20/06/2026
🚀 We just open-sourced something we've been building.
𝗘𝗰𝗵𝗼 𝗣𝗮𝘆𝗹𝗼𝗮𝗱 — a free, open-source training toolkit for making AI less wrong.
AI hallucinations aren't just annoying. In healthcare, legal research, education, and finance — they cause real damage. Wrong drug interactions. Fabricated case law. False statistics presented as fact.
We built a solution:
📦 10 fine-tuning packet modules (calibration, citation faithfulness, anti-sycophancy, injection resistance, and more)
📊 A 7-metric evaluation harness — so "safer" is measured, not just claimed
🔁 Works with every major AI model family: GPT-4o, Claude, Gemini, Llama, Mistral, Qwen, Phi, and more
✅ MIT licensed, JSONL-ready, drop it into your existing training pipeline
We built this because the baseline needs to be higher — for everyone, not just well-funded labs.
🔗 Star the repo and try it out:
https://github.com/OneiroNexus/echo-payload
10-packet fine-tuning payload for safer, lower-hallucination LLMs. 7-metric eval harness. Model-agnostic. DPO/SFT/ORPO ready. - OneiroNexus/echo-payload