change scheduler

lightllm/server/core/objs/req.py

@@ -290,13 +290,38 @@ def post_init(
         return
 
 
-class ChunkedPrefillReq(NormalReq):
+class SplitFuseReq(Req):
     _pack_ = 4
 
+    def get_tuple_tokens(self, is_busy, router_max_new_token_len):
+        has_out_len = self.shm_cur_output_len
+        if self.sample_params.ignore_eos:
+            cur_max_new_token_len = self.sample_params.max_new_tokens
+        elif is_busy:
+            cur_max_new_token_len = self.sample_params.max_new_tokens
+        else:
+            cur_max_new_token_len = min(
+                self.sample_params.max_new_tokens, max(int(1.1 * has_out_len), router_max_new_token_len)
+            )
+
+        a_len = max(self.input_len + has_out_len + 1, self.shm_cur_kv_len + 1)
+        b_len = (
+            (self.input_len + has_out_len - self.shm_cur_kv_len + self.splitfuse_block_size - 1)
+            // self.splitfuse_block_size
+            + cur_max_new_token_len
+            - has_out_len
+            - 1
+        )
+        b_len = max(0, b_len) + ADDED_OUTPUT_LEN
+
+        return (a_len, b_len)
+
+    def get_decode_need_tokens(self):
+        """
+        splitfuse 调度模式的实现
+        """
+        return min(self.input_len + self.shm_cur_output_len - self.shm_cur_kv_len, self.splitfuse_block_size)
+
     def get_first_router_need_tokens(self):
-        need_tokens = min(self.chunked_prefill_size, self.remaining_prefill_size)
-        return need_tokens
 
-    def update_remaining_prefill_size(self):
-        self.remaining_prefill_size = max(0, self.remaining_prefill_size - self.chunked_prefill_size)
-        return
+        return min(self.input_len + self.shm_cur_output_len, self.splitfuse_block_size)

lightllm/server/router/batch.py

@@ -48,18 +48,6 @@ def filter_out_finished_req(self, shm_req_manager: ShmReqManager):
         self.id_to_reqs = {req.request_id: req for req in self.reqs}
         return
 
-    def filter_out_chunked_req(self):
-        chunked_reqs = []
-        nochunked_reqs = []  # not chunked req or chunked req which completed prefill
-        for req in self.reqs:
-            if req.remaining_prefill_size > 0:
-                chunked_reqs.append(req)
-            else:
-                nochunked_reqs.append(req)
-        self.reqs = nochunked_reqs
-        self.id_to_reqs = {req.request_id: req for req in self.reqs}
-        return chunked_reqs
-
     def pop_req(self, req_id):
         self.reqs = [req for req in self.reqs if req.request_id != req_id]
         self.id_to_reqs.pop(req_id)