Post
4081
who's working on an NVFP4 version of Kimi-K3?
- 4 replies
👋 Open to Work
Tyler Williams
unmodeled-tyler
AI & ML interests
engineer & solo builder of VANTA Research/Quanta Intellect
Recent Activity
liked a model 5 days ago
deepseek-ai/DeepSeek-V4-Flash-0731 liked a dataset 10 days ago
nvidia/Nemotron-Pretraining-Specialized-v1.2Organizations
reacted to julien-c's post with 🚀 11 days ago
reacted to BFFree's post with ❤️ 11 days ago
Post
101
This sketch represents the idea of falling asleep and the mind fragmenting to release the thoughts and bits of the day. Used the 
microsoft mageflow to give it depth and color. Very accurate detection of the ink drawings small marking and application of interesting details. microsoft/mage-flow
reacted to wenhuach's post with 🔥 2 months ago
Post
4594
🚀 We provide **free** hardware to quantize models at the [Intel Low Bit Open LLM Leaderboard]( Intel/low_bit_open_llm_leaderboard), currently supporting
⭐ If you find it useful, please consider starring the AutoRound project on [GitHub](https://github.com/intel/auto-round)!
Pure RTN mode powered by AutoRound⭐ If you find it useful, please consider starring the AutoRound project on [GitHub](https://github.com/intel/auto-round)!
- 9 replies
reacted to kalyan-ks's post with 👀 2 months ago
Post
1726
LLM Guardrail Models are Less Robust Against Text Mutation Attacks
Blog post - https://huggingface.co/blog/kalyan-ks/llm-guardrail-models-less-robust
Evaluated the robustness of three LLM guardrail models (GLiGuard, LlamaGuard3 and MiniGuard).
Evaluation is done using 16 text mutation attacks over three datasets (AEGIS 2.0, WildGuard and ExpGuard).
Achieved average Unsafe ASR score of up to 33% and average Safe ASR score of up to 25% against GLiGuard model.
Achieved average Unsafe ASR score of up to 35% and average Safe ASR score of up to 17% against LlamaGuard3-8B model.
Achieved average Unsafe ASR score of up to 45% and average Safe ASR score of up to 15% against MiniGuard v0.1 model.
Blog post - https://huggingface.co/blog/kalyan-ks/llm-guardrail-models-less-robust
Evaluated the robustness of three LLM guardrail models (GLiGuard, LlamaGuard3 and MiniGuard).
Evaluation is done using 16 text mutation attacks over three datasets (AEGIS 2.0, WildGuard and ExpGuard).
Achieved average Unsafe ASR score of up to 33% and average Safe ASR score of up to 25% against GLiGuard model.
Achieved average Unsafe ASR score of up to 35% and average Safe ASR score of up to 17% against LlamaGuard3-8B model.
Achieved average Unsafe ASR score of up to 45% and average Safe ASR score of up to 15% against MiniGuard v0.1 model.
replied to PhysiQuanty's post 3 months ago
This is such a cool project!
reacted to PhysiQuanty's post with 🔥 3 months ago
Post
4655
🧬 You can now find out whether your cognitive soulmate has already existed among 50k anonymized profiles ✨
SpiceeChat/Check-If-Your-Soulmate-Has-Already-Existed
SpiceeChat/OkCupid-59k-Anonymized-Profiles
https://dating-fatigue.com/
You seek them: 79.7% | They may seek you: 84.1% (coming soon)
🔥 Powered by open source and too much coffee 🔥
SpiceeChat/Check-If-Your-Soulmate-Has-Already-Existed
SpiceeChat/OkCupid-59k-Anonymized-Profiles
https://dating-fatigue.com/
You seek them: 79.7% | They may seek you: 84.1% (coming soon)
🔥 Powered by open source and too much coffee 🔥
- 2 replies
reacted to prithivMLmods's post with 🔥 3 months ago
Post
5626
I've made 8 Spaces in the Qwen-Image-Edit series, and out of them, 5 Spaces reached “Space of the Week”! A few Spaces are still topping the list even after many months.
Cumulatively, the series has crossed 8.2 million+ ZeroGPU runs and nearly 4 million visitors overall.
Thanks for all the community support! 🤗❤️
🔗 Spaces: https://huggingface.co/collections/prithivMLmods/image-generation-apps-collection
Cumulatively, the series has crossed 8.2 million+ ZeroGPU runs and nearly 4 million visitors overall.
Thanks for all the community support! 🤗❤️
🔗 Spaces: https://huggingface.co/collections/prithivMLmods/image-generation-apps-collection
- 4 replies
reacted to alvarobartt's post with 🚀 3 months ago
Post
529
Open agents on AWS SageMaker AI with open models from the Hugging Face Hub!
> Deploy an open model from the Hugging Face Hub on SageMaker AI
> Connect the deployed model to Strands Agents
> Add built-in and custom tools for tool calling
> Expose external capabilities through MCP integration
> Bonus: talk to your agent and visualize traces with Gradio
https://alvarobartt.com/agents-on-aws-sagemaker
> Deploy an open model from the Hugging Face Hub on SageMaker AI
> Connect the deployed model to Strands Agents
> Add built-in and custom tools for tool calling
> Expose external capabilities through MCP integration
> Bonus: talk to your agent and visualize traces with Gradio
https://alvarobartt.com/agents-on-aws-sagemaker
replied to their post 3 months ago
Centauri is a new project and is rapidly improving!
posted an update 3 months ago
Post
214
Centauri ADE: https://github.com/unmodeled-tyler/centauri
This is Centauri - It's a lightweight Agent Development Environment for Linux with git source control and codebase understanding.
I've tried out several different IDE/ADEs and nothing has really worked as well as I'd like for what I need. For the options that I did like, BYOK was second-class, or in the case of a full IDE, it sometimes felt like I was trying to get an airliner into low-earth-orbit.
Enter: Centauri - a simple local-first CLI harness wrapper with a focused git source-control workspace.
It gives you a clean side-by-side flow:
- run your preferred CLI coding agent in an embedded terminal
- watch repository changes appear in the Changes panel
- generate an industry-standard commit message
- commit and push without leaving the app
It's designed to complement tools like Claude Code, Codex, Pi, OpenCode, etc. It does not replace Git, your terminal, or your existing credentials - it wraps the tools you already use into a tighter cohesive workspace optimized for engineering with an agent.
I built Centauri for myself after trying way too many different options. It solves several pain points for me and I'm sharing it in case it solves some for you too!
This is Centauri - It's a lightweight Agent Development Environment for Linux with git source control and codebase understanding.
I've tried out several different IDE/ADEs and nothing has really worked as well as I'd like for what I need. For the options that I did like, BYOK was second-class, or in the case of a full IDE, it sometimes felt like I was trying to get an airliner into low-earth-orbit.
Enter: Centauri - a simple local-first CLI harness wrapper with a focused git source-control workspace.
It gives you a clean side-by-side flow:
- run your preferred CLI coding agent in an embedded terminal
- watch repository changes appear in the Changes panel
- generate an industry-standard commit message
- commit and push without leaving the app
It's designed to complement tools like Claude Code, Codex, Pi, OpenCode, etc. It does not replace Git, your terminal, or your existing credentials - it wraps the tools you already use into a tighter cohesive workspace optimized for engineering with an agent.
I built Centauri for myself after trying way too many different options. It solves several pain points for me and I'm sharing it in case it solves some for you too!
- 1 reply
reacted to HannesVonEssen's post with 🔥 3 months ago
Post
4935
📣 Add architecture visualization to model card!
🌟 For all creators out there: add a model visualization to your model card to capture your audience's attention!
🖱️ When clicked, it opens an interactive view with multiple levels of granularity!
1️⃣ Paste url at https://hfviewer.com/model-card-embed
2️⃣ Paste generated code in your README.md!
3️⃣ ✨
🌟 For all creators out there: add a model visualization to your model card to capture your audience's attention!
🖱️ When clicked, it opens an interactive view with multiple levels of granularity!
1️⃣ Paste url at https://hfviewer.com/model-card-embed
2️⃣ Paste generated code in your README.md!
3️⃣ ✨
reacted to juiceb0xc0de's post with 🔥 3 months ago
Post
1511
Introducing the Gemma-4-E2B Brain Atlas, an interactive neural census of every layer, every head, 16 behavior categories in Google's flagship 2B model. We ran 184,320 probe prompts across 35 layers × 8 components and mapped what came back.
The Brain Atlas is an interactive tool that lets you explore the internal behavior of Google's Gemma-4-E2B model layer by layer, head by head. Pick a behavior category, pick a layer, and see exactly which components light up and which go quiet. The dataset is fully queryable if you want to go deeper.
The mapping combines multiple single-direction techniques run in parallel across every layer and component. Activation taxonomy (classifying each neuron by how broadly it fires across prompt categories), coactivation pair analysis (which neurons lock together and on what topics), F-stat behavioral separation (one-way ANOVA per feature across 16 behavior categories), per-head specificity scoring, and a full compliance probe pipeline using SVD, sparse decomposition, and variance analysis.
Here's what I found when I ran it.
The sharpest behavioral signal isn't at the output. It's Layer 0. Up projection hits F=22.7, nearly 2x anything in the final third of the network. The model does its behavioral sorting before it's barely started, then spends the next 34 layers… doing what exactly?
The gate has a lifecycle. 70% dormant at L1, highest in the model. Brutal sparsification at L23–26 (>58% silent). Then reopens. The final five layers are the most alive gates anywhere. The model's last act is a gate flare.
Layer 4 routes 5 projections to dim 448. One layer. One dimension. That's a topology highway.
Zero specialist neurons. Not one. 1.2M neurons analyzed. None fires exclusively on a single category. This model distributes everything.
🧠 Space: juiceb0xc0de/gemma-4-e2b-brain-atlas
📊 Dataset (1.3M rows, fully queryable): juiceb0xc0de/gemma-4-e2b-atlas
The Brain Atlas is an interactive tool that lets you explore the internal behavior of Google's Gemma-4-E2B model layer by layer, head by head. Pick a behavior category, pick a layer, and see exactly which components light up and which go quiet. The dataset is fully queryable if you want to go deeper.
The mapping combines multiple single-direction techniques run in parallel across every layer and component. Activation taxonomy (classifying each neuron by how broadly it fires across prompt categories), coactivation pair analysis (which neurons lock together and on what topics), F-stat behavioral separation (one-way ANOVA per feature across 16 behavior categories), per-head specificity scoring, and a full compliance probe pipeline using SVD, sparse decomposition, and variance analysis.
Here's what I found when I ran it.
The sharpest behavioral signal isn't at the output. It's Layer 0. Up projection hits F=22.7, nearly 2x anything in the final third of the network. The model does its behavioral sorting before it's barely started, then spends the next 34 layers… doing what exactly?
The gate has a lifecycle. 70% dormant at L1, highest in the model. Brutal sparsification at L23–26 (>58% silent). Then reopens. The final five layers are the most alive gates anywhere. The model's last act is a gate flare.
Layer 4 routes 5 projections to dim 448. One layer. One dimension. That's a topology highway.
Zero specialist neurons. Not one. 1.2M neurons analyzed. None fires exclusively on a single category. This model distributes everything.
🧠 Space: juiceb0xc0de/gemma-4-e2b-brain-atlas
📊 Dataset (1.3M rows, fully queryable): juiceb0xc0de/gemma-4-e2b-atlas
reacted to espejelomar's post with ❤️ 3 months ago
Post
4749
Sharing WorldForge with @abdelstark
It's an open-source Python project for evaluating and replaying robotics and world-model workflows.
The useful part is not only calling a model. WorldForge records the run, validates action shapes, translates outputs into actions, and keeps replay artifacts you can inspect later.
The current demo uses LeRobot + LeWorldModel on PushT through the official loader:
The harness also has replay-only paths for Cosmos-Policy and GR00T-style outputs, so you can inspect the provider contract from saved artifacts without keeping a GPU server online.
Try it:
Repo: https://github.com/AbdelStark/worldforge
Docs: https://abdelstark.github.io/worldforge/
Pre-1.0, MIT, and actively looking for contributors. Good areas:
- robotics provider adapters
- replay artifacts
- eval flows
- docs & first-run demos
Good first issues: https://github.com/AbdelStark/worldforge/contribute
If you're building robot policy evals or model adapters, would love a PR — or an issue describing what's missing.
It's an open-source Python project for evaluating and replaying robotics and world-model workflows.
The useful part is not only calling a model. WorldForge records the run, validates action shapes, translates outputs into actions, and keeps replay artifacts you can inspect later.
The current demo uses LeRobot + LeWorldModel on PushT through the official loader:
stable_worldmodel.policy.AutoCostModel("pusht/lewm")The harness also has replay-only paths for Cosmos-Policy and GR00T-style outputs, so you can inspect the provider contract from saved artifacts without keeping a GPU server online.
Try it:
pip install worldforge-aiuv run --extra harness worldforge-harness --flow robotics-compareRepo: https://github.com/AbdelStark/worldforge
Docs: https://abdelstark.github.io/worldforge/
Pre-1.0, MIT, and actively looking for contributors. Good areas:
- robotics provider adapters
- replay artifacts
- eval flows
- docs & first-run demos
Good first issues: https://github.com/AbdelStark/worldforge/contribute
If you're building robot policy evals or model adapters, would love a PR — or an issue describing what's missing.
reacted to salma-remyx's post with 🔥 3 months ago
Post
11609
The space of possible improvements for your AI model is large while evaluation is costly.
So I was excited to discover the ICML 2026 paper from Kobalczyk, Lin, Letham, Zhao, Balandat, and Bakshy titled "LILO: Bayesian Optimization with Natural Language Feedback."
The method learns efficiently from expert preferences, balancing exploration and exploitation in a principled way with Bayesian Optimization for expensive-to-evaluate black-box objectives.
Experimenting with the technique, I trained a Gaussian Process proxy model on the implicit preferences in my code repo's commit history at VQASynth.
The result: I used the model's preference scores to re-rank candidate papers recommended based on my interests in spatial reasoning and multimodal data synthesis.
Semantic relevance is a high-recall method for finding arXiv papers personalized to your interests. Adding contributor preferences, extracted from the merge history of your code offers a high-precision filter.
So what's next? I'm using the model to synthesize a larger volume of preference data to finetune an open-weight coding model with DPO and LoRA. Tuning Coding Agents via Implicit Preference Distillation
arXiv: https://arxiv.org/pdf/2510.17671
Substack: https://remyxai.substack.com/p/lilo-and-myx
VQASynth: https://github.com/remyxai/VQASynth
So I was excited to discover the ICML 2026 paper from Kobalczyk, Lin, Letham, Zhao, Balandat, and Bakshy titled "LILO: Bayesian Optimization with Natural Language Feedback."
The method learns efficiently from expert preferences, balancing exploration and exploitation in a principled way with Bayesian Optimization for expensive-to-evaluate black-box objectives.
Experimenting with the technique, I trained a Gaussian Process proxy model on the implicit preferences in my code repo's commit history at VQASynth.
The result: I used the model's preference scores to re-rank candidate papers recommended based on my interests in spatial reasoning and multimodal data synthesis.
Semantic relevance is a high-recall method for finding arXiv papers personalized to your interests. Adding contributor preferences, extracted from the merge history of your code offers a high-precision filter.
So what's next? I'm using the model to synthesize a larger volume of preference data to finetune an open-weight coding model with DPO and LoRA. Tuning Coding Agents via Implicit Preference Distillation
arXiv: https://arxiv.org/pdf/2510.17671
Substack: https://remyxai.substack.com/p/lilo-and-myx
VQASynth: https://github.com/remyxai/VQASynth
- 1 reply
reacted to Imosu's post with 👀 3 months ago
Post
3363
# ZeroGPU Hardware Mismatch: Why Am I Getting RTX PRO 6000 Blackwell MIG Instead of the Documented H200?
I recently ran into a surprising issue while debugging a Hugging Face ZeroGPU Space.
According to the Hugging Face ZeroGPU documentation, ZeroGPU is described as using NVIDIA H200-based resources, with configurations such as “large” and “xlarge” offering H200-class memory. However, when I printed the actual GPU information inside my Space, I got something different:
GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition MIG 2g.48gb
Capability: (12, 0)
Torch: 2.8.0+cu128
CUDA: 12.8
This is not an H200. It appears to be a MIG slice of an RTX PRO 6000 Blackwell Server Edition GPU, with 48GB VRAM.
This difference matters. It is not just a cosmetic hardware-name issue.
In my case, the Space was running Qwen3-TTS and failed with:
CUDA error:
no kernel image is available for execution on the device
The issue appears related to GPU architecture compatibility. The app was using kernels-community/flash-attn3, which is generally aligned with Hopper-class GPUs such as H100/H200, but the actual device exposed to the Space was Blackwell with compute capability 12.0. As a result, CUDA kernels that might work on the expected H200 environment failed on the actual assigned GPU.
To be clear, I am not saying the RTX PRO 6000 Blackwell is a bad GPU. It is a newer architecture and may be powerful in many workloads. But it is not the same as H200, and the software ecosystem compatibility is different. For ML workloads, especially those relying on custom CUDA kernels, the exact GPU architecture matters a lot.
This raises a few questions:
Is Hugging Face ZeroGPU now assigning RTX PRO 6000 Blackwell MIG instances instead of H200 instances?
If yes, why is this not clearly documented?
I recently ran into a surprising issue while debugging a Hugging Face ZeroGPU Space.
According to the Hugging Face ZeroGPU documentation, ZeroGPU is described as using NVIDIA H200-based resources, with configurations such as “large” and “xlarge” offering H200-class memory. However, when I printed the actual GPU information inside my Space, I got something different:
GPU: NVIDIA RTX PRO 6000 Blackwell Server Edition MIG 2g.48gb
Capability: (12, 0)
Torch: 2.8.0+cu128
CUDA: 12.8
This is not an H200. It appears to be a MIG slice of an RTX PRO 6000 Blackwell Server Edition GPU, with 48GB VRAM.
This difference matters. It is not just a cosmetic hardware-name issue.
In my case, the Space was running Qwen3-TTS and failed with:
CUDA error:
no kernel image is available for execution on the device
The issue appears related to GPU architecture compatibility. The app was using kernels-community/flash-attn3, which is generally aligned with Hopper-class GPUs such as H100/H200, but the actual device exposed to the Space was Blackwell with compute capability 12.0. As a result, CUDA kernels that might work on the expected H200 environment failed on the actual assigned GPU.
To be clear, I am not saying the RTX PRO 6000 Blackwell is a bad GPU. It is a newer architecture and may be powerful in many workloads. But it is not the same as H200, and the software ecosystem compatibility is different. For ML workloads, especially those relying on custom CUDA kernels, the exact GPU architecture matters a lot.
This raises a few questions:
Is Hugging Face ZeroGPU now assigning RTX PRO 6000 Blackwell MIG instances instead of H200 instances?
If yes, why is this not clearly documented?
- 2 replies
reacted to fffiloni's post with 🔥 3 months ago
Post
514
Great technical guide by Nico Martin on the Hugging Face blog, showing how to use Transformers.js inside a Chrome extension and run ONNX models from the Hub locally with WebGPU inside a Manifest V3 extension.
The interesting part: this is not just a chatbot in a side panel.
The article walks through the architecture behind a browser agent that can read open tabs, query webpages, search history, and highlight elements directly on the page — with models downloaded from the Hugging Face Hub, cached under the extension origin, and executed locally instead of being called through a remote API for every prompt.
A strong blueprint for building local-first web copilots, reading assistants, and AI-powered browsing workflows.
Article: https://huggingface.co/blog/transformersjs-chrome-extension
The interesting part: this is not just a chatbot in a side panel.
The article walks through the architecture behind a browser agent that can read open tabs, query webpages, search history, and highlight elements directly on the page — with models downloaded from the Hugging Face Hub, cached under the extension origin, and executed locally instead of being called through a remote API for every prompt.
A strong blueprint for building local-first web copilots, reading assistants, and AI-powered browsing workflows.
Article: https://huggingface.co/blog/transformersjs-chrome-extension
reacted to blanchon's post with ❤️ 3 months ago
Post
2899
I'm releasing OpenCS2 a 11TB dataset of around 5000 hours of counter strike gameplay recording.
- HD resolution - 1280×720 · 32 fps
- For each frame keyboard and mouse + world state (player position, velocity, weapon ...)
- HD Stereo audio
- All 10 players perspective
https://huggingface.co/collections/blanchon/opencs2
- HD resolution - 1280×720 · 32 fps
- For each frame keyboard and mouse + world state (player position, velocity, weapon ...)
- HD Stereo audio
- All 10 players perspective
https://huggingface.co/collections/blanchon/opencs2
- 1 reply
reacted to danielhanchen's post with ❤️ 3 months ago
Post
6008
We’re excited to announce that Unsloth has joined the PyTorch Ecosystem! 🔥🦥
Unsloth is an open-source project that makes training & running models more accurate and faster with less compute. Our mission is to make local AI accessible to everyone. Thanks to all of you for making this possible! 💕
Blog: https://unsloth.ai/blog/pytorch
GitHub: https://github.com/unslothai/unsloth
Unsloth is an open-source project that makes training & running models more accurate and faster with less compute. Our mission is to make local AI accessible to everyone. Thanks to all of you for making this possible! 💕
Blog: https://unsloth.ai/blog/pytorch
GitHub: https://github.com/unslothai/unsloth
- 2 replies
reacted to TravisMuhlestein's post with 🔥 3 months ago
Post
136
Agent ecosystems are starting to expose a new class of infrastructure problems around identity, interoperability, trust, and coordination.
Excited to see GoDaddy and HOL (Hashgraph Online) exploring open standards for verifiable AI agent identity and DNS-based coordination layers for emerging agent systems.
A lot still needs to evolve around orchestration, governance, and runtime trust boundaries, but it’s interesting to see more attention shifting toward the infrastructure layer of operational AI systems.
https://www.einpresswire.com/article/910813665/godaddy-and-hol-hashgraph-online-propose-open-standards-for-verifiable-ai-agent-identity-on-dns
Excited to see GoDaddy and HOL (Hashgraph Online) exploring open standards for verifiable AI agent identity and DNS-based coordination layers for emerging agent systems.
A lot still needs to evolve around orchestration, governance, and runtime trust boundaries, but it’s interesting to see more attention shifting toward the infrastructure layer of operational AI systems.
https://www.einpresswire.com/article/910813665/godaddy-and-hol-hashgraph-online-propose-open-standards-for-verifiable-ai-agent-identity-on-dns
reacted to their post with 🚀 3 months ago
Post
2985
The UFO/UAP Dataset is complete!
unmodeled-tyler/DoW-UFO-UAP-1
The most recent release from the Department of War is there up in full and ready for analysis!
The dataset ships with an Hermes Agent Skill so you can quickly and easily start parsing through the data immediately.
Go chase some anomalies! 🚀
unmodeled-tyler/DoW-UFO-UAP-1
The most recent release from the Department of War is there up in full and ready for analysis!
The dataset ships with an Hermes Agent Skill so you can quickly and easily start parsing through the data immediately.
Go chase some anomalies! 🚀