An evolutionary tree for the LLM zoo

What it does

This repository is a curated companion to the survey paper Harnessing the Power of LLMs in Practice. It collects and organizes papers, models, and data resources into practical categories—BERT-style encoders, GPT-style decoders, pretraining and finetuning datasets, and downstream NLP tasks—while flagging usage restrictions for commercial and research use. Think of it as an annotated bibliography that ships with its own evolutionary tree diagram and a decision-flow chart for choosing between LLMs and fine-tuned models.

The interesting bit

The maintainers built an evolutionary tree tracing lineage from BERT to LLaMA 2 and Claude 2, and they publish the PowerPoint source so you can edit the figure yourself. They also include a decision-flow diagram for matching NLP applications to the right model type, which is rarer than it should be in survey repos.

Key highlights

• Organizes models by architecture style (encoder/decoder-only vs. encoder-decoder) with direct paper links
• Covers the full lifecycle: pretraining data, finetuning data, test data, and real-world task benchmarks
• Explicitly catalogs usage and licensing restrictions for models and datasets
• Provides editable PowerPoint sources for the LLM evolutionary tree and decision-flow figures
• Still actively updated; latest additions include LLaMA 2 and Claude 2

Caveats

• Section headers contain visible typos (“Restractions,” “Communtity”)
• The catalog is extensive and flat; expect to scroll through long categorized lists
• It is a curated index and figure collection, not a framework or runnable toolkit

Verdict

Worth bookmarking if you are a practitioner or researcher trying to orient yourself in the LLM paper deluge; skip it if you are looking for code, training scripts, or a model zoo you can download and run.