Overview
Solid empirical gains across multiple QAT methods and datasets, with clear latency and training-cost measurements; some results hinge on choosing k/r and using a fused kernel.
Citations0
Evidence Strength0.80
Confidence0.90
Risk Signals10
Trust Signals
Findings with numeric evidence: 4/4
Findings with evidence refs: 4/4
Results with explicit delta: 6/6
Reproducibility
Status: Code + data available
Open source: Partial
At A Glance
Cost impact: 70%
Production readiness: 70%
Novelty: 60%
Why It Matters For Business
LoRAP lets teams deploy low-bit GNNs with much of the original accuracy retained while keeping memory and speed gains; it is a small, trainable add-on compatible with existing QAT pipelines.
Who Should Care
Summary TLDR
Quantizing Graph Neural Networks (GNNs) to low bit-widths hurts accuracy because feature quantization errors accumulate during neighbor aggregation. LoRAP inserts small, input-dependent low-rank prompts after aggregation (post-aggregation) to directly correct those errors. Across 4 quantization-aware training frameworks and many datasets, LoRAP + node prompts (GPF-LoRAP) consistently recovers accuracy for INT4/low-bit GNNs, sometimes exceeding FP32, while adding little compute and memory when using a fused GPU kernel.
Problem Statement
Low-bit quantization of GNNs saves memory and speed but causes large accuracy drops because quantized node features create biased aggregated messages. Pre-aggregation node prompts cannot reliably fix topology-amplified errors. We need a lightweight, input-aware way to correct aggregation-level quantization error during training.
Main Contribution
Introduce LoRAP: post-aggregation, input-dependent prompts built from a small set of low-rank basis vectors.
Show theoretically that post-aggregation prompts decouple correction from graph operator and allow node-specific bias correction.
Key Findings
GPF-LoRAP can recover severe INT4 accuracy losses on small benchmarks.
LoRAP sometimes surpasses full-precision accuracy on evaluated tasks.
Results
| Metric | Value | Baseline | Delta | Split / Dataset | Evidence | Evidence Ref |
|---|---|---|---|---|---|---|
| Accuracy | 78.5% | FP32 77.6% | +0.9% | Cora | Table 1 shows A2Q + GPF-LoRAP = 78.5% vs FP32 77.6% | Table 1 |
| Accuracy | 69.6% (GPF-LoRAP) | None 52.4% | +17.2% | Reddit-Binary | Table 2 QAT-W4A4: None 52.4% → GPF-LoRAP 69.6% | Table 2 |
What To Try In 7 Days
Run a baseline INT4 QAT pipeline on one GNN task and record accuracy/latency.
Add GPF-plus (node prompts) and LoRAP (aggregation prompts) with k≈40, r≈2 and retrain.
Measure accuracy recovery and per-layer latency; enable fused Triton kernel if available for production speedups.
Optimization Features
Infra Optimization
Model Optimization
System Optimization
Training Optimization
Inference Optimization
Reproducibility
Risks & Boundaries
Limitations
LoRAP uses high-precision prompt generation which adds small FP32 work and storage.
Approximation limited by prompt rank k; too small k may underfit quantization error.
When Not To Use
When full-precision FP32 models run fine and memory/speed are not constrained.
When you lack access to retraining or QAT pipeline to jointly optimize prompts.
Failure Modes
Poor k/r choices lead to under- or over-correction and accuracy drop.
EdgePrompt+ style additions can hurt performance if combined incorrectly.