Stop chat turns on <end_of_turn>, not <eos>

tokenizer_config declared eos_token = "<eos>" (id 1, document end), but the chat template ends a turn with <end_of_turn> (id 106) and upstream generation_config lists eos_token_id = [1, 106]. A runtime that derives its stop token from eos_token alone (swift-transformers does) never saw the turn terminator and generated to the token cap.

Render-safe: the template emits the terminator literally, not via {{ eos_token }}, so rendering is unchanged. Only this one field differs; the rest of the file is byte-identical.

hf_model/tokenizer_config.json

@@ -51335,7 +51335,7 @@
   "bos_token": "<bos>",
   "clean_up_tokenization_spaces": false,
   "eoi_token": "<end_of_image>",
-  "eos_token": "<eos>",
+  "eos_token": "<end_of_turn>",
   "extra_special_tokens": {
     "boi_token": "<start_of_image>",
     "eoi_token": "<end_of_image>",