wip

Signed-off-by: Raphael Glon <[email protected]>

app.py

@@ -4,6 +4,7 @@ import spaces
 
 import logging
 import os
+import re
 import threading
 from typing import List, Tuple, Dict
 
@@ -22,6 +23,8 @@ MAX_NEW_TOKENS = 256
 TEMPERATURE = 0.7
 TOP_P = 0.95
 
+ANALYSIS_PATTERN = analysis_match = re.compile(r'^(.*)assistantfinal', flags=re.DOTALL)
+
 # --- Silent Hub auth via env/Space Secret (no UI) ---
 HF_TOKEN = os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACEHUB_API_TOKEN") or os.getenv("HUGGINGFACE_TOKEN")
 if HF_TOKEN:
@@ -111,14 +114,25 @@ def generate_stream(message: str, history: List[Tuple[str, str]]):
     thread = threading.Thread(target=_model.generate, kwargs=gen_kwargs)
     thread.start()
 
+    analysis = ""
     output = ""
     for new_text in streamer:
         output += new_text
-        LOG.info("NEW TEXT: %s, OUTPUT: %s", new_text, output)
-        yield output
+        if not analysis:
+            m = ANALYSIS_PATTERN.match(output)
+            if m:
+                analysis = re.sub(r'^analysis\s*', '', m.group(1))
+                output = ""
+
+        LOG.info("NEW TEXT: %s, OUTPUT: %s", new_text, output.encode())
+        if not analysis:
+            answer = f"Analysis:\n{output}"
+        else:
+            answer = f"Analysis:\n{analysis}\nAnswer:\n{output}"
+        yield answer
 
 
-with gr.Blocks(title="MobileLLM-Pro — Chat") as demo:
+with gr.Blocks(title="OpenAI GPT-OSS 20B Chat") as demo:
     gr.Markdown(
         """
 # Chat