Skip to content

model : wire up Nemotron-H tensors for NVFP4 support#20561

Merged
CISC merged 3 commits intomasterfrom
cisc/nemotron-h-nvfp4-scales
Mar 16, 2026
Merged

model : wire up Nemotron-H tensors for NVFP4 support#20561
CISC merged 3 commits intomasterfrom
cisc/nemotron-h-nvfp4-scales

Conversation

@CISC
Copy link
Member

@CISC CISC commented Mar 14, 2026

prep #20539

@CISC CISC requested a review from danbev March 14, 2026 22:12
@github-actions github-actions bot added the model Model specific label Mar 14, 2026
@CISC
Copy link
Member Author

CISC commented Mar 14, 2026

As an apropos, when #20427 (comment) is addressed, an optional bias should also be added:

ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
cb(Qcur, "Qcur", il);
if (model.layers[il].bq) {
Qcur = ggml_add(ctx0, Qcur, model.layers[il].bq);
cb(Qcur, "Qcur", il);
}

cc/ @richarddd

@CISC CISC merged commit de8f01c into master Mar 16, 2026
129 of 136 checks passed
@CISC CISC deleted the cisc/nemotron-h-nvfp4-scales branch March 16, 2026 08:19
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

model Model specific

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants