We have created and fine-tuned models to: Enable farmers to check for Pest/Insects/Agri crop related issues. Ease the workflow of File Scrutiny Officer’s (FSO) in Judicial document management and examinations. Easing the fine tuning of specific models with enabling users to fine tune out specific models only by using an API call.
Key Impact Takeaways:
- The autotune structured models will cut the time to structure up code and aligning the codebase to fine tune out specific model on specific dataset and will result in easy fine tuning by any user without having knowledge of Machine Learning or any programming just by an API call request. Results in time taken to structure up every time redundant codes to nearly 0.
- The computer vision models will reduce the need of File Scrutiny Officer’s (FSO) in courts and will help in a major part of document examining. On an average an FSO spends around 10-15 minutes per case to scrutinize the documents and this can be cut to around 50%.
- Whisper Fine tuning addition along with TTS generation embedded will reduce the labor to record audio clips for fine tuning to 0 as that part will be generated by TTS models implemented.
- On the fly processing of audios in Whisper results in efficient usage of memory and GPU and prevent whole dataset to be processed initially in the memory and process them in batches on the fly.
- ONNX & Quantization to Autotune can result in model size being reduced to around 60%-at best 30% of the original size with on the fly ONNX conversion for easy and consistent workflow.
- BGE fine tuning addition to Autotune results in easy fine tuning of BGE models for retrieval task just by an API call.
- Model Versioning addition to Autotune results in easy versioning and structuring up models and their fine tuned version and additionally gets stored in MinIO buckets for storage and tree based versioning management. Not fully implemented, currently in testing and finalizing.”