Self-Training Agents: Hermes Agent, HF Traces, Skills, MCP & Finetuning
summary
Open-source models are now competitive: GLM 5.1 tops the Artificial Analysis intelligence index, and full-weight access enables quantize, fine-tune, and edge deployment. The Hugging Face flow being shown wires together inference providers, a traces repository for agent sessions, MCP and skills for job orchestration, and Hermes for memory and local integrations — Hermes agent manages memory and Slack integration. Demoed pipeline actually fine-tunes a vision‑language model where an agent computes VRAM and selects instance then launches the job; OCR batch processing of ~30k papers runs end‑to‑end.
AI-written summary · about
chapters (15)
- 0:00Introduction to Open Agent Ecosystem
- 0:39Importance of Open Source in Machine Learning
- 2:36Hugging Face Hub overview
- 3:06Agentic models and Vision-LMs
- 4:24Benchmark datasets and model filtering
- 5:16Inference providers and model routing
- 6:50Local coding agents and tools
- 7:46Hermes agents for memory management
- 9:20Traces repository for agent sessions
- 10:22Tips for finding and serving local models
- 12:07Supercharging agents with Hugging Face skills
- 13:41Live demonstration of agent-driven fine-tuning
- 14:41Training vision models (object detection/segmentation)
- 15:00Using Model Context Protocol (MCP) for agents
- 16:30Case study: OCR processing for AI papers
related talks
21mDemocratizing AI with Open Source Multi-Agent Systems
35mThe Fight for the Future of Artificial Intelligence
58mAdvanced Autonomous AI Agents: Memory, Collaboration and Evolution in Action
40mAI Agents in Practice
40mAgents in Action: Advancing Open Source AI Missions
07h 47mAI Agents Conference 2026 - Day 5
14mReliable Agentic AI, the Open Way
49mAI Agents: Collaboration in Real-Life Applications