In a Training Loop 🔄

John Smith PRO

John6666

John6666cat

AI & ML interests

None yet

Recent Activity

liked a model 15 minutes ago

Siye01/CODA-8B

liked a model 15 minutes ago

mradermacher/CODA-8B-GGUF

updated a Space 43 minutes ago

John6666/model-memory-usage-mod

View all activity

Organizations

reacted to karstenskyt's post with 👀 about 3 hours ago

Post

🚀 𝗪𝗲 𝗷𝘂𝘀𝘁 𝗼𝗽𝗲𝗻𝗲𝗱 𝘁𝗵𝗲 𝗱𝗼𝗼𝗿𝘀 𝘁𝗼 𝗹𝘂𝘅𝘂𝗿𝘆-𝗹𝗮𝗸𝗲𝗵𝗼𝘂𝘀𝗲! Our full 12-page open-source soccer analytics platform is now live and publicly accessible on Hugging Face Spaces.

We successfully migrated our entire backend Streamlit dashboard to a Docker Space. Instead of being locked behind workspace accounts, anyone can now directly explore our models, 38M+ tracking frames, and advanced metrics (xT, PAUSA, DEFCON) in the browser.

To prepare for the public release, we ran a massive optimization sprint (what used to be days of normal engineering, now condensed into a few hours of "Claude Time"):

⚡ 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲: Eliminated O(n²) bottlenecks, dropping ingestion pipeline runs from 36 minutes to <30 seconds.

🧠 𝗨𝗫: Applied a deep Cognitive Interface Audit across all surfaces to ensure the data is intuitive and human-readable.

The Hugging Face ecosystem now natively handles our entire public-facing presentation layer—Models, Datasets, Gradio Demos, and the full Streamlit App.

𝗧𝗿𝘆 𝘁𝗵𝗲 𝗹𝗶𝘃𝗲 𝗮𝗽𝗽: luxury-lakehouse/soccer-analytics-app

𝗘𝘅𝗽𝗹𝗼𝗿𝗲 𝘁𝗵𝗲 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲:

luxury-lakehouse

reacted to robtacconelli's post with 🤯 about 3 hours ago

Post

459

🧬 Midicoth: diffusion-based lossless compression — no neural net, no GPU, no training data

What if reverse diffusion could compress text — without a neural network?
Midicoth brings score-based denoising into classical compression. It treats prior smoothing as forward noise and reverses it with Tweedie's formula on a binary tree — 3 denoising steps, James-Stein shrinkage, applied after all model blending. ~2,000 lines of C, single CPU core.

Beats every dictionary compressor we tested:
enwik8 (100 MB) → 1.753 bpb (−11.9% vs xz, −15% vs Brotli, −24.5% vs bzip2)
alice29.txt → 2.119 bpb (−16.9% vs xz)
Outperforms xz, zstd, Brotli, bzip2, gzip on all inputs

PAQ/CMIX still win with hundreds of models + LSTMs. LLM compressors win with pre-trained knowledge. Midicoth closes the gap with pure statistics — no mixer, no gradient descent, just counting.
The Tweedie denoising layer adds 2.3–2.7% on every file tested — the most consistent component in the ablation. Adding SSE or logistic mixers made things worse. In the online setting, count-based beats gradient-based.
No external dependencies. Fully deterministic. Bit-exact encode/decode. ~60 KB/s throughput.
💻 Code: https://github.com/robtacconelli/midicoth
📄 Paper: Micro-Diffusion Compression -- Binary Tree Tweedie Denoising for Online Probability Estimation (2603.08771)
⭐ Space: robtacconelli/midicoth

If you ever wondered whether diffusion ideas belong in data compression — here's proof they do. ⭐ appreciated!

reacted to imnotkitty's post with 🔥 about 3 hours ago

Post

🚀 GLM-5-Turbo just dropped: The "OpenClaw Native" Model

Z.ai just shipped a 744B MoE beast that's 2-3x faster than GLM-5, with 200K context + 128K max output.

What's different:
👉 Tool call stability (no more random failures mid-chain)
👉 Complex instruction decomposition (breaks down messy prompts)
👉 Time-aware execution (understands scheduled/persistent tasks)
👉 High-throughput long-chain efficiency (doesn't choke on 50-step workflows)

📊 ZClawBench: Leads mainstream models in OpenClaw scenarios
💰 Trade-off: +20% price vs GLM-5

Anyone trying it yet? 👀

1 reply

reacted to kanaria007's post with 🧠 about 3 hours ago

Post

127

✅ Article highlight: *Federated SI* (art-60-044, v0.1)

TL;DR:
Most real systems do not live inside a single SI-Core. Cities, hospital networks, grid operators, transit systems, vendors, and neighboring institutions all run under different governance, trust, and legal boundaries.

This note sketches *Federated SI*: how multiple SI-Cores coordinate without pretending to share one brain. The focus is on portable artifacts, explicit trust boundaries, negotiated goals, limited memory exchange, and graceful failure when cooperation partially breaks.

Read:
kanaria007/agi-structural-intelligence-protocols

Why it matters:
• makes cross-operator coordination explicit instead of hiding it inside ad hoc APIs
• supports cooperation under separate trust anchors, legal regimes, and policy surfaces
• treats failure modes seriously: partitions, vetoes, degraded cooperation, partial visibility
• keeps governance portable via normalized verdicts, pinned bindings, and export-safe artifacts

What’s inside:
• why “one SI-Core sees everything” is the wrong default
• federation objects such as federated SIRs, goal surfaces, memory views, and consent records
• negotiation across cities, hospitals, utilities, and other institutional stacks
• operational labels vs exported governance verdicts (ACCEPT / DEGRADE / REJECT)
• deterministic, auditable exchange rules for cross-run / cross-vendor comparison
• failover, mutual aid, and graceful degradation when trust or connectivity breaks

Key idea:
Intelligence at institution scale is not a single runtime. It is a *federation of governed runtimes* that must negotiate, coordinate, and fail safely without collapsing auditability.

reacted to prometechinc's post with 👍 about 3 hours ago

Post

At PROMOTIONAL INSTANTANEOUS, we believe the next phase of AI isn't just about getting bigger, it's about getting smarter and more efficient. We are proud to announce the release of Cicikus_PTHS_v3_4.4B (Prometheus), representing our latest R&D milestone in the field of "High-Density Intelligence." By integrating our patented Behavioral Awareness Engine (BCE) into a compact 4.4B parameter architecture, we have developed a system capable of complex STEM reasoning and structured logic; and all this while occupying only 8 GB of local space.

This is significant because:
Native Privacy: High-level reasoning that operates completely offline.
Industrial Logic: Designed to handle complex rule sets and technical documentation with 32k context support.
Sustainable AI: Reduces computational costs and energy consumption without sacrificing logical accuracy.
Prometheus is a testament to our team's commitment to democratizing high-performance AI. We invite researchers and developers to explore the model on https://huggingface.co/collections/pthinc/cicikus-serisi-series

reacted to AINovice2005's post with 👀 about 3 hours ago

Post

I'm pleased to announce that native trackio support has landed in optuna v.4.8.0.

Now you can store HPO trial data to the 🤗 Hub as a dataset: AINovice2005/toy-optuna-hpo-dataset

or Space: AINovice2005/optuna-dashboard

Article which talks about the integration: https://medium.com/@paragekbote23/designing-and-contributing-an-optuna-integration-that-performs-well-in-practice-21a07d77ec2c

reacted to W8Yi's post with 🚀 about 3 hours ago

Post

270

I built a **TCGA WSI feature dataset using UNI2-h**.

The official release currently has incomplete coverage (see discussion):
MahmoodLab/UNI2-h-features#2

To make the features easier to use for research, I generated a new dataset:

W8Yi/tcga-wsi-uni2h-features

Key differences from the official release:

• **All detected tissue tiles are encoded** (not a sampled subset)
• **Features can be downloaded per slide** instead of large ZIP archives
• **QC overlay images** are provided for visual inspection
• **UNI2-h 1536-D tile embeddings** stored in H5 format
• Organized by TCGA project for easier use in MIL / retrieval pipelines

Example layout:

TCGA-HNSC/
  features/*.h5
  vis/*__overlay.png

Hope this helps others working on computational pathology and TCGA WSI research.

reacted to unmodeled-tyler's post with 🚀 about 24 hours ago

Post

3327

LINK: https://github.com/unmodeled-tyler/vessel-browser

Hey Hugging Face!

It's been quiet from me over here for the last few weeks, but I've been busy building! I just submitted my project to the Hermes Agent Hackathon, and wanted to share it with all of you.

This is Vessel Browser - an AI-native web browser that runs locally on Linux, and is operated by your personal AI agent via MCP server. Vessel is built from the ground up around the agent as first-class and visible UI for human-in-the-loop with 3 different levels of permissions.

Your agent finds, reads, and organizes the web for you, based on what you actually care about - not what a platform's algorithm thinks you care about.

Once your agent finds what it's looking for, it can organize bookmarked pages into custom folders with summaries for later browsing, take screenshots with highlighted text, and integrate with Obsidian for long-term browsing related-memory.

Check it out!

2 replies

reacted to AbstractPhil's post with 👀 about 24 hours ago

Post

115

Clawd breadcrumb trail AbstractPhil/geolip-hypersphere-experiments

With this I'll begin forming Clawd interface utility with the geofractal router, which will allow Clawd to form agentic clouds of utility that can be datawise trained on the go with minimal hardware requirement. This is not ready yet, but it begins very soon.

The recent experiments have solved the alignment issue that crippled collectives and forced my hand into ensemble research instead.

With those recent experiments, the geofractal router will allow modularization structural capacity after some preliminary alignment adjustment and adjudication experimentation. This will enable the full collective differentiation through codified attribution.

In other words, adding and removing modular AI elements to contribute to aligned communication streams, all speaking the same language. This is an adjacent and more powerful result than the anticipated geovocab patchwork, and it yields substantially more effective agentic solutions than moving around a bulky embedding echo-chamber.

https://github.com/AbstractEyes/geofractal

Procrustes whitening orthogonality will allow adding and removing elements from geofractal routers given a small amount of prep data, while the anchors of expectation can stay as a snap-on element.

The most inquisitive and interested researchers can follow the trail to find all of the experiments. Web crawl it with clawd and you can probably create a unified rationality pretty quickly, but I doubt you'll like what you find. The journey was extensive and the failures outweighed the successes, but I did find the lightbulb.

The represented outcomes are either in my articles in huggingface, my civit articles, my github repos, my huggingface repos, or I forgot to upload them and they're in my colab notebook heap.

As most research yields, it is mostly failures. However, there are many successes in the mix. Many. If you need solutions, you can dredge the bog.

1 reply

reacted to karstenskyt's post with 🚀 about 24 hours ago

Post

⚽ We've been building 𝚕𝚞𝚡𝚞𝚛𝚢-𝚕𝚊𝚔𝚎𝚑𝚘𝚞𝚜𝚎—an open-source soccer analytics platform—and we're using the Hugging Face Hub as our public distribution layer.

"Luxury! We used to dream of serverless!" We replaced a traditional 6-service AWS pipeline with a unified lakehouse architecture. While Databricks handles our backend Medallion architecture (processing 𝟯𝟴𝗠+ 𝘁𝗿𝗮𝗰𝗸𝗶𝗻𝗴 𝗳𝗿𝗮𝗺𝗲𝘀 from 5 vendors), we rely entirely on the HF ecosystem for public access and specific compute workloads.

I just updated our org page with a full architecture breakdown showing how we integrate 𝗛𝗙 𝗠𝗼𝗱𝗲𝗹𝘀, 𝗗𝗮𝘁𝗮𝘀𝗲𝘁𝘀, 𝗚𝗿𝗮𝗱𝗶𝗼 𝗦𝗽𝗮𝗰𝗲𝘀, and use 𝗛𝗙 𝗝𝗼𝗯𝘀 for serverless Expected Threat (xT) computation with a bidirectional Databricks sync.

Check out the full architecture details here: 👇

luxury-lakehouse

reacted to emirkaanozdemr's post with 👍 about 24 hours ago

Post

Hi everyone, my new model Fok-2B Instruct is a LoRA fine-tuned model on VNGRS AI Kumru-2B. Trained for the Turkish language, it analyzes Polar Research Station sensor data to detect potential risks and generate detailed recommendations. Please note that this is the first and smallest member of this model family.

https://huggingface.co/collections/emirkaanozdemr/fok-llm-family

reacted to Nymbo's post with 👍 about 24 hours ago

Post

4073

We should really have a release date range slider on the /models page. Tired of "trending/most downloaded" being the best way to sort and still seeing models from 2023 on the first page just because they're embedded in enterprise pipelines and get downloaded repeatedly. "Recently Created/Recently Updated" don't solve the discovery problem considering the amount of noise to sift through.

Slight caveat: Trending actually does have some recency bias, but it's not strong/precise enough.

3 replies

reacted to OzTianlu's post with 🔥 2 days ago

Post

5227

Arcade-3B — SmolReasoner
NoesisLab/Arcade-3B
Arcade-3B is a 3B instruction-following and reasoning model built on SmolLM3-3B. It is the public release from the ARCADE project at NoesisLab, which investigates the State–Constraint Orthogonality Hypothesis: standard Transformer hidden states conflate factual content and reasoning structure in the same subspace, and explicitly decoupling them improves generalization.

5 replies

reacted to EricFillion's post with 👀 2 days ago

Post

114

Here's how to perform full-parameter fine-tuning for openai/gpt-oss-20b with a single H200 GPU. Only a few lines of code are needed using my new Python package called Eric Transformer.

Article: https://www.vennify.ai/gpt-oss-20b-fine-tune/

@vennify

pip install erictransformer

from erictransformer import EricChat, EricTrainArgs

eric_chat = EricChat(model_name="openai/gpt-oss-20b")

train_args = EricTrainArgs(optim="sgd")

# see the article to learn how to format the data 

eric_chat.train("train.jsonl",  eval_path="eval.jsonl", args=train_args)

1 reply

reacted to vkatg's post with 🚀 2 days ago

Post

102

A patient's name in a clinical note. Their voice in an ASR transcript. A waveform header with their DOB. Three separate records, each one low-risk by itself. Together, they're enough to re-identify someone.

Per-record de-identification doesn't see this. It can't. It has no memory of what came before.

We built a system that does. AMPHI tracks PHI exposure across modalities and time, maintains a risk score per patient, and escalates masking automatically as exposure accumulates. When a text record and an audio record co-reference the same patient via embedding similarity, the system catches it and responds before the next record arrives.

The core results: adaptive policy holds privacy at 0.991 on high-risk bursty workloads while keeping utility at 0.847. Static redaction gets the privacy number but destroys utility. Static weak masking keeps utility but leaks on high-risk bursts. The adaptive system doesn't trade one for the other.

Full system is open-source. Five models, three datasets, two demo spaces, 141 passing tests.

Spaces: [AMPHI Demo]( vkatg/amphi-rl-dpgraph) | [DCPG Scorer]( vkatg/dcpg-scorer-demo)

Models: [DCPG Encoder]( vkatg/exposureguard-dcpg-encoder) | [Cross-Modal Risk Scorer]( vkatg/dcpg-cross-modal-phi-risk-scorer) | [PolicyNet]( vkatg/exposureguard-policynet) | [FedCRDT]( vkatg/exposureguard-fedcrdt-distill) | [DAGPlanner](https://huggingface.co/vkatg/expsoureguard-dagplanner)

Datasets: [Multimodal PHI Masking]( vkatg/multimodal-phi-masking-benchmark) | [Streaming De-ID Benchmark]( vkatg/streaming-phi-deidentification-benchmark) | [DAG Remediation Traces]( vkatg/dag_remediation_traces)

Code: [phi-exposure-guard on GitHub](https://github.com/azithteja91/phi-exposure-guard)

reacted to prithivMLmods's post with 🤗 2 days ago

Post

4763

QIE-2509-Object-Remover-Bbox-v3 is a more stable version of the Qwen Image Edit visual grounding–based object removal model. The app was previously featured in HF Spaces of the Week and is now updated with the latest Bbox-v3 LoRA adapter.

🤗 Demo: prithivMLmods/QIE-Object-Remover-Bbox
🤗 LoRA: prithivMLmods/QIE-2509-Object-Remover-Bbox-v3
🤗 Collection: https://huggingface.co/collections/prithivMLmods/qwen-image-edit-layout-bbox

To learn more, visit the app page or the respective model pages.

2 replies

reacted to kanaria007's post with 🧠 2 days ago

Post

144

✅ Article highlight: *Programming SI-Core* (art-60-043, v0.1)

TL;DR:
What do developers actually write on an SI-Core stack?

This note sketches the programming model: **SIL** for goal-native code, **DPIR** as a typed decision IR, **CPU/GSPU backends** for execution, and **SIR** for structural traces. The point is to move from prompt surgery + log spelunking toward something closer to normal, testable, compilable software engineering.

Read:
kanaria007/agi-structural-intelligence-protocols

What’s inside:
• why SI-Core programming differs from “LLM wrapper microservices”
• the mental model: **OBS → SIL → DPIR → backend → RML → SIR**
• SIL examples, DPIR sketches, and backend execution shape
• local dev loop: sandbox SIRs, si build, si test, replay, inspection
• testing strategy: unit tests, structural property tests, GCS regression, Genius Replay
• tooling: LSP, ETH/capability lints, timeline and what-if visualizers
• migration path: from plain LLM wrappers to SI-native stacks in stages

Key idea:
Treat decisions as **programs** with explicit goals, ETH checks, and structured effects — not as opaque model samples hidden behind prompts.

Related specs (/spec):
si-core-spec-v0.1.md, si-nos-design-v0.1.md, sil-compiler-spec-bundle-v0.1.md, sil-compiler-conformance-kit-v0.1.md

reacted to BibbyResearch's post with 👍 2 days ago

Post

2449

We are working on the largest Dataset and Pre-trained model for Text to Speech and Speech to text for the low-resourced language called Marwari in India.

reacted to marksverdhei's post with 🤯 3 days ago

Post

267

The hidden gem of open-source embedding models: LCO-Embedding
for text, image AND audio!

I found this model after reading the recent Massive Audio Embedding Benchmark (MAEB) paper, as it blew the other models out of the water on day zero. I've been using it personally for about a week, and searching my files by describing music, sound effects or images is both practical and entertaining. Really underrated model, would highly recommend checking it out: LCO-Embedding/LCO-Embedding-Omni-7B

PS: If you're looking you run this model on llama.cpp, i've gone ahead and quantized them for you here 👉 https://huggingface.co/collections/marksverdhei/lco-embedding-omni-gguf

2 replies

reacted to BibbyResearch's post with 👀 3 days ago

Post

1758

Used by Researchers at Allen institute, Simons foundation, Yale and other top universities.. 🤗
Researchers are using AI to write their papers.
That AI is Bibby AI.
Not GPT-5. Not Claude Opus. Not whatever wrapper your institution just paid $50k for.
Bibby.
While the AI research community spent 2025 debating whether LLMs can handle scientific writing — actual scientists at actual top-tier institutions quietly started shipping papers with it. No press release. No hype cycle. Just results.
Here's what they figured out that most people haven't:
The bottleneck in research was never the ideas. It was never the experiments. It was the 3am writing sessions where good science goes to die in a Google Doc. Writer's block, LaTex Learning frustration, formatting issues, compiler errors.
Bibby is built specifically for that gap. Citation-aware. Argument-aware. Knows when to hedge, when to assert, and — critically — knows not to hallucinate your methods section.
The institutions adopting it aren't doing it because it's trendy. They're doing it because the researcher-hours it saves are going straight back into actual research.
This is what the adoption curve looks like before the thing becomes obvious.
→ https://trybibby.com/
Who else here is using AI in their research workflow? Drop it below 👇

Best comments will receive a discounted Bibby AI subscription with a chance to win a $100 grant for their research.

John Smith PRO

AI & ML interests

Recent Activity

Organizations

John6666's activity