Update app.py

app.py

@@ -61,7 +61,10 @@ def fill_up_placeholders(txt):
         "" if len(placeholders) >= 1 else txt
     )
 
-async def chat_stream(idx, local_data, instruction_txtbox, chat_state):
+async def chat_stream(
+    idx, local_data, instruction_txtbox, chat_state,
+    global_context, res_temp, res_topk, res_rpen, res_mnts, res_sample, ctx_num_lconv
+):
     res = [
       chat_state["ppmanager_type"].from_json(json.dumps(ppm))
       for ppm in local_data
@@ -71,8 +74,18 @@ async def chat_stream(idx, local_data, instruction_txtbox, chat_state):
     ppm.add_pingpong(
         PingPong(instruction_txtbox, "")
     )
-    prompt = build_prompts(ppm, "global context", 3)
-    async for result in gen_text(prompt, hf_model=MODEL_ID, hf_token=TOKEN):
+    prompt = build_prompts(ppm, global_context, ctx_num_lconv)
+    async for result in gen_text(
+        prompt, hf_model=MODEL_ID, hf_token=TOKEN,
+        parameters={
+            'max_new_tokens': res_mnts,
+            'do_sample': res_sample,
+            'return_full_text': False,
+            'temperature': res_temp,
+            'top_k': res_topk,
+            'repetition_penalty': res_rpen           
+        }
+    ):
         ppm.append_pong(result)
         yield ppm.build_uis(), str(res)
 
@@ -198,7 +211,7 @@ with gr.Blocks(css=MODEL_SELECTION_CSS, theme='gradio/soft') as demo:
             with gr.Column():
                 with gr.Column():
                     gr.Markdown("#### Global context")
-                    with gr.Accordion("global context will persist during conversation, and it is placed at the top of the prompt", open=False):
+                    with gr.Accordion("global context will persist during conversation, and it is placed at the top of the prompt", open=True):
                         global_context = gr.Textbox(
                             "global context",
                             lines=5,
@@ -218,31 +231,12 @@ with gr.Blocks(css=MODEL_SELECTION_CSS, theme='gradio/soft') as demo:
                     
                     gr.Markdown("#### GenConfig for **response** text generation")
                     with gr.Row():
-                        res_temp = gr.Slider(0.0, 2.0, 0, step=0.1, label="temp", interactive=True)
-                        res_topp = gr.Slider(0.0, 2.0, 0, step=0.1, label="top_p", interactive=True)
-                        res_topk = gr.Slider(20, 1000, 0, step=1, label="top_k", interactive=True)
-                        res_rpen = gr.Slider(0.0, 2.0, 0, step=0.1, label="rep_penalty", interactive=True)
-                        res_mnts = gr.Slider(64, 8192, 0, step=1, label="new_tokens", interactive=True)                            
-                        res_beams = gr.Slider(1, 4, 0, step=1, label="beams")
-                        res_cache = gr.Radio([True, False], value=0, label="cache", interactive=True)
+                        res_temp = gr.Slider(0.0, 2.0, 1.0, step=0.1, label="temp", interactive=True)
+                        res_topk = gr.Slider(20, 1000, 50, step=1, label="top_k", interactive=True)
+                        res_rpen = gr.Slider(0.0, 2.0, 1.2, step=0.1, label="rep_penalty", interactive=True)
+                        res_mnts = gr.Slider(64, 8192, 512, step=1, label="new_tokens", interactive=True)
                         res_sample = gr.Radio([True, False], value=0, label="sample", interactive=True)
-                        res_eosid = gr.Number(value=0, visible=False, precision=0)
-                        res_padid = gr.Number(value=0, visible=False, precision=0)
-
-                with gr.Column(visible=False):
-                    gr.Markdown("#### GenConfig for **summary** text generation")
-                    with gr.Row():
-                        sum_temp = gr.Slider(0.0, 2.0, 0, step=0.1, label="temp", interactive=True)
-                        sum_topp = gr.Slider(0.0, 2.0, 0, step=0.1, label="top_p", interactive=True)
-                        sum_topk = gr.Slider(20, 1000, 0, step=1, label="top_k", interactive=True)
-                        sum_rpen = gr.Slider(0.0, 2.0, 0, step=0.1, label="rep_penalty", interactive=True)
-                        sum_mnts = gr.Slider(64, 8192, 0, step=1, label="new_tokens", interactive=True)
-                        sum_beams = gr.Slider(1, 8, 0, step=1, label="beams", interactive=True)
-                        sum_cache = gr.Radio([True, False], value=0, label="cache", interactive=True)
-                        sum_sample = gr.Radio([True, False], value=0, label="sample", interactive=True)
-                        sum_eosid = gr.Number(value=0, visible=False, precision=0)
-                        sum_padid = gr.Number(value=0, visible=False, precision=0)
-
+                
                 with gr.Column():
                     gr.Markdown("#### Context managements")
                     with gr.Row():
@@ -255,7 +249,8 @@ with gr.Blocks(css=MODEL_SELECTION_CSS, theme='gradio/soft') as demo:
 
     instruction_txtbox.submit(
         chat_stream,
-        [idx, local_data, instruction_txtbox, chat_state],
+        [idx, local_data, instruction_txtbox, chat_state,
+         global_context, res_temp, res_topk, res_rpen, res_mnts, res_sample, ctx_num_lconv],
         [chatbot, local_data]
     )