I want to make fully custom bras available to patients of the hospital. How is this possible? Can you outline what is done? Is it take take 6 measurements (ABTF), send to hospital, they push button and done?
Is there some information to look up? How many measurements? How many steps are automated?
I appreciate any help.
I was shocked at how many flock cameras there are. I have heard some are disabled. Is there a way of telling?
For dayton:
https://maps.deflock.org/?lat=39.7589&lng=-84.1916&zoom=11.00
For Beavercreek centric:
https://maps.deflock.org/?lat=39.7092&lng=-84.0633&zoom=11.00
I've been using Gemma, both the 26B and 31B versions, MLX optimized for MacBook Pro, and I have to say they perform well in terms of speed. But for transcription work, the results really haven't been good.
I've also tried the smaller models, and they're not useful at all. They can't even reliably remove em dashes. The 20 to 30 billion parameter models just substitute em dashes with other syntax rather than actually restructuring the prose. It's not until I go online to Sonnet that I get output that actually restructures the paragraph in a useful way, and without em dashes at all.
Has anybody had a similar experience, or am I going down the wrong path when it comes to quality transcription?
There are other videos as well…
I have MBP Max M4 48GB. Am using MLX version of Gemma 4 26b 4bit. Once I found a "good" version (gemma-4-26b-a4b-instruct) things worked great. The only issue with other models were excessive memory use. First the memory pressure window would be mostly green, then tan. Switching to this MLX optimized version resulted in very low memory pressure. There would be a climb when remote request were made followed by a sharp drop after complete. They were very quickly processed (superwhisper transcription). After a few inactivity log out / log in cycles I noticed the memory pressure stayed green but climbed dramatically with no activity on the LLM. It never went back down. It's like it started processing and just never finished when I logged in.
What am I doing wrong? Is there a memory leak? Would another MLX based LLM utility be better?
Thanks,
Paul