What's the difference between this and unsloth/DeepSeek-R1-0528-Qwen3-8B-bnb-4bit
What's the difference between this and unsloth/DeepSeek-R1-0528-Qwen3-8B-bnb-4bit
You can change the model name to whichever model you like by matching it with model's name on Hugging Face e.g. 'unsloth/llama-3.1-8b-unsloth-bnb-4bit'.
Model names ending in unsloth-bnb-4bit indicate they are Unsloth dynamic 4-bit quants. These models consume slightly more VRAM than standard BitsAndBytes 4-bit models but offer significantly higher accuracy.
If a model name ends with just bnb-4bit, without "unsloth", it refers to a standard BitsAndBytes 4-bit quantization.
Models with no suffix are in their original 16-bit or 8-bit formats. While they are the original models from the official model creators, we sometimes include important fixes - such as chat template or tokenizer fixes. So it's recommended to use our versions when available.