Large Language Models
- Artificial Analysis has a great breakdown of LLM quality vs cost vs latency vs speed
- (mini-omni2)[https://github.com/gpt-omni/mini-omni2] is a project to make something like GPT-4o with multimodal and speech duplex capability
Subtopics
Projects
Notes
Tools
- H2OGPT Local web-based interface for chatting with document db indexes etc.
- Lit-GPT - Pretrain, finetune, deploy 20+ LLMs on your own data. Uses state-of-the-art techniques: flash attention, FSDP, 4-bit, LoRA, and more.
- Instructor Embedding - Embedding specifically adapted to instructions
Finetuning
Finetuning delivers better results than RAG, but requires a training step and data prepared in the right format.
Tools
- Axolotl - Axolotl is a tool designed to streamline the fine-tuning of various AI models, offering support for multiple configurations and architectures.
Scaling Laws
Performance improvements 2023-2024:
gwern on the point of o1 in the perspective of the scaling laws
