"intelligent batching" problem at the server for Accelerated DNN Execution
Budget: $30 – $250 USD
Focus on the "intelligent batching" problem at the server for Accelerated DNN Execution
I want to do of “split” like paradigm where you determine which models are good candidates to be split.
split other models such as Resnet50, Resnet101, VGG16, VGG19, and DenseNet121... models and others
Also, think of compressing the tensors before uploading them so that communication overhead can be reduced a bit.
Primary Goal:
- This project's primary goal is to optimise speed, time, and memory usage for Accelerated DNN Execution using split with batching (batch inputs) using python
And I want to write a scientific paper on this study.
Preferred Programming Language:
- Python
Libraries/Frameworks:
- The client prefers PyTorch to be used in this project for implementing the "split with batching" approach, batch inputs.
There are two steps: first: sending different clients with the same splitting point to one server.
The second step: is sending many clients with different numbers of splitting points. We have to do the flops to see the results on a chart.
I will show you the figures to understand the work. Also, I have already the codes but need to modify them for this work as needed, After that need to start reviewing some related work and writing a paper.
Ideal Skills and Experience:
- Strong knowledge and experience optimizing speed and memory usage in DNN execution.
- Proficiency in Python programming language.
- Familiarity with PyTorch library.
- Understanding of deep learning frameworks and algorithms.
- Ability to assess and compare the performance of different approaches.
- Strong problem-solving and analytical skills.
I want to do of “split” like paradigm where you determine which models are good candidates to be split.
split other models such as Resnet50, Resnet101, VGG16, VGG19, and DenseNet121... models and others
Also, think of compressing the tensors before uploading them so that communication overhead can be reduced a bit.
Primary Goal:
- This project's primary goal is to optimise speed, time, and memory usage for Accelerated DNN Execution using split with batching (batch inputs) using python
And I want to write a scientific paper on this study.
Preferred Programming Language:
- Python
Libraries/Frameworks:
- The client prefers PyTorch to be used in this project for implementing the "split with batching" approach, batch inputs.
There are two steps: first: sending different clients with the same splitting point to one server.
The second step: is sending many clients with different numbers of splitting points. We have to do the flops to see the results on a chart.
I will show you the figures to understand the work. Also, I have already the codes but need to modify them for this work as needed, After that need to start reviewing some related work and writing a paper.
Ideal Skills and Experience:
- Strong knowledge and experience optimizing speed and memory usage in DNN execution.
- Proficiency in Python programming language.
- Familiarity with PyTorch library.
- Understanding of deep learning frameworks and algorithms.
- Ability to assess and compare the performance of different approaches.
- Strong problem-solving and analytical skills.