your playlist - add tracks one by one or batch-load a whole folder
- Retro LCD display - scrolling track info, elapsed time, the full throwback
- Full controls - play, pause, skip, shuffle, repeat
- Mobile-first - big tactile buttons, works on phone like an iPod in your pocket

No install. No GPU needed on your end. Just upload and play.

πŸ‘‰ ArtelTaleb/mp3-player

ShrijanagainΒ 
posted an update 1 day ago
view post
Post
1541
sKT-Ai-Labs


Join fast we will soon published tokens and all join and get started because we will soon off join request button if you want you can join fast guys
danielhanchenΒ 
posted an update 3 days ago
allisonandreyevΒ 
posted an update about 20 hours ago
view post
Post
332
ConfCrawler πŸ•·οΈ β€” never miss a conference deadline again

Keeping track of submission deadlines across CV, NLP, robotics, and ML conferences is a mess. ConfCrawler aggregates them in one place so you can actually plan your research calendar.

What's in it:
- Deadlines for major conferences (CVPR, ICCV, NeurIPS, ICRA, ACL, etc.)
- Updated regularly
- Filterable by field / month

Built this out of personal frustration while juggling multiple submission cycles. Hope it saves someone else the tab-hoarding.
πŸ”— https://confcrawler.vercel.app/
feedback welcome β€” open to adding more conferences if yours isn't listed!
shriarul5273Β 
posted an update 4 days ago
view post
Post
2235
πŸ” One API. 12 model families. 28 variants. Why depth_estimation makes depth research easier

Switching between depth models usually means rewriting preprocessing, adapting outputs, and dealing with different codebases.

depth_estimation removes that friction.

With the same interface, you can work with:
🌊 Depth Anything
🍎 DepthPro
🧭 MiDaS
πŸ“ ZoeDepth
🧩 MoGe
πŸ›°οΈ VGGT / OmniVGGT
and more

Change one model string, keep the rest of your workflow the same.

That makes it much easier to:
βš–οΈ compare models fairly
πŸ§ͺ prototype quickly
πŸ“ˆ benchmark consistently
πŸ› οΈ build reusable depth pipelines

GitHub: https://github.com/shriarul5273/depth_estimation

#depthestimation #research #computervision #python #machinelearning #opensource #pytorch
prabhatkrΒ 
posted an update about 21 hours ago
view post
Post
68
Why did FastMemory achieve 100% accuracy at 10 million tokens while the SOTA is stalling at 64%?

It’s not because we have a better "retriever." It’s because we changed the Topology of Truth.

Standard RAG treats your data like a pile of leaves. You search for a leaf, and you hope you find the right one. As the pile grows to 10 million leaves, you inevitably fail.

FastMemory treats your data like a Building.

Topological Isolation: We use Rust-driven Louvain clustering to atomize text into "Logic Rooms."
Crystalline Grounding: Facts don't "decay" in our context; they are locked into a deterministic graph.
Sub-Second O(1) Routing: We don't "search" the haystack; we navigate the floor plan.
If your AI is mission-critical, "mostly right" is a liability. It’s time to move beyond the RAG pile and into the topological foundation.

Verify Yourself: fastbuilderai/fastmemory-supremacy-benchmarks

#MachineLearning #GraphAI #LouvainClustering #FastBuilder #EngineeringExcellence

adamridaΒ 
posted an update 1 day ago
view post
Post
58
Open-sourced TRACER.

Many LLM classification calls in production are overkill.
For tasks like intent detection, moderation, tagging, or routing, TRACER learns which requests can be safely offloaded to a lightweight ML model trained on the LLM’s own outputs.

You keep the hard cases on the LLM, set a target quality bar, and offload the easy traffic.

On the right workloads, this can remove 90%+ of LLM calls.

GitHub:
https://github.com/adrida/tracer
mike-ravkineΒ 
posted an update 1 day ago
view post
Post
52
Gemma-4, specifically google/gemma-4-26B-A4B-it is doing something inside it's reasoning traces I have never seen before: it's recognizing that its being evaluated and spends meta-thinking tokens on understanding the evaluation regime in which it believes it find itself.

Let's see if 12/10/2023 is a more likely answer than 12/09/2023

In most AI benchmark tests (like those this prompt resembles), the simplest path is often the intended one.


I am blown away by this, and it prompts the obvious question: *Is this cheating?*

I am leaning towards no.

Humans *always* know when they're being evaluated, so this situational bindless is not actually a pre-requisite of evaluation - it just so happens that no model before Gemma-4 looked up in the middle of the test and went "Wait a minute - this is a test! I should try align my answer with the test format's expectations."

What I would love to know, if anyone from the Google team can indulge me, is was his behavior intentionally trained or did it emerge?