Search: “clem-delangue--x”

HUGGING FACE JUST AUTOMATED THEIR ENTIRE POST-TRAINING TEAM WITH AN AGENT. It reads papers, runs GPU experiments, iterates, and builds research-backed models autonomously. Pushed a benchmark from 10%

I tested @huggingface ml-intern, given the prompt 'Fine-tune a Segment Anything Model (SAM) on a useful medical dataset. Train the model, and provide a comprehensive tutorial in a Jupyter Notebook fil

Introducing ml-intern, the agent that just automated the post-training team @huggingface It's an open-source implementation of the real research loop that our ML researchers do every day. You give it

it's entirely fine and great for proprietary models to exist the issue is they try to kill open source behind scenes while pretending they support it so far they've been pretty incompetent but they'll

I’ve been using ml-intern for a while, and it genuinely changed my workflow. It's super good at: - Model/Dataset discovery. - Post-Training setup iteration. - Data processing workflows. Huge shoutout

Karpathy's autoresearch repo started an impressive trend. Agents can now train AI models to build SoTA agentic systems. And to think this is just scratching the surface. Ultimately, it boils down to g

This post celebrates RC Lens fans experiencing an euphoric moment, likely following a significant sporting victory or achievement at their Bollaert stadium. The enthusiastic French-language post with

Love this work from Aksel and the post-training team at Hugging Face! Turns out the HF ecosystem (papers, datasets, models all accessible through CLI, skills and md files) is perfect for running SOTA

open-source community will work their magic! If any inference providers could serve kimi k2.6 at 300+ tps reliably, I would be happy to switch from gpt 5.4 xhigh to kimi for daily tasks. It's a really

Share this with your representative! https://huggingface.co/blog/cybersecurity-openness I’m hearing there’s renewed lobbying in DC and in state legislatures to ban or severely restrict open-source. Li

Dataset 2 (Drift) is a 5,000-sample preview dataset hosted on Hugging Face, named after drift diffusion modeling concepts. The dataset appears to represent a novel or experimental approach to data col

dear god lol - new kimi model is a fucking beast. GPT 5.4 level coding, 76% cheaper than opus 4.7 and 100% open source / free to use, i mean look at this: > kimi k2.6 can code continuously for 12 hour

DFlash for Qwen3.6-35B-A3B just dropped ⚡ The community was running the day-1 preview before we even finished training. Now it's done: ✅ Training complete ✅ Validation passed ✅ Weights finalized ↓ Go

Free ~20-50% tok/s on a local llama.cpp setup if you already have a draft model sharing vocabulary with your main one. Local stack quietly got faster this weekend https://x.com/TechIno219886/status/20

Kimi K2.6, 1T params runs on my desktop GPU workstation. 192GB VRAM + 475GB RAM Serious contender to be the best open coding LLM on Earth. We'll see I'm checking their default harness. Goes on and on

Kimi-K2.6 is a language model that has been released on HuggingFace, a popular platform for sharing machine learning models and datasets. The announcement was made by Clem Delangue, likely indicating

Kimi K2.6 just dropped. And it crushed Claude Opus 4.6 on SWE-Bench Pro. Kimi K2.6: 58.6 GPT-5.4 xhigh: 57.7 Gemini 3.1 Pro: 54.2 Claude Opus 4.6: 53.4 An open source Chinese model is now #1 on agenti

Kimi K2.6 just entered the chat. > beats gpt 5.4 and opus 4.6 on coding. > open source. open weight. > 3x-5x cheaper. > really good for people running agents. > insane at running long tasks. > we're t

life when you discover an open-source model that runs 300 parallel agents, executes for 12+ hours straight, beats GPT-5.4 and opus 4.6 on multiple benchmarks... and the weights are on huggingface Medi

Most of our usage is now coming from agents so to monitor how they use and mention our tools, with @DAKlingbeil, we're now running regular 10k queries to all the most popular coding agents, share the

🤯QWEN 3.6 35B-A3B IS INSANE AT CODING @unslothai recently dropped Qwen3.6-35B-A3B-GGUF The first open Qwen3.6 model and it’s solid. This MoE beast is cooking on benchmarks: 🧠SWE-bench Verified: 73.4 (

Sharing an OS tool I have been building for a few weeks: http://opentraces.ai - git for agent traces. With OT, you have a local client that helps you santise, review and securely share traces from you

We're opening a Hugging Face office in Tokyo! Our goal: help open-source AI develop in Japan and grow the local community. Let's meet! ハギングフェイスの東京オフィスがオープンしました！ 私たちの目標は、日本におけるオープンソースAIの発展を支援し、ローカルコミュニ

Why? - Risk comes from capability asymmetry between attackers and defenders. Open source is what every frontier lab already trains on so defenders get the same AI firepower as attackers. With propriet

🔴 BREAKING Speculative decoding just got real for local LLMs. llama.cpp merged speculative checkpointing (PR #19493). Expect 0-50% speedups on coding tasks with `--spec-type ngram-mod`. This means: •

Good morning, everyone! TLDR: full base weights for healed 18b merge are LIVE! I am so blessed and excited by all of the support for my frankenmerge of Jackrongs models. Positive and negative feedback

This is a YouTube interview featuring Victor from Hugging Face, shared by Clem Delange on X (formerly Twitter). The specific topics discussed in the interview are not detailed in the source material p

Localmaxxing.com is a platform in private testing phase with plans for public release the same week this announcement was made, designed to simplify benchmark sharing by eliminating the need for users

I gave two MoE models the same vibe coding challenge Qwen3.6 35B A3B (31.8GB) vs Gemma4 26B A4B (23.3GB) Stack: > Unsloth Q6_K_XL > llama.cpp > Model-card recommended sampling for each 4 prompts, side

I interviewed Huggingface's head of product Victor. Huggingface is AI's core I think this is the most important platform if you're interested in learning AI, it introduced me to everything I know abou

Introducing OpenMythos An open-source, first-principles theoretical reconstruction of Claude Mythos, implemented in PyTorch. The architecture instantiates a looped transformer with a Mixture-of-Expert

releasing qwen3.6 35b a3b carnice edition (inspired by @kaiostephens) to enjoy an hermes agent tailored finetuned on unified memory hardware! safetensors and gguf versions (q8, 6,5 and 4) available! f

Who is running local models on GPUs on OpenClaw? I have started benchmarking different models this week. I am working on improving model selection and switching UX on OpenClaw, i.e. I run /model vllm/

WildChat is an amazing project. The project page seems not to be updated anymore, so people might not know that the dataset has been updated with a very large number of transcripts through July 2025:

This post explains the technical details behind common terms and tools encountered on Hugging Face for running large language models locally, including quantization formats (GGUF, Q_4_M, IQ_4XL), opti

This post discusses a Hugging Face Space that enables replaying AI agent execution traces, contrasting this capability with Netflix's replay functionality to illustrate how agent traces provide differ

You should watch this. It just shows how disconnected we are from the small group of people making decisions that will impact our future heavily. These people have so much ai psychosis. If you listen

Hugging Face is expanding its platform to support AI agents by enabling them to call upon 1 million Hugging Face Spaces, allowing agents to access diverse specialized tools and capabilities for variou

Morning everyone! I put this merge together yesterday, put it on huggingface to share with Jackrong, and a bunch of people discovered it when it got cloned into Jackrongs repo, and it’s going a bit vi

Okay this one is insane. A new 18B frankenstein model was just released on @huggingface — Beats the new Qwen3.6-35B-A3B on a 44-test suite despite requiring 12GB VRAM instead of 24GB 🤯 Runs on a SINGL

RIP Anthropic API fees. Someone just made Claude Code run 100% locally on a MacBook for $0/month. 122B parameter model. 65 tokens per second. Nothing touches the cloud. The trick everyone else missed:

It's never been easier for your agent to harness the power of Hugging Face, from account creation to getting your own finetuned model, with everything in between, without any manual action on your end

Clem Delangue, CEO of Hugging Face, is soliciting beta testers for new dataset hosting features specifically designed for robotics data. The post seeks to identify organizations and researchers curren

@opentraces 0.3 is out! Big update: a stronger security pipeline with TruffleHog and LLM review, generation-aware ingest for living traces, line-level attribution + inverse blame with a @gitbutler-sty

Sharing my current setup to run Qwen3.6 locally in a good agentic setup (Pi + llama.cpp). Should give you a good overview of how good local agents are today: # Start llama.cpp server: llama-server -hf

The post praises a coding agent tracer/viewer tool available on Hugging Face, highlighting its quality and usefulness for developers working with agents. The tool appears to be part of the pi-mono dat