InstructLab: Applying Open Source Methods to Building and Trai...
summary
InstructLab lets contributors add domain skills or knowledge to a base open model without forking by using a taxonomy-driven workflow that generates hundreds-to-thousands of examples via a teacher model, validates with a critic, and applies multi-stage tuning to a student model; it runs locally (Mac M1/M2 with ~32 GB using quantized Merlinit 7B Q4 via llama.cpp) or on heavy backends (initial tuning used 128 GPUs for 25+ hours with MLX/Kora/DeepSpeed and LoRA) and publishes tuned versions to Hugging Face, proving the pipeline works end-to-end.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:07AI and LLMs overview
- 2:34LLM tuning challenges
- 8:00The problem of contributing to LLMs
- 10:30Introducing LAB technique
- 15:27LAB skills and knowledge
- 21:06Additive model updates with LAB
- 22:12InstructLab community
- 25:45InstructLab workflows and testing
- 31:48Live demo and CLI
- 36:29Summary and future
related talks
40mInstructLab: Applying Open Source Methods to Building and Trai...
30mApplying Open Source Methods to Building and Training Large Langu...
28mA community based open source LLM and what it can do for your business
35mCreating Your Own LLM Tuning Platform with Open Source Technologies
05mKeynote: Teaching LLMs New Knowledge
01h 49mAI engineering with open access LLMs that lie, curse, and steal
29mLocal LLM fine-tuning: a practical example
44mOpen-Source and Science in the Era of Foundation Models