Loom: A Scalable Analytical Neural Computer Architecture

arXiv:2604.08816v1 Announce Type: new Abstract: We present Loom, a computer architecture that executes programs compiled from C inside a looped transformer whose weights are derived analytically. The architecture implements a 22-opcode instruction set in 8 transformer layers. Each forward pass executes one instruction; the model is applied iteratively until the program counter reaches zero. The full machine state resides in a single tensor X in R^{d imes n} of fixed size, and every step has fixed cost for fixed d and n, independent of program length or execution history. The default configuration uses d = 155 and n = 1024, yielding 4.7 million parameters and 928 instruction slots. A compact configuration at d = 146 and n = 512 suffices for a 9imes9 Sudoku solver (284 instructions). The weights are program-independent: programs live in the state tensor, and the same fixed-weight model executes any compiled program. We make Loom source code publicly available at https://github.com/mkturkcan/Loom.

Related

• Integrated electro-optic attention nonlinearities for transformers
• AFL: A Single-Round Analytic Approach for Federated Learning with Pre-trained Models
• Transformer See, Transformer Do: Copying as an Intermediate Step in Learning Analogical Reasoning
• Hierarchical Kernel Transformer: Multi-Scale Attention with an Information-Theoretic Approximation Analysis

Source: arXiv cs.LG | 2026-04-13