Update app.py

app.py

@@ -14,9 +14,8 @@ theme = gr.themes.Monochrome(
 
 TOKEN = os.getenv("USER_TOKEN")
 tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-7b-instruct")
-instruct_pipeline_falcon = pipeline(model="tiiuae/falcon-7b-instruct", tokenizer = tokenizer, torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto")
+instruct_pipeline_falcon = pipeline(model="tiiuae/falcon-7b-instruct", tokenizer = tokenizer, torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto", device=0)
 instruct_pipeline_llama = pipeline(model="HuggingFaceH4/llama-7b-ift-ds-save-test4", torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto", use_auth_token=TOKEN)
-#instruct_pipeline_12b = pipeline(model="databricks/dolly-v2-12b", torch_dtype=torch.bfloat16, trust_remote_code=True, device_map="auto")
 
 def generate(query, temperature, top_p, top_k, max_new_tokens): 
     return [instruct_pipeline_falcon(query, temperature=temperature, top_p=top_p, top_k=top_k, max_new_tokens=max_new_tokens),