docs

clarity, link to model/paper

app.py

@@ -14,9 +14,8 @@ from transformers.generation.logits_process import LogitsProcessor, LogitsProces
 # ======================
 MODEL_ID = os.getenv("MODEL_ID", "microsoft/UserLM-8b")
 DEFAULT_SYSTEM_PROMPT = (
-    "You are a user who wants to implement a special type of sequence. "
-    "The sequence sums up the two previous numbers in the sequence and adds 1 to the result. "
-    "The first two numbers in the sequence are 1 and 1."
+    "You are a user who wants to compute rolling 7-day averages over uneven time stamps. "
+    "You are suspicious of resampling magic and will accuse the assistant of witchcraft if it's not explicit."
 )
 
 # ======================
@@ -270,6 +269,7 @@ The AI plays the **user**, you play the **assistant**. Your messages appear on t
     msg = gr.Textbox(
         label="Your Reply (assistant)",
         placeholder="Type your assistant response here…",
+        info="Leave blank & press _Generate_ to create the **first user message**.",
         lines=2,
     )
 
@@ -288,8 +288,8 @@ The AI plays the **user**, you play the **assistant**. Your messages appear on t
     with gr.Accordion("Implementation Details", open=False):
         gr.Markdown(
             """
-- Decoding defaults from the model card: `temperature=1.0`, `top_p=0.8`, stop on `<|eot_id|>`, and block `<|endconversation|>`.
-- Guardrails from Appendix C.1: (1) first-token logit filter, (2) block endconversation, (3) 3–25 word length, (4) verbatim repetition filter.
+- Decoding defaults from [the model card](https://hf.co/microsoft/UserLM-8b): `temperature=1.0`, `top_p=0.8`, stop on `<|eot_id|>`, and block `<|endconversation|>`.
+- Guardrails from Appendix C.1 [of the paper](https://arxiv.org/abs/2510.06552): (1) first-token logit filter, (2) block endconversation, (3) 3–25 word length, (4) verbatim repetition filter.
             """
         )