MF-QAT: Multi-Format Quantization-Aware Training for Elastic Inference
This provides a practical solution for elastic precision scaling in AI deployments, allowing runtime format selection based on hardware constraints, though it is incremental as it builds on existing QAT methods.
The paper tackles the problem of deploying quantized models across diverse hardware by introducing multi-format quantization-aware training (QAT), which trains a single model to perform robustly across multiple quantization formats, matching single-format QAT accuracy at each precision and enabling on-the-fly conversion to unseen formats with negligible accuracy degradation.
Quantization-aware training (QAT) is typically performed for a single target numeric format, while practical deployments often need to choose numerical precision at inference time based on hardware support or runtime constraints. We study multi-format QAT, where a single model is trained to be robust across multiple quantization formats. We find that multi-format QAT can match single-format QAT at each target precision, yielding one model that performs well overall across different formats, even formats that were not seen during training. To enable practical deployment, we propose the Slice-and-Scale conversion procedure for both MXINT and MXFP that converts a high-precision representation into lower-precision formats without re-training. Building on this, we introduce a pipeline that (i) trains a model with multi-format QAT, (ii) stores a single anchor format checkpoint (MXINT8/MXFP8), and (iii) allows on-the-fly conversion to lower MXINT or MXFP formats at runtime with negligible-or no-additional accuracy degradation. Together, these components provide a practical path to elastic precision scaling and allow selecting the runtime format at inference time across diverse deployment targets.