[Ultravox] Fix Gemma instantiation, support quantization via --hf-overrides (#24131)

Signed-off-by: Peter Salas <peter@fixie.ai>
This commit is contained in:
Peter Salas
2025-09-10 22:25:34 -07:00
committed by GitHub
parent 6c8deacd72
commit f17a6aa4ec
3 changed files with 53 additions and 40 deletions
+1 -1
View File
@@ -276,7 +276,7 @@ class UltravoxProjector(nn.Module):
else:
self.act = get_act_fn(config.projector_act)
dim_out = config.text_hidden_size
dim_out = config.text_config.hidden_size
self.linear_2 = nn.Linear(dim_mid, dim_out, bias=False)
# Ultravox v0.4.1 and below use layer_norm after the second linear layer