Update app.py

app.py

@@ -14,13 +14,12 @@ headers = {
 	"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.36"
 }
 
-hf_hub_download(repo_id="LLukas22/gpt4all-lora-quantized-ggjt", filename="ggjt-model.bin", local_dir=".")
-llm = Llama(model_path="./ggjt-model.bin", n_ctx=640)
+hf_hub_download(repo_id="TheBloke/dolphin-2.5-mixtral-8x7b-GGUF", filename="dolphin-2.5-mixtral-8x7b.Q2_K.gguf", local_dir=".")
+llm = Llama(model_path="./dolphin-2.5-mixtral-8x7b.Q2_K.gguf")
 
-ins = '''### Instruction:
-{question}
-{data}
-### Response:
+ins = '''<|im_start|>user
+{question}<|im_end|>
+<|im_start|>assistant
 '''
 
 theme = gr.themes.Monochrome(
@@ -53,11 +52,12 @@ def search_ddg(question: str):
 	return output_string
 
 def generate(instruction):
-	feeding_data = "\n" + search_ddg(instruction)
-	prompt = ins.format(question=instruction, data=feeding_data)
+	# feeding_data = "\n" + search_ddg(instruction)
+	# prompt = ins.format(question=instruction, data=feeding_data)
+    prompt = ins.format(question=instruction)
 	print(prompt)
 
-	response = llm(prompt, stop=['### Instruction:', '### End'])
+	response = llm(prompt, stop=['<|im_start|>user', '<|im_end|>'])
 	result = response['choices'][0]['text']
 	return result