Why quantized model of 640MB took almost 3GB of VRAM?

[Bakht-Ullah]

I have quantized openai/whisper-medium model with W4A16 GPTQ method having 2.8GB of size to 640MB. But I am facing the issue when I load two quantized model simultaneously it utilize almost 6GB of the VRAM. Can anyone know why this utilize too much memory.

#loading model

llm = LLM(
model="Bakht123/whisper-medium-gptq-W4A16-G128",
max_model_len=448,
max_num_seqs=64,
limit_mm_per_prompt={"audio": 1},
)

llm1 = LLM(
model="Bakht123/whisper-medium-gptq-W4A16-G128",
max_model_len=448,
max_num_seqs=64,
limit_mm_per_prompt={"audio": 1},
)

#memory utilization