About this channel
Hugging Face spans hands-on AI building and discussion of machine learning research. Browser inference, local models, agent memory and training methods all appear in its videos. The channel gives…
Hugging Face spans hands-on AI building and discussion of machine learning research. Browser inference, local models, agent memory and training methods all appear in its videos. The channel gives viewers several ways to approach the subject, from constructing a specific application to examining the ideas behind how a model or agent learns. Open-source tools are a visible part of that practical emphasis.
Tutorial-style episodes include building with Pi and Telegram, running local models through llama.cpp and GGUF, and producing structured output in a browser with Transformers.js. These are concrete implementation topics, with attention to where a model runs and how its output becomes usable. An episode about WebGPU kernels adds a lower-level perspective on browser AI, while a weather prediction project offers a different application for experimentation.
Research discussion has its own place through the Hugging Face Journal Club. Topics include preference models, training AI systems to replicate research and direct on-policy distillation. Introductory material on flow matching sits alongside a series segment that moves from reward functions to training environments. An explanation of agent memory connects architecture to the broader question of what an agent needs beyond an immediate response. The range allows viewers to move between building something and investigating the concepts underneath it.
Developers experimenting with local or browser-based AI will find material relevant to implementation choices, while learners interested in model training can follow the explanatory and research-oriented episodes. The channel suits those who want to connect code, system design and research vocabulary. Different videos offer different levels of entry, so a practical project can be a starting point before moving into the training methods or architectural questions it raises.
Latest episode
HF Buckets: Upload Only What Changed with Xet
If you work with large datasets, you're probably re-uploading the same bytes every time the data changes.
In this video I upload a 191 MB NYC taxi CSV to a Hugging Face Bucket, append 100 rows, and…
If you work with large datasets, you're probably re-uploading the same bytes every time the data changes.
In this video I upload a 191 MB NYC taxi CSV to a Hugging Face Bucket, append 100 rows, and upload it again: Xet, Hugging Face's storage layer, sends only the new pieces. Then I convert the data to Parquet, prepend 100 rows, and sync again: with PyArrow's use_content_defined_chunking=True, only 8.6 MB of the 52.9 MB file goes over the wire. For data and ML engineers who push datasets to the Hub and keep them up to date.
Topics covered:
- How Xet uploads only the pieces that changed
- Hugging Face Buckets with the hf CLI (create, cp, sync)
- Content-defined Parquet blocks with PyArrow 21+
- Upload savings: ~90 MB vs ~6 MB on a 100 MB file
- Production gotchas: S3 tools, versioning, sorting, writer settings
Resources:
- Full walkthrough (commands, code, dataset): https://huggingface.co/blog/prpatel/hands-on-with-buckets
- Hugging Face forum: https://discuss.huggingface.co/
Chapters:
0:00 Stop wasting bandwidth
0:15 What you'll see
0:39 Demo: CSV and Parquet uploads to a Bucket
9:11 Xet, in three steps
10:10 The one-line fix
10:47 How much Xet saves
11:13 Before you ship it
12:27 Join the discussion
YouTube Shorts
uploads
More episodes from this channel
Build Your Own Open-Source Dots with Pi and Telegram
2.10.2026
Build your own dots-style, always-on AI assistants with Pi and pi-gateway, using your choice of models on a VPS or another always-on machine. This tutorial walks through agent workspaces, skills, Google Workspace access, and...
Predict next week's weather yourself
17.9.2026
Weather forecasting is very important, and people are moving towards using neural network models to do it. We at Hugging Face are interested in the tooling around the research and want to make it...
Transformers.js v4.3: Structured Output in the browser
16.9.2026
Transformers.js v4.3 is out, and the big one is structured output. With the new @huggingface/transformers-structured-output package you can force a model to follow an exact JSON Schema or regex, directly in the browser. No...
Flow matching for Beginners | @ariG23498
15.9.2026
Understand flow matching from ground up with more visualization and less math. Blog: https://huggingface.co/spaces/ariG23498/flow-matching Timestamps: 00:00 - Introduction 00:50 - Noise to Image 06:23 - What to train 13:23 - Train velocities 19:54 -...
We shipped 207 WebGPU Kernels for Browser AI
4.9.2026
We shipped 207 WebGPU kernels for browser AI, all open source on the Hugging Face Hub, plus @huggingface/kernels, a library that loads, validates, renders and runs them for you. This video walks through how...
Hugging Face Journal Club: AI Research Preference Models
4.9.2026
The Hugging Face research team discussed Meta's new paper on using LLMs to predict "research taste" to improve the capabilities of agents to undertake machine learning experiments autonomously. Paper link: https://huggingface.co/papers/2608.13940
Training Agents 4: From reward functions to environments.
2.9.2026
Training Agents, Session 4: from reward functions to environments. In this live session, the reward stops being a function and becomes a place the agent acts in. In Session 3 the model answered once...
Hugging Face Journal Club: Training AI Scientists to Replicate Research
21.8.2026
The Hugging Face research team discusses the Faraday-27B model that was post-trained to develop scientific reasoning to replicate AI papers, a key step towards full automation of AI R&D. Link to the paper: https://huggingface.co/papers/2608.13331
Agent Memory EXPLAINED - Complete Architecture
17.8.2026
Quick case study persistent memory for AI agents featuring Mem0. The architecture of long form memory for agents goes far beyond a vector search. This deep dive explains Mem0's memory stores, ingestion pipeline, hybrid...
Hugging Face Journal Club: Direct On-Policy Distillation
11.8.2026
The Hugging Face research team discusses the paper "Weak-to-Strong Generalization via Direct On-Policy Distillation" which proposes a cheap way to transfer the benefits of reinforcement learning from small models to much larger ones. Instead...
Prompt Caching Explained: Stop Overpaying for AI Agents
10.8.2026
Prompt caching can cut the input cost of long AI agent sessions dramatically—but only when your harness preserves reusable prompt prefixes. This video explains what prompt caching actually stores, why agent costs compound across...
How AI agents reproduced ICML 2026 papers
6.8.2026
We just wrapped up the 2026 ICML Reproduction Challenge, with more than 1,200 participants & their agents! More than 2,000 different papers were reproduced or falsified (1/3 of all of the papers at the...
Hugging Face Journal Club: Scaling Laws for Pre-training & RL
4.8.2026
The Hugging Face research team discusses the paper "Understanding Reasoning from Pretraining to Post-Training" which discovers a joint scaling law for pre-training and reinforcement learning. Paper link: https://huggingface.co/papers/2607.16097
Tau Crash Course: The Python Port of Pi
31.7.2026
Tau is a minimalist coding-agent harness inspired by Pi and implemented in Python. This crash course tours its Textual TUI, provider login and model selection, tree-structured sessions, skills, custom prompts, themes, extensions, and the...
Hugging Face Journal Club: Kimi K3
29.7.2026
The Hugging Face research team discusses the Kimi K3 tech report: https://huggingface.co/papers/2607.24653
Training Agents 3: Reinforcement Learning
22.7.2026
Training Agents, Session 3: from a fine-tuned baseline to reinforcement learning with GRPO. In this live session, we'll move past imitation and start training on outcomes. We'll take the model from the earlier sessions...
Hugging Face Journal Club: AsyncOPD and How Stale Can On-Policy Distillation Be?
21.7.2026
The Hugging Face post-training team discusses how on-policy distillation (OPD) can be made fully asynchronous to improve the throughput of training by 2-3x Paper: https://huggingface.co/papers/2606.24143
Local AI 201: Inference Engines, Hardware Stack
19.7.2026
- Link to Introduction to Local AI (101) https://www.youtube.com/watch?v=wRcByxXkJCQ - ODS https://github.com/osmantic/ods - Inference Engines blog series http://thelocalaibook.osmantic.com/ Find our speakers below: - https://x.com/TheAhmadOsman - https://x.com/MikeBradleyAI - https://x.com/alexocheema 00:00 Introduction to Local AI 201...
New Model: Inkling by Thinking Machine on Hugging Face
15.7.2026
Inkling by Thinking Machines is here, a 1 TRILLION parameter open model that natively understands images, text, AND audio, with a 1M token context window. In this video, we break down the architecture, show...
Build Small Winners Reveal
9.7.2026
After reviewing hundreds of creative projects from builders around the world, it's finally time to announce the Build Small Hackathon winners 🎉 Congratulations to all our winners, and thank you to everyone who joined...
Continue reading
You may also like
Innohub TV
Watch this content also on Innohub TV
We picked clips from Innohub TV that continue the same topic.
