sglang 0.5.2rc1__py3-none-any.whl → 0.5.2rc2__py3-none-any.whl

sglang/srt/managers/multi_tokenizer_mixin.py

@@ -23,6 +23,7 @@ import threading
 from multiprocessing import shared_memory
 from typing import Dict
 
+import setproctitle
 import zmq
 import zmq.asyncio
 
@@ -476,6 +477,9 @@ class MultiTokenizerManager(TokenizerManager, MultiTokenizerMixin):
         server_args: ServerArgs,
         port_args: PortArgs,
     ):
+        setproctitle.setproctitle(
+            f"sglang::http_server/multi_tokenizer_manager:{os.getpid()}"
+        )
         # prevent init prefill bootstrapserver again
         disaggregation_mode = server_args.disaggregation_mode
         server_args.disaggregation_mode = "null"

sglang/srt/managers/schedule_policy.py

@@ -380,8 +380,9 @@ class PrefillAdder:
         self.log_input_tokens += extend_input_len
 
     def add_chunked_req(self, req: Req):
-        truncated = req.extend_input_len > self.rem_chunk_tokens
-        req.extend_input_len = min(req.extend_input_len, self.rem_chunk_tokens)
+        _rem_tokens = min(self.rem_chunk_tokens, int(self.rem_total_tokens))
+        truncated = req.extend_input_len > _rem_tokens
+        req.extend_input_len = min(req.extend_input_len, _rem_tokens)
         req.fill_ids = req.fill_ids[: len(req.prefix_indices) + req.extend_input_len]
         self.can_run_list.append(req)
         self._update_prefill_budget(

sglang/srt/managers/scheduler.py

@@ -141,7 +141,7 @@ from sglang.srt.mem_cache.lora_radix_cache import LoRARadixCache
 from sglang.srt.mem_cache.radix_cache import RadixCache
 from sglang.srt.mem_cache.swa_radix_cache import SWARadixCache
 from sglang.srt.model_executor.forward_batch_info import ForwardMode, PPProxyTensors
-from sglang.srt.reasoning_parser import ReasoningParser
+from sglang.srt.parser.reasoning_parser import ReasoningParser
 from sglang.srt.server_args import PortArgs, ServerArgs
 from sglang.srt.speculative.spec_info import SpeculativeAlgorithm
 from sglang.srt.torch_memory_saver_adapter import TorchMemorySaverAdapter
@@ -500,6 +500,7 @@ class Scheduler(
         # Init metrics stats
         self.init_metrics(tp_rank, pp_rank, dp_rank)
         self.init_kv_events(server_args.kv_events_config)
+        self.init_dp_balance(dp_balance_meta)
 
         # Init disaggregation
         self.disaggregation_mode = DisaggregationMode(
@@ -545,15 +546,6 @@ class Scheduler(
             ]
         )
 
-        self.balance_meta = dp_balance_meta
-        if (
-            server_args.enable_dp_attention
-            and server_args.load_balance_method == "minimum_tokens"
-        ):
-            assert dp_balance_meta is not None
-
-        self.recv_dp_balance_id_this_term = []
-
     def init_tokenizer(self):
         server_args = self.server_args
         self.is_generation = self.model_config.is_generation
@@ -1126,11 +1118,7 @@ class Scheduler(
         self,
         recv_req: TokenizedGenerateReqInput,
     ):
-        if (
-            self.server_args.enable_dp_attention
-            and self.server_args.load_balance_method == "minimum_tokens"
-        ):
-            self.recv_dp_balance_id_this_term.append(recv_req.dp_balance_id)
+        self.maybe_update_dp_balance_data(recv_req)
 
         # Create a new request
         if (
@@ -1568,11 +1556,7 @@ class Scheduler(
 
         # Handle DP attention
         if need_dp_attn_preparation:
-            if (
-                self.server_args.load_balance_method == "minimum_tokens"
-                and self.forward_ct % 40 == 0
-            ):
-                self.handle_dp_balance_data(ret)
+            self.maybe_handle_dp_balance_data()
             ret = self.prepare_mlp_sync_batch(ret)
 
         return ret
@@ -1897,86 +1881,6 @@ class Scheduler(
             disable_overlap_schedule=self.server_args.disable_overlap_schedule,
         )
 
-    def handle_dp_balance_data(self, local_batch: ScheduleBatch):
-        def gather_dp_balance_info(holding_tokens_list) -> Union[None, List[List[int]]]:
-            """gather recv_dp_balance_id_this_term and holding tokens per worker for dp balance"""
-            recv_list = self.recv_dp_balance_id_this_term
-            assert len(recv_list) <= 511, (
-                "The number of requests received this round is too large. "
-                "Please increase gather_tensor_size and onfly_info_size."
-            )
-            # The maximum size of the tensor used for gathering data from all workers.
-            gather_tensor_size = 512
-
-            # recv_tensor: | holding_tokens | len(recv_dp_balance_id) | recv_dp_balance_ids
-            recv_tensor = torch.zeros(gather_tensor_size, dtype=torch.int32)
-            recv_tensor[0] = holding_tokens_list
-            recv_tensor[1] = len(
-                recv_list
-            )  # The first element is the length of the list.
-            recv_tensor[2 : len(recv_list) + 2] = torch.tensor(
-                recv_list, dtype=torch.int32
-            )
-
-            if self.tp_rank == 0:
-                gathered_list = [
-                    torch.zeros(gather_tensor_size, dtype=torch.int32)
-                    for _ in range(self.balance_meta.num_workers)
-                ]
-            else:
-                gathered_list = None
-
-            torch.distributed.gather(
-                recv_tensor, gathered_list, group=self.tp_cpu_group
-            )
-
-            gathered_id_list_per_worker = None
-            if self.tp_rank == 0:
-                gathered_id_list_per_worker = []
-                holding_tokens_list = []
-                for tensor in gathered_list:
-                    holding_tokens_list.append(tensor[0].item())
-                    list_length = tensor[1].item()
-                    gathered_id_list_per_worker.append(
-                        tensor[2 : list_length + 2].tolist()
-                    )
-
-            return gathered_id_list_per_worker, holding_tokens_list
-
-        def write_shared_dp_balance_info(new_recv_rid_lists, local_tokens):
-            meta = self.balance_meta
-
-            with meta.mutex:
-                onfly_list: List[Dict[int, int]] = meta.get_shared_onfly()
-                assert len(new_recv_rid_lists) == len(
-                    onfly_list
-                ), "num_worker not equal"
-                # 1.Check if the rid received by each worker this round is present in onfly.
-                #   If it is, remove the corresponding onfly item.
-                worker_id = 0
-                for new_recv_rids, on_fly_reqs in zip(new_recv_rid_lists, onfly_list):
-                    for new_recv_rid in new_recv_rids:
-                        assert (
-                            new_recv_rid in on_fly_reqs
-                        ), f"{new_recv_rid=} not in {worker_id=} {on_fly_reqs=}, data consistency is wrong"
-                        del on_fly_reqs[new_recv_rid]
-                    worker_id += 1
-                # 2. Atomically write local_tokens and onfly into shm under the mutex
-                meta.set_shared_onfly_info(onfly_list)
-                meta.set_shared_local_tokens(local_tokens)
-
-        holding_tokens = self.get_load()
-
-        new_recv_dp_balance_id_list, holding_token_list = gather_dp_balance_info(
-            holding_tokens
-        )
-
-        self.recv_dp_balance_id_this_term.clear()
-        if self.tp_rank == 0:  # only first worker write info
-            write_shared_dp_balance_info(
-                new_recv_dp_balance_id_list, holding_token_list
-            )
-
     @staticmethod
     def prepare_mlp_sync_batch_raw(
         local_batch: ScheduleBatch,

sglang/srt/managers/scheduler_metrics_mixin.py

@@ -1,15 +1,24 @@
+from __future__ import annotations
+
 import logging
 import time
 from collections import defaultdict
-from typing import List, Optional
+from typing import TYPE_CHECKING, Dict, List, Optional, Union
+
+import torch
 
 from sglang.srt.disaggregation.kv_events import EventPublisherFactory, KVEventBatch
 from sglang.srt.disaggregation.utils import DisaggregationMode
+from sglang.srt.managers.io_struct import TokenizedGenerateReqInput
 from sglang.srt.managers.schedule_policy import PrefillAdder
 from sglang.srt.managers.scheduler import Req, ScheduleBatch
+from sglang.srt.managers.utils import DPBalanceMeta
 from sglang.srt.metrics.collector import SchedulerMetricsCollector, SchedulerStats
 from sglang.srt.utils import get_bool_env_var
 
+if TYPE_CHECKING:
+    from sglang.srt.managers.scheduler import Scheduler
+
 logger = logging.getLogger(__name__)
 
 RECORD_STEP_TIME = get_bool_env_var("SGLANG_RECORD_STEP_TIME")
@@ -28,7 +37,9 @@ class KvMetrics:
 
 
 class SchedulerMetricsMixin:
-    def init_metrics(self, tp_rank: int, pp_rank: int, dp_rank: Optional[int]):
+    def init_metrics(
+        self: Scheduler, tp_rank: int, pp_rank: int, dp_rank: Optional[int]
+    ):
         self.last_gen_throughput: float = 0.0
         self.last_input_throughput: float = 0.0
         self.step_time_dict = defaultdict(list)  # Dict[batch size -> step time]
@@ -50,14 +61,24 @@ class SchedulerMetricsMixin:
                 labels["dp_rank"] = dp_rank
             self.metrics_collector = SchedulerMetricsCollector(labels=labels)
 
-    def init_kv_events(self, kv_events_config: Optional[str]):
+    def init_dp_balance(self: Scheduler, dp_balance_meta: Optional[DPBalanceMeta]):
+        self.balance_meta = dp_balance_meta
+        if (
+            self.server_args.enable_dp_attention
+            and self.server_args.load_balance_method == "minimum_tokens"
+        ):
+            assert dp_balance_meta is not None
+
+        self.recv_dp_balance_id_this_term = []
+
+    def init_kv_events(self: Scheduler, kv_events_config: Optional[str]):
         if self.enable_kv_cache_events:
             self.kv_event_publisher = EventPublisherFactory.create(
                 kv_events_config, self.attn_dp_rank
             )
 
     def log_prefill_stats(
-        self,
+        self: Scheduler,
         adder: PrefillAdder,
         can_run_list: List[Req],
         running_bs: int,
@@ -138,7 +159,7 @@ class SchedulerMetricsMixin:
         self._publish_kv_events()
 
     def log_decode_stats(
-        self, can_run_cuda_graph: bool, running_batch: ScheduleBatch = None
+        self: Scheduler, can_run_cuda_graph: bool, running_batch: ScheduleBatch = None
     ):
         batch = running_batch or self.running_batch
 
@@ -220,7 +241,7 @@ class SchedulerMetricsMixin:
             self._emit_kv_metrics()
         self._publish_kv_events()
 
-    def _emit_kv_metrics(self):
+    def _emit_kv_metrics(self: Scheduler):
         kv_metrics = KvMetrics()
         kv_metrics.request_active_slots = self.stats.num_running_reqs
         kv_metrics.request_total_slots = self.max_running_requests
@@ -236,9 +257,94 @@ class SchedulerMetricsMixin:
         if not self.send_metrics_from_scheduler.closed:
             self.send_metrics_from_scheduler.send_pyobj(kv_metrics)
 
-    def _publish_kv_events(self):
+    def _publish_kv_events(self: Scheduler):
         if self.enable_kv_cache_events:
             events = self.tree_cache.take_events()
             if events:
                 batch = KVEventBatch(ts=time.time(), events=events)
                 self.kv_event_publisher.publish(batch)
+
+    def maybe_update_dp_balance_data(
+        self: Scheduler, recv_req: TokenizedGenerateReqInput
+    ):
+        if (
+            self.server_args.enable_dp_attention
+            and self.server_args.load_balance_method == "minimum_tokens"
+        ):
+            self.recv_dp_balance_id_this_term.append(recv_req.dp_balance_id)
+
+    def maybe_handle_dp_balance_data(self: Scheduler):
+        if (
+            self.server_args.load_balance_method == "minimum_tokens"
+            and self.forward_ct % 40 == 0
+        ):
+            holding_tokens = self.get_load()
+
+            new_recv_dp_balance_id_list, holding_token_list = (
+                self.gather_dp_balance_info(holding_tokens)
+            )
+
+            self.recv_dp_balance_id_this_term.clear()
+            if self.tp_rank == 0:  # only first worker write info
+                self.write_shared_dp_balance_info(
+                    new_recv_dp_balance_id_list, holding_token_list
+                )
+
+    def gather_dp_balance_info(
+        self: Scheduler, holding_tokens_list
+    ) -> Union[None, List[List[int]]]:
+        """gather recv_dp_balance_id_this_term and holding tokens per worker for dp balance"""
+        recv_list = self.recv_dp_balance_id_this_term
+        assert len(recv_list) <= 511, (
+            "The number of requests received this round is too large. "
+            "Please increase gather_tensor_size and onfly_info_size."
+        )
+        # The maximum size of the tensor used for gathering data from all workers.
+        gather_tensor_size = 512
+
+        # recv_tensor: | holding_tokens | len(recv_dp_balance_id) | recv_dp_balance_ids
+        recv_tensor = torch.zeros(gather_tensor_size, dtype=torch.int32)
+        recv_tensor[0] = holding_tokens_list
+        recv_tensor[1] = len(recv_list)  # The first element is the length of the list.
+        recv_tensor[2 : len(recv_list) + 2] = torch.tensor(recv_list, dtype=torch.int32)
+
+        if self.tp_rank == 0:
+            gathered_list = [
+                torch.zeros(gather_tensor_size, dtype=torch.int32)
+                for _ in range(self.balance_meta.num_workers)
+            ]
+        else:
+            gathered_list = None
+
+        torch.distributed.gather(recv_tensor, gathered_list, group=self.tp_cpu_group)
+
+        gathered_id_list_per_worker = None
+        if self.tp_rank == 0:
+            gathered_id_list_per_worker = []
+            holding_tokens_list = []
+            for tensor in gathered_list:
+                holding_tokens_list.append(tensor[0].item())
+                list_length = tensor[1].item()
+                gathered_id_list_per_worker.append(tensor[2 : list_length + 2].tolist())
+
+        return gathered_id_list_per_worker, holding_tokens_list
+
+    def write_shared_dp_balance_info(self: Scheduler, new_recv_rid_lists, local_tokens):
+        meta = self.balance_meta
+
+        with meta.mutex:
+            onfly_list: List[Dict[int, int]] = meta.get_shared_onfly()
+            assert len(new_recv_rid_lists) == len(onfly_list), "num_worker not equal"
+            # 1.Check if the rid received by each worker this round is present in onfly.
+            #   If it is, remove the corresponding onfly item.
+            worker_id = 0
+            for new_recv_rids, on_fly_reqs in zip(new_recv_rid_lists, onfly_list):
+                for new_recv_rid in new_recv_rids:
+                    assert (
+                        new_recv_rid in on_fly_reqs
+                    ), f"{new_recv_rid=} not in {worker_id=} {on_fly_reqs=}, data consistency is wrong"
+                    del on_fly_reqs[new_recv_rid]
+                worker_id += 1
+            # 2. Atomically write local_tokens and onfly into shm under the mutex
+            meta.set_shared_onfly_info(onfly_list)
+            meta.set_shared_local_tokens(local_tokens)

sglang/srt/managers/template_manager.py

@@ -24,20 +24,20 @@ import os
 import re
 from typing import Optional
 
-from sglang.srt.code_completion_parser import (
+from sglang.srt.parser.code_completion_parser import (
     CompletionTemplate,
     FimPosition,
     completion_template_exists,
     register_completion_template,
 )
-from sglang.srt.conversation import (
+from sglang.srt.parser.conversation import (
     Conversation,
     SeparatorStyle,
     chat_template_exists,
     get_conv_template_by_model_path,
     register_conv_template,
 )
-from sglang.srt.jinja_template_utils import detect_jinja_template_content_format
+from sglang.srt.parser.jinja_template_utils import detect_jinja_template_content_format
 
 logger = logging.getLogger(__name__)
 

sglang/srt/managers/tokenizer_manager.py

@@ -329,6 +329,7 @@ class TokenizerManager:
         # Metrics
         if self.enable_metrics:
             self.metrics_collector = TokenizerMetricsCollector(
+                server_args=server_args,
                 labels={
                     "model_name": self.server_args.served_model_name,
                     # TODO: Add lora name/path in the future,

sglang/srt/mem_cache/allocator.py

@@ -283,7 +283,7 @@ class SWATokenToKVPoolAllocator(BaseTokenToKVPoolAllocator):
         self.swa_attn_allocator.clear()
         self.full_attn_allocator.clear()
         self.full_to_swa_index_mapping.fill_(0)
-        self.is_in_free_group = False
+        self.is_not_in_free_group = True
         self.free_group = []
 
 

sglang/srt/mem_cache/hicache_storage.py

@@ -27,6 +27,7 @@ class HiCacheStorageConfig:
     tp_rank: int
     tp_size: int
     is_mla_model: bool
+    is_page_first_layout: bool
     model_name: Optional[str]
     extra_config: Optional[dict] = None
 
@@ -135,18 +136,24 @@ class HiCacheFile(HiCacheStorage):
     ):
         self.file_path = os.getenv("SGLANG_HICACHE_FILE_BACKEND_STORAGE_DIR", file_path)
 
-        tp_rank, tp_size, is_mla = (
+        tp_rank, tp_size, model_name, is_mla_model = (
             storage_config.tp_rank,
             storage_config.tp_size,
+            storage_config.model_name,
             storage_config.is_mla_model,
         )
-        self.tp_suffix = f"_{tp_rank}_{tp_size}" if tp_size > 1 and not is_mla else ""
+        model_name = "-".join(model_name.split("/")) if model_name else ""
+        if is_mla_model:
+            self.config_suffix = f"_{model_name}"
+        else:
+            self.config_suffix = f"_{model_name}_{tp_rank}_{tp_size}"
+
         if not os.path.exists(self.file_path) and tp_rank == 0:
             os.makedirs(self.file_path)
             logger.info(f"Created HiCacheFile storage directory at {self.file_path}")
 
     def _get_suffixed_key(self, key: str) -> str:
-        return key + self.tp_suffix
+        return key + self.config_suffix
 
     def get(
         self,
@@ -157,13 +164,11 @@ class HiCacheFile(HiCacheStorage):
         key = self._get_suffixed_key(key)
         tensor_path = os.path.join(self.file_path, f"{key}.bin")
         try:
-            # Load directly into target_location's memory buffer
-            with open(tensor_path, "rb") as f:
-                target_location.set_(
-                    torch.frombuffer(f.read(), dtype=target_location.dtype)
-                    .reshape(target_location.shape)
-                    .untyped_storage()
-                )
+            expected = target_location.numel() * target_location.element_size()
+            with open(tensor_path, "rb", buffering=0) as f:
+                buf = memoryview(target_location.view(torch.uint8).contiguous().numpy())
+                if f.readinto(buf) != expected:
+                    raise IOError(f"Short read for {key}")
             return target_location
         except FileNotFoundError:
             logger.warning(f"Failed to fetch {key} from HiCacheFile storage.")

sglang/srt/mem_cache/hiradix_cache.py

@@ -468,9 +468,9 @@ class HiRadixCache(RadixCache):
 
         # todo: more policies for prefetch progress such as timeout
         # the current policy is to prefetch with best effort and terminate when queuing is over
-        last_host_node, token_ids, host_indices, operation = self.ongoing_prefetch.pop(
+        last_host_node, token_ids, host_indices, operation = self.ongoing_prefetch[
             req_id
-        )
+        ]
 
         if operation.host_indices is None:
             # prefetch has not been issued due to insufficient host memory
@@ -512,6 +512,7 @@ class HiRadixCache(RadixCache):
             host_indices[min_completed_tokens:completed_tokens]
         )
         last_host_node.release_host()
+        del self.ongoing_prefetch[req_id]
         self.cache_controller.prefetch_tokens_occupied -= len(token_ids)
 
         return True
@@ -775,9 +776,7 @@ class HiRadixCache(RadixCache):
         if rid not in self.ongoing_prefetch:
             return
 
-        last_host_node, token_ids, host_indices, operation = self.ongoing_prefetch.pop(
-            rid
-        )
+        last_host_node, token_ids, host_indices, operation = self.ongoing_prefetch[rid]
         if operation.host_indices is None:
             return
 
@@ -785,5 +784,6 @@ class HiRadixCache(RadixCache):
         if self.tp_world_size > 1:
             torch.distributed.barrier(group=self.tp_group)
         last_host_node.release_host()
+        del self.ongoing_prefetch[rid]
         self.cache_controller.append_host_mem_release(host_indices[:completed_tokens])
         self.cache_controller.prefetch_tokens_occupied -= len(token_ids)

sglang/srt/mem_cache/memory_pool_host.py

@@ -500,20 +500,23 @@ class MHATokenToKVPoolHost(HostKVCache):
         element_size_list = [element_size] * len(key_list)
         return key_list, ptr_list, element_size_list
 
-    def get_buffer_with_hash(self, keys, indices):
+    def get_buffer_with_hash(self, keys, indices=None):
         assert self.layout == "page_first"
-        assert len(keys) == (len(indices) // self.page_size)
+        assert indices is None or (len(keys) == (len(indices) // self.page_size))
 
         key_list = []
         buf_list = []
 
-        for key, i in zip(keys, range(0, len(indices), self.page_size)):
+        for i in range(len(keys)):
+            key = keys[i]
             key_list.append(f"{key}-k")
-            buf_list.append(self.k_buffer[i : i + self.page_size])
             key_list.append(f"{key}-v")
-            buf_list.append(self.v_buffer[i : i + self.page_size])
+            if indices is not None:
+                index = indices[i * self.page_size]
+                buf_list.append(self.k_buffer[index : index + self.page_size])
+                buf_list.append(self.v_buffer[index : index + self.page_size])
 
-        return key_list, buf_list
+        return key_list, buf_list, 2
 
 
 class MLATokenToKVPoolHost(HostKVCache):
@@ -728,13 +731,15 @@ class MLATokenToKVPoolHost(HostKVCache):
         element_size_list = [element_size] * len(key_list)
         return key_list, ptr_list, element_size_list
 
-    def get_buffer_with_hash(self, keys, indices):
+    def get_buffer_with_hash(self, keys, indices=None):
         assert self.layout == "page_first"
-        assert len(keys) == (len(indices) // self.page_size)
+        assert indices is None or (len(keys) == (len(indices) // self.page_size))
 
         buf_list = []
 
-        for i in range(0, len(indices), self.page_size):
-            buf_list.append(self.kv_buffer[i : i + self.page_size])
+        if indices is not None:
+            for i in range(len(keys)):
+                index = indices[i * self.page_size]
+                buf_list.append(self.kv_buffer[index : index + self.page_size])
 
-        return keys, buf_list
+        return keys, buf_list, 1

sglang/srt/mem_cache/storage/hf3fs/storage_hf3fs.py

@@ -128,6 +128,7 @@ class HiCacheHF3FS(HiCacheStorage):
         dtype: torch.dtype,
         metadata_client: Hf3fsMetadataInterface,
         is_mla_model: bool = False,
+        is_page_first_layout: bool = False,
     ):
         self.rank = rank
         self.file_path = file_path
@@ -138,6 +139,7 @@ class HiCacheHF3FS(HiCacheStorage):
         self.dtype = dtype
         self.metadata_client = metadata_client
         self.is_mla_model = is_mla_model
+        self.is_page_first_layout = is_page_first_layout
         self.numel = self.bytes_per_page // self.dtype.itemsize
         self.num_pages = self.file_size // self.bytes_per_page
         self.skip_backup = False
@@ -193,9 +195,13 @@ class HiCacheHF3FS(HiCacheStorage):
         )
 
         if storage_config is not None:
-            rank, is_mla_model = storage_config.tp_rank, storage_config.is_mla_model
+            rank, is_mla_model, is_page_first_layout = (
+                storage_config.tp_rank,
+                storage_config.is_mla_model,
+                storage_config.is_page_first_layout,
+            )
         else:
-            rank, is_mla_model = 0, False
+            rank, is_mla_model, is_page_first_layout = 0, False, False
 
         mla_unsupported_msg = f"MLA model is not supported without global metadata server, please refer to https://github.com/sgl-project/sglang/blob/main/python/sglang/srt/mem_cache/storage/hf3fs/docs/deploy_sglang_3fs_multinode.md"
 
@@ -213,6 +219,7 @@ class HiCacheHF3FS(HiCacheStorage):
                 entries=8,
                 dtype=dtype,
                 metadata_client=Hf3fsLocalMetadataClient(),
+                is_page_first_layout=is_page_first_layout,
             )
 
         try:
@@ -261,6 +268,7 @@ class HiCacheHF3FS(HiCacheStorage):
             dtype=dtype,
             metadata_client=metadata_client,
             is_mla_model=is_mla_model,
+            is_page_first_layout=is_page_first_layout,
         )
 
     def get(