Skip to content

[megatron, vllm] feat: NVFP4 (W4A16) QAT training support via ModelOpt #5254

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
jQizhang wants to merge 4 commits into
base: main
Choose a base branch
from
Draft

[megatron, vllm] feat: NVFP4 (W4A16) QAT training support via ModelOpt #5254

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d4e479e
support nvfp4 qat training with modelopt and megatron
jQizhang Feb 12, 2026
deb8e8f
clean some logic
jQizhang Feb 12, 2026
3dbb2d2
update MoE model logic
jQizhang Feb 12, 2026
74df2dc
add megatron patch for qat training
jQizhang Feb 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
541 changes: 541 additions & 0 deletions verl/models/mcore/qat_patch.py
View file
Open in desktop

Large diffs are not rendered by default.

6 changes: 6 additions & 0 deletions verl/trainer/config/engine/megatron.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,6 +72,12 @@ override_transformer_config:
# Attention backend to use (flash,fused,unfused,local,auto). Defaults to auto in mcore, flash in verl
attention_backend: flash

# # Quantization method. None for no quantization, "nvfp4" for NVFP4 quantization
quantization: null

# Whether to enable Quantization-Aware Training (QAT). Default False.
enable_qat: False
Comment on lines +76 to +79
Copy link
Collaborator

@ISEEKYAN ISEEKYAN Feb 10, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

better to use the same configs as in FSDP to avoid confusing the users


override_mcore_model_config: {}

# oc.select: default val for ref.megatron.use_mbridge
Expand Down
1,056 changes: 1,056 additions & 0 deletions verl/utils/modelopt_qat_utils.py
View file
Open in desktop

Large diffs are not rendered by default.

Loading