Update main.py

main.py

@@ -34,7 +34,7 @@ REQUEST_TIMEOUT = int(os.getenv("REQUEST_TIMEOUT", "120"))
 MAX_RETRIES = int(os.getenv("MAX_RETRIES", "3"))
 RETRY_BASE_DELAY = float(os.getenv("RETRY_BASE_DELAY", "1.5"))
 
-MAX_INPUT_TOKENS = int(os.getenv("MAX_INPUT_TOKENS", "80000"))  # limit w tokenach
+MAX_INPUT_TOKENS = 32000  # ustawione na stałe zgodnie z prośbą
 
 # przybliżone przeliczenie: 1 token ~ 4 znaki
 AVG_CHARS_PER_TOKEN = 4
@@ -52,7 +52,7 @@ async def get_client() -> Client:
     global _client
     if _client is None:
         log.info("Connecting to %s", HF_SPACE_URL)
-        _client = await asyncio.to_thread(Client, HF_SPACE_URL)
+        _client = await asyncio.to_thread(Client, HF_SPACE_URL, show_error=True)
         log.info("Connected.")
     return _client
 
@@ -140,12 +140,9 @@ def _condense_messages(messages: list[Message], max_tokens: int) -> str:
         tcount = _token_count(text)
 
         if tokens_so_far + tcount > max_tokens:
-            # nadmiar: spróbuj skrócić starsze wiadomości
             remaining_tokens = max_tokens - tokens_so_far
             if remaining_tokens <= 0:
-                # nie mieszczą się już starsze, pomijamy starsze części
                 continue
-            # skracamy proporcjonalnie
             approx_chars = remaining_tokens * AVG_CHARS_PER_TOKEN
             text = text[-approx_chars:]  # zachowujemy ostatnią część
             tcount = _token_count(text)
@@ -224,9 +221,20 @@ async def _call_with_retries(prompt: str, req: ChatCompletionRequest) -> str:
 
 async def _call_falcon_once(prompt: str, req: ChatCompletionRequest) -> str:
     client = await get_client()
-    settings = {"model": req.model, "temperature": req.temperature, "max_new_tokens": req.max_tokens, "top_p": req.top_p}
+    settings = {
+        "model": req.model,
+        "temperature": req.temperature,
+        "max_new_tokens": req.max_tokens,
+        "top_p": req.top_p,
+    }
 
-    await asyncio.to_thread(client.predict, api_name="/new_chat")
+    # inicjalizacja nowego chatu z promptem
+    await asyncio.to_thread(
+        client.predict,
+        input_value=prompt,
+        settings_form_value=settings,
+        api_name="/new_chat",
+    )
 
     result = await asyncio.to_thread(
         client.predict,