Hidden

CV+Pytorch engineer

Scaletorch has built a product called the Deep Learning Offload Processor, which speeds up AI model training and fine-tuning by a factor of 10x-60x by implementing a number of low level optimizations. It achieves this speed up with zero code changes as well as no changes to model quality. 

Tasks:

We intend to expand the scope of this product to the following:

1. Work on approximation and optimization techniques to help speed up model training - for example using sparsity in point clouds, use of selective backpropagation, etc.
2. Work on a combination of NAS and quantization aware training 
3. Work on efficient methods to do multi-model training
4. Work on the design and implementation of algorithms that can reduce the need for data labelling 
5. Run benchmarks

Skillset Required:

1. PyTorch (in-depth knowledge)
2. Python (in-depth knowledge) 
3. Computer Vision - 2D/video preferred (3D and other areas is a plus)
4. CUDA/C++ - bonus, not a requirement

Location is remote (can be planet Mars as long as the internet is good). 

The candidates can select to work as employees or as independent contractors/freelancer. 

May require working from Moscow for the first month. After that, the candidate can work from a location of their choice. 

Send CVs to the email in contacts.