Try to fix prompts

code/app.py

@@ -12,10 +12,11 @@ mdlpath = hf_hub_download(repo_id="TheBloke/Starling-LM-7B-alpha-GGUF", filename
 
 #lcpp_model = Llama(model_path=model_path, n_ctx=8192)
 llm = AutoModelForCausalLM.from_pretrained(model_path_or_repo_id=model_path, local_files_only=True, model_type="mistral")
-global otxt, txtinput, txtoutput, running, result
+global otxt, txtinput, txtoutput, running, result, echoed
 otxt = ""
 running = False
 result = None
+echoed = False
 
 def stowtext(curr, inp):
     if len(curr) <= 1:
@@ -63,14 +64,15 @@ def jsn2prompt(jsn):
     return txt
 
 def talk(txt, jsn):
-    global running, result
+    global running, result, echoed
     if not jsn:
         jsn = txt
     if not running:
         #result = lcpp_model.create_chat_completion(messages=txt,stream=True,stop=["GPT4 Correct User: ", "<|end_of_turn|>", "</s>"], max_tokens=64, )
         #result = lcpp_model(prompt=jsn2prompt(txt), stream=True, stop=["GPT4 Correct User: ", "<|end_of_turn|>", "</s>"], max_tokens=64, echo=False)
-        result = llm(prompt=jsn2prompt(txt), stream=True, stop=["GPT4 Correct User: ", "<|end_of_turn|>", "</s>"], echo=False)
+        result = llm(prompt=jsn2prompt(txt), stream=True, stop=["GPT4 Correct User: ", "<|end_of_turn|>", "</s>"])
         running = True
+        echoed = False
     for r in result:
         print("GOT RESULT:", r)
         txt2 = None
@@ -80,7 +82,14 @@ def talk(txt, jsn):
             running = False
             yield txt
         if txt2 is not None:
+            txt3 = txt
             txt = stowchunk(txt, txt2)
+            if not echoed and txt[-1]["content"].contains(jsn2prompt([txt3[-1]])):
+                echoed = True
+                yield txt
+            elif not echoed and txt[-1]["role"] != "assistant":
+                txt[-1]["content"] = "*Thinking*"
+                yield txt
             yield txt
     yield txt