Of course, I compressed it to Q2 or Q4, but the flash model is running.
For fun, I tried running it on DGX SPARK,
I confirmed that it runs, and since I came back to the original task, I can't evaluate it,
but it seems like a fun attempt.
https://github.com/antirez/ds4
Recently released models are all at a level where the parameter values are changing units, but if these fun attempts become active,
I hope that improvements and development will be made so that other large models can also run.
