Update app.py

app.py

@@ -11,8 +11,13 @@ openai.api_key = os.environ.get("OPENAI_API_KEY")
 BASE_SYSTEM_MESSAGE = """"""
 
 def make_prediction(prompt, max_tokens=None, temperature=None, top_p=None, top_k=None, repetition_penalty=None):
-    completion = openai.Completion.create(model="/workspace/text-generation-webui/models/wizardcoder-python-34b-v1.0.Q5_K_M.gguf", prompt=prompt, max_tokens=max_tokens, temperature=temperature, top_p=top_p, top_k=top_k, repetition_penalty=repetition_penalty, stream=True, stop=["</s>", "<|im_end|>"])
-    for chunk in completion:
+    openai.api_base = os.environ.get("OPENAI_API_BASE")
+    completion1 = openai.Completion.create(model="wizardcoder-python-34b-v1.0.Q5_K_M.gguf", prompt=prompt, max_tokens=max_tokens, temperature=temperature, top_p=top_p, top_k=top_k, repetition_penalty=repetition_penalty, stream=True, stop=["</s>", "<|im_end|>"])
+    openai.api_base = os.environ.get("OPENAI_API_BASE2")
+    completion2 = openai.Completion.create(model="wizardcoder-python-34b-v1.0.Q5_K_M.gguf", prompt=prompt, max_tokens=max_tokens, temperature=temperature, top_p=top_p, top_k=top_k, repetition_penalty=repetition_penalty, stream=True, stop=["</s>", "<|im_end|>"])
+    for chunk in completion1:
+        yield chunk["choices"][0]["text"]
+    for chunk in completion2:
         yield chunk["choices"][0]["text"]