upd

app.py

@@ -71,7 +71,7 @@ def load_lora(lora_path, base_model="decapoda-research/llama-7b-hf"):
         # load_in_8bit=True,
         # device_map=device_map,
         low_cpu_mem_usage=True,
-        torch_type=torch.float16,
+        # torch_type=torch.float16,
     )
     print("Loading LoRA...")
     lora = PeftModel.from_pretrained(