[minor] clean up docs and eos id (#2622)

python/sglang/srt/configs/model_config.py

@@ -15,7 +15,6 @@
 import json
 import logging
 from enum import IntEnum, auto
-from functools import lru_cache
 from typing import List, Optional, Set, Union
 
 import torch
@@ -48,6 +47,7 @@ class ModelConfig:
         self.model_path = model_path
         self.revision = revision
         self.quantization = quantization
+
         # Parse args
         self.model_override_args = json.loads(model_override_args)
         self.hf_config = get_config(
@@ -131,6 +131,9 @@ class ModelConfig:
         # Veirfy quantization
         self._verify_quantization()
 
+        # Text attrs
+        self.hf_eos_token_id = self.get_hf_eos_token_id()
+
         # Multimodel attrs
         self.image_token_id = getattr(self.hf_config, "image_token_id", None)
 
@@ -272,7 +275,6 @@ class ModelConfig:
                     self.quantization,
                 )
 
-    @lru_cache()
     def get_hf_eos_token_id(self) -> Optional[Set[int]]:
         eos_ids = getattr(self.hf_config, "eos_token_id", None)
         if eos_ids:

python/sglang/srt/managers/scheduler.py

@@ -517,7 +517,7 @@ class Scheduler:
                 stream=recv_req.stream,
                 lora_path=recv_req.lora_path,
                 input_embeds=recv_req.input_embeds,
-                eos_token_ids=self.model_config.get_hf_eos_token_id(),
+                eos_token_ids=self.model_config.hf_eos_token_id,
             )
             req.tokenizer = self.tokenizer