Need only Coder for research paper results
Budget: ₹1,500 – ₹12,500 INR
I am searching for my AI/ML research paper and need an experienced PyTorch developer to take care of the model training and evaluation phase. I only need the developing part, not the writing, so please apply only if you can do this. We need Sci/Scopus label results. The core architecture and data pipeline are drafted; what remains is to turn those ideas into reproducible code, run rigorous experiments, and package the results so they drop cleanly into the paper.
Scope
• Build or adapt the proposed network in PyTorch, keeping the code modular and well-commented.
• Handle train/val/test splits, loaders, and any data augmentation that improves generalization.
• Design a training loop with appropriate loss functions, schedulers, mixed-precision where helpful, and solid logging (TensorBoard or equivalent).
• Produce quantitative metrics (accuracy, F1, AUROC, t-SNE as relevant) and clear plots that can go straight into LaTeX.
• Save checkpoints and provide a short inference script so reviewers can replicate results.
Acceptance criteria
– Code runs end-to-end on my GPU instance with a single command.
– Reported metrics match those submitted in the final jpg
– All dependencies are captured in environment.yml or requirements.txt.
The timeline is within a week, maximum two weeks.
Scope
• Build or adapt the proposed network in PyTorch, keeping the code modular and well-commented.
• Handle train/val/test splits, loaders, and any data augmentation that improves generalization.
• Design a training loop with appropriate loss functions, schedulers, mixed-precision where helpful, and solid logging (TensorBoard or equivalent).
• Produce quantitative metrics (accuracy, F1, AUROC, t-SNE as relevant) and clear plots that can go straight into LaTeX.
• Save checkpoints and provide a short inference script so reviewers can replicate results.
Acceptance criteria
– Code runs end-to-end on my GPU instance with a single command.
– Reported metrics match those submitted in the final jpg
– All dependencies are captured in environment.yml or requirements.txt.
The timeline is within a week, maximum two weeks.