NVFP4 QAT (Quantization-Aware Training) in verl

Last updated: 04/02/2026

verl supports NVFP4 Quantization-Aware Training (QAT), which applies fake quantization during training so the model learns to tolerate NVFP4 quantization error. At rollout time, weights are packed into real NVFP4 format for vLLM inference. This closes the precision gap between training and inference, preventing KL divergence explosion.

Training Backend	Training Precision	Rollout Precision	vLLM Quant Method
FSDP	BF16 + fake quantization	NVFP4 W4A16	`compressed-tensors`
Megatron	BF16 + fake quantization	NVFP4 W4A16	`modelopt`

[!TIP] For ready-to-run scripts, environment setup, and experimental results, see the QAT recipe.

Key Configuration

FSDP Backend

Configured under actor_rollout_ref.actor.fsdp_config.qat:

actor_rollout_ref:
  actor:
    fsdp_config:
      qat:
        enable: true
        mode: "w4a16"
        group_size: 16
        ignore_patterns:
          - "lm_head"
          - "embed_tokens"
          - "re:.*mlp.gate$"
        quantization_config_path: "recipe/qat/config/nvfp4_w4a16.json"

Parameter	Description	Default
`fsdp_config.qat.enable`	Enable QAT	`False`
`fsdp_config.qat.mode`	Quantization mode	`"w4a16"`
`fsdp_config.qat.group_size`	Quantization group size	`16`
`fsdp_config.qat.ignore_patterns`	Layers to skip. Supports `re:` prefix for regex, otherwise substring match	`["lm_head", "embed_tokens", "re:.*mlp.gate$"]`
`fsdp_config.qat.quantization_config_path`	vLLM quantization config JSON path	Required

Megatron Backend

Configured under actor_rollout_ref.actor.megatron.qat:

actor_rollout_ref:
  actor:
    megatron:
      qat:
        enable: true
        mode: "w4a16"
        group_size: 16
        ignore_patterns:
          - "lm_head"
          - "*mlp.gate"
        quantization_config_path: "recipe/qat/config/nvfp4_w4a16_megatron.json"

Parameter	Description	Default
`megatron.qat.enable`	Enable QAT	`False`
`megatron.qat.mode`	Quantization mode	`"w4a16"`
`megatron.qat.group_size`	Quantization group size	`16`
`megatron.qat.ignore_patterns`	Layers to skip. Uses `fnmatch` glob syntax	`["lm_head", "*mlp.gate"]`
`megatron.qat.quantization_config_path`	vLLM quantization config JSON path	Required

Support Matrix

NVFP4 W4A16 (weight-only FP4 quantization)
Dense models and MoE models
FSDP and Megatron training backends
Full quantization and FFN-only quantization strategies
Verified on Qwen3-8B-Base and Qwen3-30B-A3B-Base

Notes

FSDP backend has scalability limitations for very large models. For large-scale training, use the Megatron backend.
FSDP uses re: prefix regex for ignore_patterns, while Megatron uses fnmatch glob syntax. The two are not interchangeable.