# Multi-Task Learning

Harmon is an innovative unified multimodal understanding and generation framework that coordinates visual representations for understanding and generation through a shared MAR encoder, demonstrating excellent performance in text-to-image generation and multimodal understanding tasks.

A financial domain model developed based on Llama3-8B, enhanced with an instruction pretraining framework to improve domain adaptability, achieving or even surpassing the performance of Llama3-70B on financial tasks.

A text processing model fine-tuned based on the T5-small architecture, focusing on text generation and transformation for specific tasks

This is a model based on the Keras framework, with unspecified functionality, possibly used for image classification or spam detection tasks

A multi-task model for joint voice activity detection, speech signal-to-noise ratio, and C50 room acoustic parameter estimation

Based on the mengzi-bert-base model, further trained with 20G of financial news and research report data, focusing on natural language processing tasks in the Chinese financial domain.