Definition
Tensor networks are mathematical factorizations (e.g., Matrix Product Operators) used to represent high-dimensional tensors with controlled bond dimension. In AI, they underpin quantum-inspired model-compression methods such as compactifai.
Key Points
- multiverse-computing CompactifAI applies tensor networks to LLM weight matrices
- Vendor claim: up to 80–95% size reduction with immaterial accuracy loss — not independently audited in press coverage
- Academic CompactifAI paper reported large memory cuts on LLaMA-2 7B when combined with quantization
- Enables edge-ai / on-prem deployment where hyperscaler round-trips are costly or disallowed