Split the __init__ of scheduler as smaller functions. Improve the eagle tests (#4128)

2025-03-06 00:13:20 -08:00
parent 0804dd11a0
commit fcc2e37f69
7 changed files with 279 additions and 341 deletions
--- a/python/sglang/srt/managers/tokenizer_manager.py
+++ b/python/sglang/srt/managers/tokenizer_manager.py
@@ -1068,6 +1068,7 @@ class TokenizerManager:
            self.metrics_collector.observe_one_finished_request(
                recv_obj.prompt_tokens[i],
                completion_tokens,
+                recv_obj.cached_tokens[i],
                state.finished_time - state.created_time,
            )