[Refactor] [6/N] to simplify the vLLM openai chat_completion serving architecture (#32240)

Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>

vllm/tool_parsers/deepseekv32_tool_parser.py

@@ -8,8 +8,10 @@ from typing import Any
 
 import regex as re
 
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/deepseekv3_tool_parser.py

@@ -6,8 +6,10 @@ from collections.abc import Sequence
 import regex as re
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/ernie45_tool_parser.py

@@ -6,8 +6,10 @@ from collections.abc import Sequence
 
 import regex as re
 
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/functiongemma_tool_parser.py

@@ -7,8 +7,10 @@ from collections.abc import Sequence
 import regex as re
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/gigachat3_tool_parser.py

@@ -7,8 +7,10 @@ from collections.abc import Sequence
 import regex as re
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/glm4_moe_tool_parser.py

@@ -8,9 +8,11 @@ from typing import Any
 
 import regex as re
 
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
     ChatCompletionToolsParam,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/granite_20b_fc_tool_parser.py

@@ -10,8 +10,10 @@ import regex as re
 from partial_json_parser.core.options import Allow
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/granite_tool_parser.py

@@ -8,8 +8,10 @@ import partial_json_parser
 from partial_json_parser.core.options import Allow
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/hermes_tool_parser.py

@@ -9,8 +9,10 @@ import regex as re
 from partial_json_parser.core.options import Allow
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/hunyuan_a13b_tool_parser.py

@@ -8,8 +8,10 @@ from typing import Any
 
 import regex as re
 
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/internlm2_tool_parser.py

@@ -8,8 +8,10 @@ import partial_json_parser
 from partial_json_parser.core.options import Allow
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/jamba_tool_parser.py

@@ -9,8 +9,10 @@ import regex as re
 from partial_json_parser.core.options import Allow
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/kimi_k2_tool_parser.py

@@ -6,8 +6,10 @@ from collections.abc import Sequence
 
 import regex as re
 
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/llama4_pythonic_tool_parser.py

@@ -9,8 +9,10 @@ import regex as re
 from transformers import PreTrainedTokenizerBase
 
 import vllm.envs as envs
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/llama_tool_parser.py

@@ -11,8 +11,10 @@ from transformers import PreTrainedTokenizerBase
 
 import vllm.envs as envs
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/minimax_m2_tool_parser.py

@@ -8,8 +8,10 @@ from typing import Any
 
 import regex as re
 
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/minimax_tool_parser.py

@@ -8,8 +8,10 @@ from typing import Any
 import regex as re
 
 from vllm.entrypoints.chat_utils import make_tool_call_id
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/mistral_tool_parser.py

@@ -12,8 +12,10 @@ import ijson
 import regex as re
 from pydantic import Field
 
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/olmo3_tool_parser.py

@@ -9,8 +9,10 @@ import regex as re
 from transformers import PreTrainedTokenizerBase
 
 import vllm.envs as envs
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaFunctionCall,
     DeltaMessage,
     DeltaToolCall,

vllm/tool_parsers/openai_tool_parser.py

@@ -4,14 +4,16 @@ import json
 from collections.abc import Sequence
 from typing import TYPE_CHECKING
 
-from vllm.entrypoints.openai.parser.harmony_utils import parse_output_into_messages
-from vllm.entrypoints.openai.protocol import (
+from vllm.entrypoints.openai.chat_completion.protocol import (
     ChatCompletionRequest,
+)
+from vllm.entrypoints.openai.engine.protocol import (
     DeltaMessage,
     ExtractedToolCallInformation,
     FunctionCall,
     ToolCall,
 )
+from vllm.entrypoints.openai.parser.harmony_utils import parse_output_into_messages
 from vllm.logger import init_logger
 from vllm.tool_parsers.abstract_tool_parser import (
     ToolParser,