Controller usually misunderstand uri /worker_generate_stream belongs to tool workers

[tranvanhoa533]

Describe the issue

Issue: When start model worker and tool worker (example grounding_dino), usually controller call api /worker_generate_stream to tool worker and return 404 in tool worker

Log:

2023-11-14 14:08:57 | INFO | /data/miniconda3/envs/llava/lib/python3.10/os.py | Register to controller
2023-11-14 14:09:42 | INFO | /data/miniconda3/envs/llava/lib/python3.10/os.py | Send heart beat. Models: ['grounding_dino']. Semaphore: None. global_counter: 0. worker_id: 029142.
2023-11-14 14:09:43 | INFO | stdout | INFO:     ::1:21820 - "POST /worker_get_status HTTP/1.1" 200 OK
2023-11-14 14:09:47 | INFO | stdout | INFO:     ::1:21824 - "POST /worker_generate_stream HTTP/1.1" 404 Not Found
2023-11-14 14:10:02 | INFO | stdout | INFO:     ::1:3926 - "POST /worker_get_status HTTP/1.1" 200 OK
2023-11-14 14:10:06 | INFO | stdout | INFO:     ::1:3940 - "POST /worker_generate_stream HTTP/1.1" 404 Not Found

[fearnworks]

Also get this issue very consistently. Slightly different form, but it has a lot of problems with the api calls :

requests.exceptions.MissingSchema: Invalid URL '/worker_generate': No scheme supplied. Perhaps you meant https:///worker_generate?

[liufangtao]

"POST /worker_generate_stream HTTP/1.1" 404t Found ---在web端运行除llava_plus其他sever/***_work.py时出现这种404

[pedramaghazadeh]

If you're using the Gradio web app, you need to select llava (or other base llms you're using) from the drop down menu on the top left corner. Otherwise, it might randomly select a tool and send your prompt and images as input to that tool, which is wrong.