Create app.py

app.py

@@ -0,0 +1,29 @@
+import os
+import gradio as gr
+import spaces
+from transformers import AutoTokenizer, AutoModel
+
+
+model_name = "teknium/OpenHermes-2.5-Mistral-7B"
+token = os.environ['hf_token']
+
+model = AutoModel.from_pretrained(model_name).cuda()
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+
+system_prompt = '''You are given an input text for a chat interface. Propose auto-completion to the text. You have several roles:
+- Fight under-specification: if the user does not provide sufficient context, propose them a set of relevant suggestions.
+- Complete text: The text provided to you is in the making. If you have a good idea for how to complete - make suggestions.
+
+Make sure the suggestions are valid completions of the text! No need for them to complete the text completely. 
+Suggest only up to 5 works ahead. 
+'''
+
+@spaces.GPU
+def generate(text):
+    data = [
+        {'role': 'system', 'content': system_prompt},
+        {'role': 'user', 'content': text}
+    ]
+    tokenized = tokenizer.apply_chat_template(data, return_tensors='pt')
+    return tokenizer.deocode(model.generate(**tokenized).squeeze(0))