Fix lora inference

fishaudio · Dec 29, 2023 · 982f787 · 982f787
1 parent cdecc2a
commit 982f787
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/tools/llama/generate.py b/tools/llama/generate.py
@@ -323,7 +323,7 @@ def load_model(config_name, checkpoint_path, device, precision):
         cfg = compose(config_name=config_name)
 
     with torch.device("meta"):
-        model: Transformer = instantiate(cfg.model.model)
+        model: Transformer = instantiate(cfg.model).model
 
     if "int8" in str(checkpoint_path):
         logger.info("Using int8 weight-only quantization!")