Investigate ASR being slow and hanging

[rossdanlm]

Use the Gradio cookbook to create a new ASR prompt. Investigate into using faster models perhaps?

Testing

See general instructions and test plan in

• setting up server: [HF][streaming][4/n] Image2Text (no streaming, but lots of fixing) #855
• using ASR: [HF][fix] Small nit fixes to ASR: Remove "model" from completion params, get_output_text in right format #883

[rossdanlm]

Closing because we switched to remote inference instead of local