Release Notes
15 April 2024, Version - 0.1.0.
First Launch Release!
New features
- Kompress - A model compression utility that can create data-aware compressed variants of your deep learning models. Users can simply import their data, choose a target model and compress via - Network Pruning, Knowledge Distillation or Quantization! Currently, the following tasks are supported -
- Image Classification
- Object Detection
- LLMs
- Adapt - An efficient model fine-tuning utility that can help users fine-tune any deep learning model on your domain-specific data. Adapt currently has LoRA, DoRA and SSF with quantization support along with full-finetuning. Currently, the following tasks are supported -
- Image Classification
- Object Detection
- Instance Segmentation
- Pose Estimation
- Text Classification
- Text Translation
- Text Summarization
- Question Answering
- LLM
Detailed information on the supported models and tasks can be found here
Key Pointers
- Nyun Zero never stores custom model weights or data files and hence users can use their own compute to host the services of Nyun Zero. Specifically, AWS EC2, GCP VM and Azure VM are fully supported; additionally, users can also use their own local compute hardware with a simple SSH-based connection. The exact connection details can be found here
- Multiple Data formats are supported for each task, users need to organize the data as per the compatible formats. More details on the supported data formats can be found here
- Output models and logs can be directly downloaded on your local machine or even accessed in the connected infrastructure.