I am seeking help building a model using the PyTorch and the Hugging Face transformers library. I am a Senior Research Data Scientist at Columbia University, and I have a background in machine learning. Unfortunately, due to a time-crunch, I need help building a model that requires customizations I don't know how to implement. I am seeking an expert in deep learning who can pair-code with me to get the first iteration of this model working. Pay will depend on your level of experience and depth of knowledge of the specific tools I plan to use.
Here is a brief description of the issues I am running into with my model.
1. I need to train three models with shared components. I need to save checkpoints for each component separately for transfer learning.
2. I need to be able to integrate two or more pretrained models from the Hugging Face `transformers` library into one larger model. I am unsure of how to implement this kind of design using the `transformers` abstractions, and I don't know how to go under the hood and customize things.
3. I need to incorporate a LLM into the CTC inference function for one of the models, and I am unsure of how to do this with a custom model.
4. I will start with small versions of the pretrained models that will be incorporated into the custom models. However, I need to ensure things are implemented such that they can scale up and be deployed on a cluster after the initial pipeline is developed. I would prefer to avoid design decisions that will make scaling harder in the future.
This job is already closed and no longer accepting applicants, sorry.