Update app.py

app.py

@@ -23,6 +23,7 @@ client = InferenceClient(
     "mistralai/Mixtral-8x7B-Instruct-v0.1"
 )
 history = []
+max_history=1000
 
 def format_prompt(message, history):
   prompt = "<s>"
@@ -32,12 +33,29 @@ def format_prompt(message, history):
   prompt += f"[INST] {message} [/INST]"
   return prompt
 
+def condense(in_prompt):
+    seed = random.randint(1,1111111111111111)
+    print (seed)
+    generate_kwargs = dict(
+        temperature=1.0,
+        max_new_tokens=512,
+        top_p=0.99,
+        repetition_penalty=1.0,
+        do_sample=True,
+        seed=seed,
+    )
+    content = agent.CONDENSE_PROMPT + prompt
+    print(f'CONDENSED:: {content}')
 
+    stream = client.text_generation(content, **generate_kwargs, stream=True, details=True, return_full_text=False)
+    resp = ""
+    for response in stream:
+        resp += response.token.text
+    return resp
 
-def run_gpt(
-    in_prompt,
-    history,
-):
+def run_gpt(in_prompt,history,):
+    if len(in_prompt)>max_prompt:
+        in_prompt = condense(in_prompt)
     print(f'history :: {history}')
     prompt=format_prompt(in_prompt,history)
     seed = random.randint(1,1111111111111111)
@@ -50,15 +68,8 @@ def run_gpt(
         do_sample=True,
         seed=seed,
     )
-
-    
     content = agent.GENERATE_PROMPT + prompt
-
     print(content)
-    
-    #formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history)
-    #formatted_prompt = format_prompt(f'{content}', history)
-
     stream = client.text_generation(content, **generate_kwargs, stream=True, details=True, return_full_text=False)
     resp = ""
     for response in stream: