biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Fix Gemma3n audio encoder for Transformers v5 (#33673)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-02-03 13:49:49 +00:00
committed by GitHub
parent a3acfa1071
commit 2a8d84e66d
1 changed files with 9 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
vllm/model_executor/models/gemma3n_mm.py
View File

@@ -621,10 +621,15 @@ class Gemma3nForConditionalGeneration(
# Run on padded features to enable batching # Run on padded features to enable batching
input_features = audio_input["input_features_padded"].squeeze(1) input_features = audio_input["input_features_padded"].squeeze(1)
input_features_mask = audio_input["input_features_mask"].squeeze(1) input_features_mask = audio_input["input_features_mask"].squeeze(1)
audio_outputs, audio_mask = self.audio_tower( audio_outputs = self.audio_tower(input_features, ~input_features_mask)
input_features, ~input_features_mask if isinstance(audio_outputs, tuple):
) # Transformers v4
audio_features = self.embed_audio(inputs_embeds=audio_outputs) audio_encodings, audio_mask = audio_outputs
else:
# Transformers v5
audio_encodings = audio_outputs.last_hidden_state
audio_mask = audio_outputs.audio_mel_mask
audio_features = self.embed_audio(inputs_embeds=audio_encodings)
# The Gemma3nProcessor expects all audio will be 30s in length and # The Gemma3nProcessor expects all audio will be 30s in length and
# inserts 188 audio soft tokens into the text to account for this. # inserts 188 audio soft tokens into the text to account for this.