Revert removing the unused imports

python/sglang/api.py

@@ -3,7 +3,11 @@
 import re
 from typing import Callable, List, Optional, Union
 
+from sglang.backend.anthropic import Anthropic
 from sglang.backend.base_backend import BaseBackend
+from sglang.backend.openai import OpenAI
+from sglang.backend.runtime_endpoint import RuntimeEndpoint
+from sglang.backend.vertexai import VertexAI
 from sglang.global_config import global_config
 from sglang.lang.ir import (
  SglExpr,

python/sglang/backend/anthropic.py

@@ -1,3 +1,7 @@
+from typing import List, Optional, Union
+
+import numpy as np
+
 from sglang.backend.base_backend import BaseBackend
 from sglang.lang.chat_template import get_chat_template
 from sglang.lang.interpreter import StreamExecutor

python/sglang/backend/base_backend.py

@@ -1,4 +1,4 @@
-from typing import List, Optional, Union
+from typing import Callable, List, Optional, Union
 
 from sglang.lang.chat_template import get_chat_template
 from sglang.lang.interpreter import StreamExecutor

python/sglang/backend/openai.py

@@ -1,6 +1,6 @@
 import logging
 import time
-from typing import List, Optional
+from typing import Callable, List, Optional, Union
 
 import numpy as np
 

python/sglang/backend/runtime_endpoint.py

@@ -1,14 +1,15 @@
 import json
-from typing import List, Optional
+from typing import Callable, List, Optional, Union
 
 import numpy as np
+import requests
 
 from sglang.backend.base_backend import BaseBackend
 from sglang.global_config import global_config
 from sglang.lang.chat_template import get_chat_template_by_model_path
 from sglang.lang.interpreter import StreamExecutor
-from sglang.lang.ir import SglSamplingParams
-from sglang.utils import find_printable_text, http_request
+from sglang.lang.ir import SglArgument, SglSamplingParams
+from sglang.utils import encode_image_base64, find_printable_text, http_request
 
 
 class RuntimeEndpoint(BaseBackend):

python/sglang/backend/vertexai.py

@@ -1,5 +1,8 @@
 import os
 import warnings
+from typing import List, Optional, Union
+
+import numpy as np
 
 from sglang.backend.base_backend import BaseBackend
 from sglang.lang.chat_template import get_chat_template

python/sglang/lang/chat_template.py

@@ -1,6 +1,6 @@
-from dataclasses import dataclass
+from dataclasses import dataclass, field
 from enum import Enum, auto
-from typing import Callable, Dict, List, Tuple
+from typing import Callable, Dict, List, Optional, Tuple
 
 
 class ChatTemplateStyle(Enum):

python/sglang/lang/compiler.py

@@ -5,7 +5,13 @@
 
 from sglang.global_config import global_config
 from sglang.lang.interpreter import ProgramState, StreamExecutor, pin_program
-from sglang.lang.ir import SglArgument, SglExpr, SglSamplingParams, SglVariable
+from sglang.lang.ir import (
+ SglArgument,
+ SglConstantText,
+ SglExpr,
+ SglSamplingParams,
+ SglVariable,
+)
 
 
 def compile_func(function, backend):

python/sglang/lang/interpreter.py

@@ -7,7 +7,7 @@
 import uuid
 from concurrent.futures import ThreadPoolExecutor
 from contextlib import contextmanager
-from typing import Any, Callable, Dict, List, Optional
+from typing import Any, Callable, Dict, List, Optional, Union
 
 import tqdm
 
@@ -18,6 +18,7 @@
  SglConstantText,
  SglExpr,
  SglExprList,
+ SglFunction,
  SglGen,
  SglImage,
  SglRoleBegin,

python/sglang/lang/ir.py

@@ -472,4 +472,4 @@ def __init__(self):
  super().__init__()
 
  def __repr__(self):
- return "CommitLazy()"
+ return f"CommitLazy()"

python/sglang/lang/tracer.py

@@ -1,16 +1,20 @@
 """Tracing a program."""
 
 import uuid
-from typing import Any, Dict, List, Optional
+from typing import Any, Callable, Dict, List, Optional, Union
 
 from sglang.backend.base_backend import BaseBackend
+from sglang.global_config import global_config
 from sglang.lang.interpreter import ProgramState, ProgramStateGroup
 from sglang.lang.ir import (
  SglArgument,
+ SglCommitLazy,
+ SglConcateAndAppend,
  SglConstantText,
  SglExpr,
  SglExprList,
  SglFork,
+ SglFunction,
  SglGen,
  SglGetForkItem,
  SglRoleBegin,

python/sglang/srt/hf_transformers_utils.py

@@ -3,7 +3,7 @@
 import json
 import os
 import warnings
-from typing import Optional, Union
+from typing import List, Optional, Tuple, Union
 
 from huggingface_hub import snapshot_download
 from transformers import (

python/sglang/srt/managers/detokenizer_manager.py

@@ -84,7 +84,7 @@ def start_detokenizer_process(
 ):
  try:
  manager = DetokenizerManager(server_args, port_args)
- except Exception:
+ except Exception as e:
  pipe_writer.send(get_exception_traceback())
  raise
  pipe_writer.send("init ok")

python/sglang/srt/managers/router/radix_cache.py

@@ -1,6 +1,8 @@
 import heapq
 import time
 from collections import defaultdict
+from dataclasses import dataclass
+from typing import Tuple
 
 import torch
 

python/sglang/srt/models/commandr.py

@@ -20,7 +20,7 @@
 
 # This file is based on the LLama model definition file in transformers
 """PyTorch Cohere model."""
-from typing import Optional, Tuple
+from typing import List, Optional, Tuple
 
 import torch
 import torch.utils.checkpoint

python/sglang/srt/models/llama2.py

@@ -1,7 +1,7 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/671af2b1c0b3ed6d856d37c21a561cc429a10701/vllm/model_executor/models/llama.py#L1
 """Inference-only LLaMA model compatible with HuggingFace weights."""
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Dict, List, Optional, Tuple
 
 import torch
 from torch import nn

python/sglang/srt/models/llava.py

@@ -5,7 +5,7 @@
 import numpy as np
 import torch
 from torch import nn
-from transformers import CLIPVisionModel, LlavaConfig
+from transformers import CLIPVisionModel, LlamaConfig, LlavaConfig
 from transformers.models.llava.modeling_llava import LlavaMultiModalProjector
 from vllm.model_executor.layers.linear import LinearMethodBase
 from vllm.model_executor.weight_utils import (

python/sglang/srt/models/mixtral.py

@@ -1,7 +1,7 @@
 # Adapted from
 # https://github.com/vllm-project/vllm/blob/d0215a58e78572d91dadafe9d832a2db89b09a13/vllm/model_executor/models/mixtral.py#L1
 """Inference-only Mixtral model."""
-from typing import Optional
+from typing import List, Optional, Tuple
 
 import numpy as np
 import torch

python/sglang/srt/models/qwen.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict, Optional
+from typing import Any, Dict, List, Optional, Tuple
 
 import torch
 from torch import nn

python/sglang/srt/models/qwen2.py

@@ -1,7 +1,7 @@
 # Adapted from llama2.py
 # Modify details for the adaptation of Qwen2 model.
 """Inference-only Qwen2 model compatible with HuggingFace weights."""
-from typing import Any, Dict, Optional, Tuple
+from typing import Any, Dict, List, Optional, Tuple
 
 import torch
 from torch import nn

python/sglang/srt/models/yivl.py

@@ -1,6 +1,7 @@
 """Inference-only Yi-VL model."""
 
-from typing import Optional
+import os
+from typing import List, Optional
 
 import torch
 import torch.nn as nn
@@ -12,6 +13,7 @@
 
 from sglang.srt.models.llava import (
  LlavaLlamaForCausalLM,
+ clip_vision_embed_forward,
  monkey_path_clip_vision_embed_forward,
 )
 

python/sglang/srt/server.py

@@ -10,6 +10,9 @@
 import time
 from typing import List, Optional, Union
 
+# Fix a Python bug
+setattr(threading, "_register_atexit", lambda *args, **kwargs: None)
+
 import aiohttp
 import psutil
 import pydantic
@@ -55,9 +58,6 @@
 from sglang.srt.server_args import PortArgs, ServerArgs
 from sglang.srt.utils import enable_show_time_cost, handle_port_init
 
-# Fix a Python bug
-setattr(threading, "_register_atexit", lambda *args, **kwargs: None)
-
 asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
 
 API_KEY_HEADER_NAME = "X-API-Key"
@@ -619,7 +619,7 @@ def _wait_and_warmup():
  try:
  requests.get(url + "/get_model_info", timeout=5, headers=headers)
  break
- except requests.exceptions.RequestException:
+ except requests.exceptions.RequestException as e:
  pass
  else:
  if pipe_finish_writer is not None:

python/sglang/srt/utils.py

@@ -157,6 +157,7 @@ def get_exception_traceback():
 
 
 def get_int_token_logit_bias(tokenizer, vocab_size):
+ from transformers import LlamaTokenizer, LlamaTokenizerFast
 
  # a bug when model's vocab size > tokenizer.vocab_size
  vocab_size = tokenizer.vocab_size

test/lang/run_all.py

@@ -1,6 +1,7 @@
 import argparse
 import glob
 import multiprocessing
+import os
 import time
 import unittest
 

test/lang/test_anthropic_backend.py

@@ -1,3 +1,4 @@
+import json
 import unittest
 
 from sglang import Anthropic, set_default_backend

test/lang/test_srt_backend.py

@@ -2,6 +2,7 @@
 python3 -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000
 """
 
+import json
 import unittest
 
 import sglang as sgl
@@ -12,6 +13,8 @@
  test_few_shot_qa,
  test_mt_bench,
  test_parallel_decoding,
+ test_parallel_encoding,
+ test_react,
  test_regex,
  test_select,
  test_stream,

test/lang/test_tracing.py

@@ -110,7 +110,7 @@ def tip_suggestion(s):
  forks = s.fork(3)
  for i in range(3):
  forks[i] += f"Now, expand tip {i+1} into a paragraph:\n"
- forks[i] += sgl.gen("detailed_tip")
+ forks[i] += sgl.gen(f"detailed_tip")
 
  s += "Tip 1:" + forks[0]["detailed_tip"] + "\n"
  s += "Tip 2:" + forks[1]["detailed_tip"] + "\n"

test/srt/model/reference_hf.py

@@ -1,4 +1,5 @@
 import argparse
+import os
 
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer

test/srt/model/test_llama_extend.py

@@ -1,6 +1,10 @@
 import multiprocessing
 import os
+import time
 
+import numpy as np
+import torch
+import torch.distributed as dist
 import transformers
 
 from sglang.srt.managers.router.infer_batch import Batch, ForwardMode, Req

test/srt/model/test_llava_low_api.py

@@ -1,10 +1,13 @@
 import multiprocessing
+import time
 
 import numpy as np
 import torch
+import torch.distributed as dist
 
 from sglang.srt.hf_transformers_utils import get_processor
-from sglang.srt.managers.router.model_runner import ModelRunner
+from sglang.srt.managers.router.infer_batch import ForwardMode
+from sglang.srt.managers.router.model_runner import InputMetadata, ModelRunner
 from sglang.srt.model_config import ModelConfig
 from sglang.srt.utils import load_image
 

test/srt/test_httpserver_concurrent.py

@@ -9,8 +9,11 @@
 
 import argparse
 import asyncio
+import json
+import time
 
 import aiohttp
+import requests
 
 
 async def send_request(url, data, delay=0):

test/srt/test_httpserver_llava.py

@@ -10,6 +10,7 @@
 import argparse
 import asyncio
 import json
+import time
 
 import aiohttp
 import requests

test/srt/test_httpserver_reuse.py

@@ -6,6 +6,7 @@
 """
 
 import argparse
+import time
 
 import requests