From 25f1a6ec9e8a487c282d1767cc9a174b3c1f2b99 Mon Sep 17 00:00:00 2001
From: Andreew Gregory <grinadand@gmail.com>
Date: Sun, 29 Mar 2026 14:41:39 +0300
Subject: [PATCH] infinite token limit

---
 dedicated_ai_server/server.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/dedicated_ai_server/server.py b/dedicated_ai_server/server.py
index d5ed84f..ba6e0a1 100644
--- a/dedicated_ai_server/server.py
+++ b/dedicated_ai_server/server.py
@@ -14,9 +14,6 @@ from config import Config, read_config
 from secret_stream_socket import ProtocolError, SecretStreamSocket, wrap_connection_socket
 
 
-MAX_NEW_TOKENS = 256
-
-
 @dataclass
 class MessagePiece:
     piece: str = ""
@@ -119,8 +116,9 @@ def generate_llm_pieces(bundle: ModelBundle, messages: list) -> Iterable[str]:
         eos_token_ids = set(int(x) for x in eos_token_id)
     else:
         eos_token_ids = {int(eos_token_id)}
+    print(f"[debug] eos_token_ids={sorted(eos_token_ids)}", flush=True)
 
-    for _ in range(MAX_NEW_TOKENS):
+    while True:
         with torch.inference_mode():
             outputs = bundle.model(
                 input_ids=input_ids,