Does Model Optimizer support QAT for GLM-5 with FP8, INT4, and NVFP4 quantization?

[hibayesian]

How would you like to use ModelOpt

I would like to perform Quantization-Aware Training (QAT) on the GLM-5 model (744B parameters, MoE architecture) using Model-Optimizer, targeting the following low‑precision formats:

• FP8
• INT4 (full QAT, both weights and activations)
• NVFP4

Specifically, I want to verify official support and obtain any available examples or configuration templates for GLM-5 QAT with these precisions.

System information

CUDA version: 13
GPU: Blackwell

[LinearPro]

I also want to know!

[ChenhanYu]

QAT/D support for none-Nemotron models will be coming to Megatron-Bridge (with ModelOpt fakequant and logits distillation). The first support targets NVFP4 (where FP8 should also work) on Qwen3.6 and GPT-OSS. Details can be found in NVIDIA-NeMo/Megatron-Bridge#4512

Megatron-Bridge has GLM-5 support. It maybe our next recipe support target if there is no OSS contribution by then.

[kevalmorabia97]

In terms of software support, PTQ and QAD for LLMs in Megatron-Bridge framework is already supported but we have not published any reproducible recipes yet. You can refer to the documentation here if you want to give it a try: https://github.com/NVIDIA/Model-Optimizer/tree/main/examples/megatron_bridge/