Skip to content

mtmd: fix gemma 4 audio rms norm eps #23815

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

Merged
ngxson merged 2 commits into from
May 28, 2026
Merged

mtmd: fix gemma 4 audio rms norm eps #23815

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2528c65
mtmd: fix gemma 4 audio rms norm eps
ngxson May 28, 2026
c807c2c
Update tools/mtmd/clip.cpp
ngxson May 28, 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
11 changes: 6 additions & 5 deletions conversion/gemma.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -786,14 +786,15 @@ def set_gguf_parameters(self):
super().set_gguf_parameters()

# vision params
assert self.hparams_vision is not None
self.gguf_writer.add_clip_vision_projector_type(gguf.VisionProjectorType.GEMMA4V)
self.gguf_writer.add_vision_attention_layernorm_eps(self.hparams.get("layer_norm_eps", 1e-6))
self.gguf_writer.add_vision_attention_layernorm_eps(self.hparams_vision.get("layer_norm_eps", 1e-6))

# audio params
if self.hparams_audio:
self.gguf_writer.add_clip_audio_projector_type(gguf.VisionProjectorType.GEMMA4A)
self.gguf_writer.add_audio_num_mel_bins(self.hparams_audio["feat_in"])
self.gguf_writer.add_audio_attention_layernorm_eps(1e-5)
assert self.hparams_audio is not None
self.gguf_writer.add_clip_audio_projector_type(gguf.VisionProjectorType.GEMMA4A)
self.gguf_writer.add_audio_num_mel_bins(self.hparams_audio["feat_in"])
self.gguf_writer.add_audio_attention_layernorm_eps(self.hparams_audio.get("layer_norm_eps", 1e-6))

def is_audio_tensor(self, name: str) -> bool:
return "audio_tower" in name or "embed_audio" in name
Expand Down
3 changes: 3 additions & 0 deletions tools/mtmd/clip.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1552,6 +1552,9 @@ struct clip_model_loader {
hparams.audio_n_fft = 512;
hparams.audio_window_len = 320; // 20ms frame (NOT 25ms/400)
hparams.audio_hop_len = 160;
// due to a mistake in the original converstion code, rms_norm_eps is set to a wrong value
Comment thread
ngxson marked this conversation as resolved.
Outdated
// since all gemma4a models use 1e-6, we just hardcode it here to avoid re-conversion
hparams.eps = 1e-6f;
} break;
case PROJECTOR_TYPE_GRANITE_SPEECH:
{
Expand Down
Loading