fix chat history, update settings to use GPU

chat.py

@@ -38,7 +38,7 @@ def chat(history, system_message):
     history[-1][1] = ""
     for output in llm(messages, max_tokens=512, stop=["</s>", "<unk>", "### User:"], echo=False, stream=True):
         answer = output['choices'][0]['text']
-        history[-1][1] = answer
+        history[-1][1] += answer
 
         yield history, history
 

config.yml

@@ -5,3 +5,4 @@ file: wizard-vicuna-13B.ggml.q5_1.bin
 base_model: junelee/wizard-vicuna-13b
 llama_cpp:
   n_ctx: 1024
+  n_gpu_layers: 40  # llama 13b has 40 layers