config.json tokenizer.json model.safetensors
Throughput and scores from a Qwen3 32B benchmark
Full precision is technically 32 bits, but 16 bits loses almost no accuracy, so people loosely call 16 bits full precision.
Other common labels are Chat or Instruct for a fine tune and A3B for 3B active parameters.
So then why aren't the LLMs we use every day incredibly slow?