Deep learning researcher with custom research to speed up models 20-90 percent

  • Thread starter Thread starter mrrobot
  • Start date Start date
Joined
7/1/25
Messages
6
Points
1
I have custom research in ml , deep learning I have my own custom research that can speed up matrix multiplications using approximation. Error are less compared to other methods like pruning. It can also be used in other quant model where few percent accuracy is not issue. Any firm that wants to use this research please contact me . I have not published my research as to maintain secrecy . Message me if interested. I can do one or two model optimization for sample .
 
📢

Fast matrix computation is not rocket science any more.
What's the meat?
 
📢

Fast matrix computation is not rocket science any more.
What's the meat?
I can get most fastest method at a given accuracy level. It is way faster than what is used in current llms and deep learning system . I have reduced latency of llm 1 b model by 45 percent. With almost no loss in accuracy. It is mathematical method not code optimization
 
Last edited:
I thought you could even do it with quantum computers
It can do if matrix satisfies certain conditions (unitary and hermitian) and it is not fast when it comes to matrix size of few thousand by few thousands . it is faster if matrix size is more than billion by billion
 
Back
Top Bottom