Let intelligence learn its shape.
Advancing next-generation AI through brain-inspired sparse intelligence, powered by original CHT dynamic sparsity science.
Structural learning raises intelligence density, unlocking efficiency potential across models and computing systems.
Neurons that activate together can form stronger connections.
PUBLIC MODEL
From original brain-inspired science to a model the global community can explore. SparseMind’s first public 8B sparse language model makes structural learning tangible and testable.
This milestone adapts Qwen3-8B with structured sparsity and publishes model weights and a release manifest, inviting researchers, developers and hardware partners to take the next step.
Structured sparsity in targeted weight matrices
Public weights · adapted from Qwen3-8B
The release manifest uses dense BF16 execution with fixed 2:4 zeros. System efficiency depends on the model, hardware and workload. Graphic illustrates the format.
Intelligence lives in the strength of connections.
And in the way they are organized.
CHT lets networks explore their structure.
Sparse models, built for a world of diverse hardware.
From open models to sparse algorithm and silicon co-design, build a verifiable path for real workloads.
Bring structural learning to deployable language models.
Evolve sparse silicon and model structure together.
Public artifacts, engineering validation and research targets, clearly distinguished.
Quality adaptation and hardware-path evaluation. Perplexity changes include adaptation, sparsification and training.
Sparse models meet compiler execution. Results from different model sizes, precisions and workloads remain separate.
Advancing GaN physics simulation with sparse intelligence. Accuracy and sparsity figures remain acceptance targets.
CARLO VITTORIO CANNISTRACICHT emerged from Professor Carlo Vittorio Cannistraci and his research team’s original work on brain-network organization and structural learning. SparseMind connects that insight with model, hardware and business expertise.
Academic research and company engineering progress are distinguished. Authorship and institutional attribution follow the original papers.
Meet the team and scientific rootsFrom ICLR to NeurIPS and PMLR.
Advancing structure learning and hardware efficiency.
Cannistraci-Hebb Channel-wise Dynamic Sparse Training of Convolutional Neural Networks with Contextual Modulation, poster.
SCHTs: A Semi-Structured Dynamic Sparse Training Framework for Hardware-Efficient Deep Learning, poster.
Yingtao Zhang et al. · How network organization supports learning.
Yingtao Zhang et al. · Soft connection sampling and gradual density decay for dynamic sparse training.
Jialin Zhao et al. · Adaptive topology rules for link prediction in complex networks.
Jiaqing Lyu et al. · CHTs24 and sparse retraining for hardware-friendly 2:4 formats.
Explore weights, release manifests and research.
Explore the modelDefine your model, hardware and real workload.
Discuss your challengeConnect original science, public milestones and commercial direction.
Discover SparseMind