ok? hope...

app.py

@@ -18,6 +18,14 @@ def C(text, color="yellow"):
         f"{color_dict.get(color, None)}"
         f"{text}{color_dict[None]}")
 
+def stcache():
+    from packaging import version
+    if version.parse(st.__version__) < version.parse("1.18"):
+        return lambda f: st.cache(suppress_st_warning=True)(f)
+    return lambda f: st.cache_resource()(f)
+    
+st.title("`ckip-joint/bloom-1b1-zh` demo")
+
 # ------------------ ENVIORNMENT ------------------- #
 os.environ["HF_ENDPOINT"] = "https://huggingface.co"
 device = ("cuda" 
@@ -25,9 +33,8 @@ device = ("cuda"
 logging.info(C("[INFO] "f"device = {device}"))
 
 # ------------------ INITITALIZE ------------------- #
-@st.cache(
-        suppress_st_warning=True
-)
+stdec = stcache()
+@stdec
 def model_init():
     
     logging.info(C("[INFO] "f"Model init start!"))
@@ -60,25 +67,72 @@ def model_init():
 
 tokenizer, model = model_init()
 
-try:
-    # ===================== INPUT ====================== #
-    prompt = st.text_input("Prompt: ")
 
-    # =================== INFERENCE ==================== #
-    if prompt:
-        st.title(prompt)
-        with torch.no_grad():
-            [texts_out] = model.generate(
-                **tokenizer(
-                    prompt, return_tensors="pt",
-                    
-                ).to(device),
-                max_new_tokens=100,
+if 1:
+    try:
+        # ===================== INPUT ====================== #
+        prompt = st.text_input("Prompt: ")
+
+        # =================== INFERENCE ==================== #
+        if prompt:
+            # placeholder = st.empty()
+            # st.title(prompt)
+            with st.container():
+                st.markdown(f""
+                    f":violet[{prompt}]⋯⋯"
+                )
+                # st.empty()
+
+                with torch.no_grad():
+                    [texts_out] = model.generate(
+                        **tokenizer(
+                            prompt, return_tensors="pt",
+                            
+                        ).to(device),
+                        min_new_tokens=0,
+                        max_new_tokens=100,
+                    )
+                output_text = tokenizer.decode(texts_out, 
+                    skip_special_tokens=True,
+                )
+                st.empty()
+            if output_text.startswith(prompt):
+                out_gens = output_text[len(prompt):]
+                assert prompt + out_gens == output_text
+            else:
+                out_gens = output_text
+                prompt = ""
+            st.balloons()
+            
+            def multiline(string):
+                lines = string.split('\n')
+                return '\\\n'.join([f"**:red[{l}]**"
+                for l in lines])
+                
+                
+            
+            # st.empty()
+            st.caption("Result: ")
+            st.markdown(f""
+                f":blue[{prompt}]**:red[{multiline(out_gens)}]**"
             )
-        output_text = tokenizer.decode(texts_out)
-        st.balloons()
-        st.markdown(output_text)
+            # st.text(repr(out_gens0))
+
+    except Exception as err:
+        st.write(str(err))
+        st.snow()
+
+
+    # import streamlit as st
 
-except Exception as err:
-    st.write(str(err))
-    st.snow()
+    # st.markdown('Streamlit is **_really_ cool**.')
+    # st.markdown("This text is :red[colored red], and this is **:blue[colored]** and bold.")
+    # st.markdown(":green[$\sqrt{x^2+y^2}=1$] is a Pythagorean identity. :pencil:")
+# def multiline(string):
+#     lines = string.split('\n')
+#     return '\\\n'.join([f"**:red[{l}]**"
+#     for l in lines])
+# st.markdown(multiline("1234 \n5616"))
+# st.markdown("1234\\\n5616")
+# https://docs.streamlit.io/library/api-reference/status/st.spinner
+# https://stackoverflow.com/questions/32402502/how-to-change-the-time-zone-in-python-logging

requirements.txt

@@ -1,4 +1,5 @@
 torch
 transformers
 streamlit==1.17.0
-gradio==3.19.1
+gradio==3.19.1
+packaging