More experimenting

app.py

@@ -22,11 +22,15 @@ tokenizer = AutoTokenizer.from_pretrained("bigcode/starcoderbase-3b")
 vardecoder_model = AutoModelForCausalLM.from_pretrained(
     "ejschwartz/resym-vardecoder", 
     torch_dtype=torch.bfloat16,
-).to("cuda")
+    low_cpu_mem_usage=True,
+    device_map="cuda",
+)
 fielddecoder_model = AutoModelForCausalLM.from_pretrained(
     "ejschwartz/resym-fielddecoder", 
     torch_dtype=torch.bfloat16,
-).to("cuda")
+    low_cpu_mem_usage=True,
+    device_map="cuda",
+)
 
 make_gradio_client = lambda: Client("https://ejschwartz-resym-field-helper.hf.space/")