Tiny move files to utils folder (#11166)

benchmark/json_schema/bench_sglang.py

@@ -8,7 +8,7 @@ from datasets import load_dataset
 
 import sglang as sgl
 from sglang.global_config import global_config
-from sglang.srt.hf_transformers_utils import get_tokenizer
+from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 from sglang.test.test_utils import (
     add_common_sglang_args_and_parse,
     select_sglang_backend,

benchmark/multi_turn_chat/long_prompt_multi_turn.py

@@ -7,7 +7,7 @@ from pathlib import Path
 from tqdm import tqdm
 
 import sglang as sgl
-from sglang.srt.hf_transformers_utils import get_tokenizer
+from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 from sglang.test.test_utils import (
     add_common_sglang_args_and_parse,
     select_sglang_backend,

examples/runtime/token_in_token_out/token_in_token_out_llm_engine.py

@@ -3,7 +3,7 @@ This example demonstrates how to provide tokenized ids to LLM as input instead o
 """
 
 import sglang as sgl
-from sglang.srt.hf_transformers_utils import get_tokenizer
+from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 
 MODEL_PATH = "meta-llama/Llama-3.1-8B-Instruct"
 

examples/runtime/token_in_token_out/token_in_token_out_llm_server.py

@@ -7,7 +7,7 @@ python token_in_token_out_llm_server.py
 
 import requests
 
-from sglang.srt.hf_transformers_utils import get_tokenizer
+from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 from sglang.test.test_utils import is_in_ci
 from sglang.utils import terminate_process, wait_for_server
 

python/sglang/bench_one_batch.py

@@ -60,7 +60,6 @@ import torch.distributed as dist
 from sglang.srt.configs.model_config import ModelConfig
 from sglang.srt.distributed.parallel_state import destroy_distributed_environment
 from sglang.srt.entrypoints.engine import _set_envs_and_config
-from sglang.srt.hf_transformers_utils import get_tokenizer
 from sglang.srt.layers.moe import initialize_moe_config
 from sglang.srt.managers.schedule_batch import Req, ScheduleBatch
 from sglang.srt.managers.scheduler import Scheduler
@@ -78,6 +77,7 @@ from sglang.srt.utils import (
     set_gpu_proc_affinity,
     suppress_other_loggers,
 )
+from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 
 
 @dataclasses.dataclass

python/sglang/bench_serving.py

@@ -635,7 +635,7 @@ def get_tokenizer(
     if pretrained_model_name_or_path.endswith(
         ".json"
     ) or pretrained_model_name_or_path.endswith(".model"):
-        from sglang.srt.hf_transformers_utils import get_tokenizer
+        from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 
         return get_tokenizer(pretrained_model_name_or_path)
 

python/sglang/lang/backend/runtime_endpoint.py

@@ -433,7 +433,7 @@ class Runtime:
         self.endpoint.cache_prefix(prefix)
 
     def get_tokenizer(self):
-        from sglang.srt.hf_transformers_utils import get_tokenizer
+        from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 
         return get_tokenizer(
             self.server_args.tokenizer_path,

python/sglang/srt/configs/model_config.py

@@ -23,16 +23,16 @@ import torch
 from transformers import PretrainedConfig
 
 from sglang.srt.environ import envs
-from sglang.srt.hf_transformers_utils import (
+from sglang.srt.layers.quantization import QUANTIZATION_METHODS
+from sglang.srt.server_args import ServerArgs
+from sglang.srt.utils import is_hip, retry
+from sglang.srt.utils.hf_transformers_utils import (
     get_config,
     get_context_length,
     get_generation_config,
     get_hf_text_config,
     get_sparse_attention_config,
 )
-from sglang.srt.layers.quantization import QUANTIZATION_METHODS
-from sglang.srt.server_args import ServerArgs
-from sglang.srt.utils import is_hip, retry
 from sglang.utils import is_in_ci
 
 logger = logging.getLogger(__name__)

python/sglang/srt/lora/lora.py

@@ -26,12 +26,12 @@ import torch
 from torch import nn
 
 from sglang.srt.configs.load_config import LoadConfig
-from sglang.srt.hf_transformers_utils import AutoConfig
 from sglang.srt.lora.backend.base_backend import BaseLoRABackend
 from sglang.srt.lora.backend.chunked_backend import ChunkedSgmvLoRABackend
 from sglang.srt.lora.backend.triton_backend import TritonLoRABackend
 from sglang.srt.lora.lora_config import LoRAConfig
 from sglang.srt.model_loader.loader import DefaultModelLoader
+from sglang.srt.utils.hf_transformers_utils import AutoConfig
 
 logger = logging.getLogger(__name__)
 

python/sglang/srt/lora/lora_manager.py

@@ -21,7 +21,6 @@ from typing import Dict, Iterable, List, Optional, Set, Tuple
 import torch
 
 from sglang.srt.configs.load_config import LoadConfig
-from sglang.srt.hf_transformers_utils import AutoConfig
 from sglang.srt.lora.backend.base_backend import BaseLoRABackend, get_backend_from_name
 from sglang.srt.lora.layers import BaseLayerWithLoRA, get_lora_layer
 from sglang.srt.lora.lora import LoRAAdapter
@@ -39,6 +38,7 @@ from sglang.srt.managers.io_struct import LoRAUpdateOutput
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch
 from sglang.srt.server_args import ServerArgs
 from sglang.srt.utils import replace_submodule
+from sglang.srt.utils.hf_transformers_utils import AutoConfig
 
 logger = logging.getLogger(__name__)
 

python/sglang/srt/lora/mem_pool.py

@@ -4,7 +4,6 @@ from typing import Callable, Dict, Iterable, List, Optional, Set, Tuple, Union
 import torch
 
 from sglang.srt.distributed import divide
-from sglang.srt.hf_transformers_utils import AutoConfig
 from sglang.srt.lora.layers import BaseLayerWithLoRA
 from sglang.srt.lora.lora import LoRAAdapter
 from sglang.srt.lora.lora_config import LoRAConfig
@@ -17,6 +16,7 @@ from sglang.srt.lora.utils import (
     get_stacked_multiply,
     get_target_module_name,
 )
+from sglang.srt.utils.hf_transformers_utils import AutoConfig
 
 logger = logging.getLogger(__name__)
 

python/sglang/srt/lora/utils.py

@@ -5,7 +5,7 @@ from typing import Iterable, Optional, Set, Tuple
 
 import torch
 
-from sglang.srt.hf_transformers_utils import AutoConfig
+from sglang.srt.utils.hf_transformers_utils import AutoConfig
 
 
 @dataclass

python/sglang/srt/managers/detokenizer_manager.py

@@ -24,7 +24,6 @@ import psutil
 import setproctitle
 import zmq
 
-from sglang.srt.hf_transformers_utils import get_tokenizer
 from sglang.srt.managers.io_struct import (
     BatchEmbeddingOutput,
     BatchMultimodalDecodeReq,
@@ -42,6 +41,7 @@ from sglang.srt.utils import (
     get_zmq_socket,
     kill_itself_when_parent_died,
 )
+from sglang.srt.utils.hf_transformers_utils import get_tokenizer
 from sglang.utils import (
     TypeBasedDispatcher,
     find_printable_text,

python/sglang/srt/managers/scheduler.py

@@ -60,11 +60,6 @@ from sglang.srt.disaggregation.utils import (
 )
 from sglang.srt.distributed import get_pp_group, get_world_group
 from sglang.srt.eplb.expert_distribution import get_global_expert_distribution_recorder
-from sglang.srt.hf_transformers_utils import (
-    get_processor,
-    get_tokenizer,
-    get_tokenizer_from_processor,
-)
 from sglang.srt.layers.dp_attention import compute_dp_attention_world_info
 from sglang.srt.layers.logits_processor import LogitsProcessorOutput
 from sglang.srt.layers.moe import initialize_moe_config
@@ -190,6 +185,11 @@ from sglang.srt.utils import (
     set_random_seed,
     suppress_other_loggers,
 )
+from sglang.srt.utils.hf_transformers_utils import (
+    get_processor,
+    get_tokenizer,
+    get_tokenizer_from_processor,
+)
 from sglang.utils import TypeBasedDispatcher, get_exception_traceback
 
 logger = logging.getLogger(__name__)

python/sglang/srt/managers/scheduler_input_blocker.py

@@ -17,7 +17,7 @@ from enum import Enum, auto
 from typing import Any, List, Optional
 
 from sglang.srt.managers.io_struct import BlockReqInput, BlockReqType
-from sglang.srt.poll_based_barrier import PollBasedBarrier
+from sglang.srt.utils.poll_based_barrier import PollBasedBarrier
 
 logger = logging.getLogger(__name__)
 

python/sglang/srt/managers/scheduler_profiler_mixin.py

@@ -204,7 +204,7 @@ class SchedulerProfilerMixin:
 
             torch.distributed.barrier(self.tp_cpu_group)
             if self.tp_rank == 0:
-                from sglang.srt.rpd_utils import rpd_to_chrome_trace
+                from sglang.srt.utils.rpd_utils import rpd_to_chrome_trace
 
                 rpd_to_chrome_trace("trace.rpd", self.rpd_profile_path)
             self.rpd_profiler = None

python/sglang/srt/managers/tokenizer_manager.py

@@ -43,11 +43,6 @@ from fastapi import BackgroundTasks
 from sglang.srt.aio_rwlock import RWLock
 from sglang.srt.configs.model_config import ModelConfig
 from sglang.srt.disaggregation.utils import DisaggregationMode
-from sglang.srt.hf_transformers_utils import (
-    get_processor,
-    get_tokenizer,
-    get_tokenizer_from_processor,
-)
 from sglang.srt.lora.lora_registry import LoRARegistry
 from sglang.srt.managers.async_dynamic_batch_tokenizer import AsyncDynamicbatchTokenizer
 from sglang.srt.managers.disagg_service import start_disagg_service
@@ -99,6 +94,11 @@ from sglang.srt.utils import (
     get_zmq_socket,
     kill_process_tree,
 )
+from sglang.srt.utils.hf_transformers_utils import (
+    get_processor,
+    get_tokenizer,
+    get_tokenizer_from_processor,
+)
 from sglang.utils import TypeBasedDispatcher, get_exception_traceback
 
 asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())

python/sglang/srt/managers/tp_worker.py

@@ -22,11 +22,6 @@ import torch
 
 from sglang.srt.configs.model_config import ModelConfig
 from sglang.srt.distributed import get_pp_group, get_world_group
-from sglang.srt.hf_transformers_utils import (
-    get_processor,
-    get_tokenizer,
-    get_tokenizer_from_processor,
-)
 from sglang.srt.layers.logits_processor import LogitsProcessorOutput
 from sglang.srt.managers.io_struct import (
     DestroyWeightsUpdateGroupReqInput,
@@ -49,9 +44,14 @@ from sglang.srt.model_executor.forward_batch_info import (
     PPProxyTensors,
 )
 from sglang.srt.model_executor.model_runner import ModelRunner
-from sglang.srt.patch_torch import monkey_patch_torch_reductions
 from sglang.srt.server_args import ServerArgs
 from sglang.srt.utils import MultiprocessingSerializer, broadcast_pyobj, set_random_seed
+from sglang.srt.utils.hf_transformers_utils import (
+    get_processor,
+    get_tokenizer,
+    get_tokenizer_from_processor,
+)
+from sglang.srt.utils.patch_torch import monkey_patch_torch_reductions
 
 if TYPE_CHECKING:
     from sglang.srt.managers.cache_controller import LayerDoneCounter

python/sglang/srt/model_executor/cpu_graph_runner.py

@@ -34,7 +34,6 @@ from sglang.srt.model_executor.forward_batch_info import (
     ForwardMode,
     PPProxyTensors,
 )
-from sglang.srt.patch_torch import monkey_patch_torch_compile
 from sglang.srt.speculative.spec_info import SpeculativeAlgorithm
 from sglang.srt.utils import (
     log_info_on_rank0,
@@ -43,6 +42,7 @@ from sglang.srt.utils import (
     require_mlp_sync,
     require_mlp_tp_gather,
 )
+from sglang.srt.utils.patch_torch import monkey_patch_torch_compile
 
 logger = logging.getLogger(__name__)
 

python/sglang/srt/model_executor/cuda_graph_runner.py

@@ -48,7 +48,6 @@ from sglang.srt.model_executor.forward_batch_info import (
     PPProxyTensors,
     enable_num_token_non_padded,
 )
-from sglang.srt.patch_torch import monkey_patch_torch_compile
 from sglang.srt.two_batch_overlap import TboCudaGraphRunnerPlugin
 from sglang.srt.utils import (
     empty_context,
@@ -62,6 +61,7 @@ from sglang.srt.utils import (
     require_mlp_sync,
     require_mlp_tp_gather,
 )
+from sglang.srt.utils.patch_torch import monkey_patch_torch_compile
 
 _is_hip = is_hip()