[Feature] Support Paddle-OCR (#4396)

* init

* update code

* fix code style & disable thinking

* adapt for common_engine.update_mm_requests_chunk_size

* use 3d rope

* use flash_attn_unpadded

* opt siglip

* update to be compatible with the latest codebase

* fix typo

* optim OCR performance

* fix bug

* fix bug

* fix bug

* fix bug

* normlize name

* modify xpu rope

* revert logger

* fix bug

* fix bug

* fix bug

* support default_v1

* optim performance

* fix bug

---------

Co-authored-by: root <root@szzj-acg-tge1-fdda9.szzj.baidu.com>
Co-authored-by: zhangyue66 <zhangyue66@baidu.com>

fastdeploy/engine/sched/resource_manager_v1.py

@@ -27,6 +27,7 @@ from typing import Union
 import numpy as np
 import paddle
 
+from fastdeploy import envs
 from fastdeploy.engine.request import Request, RequestOutput, RequestStatus, RequestType
 from fastdeploy.engine.resource_manager import ResourceManager
 from fastdeploy.inter_communicator import IPCSignal
@@ -153,6 +154,7 @@ class ResourceManagerV1(ResourceManager):
         # Priority queues for requests.
         self.waiting: deque[Request] = deque()
         self.running: list[Request] = []
+        self.enable_max_prefill = envs.FD_ENABLE_MAX_PREFILL
         self.finish_execution_pool = ThreadPoolExecutor(max_workers=1)
         self.lock = threading.Lock()
         self.to_be_rescheduled_request_id_set = set()
@@ -544,8 +546,9 @@ class ResourceManagerV1(ResourceManager):
                 while self.waiting and token_budget > 0:
                     if len(self.running) == self.max_num_seqs:
                         break
-                    if (self.config.model_config.enable_mm or paddle.is_compiled_with_xpu()) and self.exist_prefill(
-                        scheduled_reqs
+                    if not self.enable_max_prefill and (
+                        (self.config.model_config.enable_mm or paddle.is_compiled_with_xpu())
+                        and self.exist_prefill(scheduled_reqs)
                     ):
                         break
                     request = self.waiting[0]