Try device map

app.py

@@ -16,11 +16,15 @@ huggingface_hub.login(token=hf_key)
 
 tokenizer = AutoTokenizer.from_pretrained("bigcode/starcoderbase-3b")
 vardecoder_model = AutoModelForCausalLM.from_pretrained(
-    "ejschwartz/resym-vardecoder", torch_dtype=torch.bfloat16
-).cuda()
+    "ejschwartz/resym-vardecoder", 
+    torch_dtype=torch.bfloat16,
+    device_map="auto"
+)
 fielddecoder_model = AutoModelForCausalLM.from_pretrained(
-    "ejschwartz/resym-fielddecoder", torch_dtype=torch.bfloat16
-).cuda()
+    "ejschwartz/resym-fielddecoder", 
+    torch_dtype=torch.bfloat16,
+    device_map="auto"
+)
 
 make_gradio_client = lambda: Client("https://ejschwartz-resym-field-helper.hf.space/")