[Bugfix] Fix prefix strings for quantized VLMs (#9772)

2024-10-29 19:02:59 -04:00
parent 8d7724104a
commit bc73e9821c
20 changed files with 288 additions and 97 deletions
--- a/vllm/model_executor/models/ultravox.py
+++ b/vllm/model_executor/models/ultravox.py
@@ -357,7 +357,10 @@ class UltravoxModel(nn.Module, SupportsMultiModal, SupportsPP):
                ))
        self.multi_modal_projector = UltravoxProjector(config)
        self.language_model = init_vllm_registered_model(
-            config.text_config, cache_config, quant_config)
+            config.text_config,
+            cache_config,
+            quant_config,
+            prefix="language_model")
        if config.text_model_id is not None:
            self.secondary_weights.append(
                DefaultModelLoader.Source(model_or_path=config.text_model_id,