mlc-llm
mlc-llm copied to clipboard
mlc-ai
Wrong answers in simple calculations
Hello,
In mlc-chat-cli, Vicuna does not is able to answer simple questions like square root. Any clue about ? Thank you guys (Windows 10, vicuna-v1-7b-q3f16_0)
Examples....
USER: What is the square root of 36 ? ASSISTANT: The square root of 36 is 4.
USER: what is the square root of 49 ? ASSISTANT: The square root of 49 is 7.
USER: what is the square root of 36 ? ASSISTANT: The square root of 36 is 7.
...
Obs.: I'm a simple enthusiast of AI, perhaps a future user of such tech. I reported this observation about wrong square root here just to developers of these brilliant tools know - & perhaps adjust some bolts in this system to avoid so primary mistakes in their answers. Thanks a lot
MLC LLM didn't make the model, so this is a question for the makers of Vicuna (https://huggingface.co/lmsys/vicuna-13b-delta-v1.1).
MLC LLM makes it easier to run these models on various types of hardware.
You might get better results by using a version of Vicuna with more parameters, but that would also require more memory.
