Hugging Face Transformers
Serves as the base ecosystem for integration; Unsloth builds on it to optimize fine-tuning performance.
BitsAndBytes
Focuses on quantization techniques like 4-bit; Unsloth offers dynamic quantization with better VRAM efficiency and accuracy.
llama.cpp
Primarily a deployment and inference tool; Unsloth supports exporting models compatible with llama.cpp after fine-tuning.
vLLM
Used for efficient inference; Unsloth exports fine-tuned models to vLLM format for deployment.