Requant Network research: verifiable LoRA fine-tuning step - exact integer forward+backward with Freivalds/Fiat-Shamir certificates and an independent CPU verifier (SM120)
-
Updated
Nov 19, 2026 - Python
Requant Network research: verifiable LoRA fine-tuning step - exact integer forward+backward with Freivalds/Fiat-Shamir certificates and an independent CPU verifier (SM120)
Requant Network GPU research: exact RNS/NTT arithmetic, real CKKS encryption and slim bootstrapping on RTX 5060 Ti 16 GB.
Requant Network research: low-precision GEMM accuracy, quantization costs and operand-reuse trade-offs on RTX 5060 Ti (SM120).
Requant Network research: complete sparse GF(2) GPU solves, Block Wiedemann, independent certificates and representation/setup trade-offs.
Requant Network research: exact GPU Wagner/Equihash solvers, independently verified collision trees and measured time-memory trade-offs.
Requant Network research: batch-invariant deterministic inference kernels on SM120 — measuring batch-induced nondeterminism and its cost
Requant Network research: certifiable GF(2) matrix-multiplication search on consumer GPUs, with exact rank-47 reproduction and replicated policy benchmarks.
Requant Network research: ternary (BitNet b1.58-style) weight GEMM on SM120 Blackwell tensor cores — exact NVFP4/FP8/INT8 paths, throughput/accuracy trade-off
Post-quantum NTT (ML-KEM q=3329 / ML-DSA q=8380417) on SM120: SIMT vs tensor-core INT8 limb GEMM — exact, measured on RTX 5060 Ti
Requant Network research: exact modular arithmetic on SM120 Tensor Cores, FP4 encodings, integer baselines and full-pipeline measurements.
Requant Network research: MSM bucket accumulation over BN254 via exact INT8-limb tensor-core GEMM (Blackwell SM120)
Requant Network research: SM120 inference-prover support, complete synthetic-model proofs, independent CPU verification and measured limits.
Requant Network GPU research: exact GF(2^128), CLMAD, sumcheck transcript graphs, CPU oracles and reproducible RTX 5060 Ti measurements.
Requant Network GPU research: Pearl FP8 useful-work conformance, exact BF16 norms and offline GPU candidate to native plain-proof verification.
Requant Network research: exact GPU Reed-Solomon coding with CUDA CLMAD, ISA-L multicore baselines and replicated resident/PCIe transfer measurements.
To associate your repository with the requant-network topic, visit your repo's landing page and select "manage topics."