Base LLM
Base LLM is a full-stack Chinese-language course going from traditional NLP foundations, word vectors and RNNs, through the Transformer architecture, to modern LLMs, covering hand-written BERT, GPT and Llama 2, parameter-efficient fine-tuning, RLHF and deployment.
Base LLM is a Chinese-language, full-stack tutorial that takes a reader from traditional natural language processing through to large language models. It argues that developers who jump straight to calling an API or fine-tuning a model often skip the underlying NLP foundations, and works systematically from tokenization and word vectors through RNNs and LSTMs to the Transformer architecture and milestone pretrained models such as BERT, GPT and T5.
Its advanced section has the reader hand-write a Llama 2 model from scratch and covers parameter-efficient fine-tuning methods such as LoRA and RLHF, while later material covers model quantization, inference acceleration, and deploying a model as a service with Docker and FastAPI, plus a look at model safety, ethics and multimodal models. Practical projects include text classification, named entity recognition and fine-tuning Qwen2.5 on private data, and the course is aimed at students, engineers moving from classical machine learning into large language models, and researchers.