Hugging Face Demo

[Mandalorian007]

Motivation & Examples

This project looks absolutely incredible, but I have been really struggling to get it running locally due to not having a dedicated GPU and remote on a colab notebook is a bit messy trying to upload content through tunnels with gradio. I was thinking it would be pretty epic if this project had a Hugging Face style demo to actually try out the project easier. No pressure, but just some feedback from a novice AI vision user.

All the same incredible work this has to be one of the better looking promising projects that has potential to incorporate into other software.

[luca-medeiros]

Appreciate the kind words.
Yea, thats something I thought about it before. I'll update you once I find some time to set up a HF space.

In the meantime, you might find lightning ai's studio a good option to try it out out of the box: https://lightning.ai/studios?section=featured

[Mandalorian007]

Appreciate the kind words. Yea, thats something I thought about it before. I'll update you once I find some time to set up a HF space.

In the meantime, you might find lightning ai's studio a good option to try it out out of the box: https://lightning.ai/studios?section=featured

Thank you I will definitely try this out!

[ld-xy]

Hello, I followed the instructions to install it on the docker server. After the installation is complete, I run python app.py, but I cannot access it on the local browser? Why?

[Mandalorian007]

Appreciate the kind words. Yea, thats something I thought about it before. I'll update you once I find some time to set up a HF space.

In the meantime, you might find lightning ai's studio a good option to try it out out of the box: https://lightning.ai/studios?section=featured

Hey Luca I hope your holidays are going great.

I took your advice and have been trying to get this project up and running on lightning.ai like suggested, but I have been hitting some issues I am struggling to debug.

After swapping env to python 3.11.11 and following installation instructions on repo I leverage the gradio app to run the app.py file (I change the port in the startup to 8000) and then it loads up and I get the litserve running on the base "/" url.

When I copy the public url and go to "gradio" I keep getting http 307s. Any idea what I should check on?

GRADIO_SERVER_PORT=8000 gradio lang-segment-anything/app.py
⚡ ~ GRADIO_SERVER_PORT=8000 gradio lang-segment-anything/app.py

Warning: Cannot statically find a gradio demo called demo. Reload work may fail.
Watching: '/home/zeus/miniconda3/envs/cloudspace/lib/python3.11/site-packages/gradio', '/teamspace/studios/this_studio/lang-segment-anything'

WARNING:root:No GPU found, using CPU instead
WARNING:root:No GPU found, using CPU instead
Starting LitServe and Gradio server on port 8000...
INFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)
Swagger UI is available at http://0.0.0.0:8000/docs
INFO:     Started server process [20226]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
IMPORTANT: You are using gradio version 3.50.2, however version 4.44.1 is available, please upgrade.
--------
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
WARNING:root:No GPU found, using CPU instead
WARNING:root:No GPU found, using CPU instead
IMPORTANT: You are using gradio version 3.50.2, however version 4.44.1 is available, please upgrade.
--------
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
LangSAM model initialized.
Setup complete for worker 0.
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:46360 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:23616 - "GET /gradio HTTP/1.1" 307 Temporary Redirect
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:23616 - "GET / HTTP/1.1" 200 OK

[luca-medeiros]

You seem to be running GRADIO_SERVER_PORT=8000 gradio lang-segment-anything/app.py, mind trying python lang-segment-anything/app.py?

You may find better reproducibility by using docker and Dockefile on root.

[luca-medeiros]

does /docs work?

[Mandalorian007]

does /docs work?

That was a good catch I guess the gradio plugin enforces the gradio start. I swapped the application to be fully setup by terminal and then using the Port Viewer Plugin, but I am seeing this behavior:

⚡ ~ python lang-segment-anything/app.py
/home/zeus/miniconda3/envs/cloudspace/lib/python3.11/site-packages/sam2/modeli
/sam/transformer.py:23: UserWarning: Flash Attention is disabled as it require
a GPU with Ampere (8.0) CUDA capability.
  OLD_GPU, USE_FLASH_ATTN, MATH_KERNEL_ON = get_sdpa_settings()
Starting LitServe and Gradio server on port 8000...
INFO:     Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)
Swagger UI is available at http://0.0.0.0:8000/docs
INFO:     Started server process [5365]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
/home/zeus/miniconda3/envs/cloudspace/lib/python3.11/site-packages/sam2/modeli
/sam/transformer.py:23: UserWarning: Flash Attention is disabled as it require
a GPU with Ampere (8.0) CUDA capability.
  OLD_GPU, USE_FLASH_ATTN, MATH_KERNEL_ON = get_sdpa_settings()
LangSAM model initialized.
Setup complete for worker 0.
INFO:     130.211.236.75:48621 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:48621 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:48621 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:48621 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:48621 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:48621 - "GET /gradio HTTP/1.1" 307 Temporary Redirect
INFO:     130.211.236.75:47700 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:47700 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:47700 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:47700 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:47700 - "GET / HTTP/1.1" 200 OK

Using the /docs endpoint seems to work correctly giving me a FastAPI Swagger page:

INFO:     130.211.236.75:46487 - "GET /openapi.json HTTP/1.1" 200 OK
INFO:     130.211.236.75:46974 - "GET / HTTP/1.1" 200 OK
INFO:     130.211.236.75:46974 - "GET / HTTP/1.1" 200 OK