Text Generation
Transformers
Safetensors
imp
custom_code

Problem with Vulkan Demo

#3
by Fcsalvagnini - opened

Hi.

Great work you did, guys. I am facing a problem here with running MLC-Imp on Linux. Could you guys provide me some guidance?

I am trying your model using the MLC-imp repository. By following exactly what is in the markdown, I could compile the tvm and mlc_llm, generating a vulkan so (imp-v1.5-3B-196-q4f16_1-vulkan.so). However, when I try the Python snippet from your repository, I got the following answer:

advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advisesadvises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises

Hi.

Great work you did, guys. I am facing a problem here with running MLC-Imp on Linux. Could you guys provide me some guidance?

I am trying your model using the MLC-imp repository. By following exactly what is in the markdown, I could compile the tvm and mlc_llm, generating a vulkan so (imp-v1.5-3B-196-q4f16_1-vulkan.so). However, when I try the Python snippet from your repository, I got the following answer:

advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advisesadvises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises advises

Thank you for your issue. Could you please provide your prompt and image?

I am using the bus image, and the standard prompt:

output = cm.generate(
prompt="\nWhat are the colors of the bus in the image?",
pixel_values=image_features
)

I am using the bus image, and the standard prompt:

output = cm.generate(
prompt="\nWhat are the colors of the bus in the image?",
pixel_values=image_features
)

I have encountered similar errors before because of poor device performance. What model is your graphics card?

Sign up or log in to comment