Update app.py

app.py

@@ -16,18 +16,23 @@ def format_prompt(message, history):
     prompt += f"[INST] {message} [/INST]"
     return prompt
 
-def compress_history(purpose, task, history):
-    resp = run_gpt(
-        COMPRESS_HISTORY,
-        stop_tokens=["observation:", "task:", "action:", "thought:"],
-        max_tokens=512,
-        seed=random.randint(1,1000000000),
-        purpose=purpose,
-        task=task,
-        history=history,
+def compress_history(history,temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0):
+    formatted_prompt=f"{COMPRESS_HISTORY.format(history=history)}"
+    generate_kwargs = dict(
+        temperature=temperature,
+        max_new_tokens=max_new_tokens,
+        top_p=top_p,
+        repetition_penalty=repetition_penalty,
+        do_sample=True,
+        seed=random.randint(1,99999999999)
+        #seed=42,
     )
-    history = resp
-    return history
+    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
+    output = ""
+    for response in stream:
+        output += response.token.text
+    return output
+
 MAX_HISTORY=100
 
 def generate(
@@ -55,7 +60,7 @@ def generate(
             cnt+=len(l.split("\n"))
     print(f'cnt:: {cnt}')
     if cnt > MAX_HISTORY:
-        history = compress_history(history, prompt, history)
+        history = compress_history(history, temperature, max_new_tokens, top_p, repetition_penalty)
     formatted_prompt = format_prompt(f"{GAME_MASTER.format(history=history)}, {prompt}", history)
     stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
     output = ""