gpt4all run gguf model cost much less memory than llama.cpp, amazing, how can you do this

[javalover123]

Bug Report

gpt4all run gguf model cost much less memory than llama.cpp, amazing, how can you do this

Steps to Reproduce

1. Download DeepSeek-R1-Distill-Qwen-1.5B-Q8_0.gguf, the file is about 1.8GB
2. llama-cli -m DeepSeek-R1-Distill-Qwen-1.5B-Q8_0.gguf, cost about 1.9GB memory
3. gpt4all just cost 260MB memory, amazing, how can you do this

Expected Behavior

Can you contribute this feature to llama.cpp?

Your Environment

• GPT4All version (if applicable): 3.9.0
• Operating System: Win 10
• Chat model used (if applicable): DeepSeek-R1-Distill-Qwen-1.5B-Q8_0.gguf