AC-Transformer-DS: AI Accelerator Chip for Multimodal Large Models

Technology Edges and Chip Highlights

Based on algorithm-hardware co-design strategy and leveraging our self-developed AI chip design automation toolset, create high-efficiency AI accelerator chips for large multimodal models
Support rapid reconfiguration of parallel dimensions, memory hierarchy, and other hyperparameters according to the architecture search results from the toolchain software, enabling optimization for diverse workloads
Support more than 40 core instructions and provides a scalable instruction framework that allows customized development for specialized operations or emerging algorithms

Large Language Models
DeepSeek-1.5B, DeepSeek-7B, Llama3-8B, Gemma3-4B, Qwen3-1.7B, Qwen3-8B, …

Generative Models
Grounding DINO, Stable Diffusion, ...

Vision-Language Models
Llava-7b, Qwen-VL-7B, DeepSeek-VL-7B, ...

Vision Transformers
EffViT, SegFormer, MobileViT, SwinT, HrFormer, …

Convolution Neural Networks
ResNet, MobileNet, Yolo, Openpose, Inception, MTCNN, UNet, ...