fix

utils/chatglm.py

@@ -81,11 +81,22 @@ class ChatGLM(BasePredictor):
             trust_remote_code=True,
             resume_download=True
         )
-        model = AutoModel.from_pretrained(
-            model_name,
-            trust_remote_code=True,
-            resume_download=True
-        ).half().to(self.device)
+        if self.device == 'cuda':
+            model = AutoModel.from_pretrained(
+                model_name,
+                trust_remote_code=True,
+                resume_download=True
+            ).half().to(self.device)
+        else:
+            model = AutoModel.from_pretrained(
+                "THUDM/chatglm-6b-int8",
+                trust_remote_code=True,
+                resume_download=True,
+                low_cpu_mem_usage=True,
+                torch_dtype=torch.float16
+                if self.device == 'cuda' else torch.float32,
+                device_map={'': self.device}
+            )
 
         model = model.eval()
         self.model = model