Inferencing on Fine-Tuned Microsoft Multi-modal SLM Phi3-Vision on Azure GPU Compute (NC-series)

This video illustrates the inferencing process for fine-tuned Microsoft Multi-modal SLM Phi3-Vision based on the standard fine-tuning recipe available in Phi3 Cookbook (public GitHub) with a sample image from Hateful Memes canonical HuggingFaceM4 dataset and using Standard_NC96ads_A100_v4 NC-series Azure GPU Compute (partly leveraging one of 4 A100 (Ampere) GPUs with 80GB VRAM)

Phi3-Vision model on HuggingFace: https://huggingface.co/microsoft/Phi-3-vision-128k-instruct
Hateful Memes HuggingFaceM4 dataset: https://huggingface.co/datasets/HuggingFaceM4/the_cauldron/tree/main/hateful_memes
Phi3 Cookbook on GitHub: https://github.com/microsoft/Phi-3CookBook

Видео Inferencing on Fine-Tuned Microsoft Multi-modal SLM Phi3-Vision on Azure GPU Compute (NC-series) канала Alejandro Buendía