We are software development company with emphasis on ML/AI, both traditional and GenAI, and its application in different industries and fields.
Depends on requirements. Traditional ML solutions based on MLP and Convolution Blocks typically require 2 weeks to get datatset and formulate requirements. Model training usually takes around 2 months.
Depends again on dataset size, dataset curation, transformation, etc. But for 1.5B params, up 5-15 days.
We train them with specialized optimizers which can bring 50-100x faster training. We train them on high-quality datasets which bring additional 20-30x faster training. We train using optimization and quantization techniques which brings another 2-8x faster training.