hmm

app.py

@@ -15,6 +15,7 @@ tokenizer = AutoTokenizer.from_pretrained(
 vardecoder_model = AutoModelForCausalLM.from_pretrained(
     "ejschwartz/resym-vardecoder", torch_dtype=torch.bfloat16, device_map="auto"
 )
+print(vardecoder_model.device_map)
 
 zero = torch.Tensor([0]).cuda()
 print(zero.device)  # <-- 'cpu' 🤔

requirements.txt

@@ -1 +1,2 @@
 transformers
+accelerate