added printer generated question

app.py

@@ -1,6 +1,7 @@
 import datetime
 from google.protobuf import message
 import torch
+import json
 import time
 import threading
 import streamlit as st
@@ -23,6 +24,55 @@ if device == "cpu":
     # check if mps is available
     device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
 
+printer_models = [
+    "HP Smart Tank 750",
+    "HP LaserJet Pro",
+    "HP LaserJet 4100",
+    "HP LaserJet 4000",
+    "HP Photosmart C4635",
+    "HP OfficeJet Pro 9015",
+    "HP Envy 6055",
+    "HP DeskJet 3755",
+    "HP Color LaserJet MFP M283fdw",
+    "HP DesignJet T630",
+    "HP PageWide Pro 477dw",
+    "HP LaserJet Enterprise M506",
+    "HP OfficeJet 5255",
+    "HP Envy Photo 7855",
+    "HP LaserJet Pro M404dn",
+    "HP DeskJet Plus 4155",
+    "HP LaserJet Enterprise MFP M528f",
+    "HP Neverstop Laser 1001nw",
+    "HP Tango X",
+    "HP Color LaserJet Pro M255dw",
+    "HP Smart Tank Plus 651",
+    "HP LaserJet Pro MFP M428fdw",
+    "HP OfficeJet Pro 8035",
+    "HP Envy 6075",
+    "HP DeskJet 2622",
+    "HP LaserJet Pro M15w"
+]
+
+def generate_printer_prompt(prompt_instructions):
+    """Encode multiple prompt instructions into a single string."""
+
+    prompt = """
+Come up with a printer related task or question that a person might ask for support.
+no further text/explanation, no additional information.
+Ensure the tasks/questions should follow the same style and complexity
+Examples:
+"""
+    for idx, instruction in enumerate(prompt_instructions):
+        instruction = re.sub(r"\s+", " ", instruction).strip().rstrip(":")
+        # pick one random printer model to replace the placeholder
+        printer_model = random.choice(printer_models)
+        instruction = re.sub(r"<\|hp-printer\|>", printer_model, instruction)
+        prompt += f"Q: {instruction}\n\n"
+    # prompt += f"{len(prompt_instructions) + 1}. Q:"
+    prompt += "Now it's your turn, come up with a printer task/question that a person might ask for support.\n"
+    prompt += "Q: (your task/question)"
+    return prompt
+
 def get_model_tokenizer(sota_model_name):
     model, tokenizer = FastLanguageModel.from_pretrained(
         model_name = "jed-tiotuico/twitter-llama",
@@ -293,6 +343,28 @@ if st.button("your website is straight up garbage. how do you sell high end tech
         )
     )
 
+if st.button("Generate printer task/question"):
+    seed_tasks = [json.loads(l) for l in open("printer-seed.jsonl", "r")]
+    seed_instructions = [t["text"] for t in seed_tasks]
+    prompt_instructions = []
+    prompt_instructions += random.sample(seed_instructions, num_prompt_instructions - len(prompt_instructions))
+    random.shuffle(prompt_instructions)
+    customer_msg = generate_printer_prompt(prompt_instructions)
+    st.session_state["user_msg_as_prompt"] = customer_msg
+    write_user_chat_message(user_chat, customer_msg)
+    model, tokenizer = get_model_tokenizer(sota_model_name)
+    input_text = alpaca_input_text_format.format(customer_msg)
+    st.write(f"```\n{input_text}```")
+    assistant_chat.write_stream(
+        stream_generation(
+            input_text,
+            show_prompt=False,
+            tokenizer=tokenizer,
+            model=model,
+            temperature=0.5,
+        )
+    )
+
 # - Generate Customer Tweet
 if st.button("Generate Customer Message using Few Shots"):
     model, tokenizer = get_mistral_model_tokenizer(sota_model_name)