Fix rid state map leak + Refractor .finished (#505)

Co-authored-by: ZX <zx@lbx.dev>

python/sglang/srt/managers/controller/dp_worker.py

@@ -10,6 +10,7 @@ import zmq
 
 from sglang.global_config import global_config
 from sglang.srt.managers.controller.tp_worker import ModelTpClient
+from sglang.srt.managers.io_struct import BatchTokenIDOut
 from sglang.srt.server_args import PortArgs, ServerArgs
 from sglang.utils import get_exception_traceback
 
@@ -44,6 +45,8 @@ class DataParallelWorkerThread(threading.Thread):
             requests = []
             while not self.request_queue.empty():
                 requests.append(self.request_queue.get())
+
+            out_pyobjs: List[BatchTokenIDOut] = []
             try:
                 out_pyobjs = await self.step(requests)
             except Exception:
@@ -61,7 +64,7 @@ class DataParallelWorkerThread(threading.Thread):
 
             # async sleep for receiving the subsequent request and avoiding cache miss
             if len(out_pyobjs) != 0:
-                has_finished = any([obj.finished for obj in out_pyobjs])
+                has_finished = any([obj.finished_reason is not None for obj in out_pyobjs])
                 if has_finished:
                     await asyncio.sleep(self.request_dependency_delay)
             await asyncio.sleep(global_config.wait_for_new_request_delay)

python/sglang/srt/managers/controller/infer_batch.py

@@ -15,25 +15,47 @@ class ForwardMode(IntEnum):
     EXTEND = auto()
     DECODE = auto()
 
+class BaseFinishReason:
+    def __init__(self, is_error: bool = False):
+        self.is_error = is_error
 
-class FinishReason(IntEnum):
-    EOS_TOKEN = auto()
-    LENGTH = auto()
-    STOP_STR = auto()
-    ABORT = auto()
-
-    @staticmethod
-    def to_str(reason):
-        if reason == FinishReason.EOS_TOKEN:
-            return None
-        elif reason == FinishReason.LENGTH:
-            return "length"
-        elif reason == FinishReason.STOP_STR:
-            return "stop"
-        elif reason == FinishReason.ABORT:
-            return "abort"
-        else:
-            return None
+    def __str__(self):
+        raise NotImplementedError("Subclasses must implement this method")
+
+
+class FINISH_MATCHED_TOKEN(BaseFinishReason):
+    def __init__(self, matched: int | List[int]):
+        super().__init__()
+        self.matched = matched
+
+    def __str__(self) -> str:
+        return f"FINISH_MATCHED_TOKEN: {self.matched}"
+
+
+class FINISH_LENGTH(BaseFinishReason):
+    def __init__(self, length: int):
+        super().__init__()
+        self.length = length
+
+    def __str__(self) -> str:
+        return f"FINISH_LENGTH: {self.length}"
+
+
+class FINISH_MATCHED_STR(BaseFinishReason):
+    def __init__(self, matched: str):
+        super().__init__()
+        self.matched = matched
+
+    def __str__(self) -> str:
+        return f"FINISH_MATCHED_STR: {self.matched}"
+
+
+class FINISH_ABORT(BaseFinishReason):
+    def __init__(self):
+        super().__init__(is_error=True)
+
+    def __str__(self) -> str:
+        return "FINISH_ABORT"
 
 
 class Req:
@@ -61,11 +83,10 @@ class Req:
         self.sampling_params = None
         self.stream = False
 
-        # Check finish
         self.tokenizer = None
-        self.finished = False
-        self.finish_reason = None
-        self.hit_stop_str = None
+
+        # Check finish
+        self.finished_reason = None
 
         # Prefix info
         self.extend_input_len = 0
@@ -90,6 +111,10 @@ class Req:
         self.regex_fsm_state = 0
         self.jump_forward_map = None
 
+    # whether request reached finished condition
+    def finished(self) -> bool:
+        return self.finished_reason is not None
+
     def partial_decode(self, ids):
         first_token = self.tokenizer.convert_ids_to_tokens(ids[0])
         first_token = (
@@ -101,23 +126,21 @@ class Req:
         return self.sampling_params.max_new_tokens
 
     def check_finished(self):
-        if self.finished:
+        if self.finished():
             return
 
         if (
             len(self.prev_output_ids) + len(self.output_ids)
             >= self.sampling_params.max_new_tokens
         ):
-            self.finished = True
-            self.finish_reason = FinishReason.LENGTH
+            self.finished_reason = FINISH_LENGTH(len(self.prev_output_ids) + len(self.output_ids))
             return
 
         if (
             self.output_ids[-1] == self.tokenizer.eos_token_id
-            and self.sampling_params.ignore_eos == False
+            and not self.sampling_params.ignore_eos
         ):
-            self.finished = True
-            self.finish_reason = FinishReason.EOS_TOKEN
+            self.finished_reason = FINISH_MATCHED_TOKEN(matched=self.tokenizer.eos_token_id)
             return
 
         if len(self.sampling_params.stop_strs) > 0:
@@ -128,9 +151,7 @@ class Req:
             for stop_str in self.sampling_params.stop_strs:
                 # FIXME: (minor) try incremental match in prev_output_str
                 if stop_str in tail_str or stop_str in self.prev_output_str:
-                    self.finished = True
-                    self.finish_reason = FinishReason.STOP_STR
-                    self.hit_stop_str = stop_str
+                    self.finished_reason = FINISH_MATCHED_STR(matched=stop_str)
                     return
 
     def jump_forward_and_retokenize(self, jump_forward_str, next_state):

python/sglang/srt/managers/controller/manager_single.py

@@ -45,7 +45,7 @@ class ControllerSingle:
             # async sleep for receiving the subsequent request and avoiding cache miss
             slept = False
             if len(out_pyobjs) != 0:
-                has_finished = any([obj.finished for obj in out_pyobjs])
+                has_finished = any([obj.finished_reason is not None for obj in out_pyobjs])
                 if has_finished:
                     if self.request_dependency_delay > 0:
                         slept = True

python/sglang/srt/managers/controller/tp_worker.py

@@ -19,7 +19,7 @@ from sglang.srt.managers.io_struct import (
     FlushCacheReq,
     TokenizedGenerateReqInput,
 )
-from sglang.srt.managers.controller.infer_batch import Batch, FinishReason, ForwardMode, Req
+from sglang.srt.managers.controller.infer_batch import BaseFinishReason, Batch, FINISH_ABORT, ForwardMode, Req
 from sglang.srt.managers.controller.model_runner import ModelRunner
 from sglang.srt.managers.controller.radix_cache import RadixCache
 from sglang.srt.managers.controller.schedule_heuristic import ScheduleHeuristic
@@ -595,20 +595,19 @@ class ModelTpServer:
         output_rids = []
         prev_output_strs = []
         output_tokens = []
-        output_hit_stop_str = []
         output_skip_special_tokens = []
         output_spaces_between_special_tokens = []
         output_meta_info = []
-        output_finished = []
+        output_finished_reason: List[BaseFinishReason] = []
         finished_indices = []
         unfinished_indices = []
         for i, req in enumerate(batch.reqs):
-            if req.finished:
+            if req.finished():
                 finished_indices.append(i)
             else:
                 unfinished_indices.append(i)
 
-            if req.finished or (
+            if req.finished() or (
                 (
                     req.stream
                     and (
@@ -620,7 +619,6 @@ class ModelTpServer:
                 output_rids.append(req.rid)
                 prev_output_strs.append(req.prev_output_str)
                 output_tokens.append(req.output_ids)
-                output_hit_stop_str.append(req.hit_stop_str)
                 output_skip_special_tokens.append(
                     req.sampling_params.skip_special_tokens
                 )
@@ -632,8 +630,7 @@ class ModelTpServer:
                     "prompt_tokens": len(req.origin_input_ids),
                     "completion_tokens": len(req.prev_output_ids) + len(req.output_ids),
                     "completion_tokens_wo_jump_forward": req.completion_tokens_wo_jump_forward,
-                    "finish_reason": FinishReason.to_str(req.finish_reason),
-                    "hit_stop_str": req.hit_stop_str,
+                    "finish_reason": str(req.finished_reason),
                 }
                 if req.return_logprob:
                     (
@@ -650,7 +647,7 @@ class ModelTpServer:
                         req.normalized_prompt_logprob,
                     )
                 output_meta_info.append(meta_info)
-                output_finished.append(req.finished)
+                output_finished_reason.append(req.finished_reason)
 
         # Send to detokenizer
         if output_rids:
@@ -659,11 +656,10 @@ class ModelTpServer:
                     output_rids,
                     prev_output_strs,
                     output_tokens,
-                    output_hit_stop_str,
                     output_skip_special_tokens,
                     output_spaces_between_special_tokens,
                     output_meta_info,
-                    output_finished,
+                    output_finished_reason,
                 )
             )
 
@@ -720,8 +716,7 @@ class ModelTpServer:
         if self.running_batch:
             for req in self.running_batch.reqs:
                 if req.rid == recv_req.rid:
-                    req.finished = True
-                    req.finish_reason = FinishReason.ABORT
+                    req.finished_reason = FINISH_ABORT()
                     break
 
 

python/sglang/srt/managers/detokenizer_manager.py

@@ -9,6 +9,7 @@ from sglang.srt.hf_transformers_utils import get_tokenizer
 from sglang.srt.managers.io_struct import BatchStrOut, BatchTokenIDOut
 from sglang.srt.server_args import PortArgs, ServerArgs
 from sglang.utils import get_exception_traceback, graceful_registry
+from sglang.srt.managers.controller.infer_batch import FINISH_MATCHED_STR
 
 asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
 
@@ -34,49 +35,47 @@ class DetokenizerManager:
 
     async def handle_loop(self):
         while True:
-            recv_obj = await self.recv_from_router.recv_pyobj()
-
-            if isinstance(recv_obj, BatchTokenIDOut):
-                output_tokens = recv_obj.output_tokens
-
-                # TODO(lmzheng): handle skip_special_tokens/spaces_between_special_tokens per request
-                output_strs = self.tokenizer.batch_decode(
-                    output_tokens,
-                    skip_special_tokens=recv_obj.skip_special_tokens[0],
-                    spaces_between_special_tokens=recv_obj.spaces_between_special_tokens[
-                        0
-                    ],
-                )
-
-                # Trim stop str
-                # TODO(lmzheng): handle the case where multiple stop strs are hit
-                for i in range(len(output_strs)):
-                    if len(output_tokens[i]) > 0:
-                        first_token = self.tokenizer.convert_ids_to_tokens(
-                            int(output_tokens[i][0])
-                        )
-                        if not isinstance(first_token, str):
-                            first_token = first_token.decode("utf-8", errors="ignore")
-                        if first_token.startswith("▁"):
-                            output_strs[i] = " " + output_strs[i]
-
-                    output_strs[i] = recv_obj.prev_output_strs[i] + output_strs[i]
-
-                    if recv_obj.hit_stop_str[i] is not None:
-                        pos = output_strs[i].find(recv_obj.hit_stop_str[i])
-                        if pos != -1:
-                            output_strs[i] = output_strs[i][:pos]
-
-                self.send_to_tokenizer.send_pyobj(
-                    BatchStrOut(
-                        recv_obj.rids,
-                        output_strs,
-                        recv_obj.meta_info,
-                        recv_obj.finished,
+            recv_obj: BatchTokenIDOut = await self.recv_from_router.recv_pyobj()
+            assert isinstance(recv_obj, BatchTokenIDOut)
+
+            output_tokens = recv_obj.output_tokens
+
+            # TODO(lmzheng): handle skip_special_tokens/spaces_between_special_tokens per request
+            output_strs = self.tokenizer.batch_decode(
+                output_tokens,
+                skip_special_tokens=recv_obj.skip_special_tokens[0],
+                spaces_between_special_tokens=recv_obj.spaces_between_special_tokens[
+                    0
+                ],
+            )
+
+            # Trim stop str
+            # TODO(lmzheng): handle the case where multiple stop strs are hit
+            for i in range(len(output_strs)):
+                if len(output_tokens[i]) > 0:
+                    first_token = self.tokenizer.convert_ids_to_tokens(
+                        int(output_tokens[i][0])
                     )
+                    if not isinstance(first_token, str):
+                        first_token = first_token.decode("utf-8", errors="ignore")
+                    if first_token.startswith("▁"):
+                        output_strs[i] = " " + output_strs[i]
+
+                output_strs[i] = recv_obj.prev_output_strs[i] + output_strs[i]
+
+                if isinstance(recv_obj.finished_reason[i], FINISH_MATCHED_STR):
+                    pos = output_strs[i].find(recv_obj.finished_reason[i].matched)
+                    if pos != -1:
+                        output_strs[i] = output_strs[i][:pos]
+
+            self.send_to_tokenizer.send_pyobj(
+                BatchStrOut(
+                    rids=recv_obj.rids,
+                    output_str=output_strs,
+                    meta_info=recv_obj.meta_info,
+                    finished_reason=recv_obj.finished_reason,
                 )
-            else:
-                raise ValueError(f"Invalid object: {recv_obj}")
+            )
 
 
 def start_detokenizer_process(

python/sglang/srt/managers/io_struct.py

@@ -3,6 +3,7 @@ from dataclasses import dataclass
 from typing import Dict, List, Optional, Union
 
 from sglang.srt.sampling_params import SamplingParams
+from sglang.srt.managers.controller.infer_batch import BaseFinishReason
 
 
 @dataclass
@@ -105,21 +106,19 @@ class TokenizedGenerateReqInput:
 @dataclass
 class BatchTokenIDOut:
     rids: List[str]
-    prev_output_strs : List[str]
+    prev_output_strs: List[str]
     output_tokens: List[List[int]]
-    hit_stop_str: List[Optional[str]]
     skip_special_tokens: List[bool]
     spaces_between_special_tokens: List[bool]
     meta_info: List[Dict]
-    finished: List[bool]
-
+    finished_reason: List[BaseFinishReason]
 
 @dataclass
 class BatchStrOut:
     rids: List[str]
     output_str: List[str]
     meta_info: List[Dict]
-    finished: List[bool]
+    finished_reason: List[BaseFinishReason]
 
 
 @dataclass
@@ -134,4 +133,4 @@ class AbortReq:
 
 @dataclass
 class DetokenizeReqInput:
-    input_ids: List[int]
\ No newline at end of file
+    input_ids: List[int]

python/sglang/srt/managers/tokenizer_manager.py

@@ -4,7 +4,7 @@ import dataclasses
 import logging
 import multiprocessing as mp
 import os
-from typing import List
+from typing import List, Dict
 
 import numpy as np
 import transformers
@@ -26,6 +26,7 @@ from sglang.srt.managers.io_struct import (
     GenerateReqInput,
     TokenizedGenerateReqInput,
 )
+from sglang.srt.managers.io_struct import BatchTokenIDOut
 from sglang.srt.mm_utils import expand2square, process_anyres_image
 from sglang.srt.sampling_params import SamplingParams
 from sglang.srt.server_args import PortArgs, ServerArgs
@@ -89,7 +90,7 @@ class TokenizerManager:
             )
 
         self.to_create_loop = True
-        self.rid_to_state = {}  # Dict[str -> ReqState]
+        self.rid_to_state: Dict[str, ReqState] = {} 
 
     async def get_pixel_values(self, image_data):
         aspect_ratio = getattr(self.hf_config, "image_aspect_ratio", None)
@@ -183,12 +184,17 @@ class TokenizerManager:
                 if self.server_args.log_requests and state.finished:
                     logger.info(f"in={obj.text}, out={out}")
 
-                yield out
                 state.out_list = []
                 if state.finished:
                     del self.rid_to_state[rid]
+
+                    yield out
+
                     break
+
                 event.clear()
+
+                yield out
         else:
             if obj.stream:
                 raise ValueError("Do not support stream for batch mode.")
@@ -298,24 +304,23 @@ class TokenizerManager:
 
     async def handle_loop(self):
         while True:
-            recv_obj = await self.recv_from_detokenizer.recv_pyobj()
+            recv_obj: BatchTokenIDOut = await self.recv_from_detokenizer.recv_pyobj()
+            assert isinstance(recv_obj, BatchStrOut)
 
-            if isinstance(recv_obj, BatchStrOut):
-                for i, rid in enumerate(recv_obj.rids):
-                    state = self.rid_to_state.get(rid, None)
-                    if state is None:
-                        continue
+            for i, rid in enumerate(recv_obj.rids):
+                state = self.rid_to_state.get(rid, None)
+                if state is None:
+                    continue
+
+                recv_obj.meta_info[i]["id"] = rid
+                out_dict = {
+                    "text": recv_obj.output_str[i],
+                    "meta_info": recv_obj.meta_info[i],
+                }
+                state.out_list.append(out_dict)
+                state.finished = recv_obj.finished_reason[i] is not None
+                state.event.set()
 
-                    recv_obj.meta_info[i]["id"] = rid
-                    out_dict = {
-                        "text": recv_obj.output_str[i],
-                        "meta_info": recv_obj.meta_info[i],
-                    }
-                    state.out_list.append(out_dict)
-                    state.finished = recv_obj.finished[i]
-                    state.event.set()
-            else:
-                raise ValueError(f"Invalid object: {recv_obj}.")
 
     def convert_logprob_style(
         self, ret, return_logprob, top_logprobs_num, return_text_in_logprobs