Test 6

code/app.py

@@ -17,15 +17,17 @@ model_path = "./starling-lm-7b-alpha.Q6_K.gguf"
 mdlpath = hf_hub_download(repo_id="TheBloke/Starling-LM-7B-alpha-GGUF", filename=model_path, local_dir="./")
 
 lcpp_model = Llama(model_path=model_path)
-global otxt, txtinput, txtoutput
+global otxt, txtinput, txtoutput, running, result
 otxt = ""
+running = False
+result = None
 
 def stowtext(curr, inp):
     curr.append({
         "role": "user",
         "content": inp,
     })
-    return [curr, curr]
+    return curr
 
 def stowchunk(curr, inp):
     first = curr[-1]["role"] == "user"
@@ -49,21 +51,27 @@ def printfmt(jsn):
             txt += "# " + msg["content"] + "\n\n"
     return txt
 
-def talk(txt):
-    result = lcpp_model.create_chat_completion(messages=txt, stop=["</s>", "<|end_of_text|>", "GPT4 User: ", "<|im_sep|>", "\n\n"], stream=True)
+def talk(txt, jsn):
+    global running, result
+    if not jsn:
+        jsn = txt
+    if not running:
+        result = lcpp_model.create_chat_completion(messages=txt,stream=True)
+        running = True
     for r in result:
         txt2 = None
         if "content" in r["choices"][0]["delta"]:
             txt2 = r["choices"][0]["delta"]["content"]
-            if txt2.startswith("\n"):
-                txt2 = txt2[1:]
+        elif not "content" in r["choices"][0]["delta"] and not "role" in r["choices"][0]["delta"]:
+            running = False
+            yield txt
         if txt2 is not None:
             txt = stowchunk(txt, txt2)
-            yield [printfmt(txt), txt]
-    yield [printfmt(txt), txt]
+            yield txt
+    yield txt
 
 def main():
-    global otxt, txtinput
+    global otxt, txtinput, running
     logging.basicConfig(level=logging.INFO)
 
     with gr.Blocks() as demo:
@@ -76,12 +84,17 @@ def main():
         with gr.Row(variant="panel"):
             talk_btn = gr.Button("Send")
         with gr.Row(variant="panel"):
-            jsn = gr.JSON(visible=False, value="[]")
-            jsn2 = gr.JSON(visible=False, value="[]")
+            jsn = gr.JSON(visible=True, value="[]")
+            jsn2 = gr.JSON(visible=True, value="[]")
 
-        talk_btn.click(stowtext, inputs=[jsn2, txtinput], outputs=[jsn, jsn2], api_name="talk")
+        talk_btn.click(stowtext, inputs=[jsn2, txtinput], outputs=jsn, api_name="talk")
+        talk_btn.click(lambda x: gr.update(visible=False), inputs=talk_btn, outputs=talk_btn)
         talk_btn.click(lambda x: gr.update(value=""), inputs=txtinput, outputs=txtinput)
-        jsn.change(talk, inputs=jsn, outputs=[talk_output, jsn2], api_name="talk")
+        talk_btn.click(lambda x: gr.update(value="[]"), inputs=jsn2, outputs=jsn2)
+        jsn.change(talk, inputs=[jsn, jsn2], outputs=jsn2, api_name="talk")
+        jsn2.change(lambda x: gr.update(value=printfmt(x)), inputs=jsn2, outputs=talk_output)
+        jsn2.change(lambda x: gr.update(visible=not running), inputs=jsn2, outputs=talk_btn)
+        #jsn2.change(lambda x: gr.update(value=x), inputs=jsn2, outputs=jsn)
 
     demo.queue().launch(server_name="0.0.0.0", server_port=7860, share=True)