Local Large Model and Embedding Experiments
Experiments on local LLMs and embedding models with consumer GPU hardware.
This project records experiments on running local LLMs and embedding models on consumer GPU hardware.
Highlights
- Explored local deployment of Qwen-series models and BGE embedding models.
- Studied Word2Vec, FastText, BGE training stages, contrastive learning, hard negatives, and RAG-oriented embeddings.
- Hardware context: RTX 5060 8GB on Linux / CachyOS.
Context
The aim is to understand the full path from representation learning concepts to practical retrieval systems, especially when constrained by local hardware and deployment limits.