Update app.py

app.py

@@ -86,20 +86,40 @@ def build_chat():
     prompt = conv.get_prompt() 
     return prompt
 
+
+import re
+
+def replace_repeated_spaces_and_newlines(text):
+    # Replace repeated spaces with a single space
+    text = re.sub(r'\s+', ' ', text)
+
+    # Replace repeated newlines with a single newline
+    text = re.sub(r'\n+', '\n', text)
+
+    return text
+
 from fastchat.model import get_conversation_template
 
 @spaces.GPU
 def generate(
     message: str,
-    chat_history: list[tuple[str, str]],
+    chat_history,
     system_prompt: str,
+    input_pdf: BytesIO = None,
     max_new_tokens: int = 1024,
-    temperature: float = 0.6,
-    top_p: float = 0.9,
+    temperature: float = 0.7,
+    top_p: float = 1.0,
     top_k: int = 50,
-    repetition_penalty: float = 1.2,
+    repetition_penalty: float = 1.0,
 ) -> Iterator[str]:
+    if input_pdf is not None:
+        pdf_text = process_pdf(input_pdf)
+        # print(pdf_text)
+        pdf_text = replace_repeated_spaces_and_newlines(pdf_text)
+        message += f"\nThis is the beginning of a pdf\n{pdf_text}This is the end of a pdf\n"
     conv = get_conversation_template("vicuna")
+    if system_prompt is not None:
+        conv.set_system_message(system_prompt)
     conv.append_message(conv.roles[0], message)
     conv.append_message(conv.roles[1], None)
     prompt = conv.get_prompt()
@@ -108,7 +128,7 @@ def generate(
     # for user, assistant in chat_history:
     #     conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
     # conversation.append({"role": "user", "content": message})
-
+    print(prompt[500:1000])
     # chat = tokenizer.apply_chat_template(conversation, tokenize=False)
     inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=False).to("cuda")
     if len(inputs) > MAX_INPUT_TOKEN_LENGTH:
@@ -136,34 +156,34 @@ def generate(
         yield "".join(outputs)
 
 
-def generate_with_pdf(
-    message: str,
-    chat_history: list[tuple[str, str]],
-    system_prompt: str,
-    input_pdf: BytesIO = None,
-    max_new_tokens: int = 1024,
-    temperature: float = 0.6,
-    top_p: float = 0.9,
-    top_k: int = 50,
-    repetition_penalty: float = 1.2,
-) -> Iterator[str]:
-    if input_pdf is not None:
-        pdf_text = process_pdf(input_pdf)
-        # print(pdf_text)
-        message += f"\nThis is the beginning of a pdf\n{pdf_text}This is the end of a pdf\n"
-    yield from generate(
-        message,
-        chat_history,
-        system_prompt,
-        max_new_tokens,
-        temperature,
-        top_p,
-        top_k,
-        repetition_penalty
-    )
+# def generate_with_pdf(
+#     message: str,
+#     chat_history,
+#     system_prompt: str,
+#     input_pdf: BytesIO = None,
+#     max_new_tokens: int = 1024,
+#     temperature: float = 0.6,
+#     top_p: float = 0.9,
+#     top_k: int = 50,
+#     repetition_penalty: float = 1.2,
+# ) -> Iterator[str]:
+#     if input_pdf is not None:
+#         pdf_text = process_pdf(input_pdf)
+#         # print(pdf_text)
+#         message += f"\nThis is the beginning of a pdf\n{pdf_text}This is the end of a pdf\n"
+#     yield from generate(
+#         message,
+#         chat_history,
+#         system_prompt,
+#         max_new_tokens,
+#         temperature,
+#         top_p,
+#         top_k,
+#         repetition_penalty
+#     )
 
 chat_interface = gr.ChatInterface(
-    fn=generate_with_pdf,
+    fn=generate,
     additional_inputs=[
         gr.Textbox(label="System prompt", lines=6),
         gr.File(label="PDF File", accept=".pdf"),
@@ -179,14 +199,14 @@ chat_interface = gr.ChatInterface(
             minimum=0.1,
             maximum=4.0,
             step=0.1,
-            value=0.6,
+            value=0.7,
         ),
         gr.Slider(
             label="Top-p (nucleus sampling)",
             minimum=0.05,
             maximum=1.0,
             step=0.05,
-            value=0.9,
+            value=1.0,
         ),
         gr.Slider(
             label="Top-k",
@@ -200,7 +220,7 @@ chat_interface = gr.ChatInterface(
             minimum=1.0,
             maximum=2.0,
             step=0.05,
-            value=1.2,
+            value=1.0,
         ),
     ],
     stop_btn=None,