Add application file

app.py

@@ -7,7 +7,7 @@ from threading import Thread
 #model = AutoModelForCausalLM.from_pretrained("togethercomputer/RedPajama-INCITE-Chat-3B-v1", torch_dtype=torch.float16)
 
 tokenizer = AutoTokenizer.from_pretrained("aaditya/OpenBioLLM-Llama3-8B-GGUF")
-model = AutoModelForCausalLM.from_pretrained("aaditya/openbiollm-llama3-8b.Q5_K_M", torch_dtype=torch.float16)
+model = AutoModelForCausalLM.from_pretrained("aaditya/openbiollm-llama3-8b.Q5_K_M.gguf", torch_dtype=torch.float16)
 
 class StopOnTokens(StoppingCriteria):
     def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool: