Skip to content

LLM inference in C/C++

C++ ggml

convert : fix Gemma4 with no audio encoder (#24242)

Sigbjørn Skjæretf71af352a52b
diff --git a/conversion/gemma.py b/conversion/gemma.py
index 379876629f..1258428b04 100644
--- a/conversion/gemma.py
+++ b/conversion/gemma.py
@@ -812,10 +812,11 @@ class Gemma4VisionAudioModel(MmprojModel):
self.gguf_writer.add_vision_attention_layernorm_eps(self.hparams_vision.get("layer_norm_eps", 1e-6))
# audio params
- 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))
+ if self.has_audio_encoder:
+ 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