Merge pull request #3 from dev-strender/apply-chatui

채팅과 유사한 UI적용, 약간의 css 추가

Author: hunkim

app.py

@@ -41,18 +41,52 @@ async def summary(request: Request):
         resp = es.gpt_answer(q, text_results=text_results)
     else:
         es_results = es.search(q)
-        
+
         if es_results:
             # Generate summaries of the search results
             resp = es.gpt_answer(q, es_results=es_results)
         else:
             resp = es.gpt_answer(q, text_results="No results found")
-            
+
+    if resp.status_code != 200:
+        raise HTTPException(resp.status_code, resp.text)
+
+    return StreamingResponse(resp.iter_content(1),
+                             media_type="text/event-stream")
+
+## 추가를 해보았는데 막상 제대로 동작하지 않는 것 같아서 우선은 사용하지 않습니다.
+@app.post("/question-suggestion")
+async def question_suggestion(request: Request):
+    payload = await request.json()
+    text_results = payload.get("text_results", "")
+
+    if text_results is not None:
+        # Generate further question by first relevent abstract text
+        resp = es.gpt_question_generator(text_results=text_results)
+
+        if resp.status_code != 200:
+            raise HTTPException(resp.status_code, resp.text)
+        else:
+            return StreamingResponse(resp.iter_content(1), media_type="text/event-stream")
+    else:
+        raise HTTPException(400, "Unexpected Error")
+
+@app.post("/question")
+async def question(request: Request):
+    payload = await request.json()
+    q = payload.get("q", "")
+
+    if q:
+        resp = es.gpt_direct_answer(q)
+    else:
+        raise HTTPException(400, "Unexpected Error")
+
     if resp.status_code != 200:
         raise HTTPException(resp.status_code, resp.text)
 
     return StreamingResponse(resp.iter_content(1),
                              media_type="text/event-stream")
+
 # Define the static files route
 # Need to set html=True to serve index.html
 # Need to put at the end of the routes

es_gpt.py

@@ -35,6 +35,10 @@ def __init__(self, index_name):
         # Load the cl100k_base tokenizer which is designed to work with the ada-002 model
         self.tokenizer = tiktoken.get_encoding("cl100k_base")
 
+        self.answer_generation_prompt = "Based on the context below\"\n\nContext: {}\n\n---\n\nPlease provide concise answer for this questions: {}"
+        self.question_suggestion_prompt = "Based on the context below\"\n\nContext: {}\n\n---\n\nPlease recommend 3 more questions to be curious about {}"
+        self.just_question_prompt = "{}{}"
+
     def index(self, doc_id, doc, text):
         doc["embeddings_dict_list"] = self._create_emb_dict_list(text)
         self.es.index(index=self.index_name,
@@ -146,10 +150,17 @@ def _create_context(self, question, df):
         # Return the context and the length of the context
         return "\n\n###\n\n".join(returns), cur_len
 
-    def _gpt_api_call(self, query, input_token_len, context):
+    def _gpt_api_call(self, query, input_token_len, context, call_type):
+        if call_type == "answer":
+            prompt = self.answer_generation_prompt
+        elif call_type == "question":
+            prompt = self.just_question_prompt
+        else:
+            prompt = self.question_suggestion_prompt
+
         body = {
             "model": self.model_engine,
-            "prompt": f"Based on the context below\"\n\nContext: {context}\n\n---\n\nPlease provide concise answer for this questions: {query}",
+            "prompt": prompt.format(context, query),
             "max_tokens": self.model_max_tokens - input_token_len,
             "n": 1,
             "temperature": 0.5,
@@ -165,6 +176,7 @@ def _gpt_api_call(self, query, input_token_len, context):
                              stream=True)
         return resp
 
+
     def gpt_answer(self, query, es_results=None, text_results=None):
         # Generate summaries for each search result
         if text_results:
@@ -204,7 +216,32 @@ def gpt_answer(self, query, es_results=None, text_results=None):
         else:
             assert False, "Must provide either es_results or text_results"
 
-        return self._gpt_api_call(query, input_token_len, context)
+        return self._gpt_api_call(query, input_token_len, context, call_type="answer")
+
+    def gpt_question_generator(self, text_results=None):
+        if text_results:
+            input_token_len = len(self.tokenizer.encode(text_results))
+            if input_token_len < self.max_tokens:
+                context = text_results
+            else:
+                context = text_results[:self.max_tokens]
+                input_token_len = self.max_tokens
+        else:
+            assert False, "Text results are not found"
+
+        return self._gpt_api_call("", input_token_len, context, call_type="suggestion")
+
+    def gpt_direct_answer(self, q):
+        input_token_len = len(self.tokenizer.encode(q))
+        if input_token_len < self.max_tokens:
+            query = q
+        else:
+            query = q[:self.max_tokens]
+            input_token_len = self.max_tokens
+        return self._gpt_api_call(q, input_token_len, "", call_type="question")
+
+
+
 
 
 # Example usage