Update app.py

app.py

@@ -8,8 +8,8 @@ from transformers import StoppingCriteria, TextIteratorStreamer, StoppingCriteri
 
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
-model = AutoModel.from_pretrained("unum-cloud/uform-gen2-qwen-halfB", trust_remote_code=True).to(device)
-processor = AutoProcessor.from_pretrained("unum-cloud/uform-gen2-qwen-halfB", trust_remote_code=True)
+model = AutoModel.from_pretrained("unum-cloud/uform-gen2-qwen-500m", trust_remote_code=True).to(device)
+processor = AutoProcessor.from_pretrained("unum-cloud/uform-gen2-qwen-500m", trust_remote_code=True)
 
 class StopOnTokens(StoppingCriteria):
     def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool: