[Bug Fix] fix outdated doc and disable mm model prefix caching (#4425)

* fix outdated doc and disable mm model prefix caching

* fix outdated doc and disable mm model prefix caching

---------

Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com>

docs/zh/offline_inference.md

@@ -92,7 +92,7 @@ from PIL import Image
 
 from fastdeploy.entrypoints.llm import LLM
 from fastdeploy.engine.sampling_params import SamplingParams
-from fastdeploy.input.ernie_tokenizer import Ernie4_5Tokenizer
+from fastdeploy.input.ernie4_5_tokenizer import Ernie4_5Tokenizer
 
 PATH = "baidu/ERNIE-4.5-VL-28B-A3B-Paddle"
 tokenizer = Ernie4_5Tokenizer.from_pretrained(PATH)

fastdeploy/config.py

@@ -1257,6 +1257,8 @@ class FDConfig:
 
         self.cache_config.postprocess(self.max_num_batched_tokens, self.max_num_seqs)
         self.cache_config.max_block_num_per_seq = int(self.max_model_len // self.cache_config.block_size)
+        if self.model_config is not None and self.model_config.enable_mm:
+            self.cache_config.enable_prefix_caching = False
 
         if self.guided_decoding_backend == "auto":
             if self.model_config.enable_mm:

fastdeploy/engine/args_utils.py

@@ -394,8 +394,6 @@ class EngineArgs:
             self.enable_prefix_caching = False
         if self.speculative_config is not None:
             self.enable_prefix_caching = False
-        if self.enable_mm:
-            self.enable_prefix_caching = False
         if not current_platform.is_cuda():
             self.enable_prefix_caching = False
         # if self.dynamic_load_weight: