Improve: Extract generation_manager from tokenizer_manager (#3115)

3f41b184 · fzyzcjy · GitHub · 45205d88 · 3f41b184 · 3f41b184
Unverified Commit 3f41b184 authored Feb 23, 2025 by fzyzcjy Committed by GitHub Feb 22, 2025
3 changed files
--- a/python/sglang/srt/entrypoints/engine.py
+++ b/python/sglang/srt/entrypoints/engine.py
@@ -463,5 +463,5 @@ def _launch_subprocesses(server_args: ServerArgs) -> Tuple[TokenizerManager, Dic

    # Assume all schedulers have the same scheduler_info
    scheduler_info = scheduler_infos[0]
-    tokenizer_manager.max_req_input_len = scheduler_info["max_req_input_len"]
+    tokenizer_manager.configure_max_req_input_len(scheduler_info["max_req_input_len"])
    return tokenizer_manager, scheduler_info
--- a/python/sglang/srt/managers/generation_manager.py
+++ b/python/sglang/srt/managers/generation_manager.py
--- a/python/sglang/srt/managers/tokenizer_manager.py
+++ b/python/sglang/srt/managers/tokenizer_manager.py