Bonjour
[𝗟𝗮𝗯𝗩𝗜𝗘𝗪 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜] The next generation of the LabVIEW Deep Learning Module will enable seamless execution of open-source LLMs locally, including fine-tuning and quantizing them to integrate into any industrial architecture.
As an initial demonstration, in this video, we show a simple execution of a 𝗟𝗹𝗮𝗺𝗮𝟯 𝟴𝗕 𝗺𝗼𝗱𝗲𝗹 𝘂𝗻𝗱𝗲𝗿 𝗟𝗮𝗯𝗩𝗜𝗘𝗪 using the HAIBAL 2.0 architecture.
We also take this occasion to announce a new tool named LM Studio, part of the SOTA suite, which aims to offer a comprehensive framework like Langchain for language models but within LabVIEW using HAIBAL! Naturally, we will provide architectures for building RAGs, vector databases, and mixtures of experts for chatbots. A simple and complete tool is on the horizon!
Our philosophy is to make Generative AI easy and allow you to choose between open-source LLMs or OpenAI with LabVIEW (both will be available).
Among the new features of the next generation of HAIBAL:
𝗙𝘂𝗹𝗹 𝗰𝗼𝗺𝗽𝗮𝘁𝗶𝗯𝗶𝗹𝗶𝘁𝘆 𝘄𝗶𝘁𝗵 𝗮𝗹𝗹 𝗲𝘅𝗶𝘀𝘁𝗶𝗻𝗴 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸𝘀: Keras, TensorFlow, PyTorch, ONNX.
𝗦𝘁𝘂𝗻𝗻𝗶𝗻𝗴 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲: HAIBAL 2 will be 50 times faster than the first HAIBAL generation and 20% faster than PyTorch.
𝟯𝟮 𝗮𝗻𝗱 𝟲𝟰-𝗯𝗶𝘁 𝘀𝘂𝗽𝗽𝗼𝗿𝘁: Deployment on 32-bit systems with limited functionality.
Hardware support: Support for CUDA, TensorRT for NVIDIA, Rocm for AMD, OneAPI for Intel.
𝗠𝗮𝘅𝗶𝗺𝘂𝗺 𝗺𝗼𝗱𝘂𝗹𝗮𝗿𝗶𝘁𝘆: Define your own layers and loss functions.
𝗚𝗿𝗮𝗽𝗵 𝗡𝗲𝘂𝗿𝗮𝗹 𝗡𝗲𝘁𝘄𝗼𝗿𝗸𝘀: Complete and advanced integration.
Annotation tools: An annotator as efficient as Roboflow, integrated into our software suite.
𝗠𝗼𝗱𝗲𝗹 𝘃𝗶𝘀𝘂𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻: Utilizing Netron for graphical model summaries.
In summary, we are targeting a release for July. Alright, that's all for now. Back to work to make this happen!
Follow us on YouTube and LinkedIn to get the latest updates. This summer is going to be exciting! 🤠
Youtube channel https://www.youtube.com/@graiphic
Linkedin channel https://www.linkedin.com/company/graiphic
