sglang 0.5.1.post1__py3-none-any.whl → 0.5.1.post3__py3-none-any.whl

sglang/srt/entrypoints/openai/protocol.py

@@ -35,6 +35,8 @@ from pydantic import (
 )
 from typing_extensions import Literal
 
+DEFAULT_MODEL_NAME = "default"
+
 
 class ModelCard(BaseModel):
     """Model cards."""
@@ -108,6 +110,23 @@ class JsonSchemaResponseFormat(BaseModel):
     strict: Optional[bool] = False
 
 
+class ResponseFormat(BaseModel):
+    type: Literal["text", "json_object", "json_schema"]
+    json_schema: Optional[JsonSchemaResponseFormat] = None
+
+
+class StructuresResponseFormat(BaseModel):
+    begin: str
+    schema_: Optional[Dict[str, object]] = Field(alias="schema", default=None)
+    end: str
+
+
+class StructuralTagResponseFormat(BaseModel):
+    type: Literal["structural_tag"]
+    structures: List[StructuresResponseFormat]
+    triggers: List[str]
+
+
 class FileRequest(BaseModel):
     # https://platform.openai.com/docs/api-reference/files/create
     file: bytes  # The File object (not file name) to be uploaded
@@ -166,7 +185,7 @@ class BatchResponse(BaseModel):
 class CompletionRequest(BaseModel):
     # Ordered by official OpenAI API documentation
     # https://platform.openai.com/docs/api-reference/completions/create
-    model: str
+    model: str = DEFAULT_MODEL_NAME
     prompt: Union[List[int], List[List[int]], str, List[str]]
     best_of: Optional[int] = None
     echo: bool = False
@@ -200,6 +219,7 @@ class CompletionRequest(BaseModel):
     skip_special_tokens: bool = True
     lora_path: Optional[Union[List[Optional[str]], Optional[str]]] = None
     session_params: Optional[Dict] = None
+    response_format: Optional[Union[ResponseFormat, StructuralTagResponseFormat]] = None
 
     # For PD disaggregation
     bootstrap_host: Optional[Union[List[str], str]] = None
@@ -327,7 +347,7 @@ class ToolCall(BaseModel):
 
 
 class ChatCompletionMessageGenericParam(BaseModel):
-    role: Literal["system", "assistant", "tool"]
+    role: Literal["system", "assistant", "tool", "function"]
     content: Union[str, List[ChatCompletionMessageContentTextPart], None] = Field(
         default=None
     )
@@ -341,9 +361,9 @@ class ChatCompletionMessageGenericParam(BaseModel):
     def _normalize_role(cls, v):
         if isinstance(v, str):
             v_lower = v.lower()
-            if v_lower not in {"system", "assistant", "tool"}:
+            if v_lower not in {"system", "assistant", "tool", "function"}:
                 raise ValueError(
-                    "'role' must be one of 'system', 'assistant', or 'tool' (case-insensitive)."
+                    "'role' must be one of 'system', 'assistant', 'tool', or 'function' (case-insensitive)."
                 )
             return v_lower
         raise ValueError("'role' must be a string")
@@ -359,23 +379,6 @@ ChatCompletionMessageParam = Union[
 ]
 
 
-class ResponseFormat(BaseModel):
-    type: Literal["text", "json_object", "json_schema"]
-    json_schema: Optional[JsonSchemaResponseFormat] = None
-
-
-class StructuresResponseFormat(BaseModel):
-    begin: str
-    schema_: Optional[Dict[str, object]] = Field(alias="schema", default=None)
-    end: str
-
-
-class StructuralTagResponseFormat(BaseModel):
-    type: Literal["structural_tag"]
-    structures: List[StructuresResponseFormat]
-    triggers: List[str]
-
-
 class Function(BaseModel):
     """Function descriptions."""
 
@@ -409,7 +412,7 @@ class ChatCompletionRequest(BaseModel):
     # Ordered by official OpenAI API documentation
     # https://platform.openai.com/docs/api-reference/chat/create
     messages: List[ChatCompletionMessageParam]
-    model: str
+    model: str = DEFAULT_MODEL_NAME
     frequency_penalty: float = 0.0
     logit_bias: Optional[Dict[str, float]] = None
     logprobs: bool = False
@@ -571,7 +574,7 @@ class EmbeddingRequest(BaseModel):
     # Ordered by official OpenAI API documentation
     # https://platform.openai.com/docs/api-reference/embeddings/create
     input: EmbeddingInput
-    model: str
+    model: str = DEFAULT_MODEL_NAME
     encoding_format: str = "float"
     dimensions: Optional[int] = None
     user: Optional[str] = None
@@ -605,7 +608,7 @@ class ScoringRequest(BaseModel):
     )
     apply_softmax: bool = False
     item_first: bool = False
-    model: str
+    model: str = DEFAULT_MODEL_NAME
 
 
 class ScoringResponse(BaseModel):

sglang/srt/entrypoints/openai/serving_chat.py

@@ -148,6 +148,16 @@ class OpenAIServingChat(OpenAIServingBase):
         self, request: ChatCompletionRequest, is_multimodal: bool
     ) -> MessageProcessingResult:
         """Process chat messages and apply chat template"""
+        is_gpt_oss = (
+            hasattr(self.tokenizer_manager.model_config, "hf_config")
+            and hasattr(self.tokenizer_manager.model_config.hf_config, "model_type")
+            and self.tokenizer_manager.model_config.hf_config.model_type == "gpt_oss"
+        )
+
+        # GptOss model needs to keep special tokens for harmony parsing
+        if is_gpt_oss:
+            request.skip_special_tokens = False
+
         tool_call_constraint = None
 
         # Apply chat template and its stop strings
@@ -207,6 +217,25 @@ class OpenAIServingChat(OpenAIServingBase):
                 audio_data,
                 modalities,
             )
+
+            # per the Transformers docs & maintainers, tool call arguments in
+            # assistant-role messages with tool_calls need to be dicts not JSON str -
+            # this is how tool-use chat templates will expect them moving forwards
+            # so, for messages that have tool_calls, parse the string (which we get
+            # from openAI format) to dict
+            if (
+                processed_msg["role"] == "assistant"
+                and "tool_calls" in processed_msg
+                and isinstance(processed_msg["tool_calls"], list)
+            ):
+                for item in processed_msg["tool_calls"]:
+                    if "arguments" in item["function"] and isinstance(
+                        item["function"]["arguments"], str
+                    ):
+                        item["function"]["arguments"] = json.loads(
+                            item["function"]["arguments"]
+                        )
+
             openai_compatible_messages.append(processed_msg)
 
         # Handle assistant prefix for continue_final_message
@@ -806,15 +835,23 @@ class OpenAIServingChat(OpenAIServingBase):
                 finish_reason["matched"] = None
             try:
                 text, call_info_list = parser.parse_non_stream(text)
-                tool_calls = [
-                    ToolCall(
-                        id=f"call_{uuid.uuid4().hex[:24]}",
-                        function=FunctionResponse(
-                            name=call_info.name, arguments=call_info.parameters
-                        ),
+                tool_calls = []
+                for call_info in call_info_list:
+                    # For Kimi-K2, align tool_call_id with the model format: functions.{name}:{index}
+                    if tool_call_parser == "kimi_k2" and call_info.name is not None:
+                        tool_id = f"functions.{call_info.name}:{call_info.tool_index}"
+                    else:
+                        tool_id = f"call_{uuid.uuid4().hex[:24]}"
+
+                    tool_calls.append(
+                        ToolCall(
+                            id=tool_id,
+                            index=getattr(call_info, "tool_index", None),
+                            function=FunctionResponse(
+                                name=call_info.name, arguments=call_info.parameters
+                            ),
+                        )
                     )
-                    for call_info in call_info_list
-                ]
                 return tool_calls, text, finish_reason
             except Exception as e:
                 logger.error(f"Tool call parsing error: {e}")
@@ -925,7 +962,11 @@ class OpenAIServingChat(OpenAIServingBase):
             # Tool call ID should be generated only once per tool call
             if call_item.name:
                 # First chunk: include ID and function name
-                tool_call_id = f"call_{uuid.uuid4().hex[:24]}"
+                if self.tokenizer_manager.server_args.tool_call_parser == "kimi_k2":
+                    # Align with Kimi-K2 format: functions.{name}:{index}
+                    tool_call_id = f"functions.{call_item.name}:{call_item.tool_index}"
+                else:
+                    tool_call_id = f"call_{uuid.uuid4().hex[:24]}"
                 function_name = call_item.name
             else:
                 # Subsequent chunks: null ID and name for argument deltas

sglang/srt/entrypoints/openai/serving_completions.py

@@ -23,6 +23,7 @@ from sglang.srt.entrypoints.openai.utils import (
 from sglang.srt.managers.io_struct import GenerateReqInput
 from sglang.srt.managers.template_manager import TemplateManager
 from sglang.srt.managers.tokenizer_manager import TokenizerManager
+from sglang.utils import convert_json_schema_to_str
 
 logger = logging.getLogger(__name__)
 
@@ -125,6 +126,20 @@ class OpenAIServingCompletion(OpenAIServingBase):
             "logit_bias": request.logit_bias,
         }
 
+        # Handle response_format constraints
+        if request.response_format and request.response_format.type == "json_schema":
+            sampling_params["json_schema"] = convert_json_schema_to_str(
+                request.response_format.json_schema.schema_
+            )
+        elif request.response_format and request.response_format.type == "json_object":
+            sampling_params["json_schema"] = '{"type": "object"}'
+        elif (
+            request.response_format and request.response_format.type == "structural_tag"
+        ):
+            sampling_params["structural_tag"] = convert_json_schema_to_str(
+                request.response_format.model_dump(by_alias=True)
+            )
+
         return sampling_params
 
     async def _handle_streaming_request(

sglang/srt/entrypoints/tool.py

@@ -4,6 +4,8 @@ import os
 from abc import ABC, abstractmethod
 from typing import TYPE_CHECKING, Any
 
+from sglang.srt.utils import print_info_once, print_warning_once
+
 if TYPE_CHECKING:
     # Avoid circular import.
     from sglang.srt.entrypoints.context import ConversationContext
@@ -25,7 +27,7 @@ class HarmonyBrowserTool(Tool):
         exa_api_key = os.getenv("EXA_API_KEY")
         if not exa_api_key:
             self.enabled = False
-            logger.warning_once("EXA_API_KEY is not set, browsing is disabled")
+            print_warning_once("EXA_API_KEY is not set, browsing is disabled")
             return
 
         try:
@@ -33,12 +35,12 @@ class HarmonyBrowserTool(Tool):
             from gpt_oss.tools.simple_browser.backend import ExaBackend
         except ImportError:
             self.enabled = False
-            logger.warning_once("gpt_oss is not installed, browsing is disabled")
+            print_warning_once("gpt_oss is not installed, browsing is disabled")
             return
 
         browser_backend = ExaBackend(source="web", api_key=exa_api_key)
         self.browser_tool = SimpleBrowserTool(backend=browser_backend)
-        logger.info_once("Browser tool initialized")
+        print_info_once("Browser tool initialized")
 
     async def get_result(self, context: "ConversationContext") -> Any:
         from sglang.srt.entrypoints.context import HarmonyContext
@@ -64,13 +66,11 @@ class HarmonyPythonTool(Tool):
             from gpt_oss.tools.python_docker.docker_tool import PythonTool
         except ImportError:
             self.enabled = False
-            logger.warning_once(
-                "gpt_oss is not installed, code interpreter is disabled"
-            )
+            print_warning_once("gpt_oss is not installed, code interpreter is disabled")
             return
 
         self.python_tool = PythonTool()
-        logger.info_once("Code interpreter tool initialized")
+        print_info_once("Code interpreter tool initialized")
 
     async def get_result(self, context: "ConversationContext") -> Any:
         from sglang.srt.entrypoints.context import HarmonyContext

sglang/srt/function_call/deepseekv31_detector.py

@@ -0,0 +1,222 @@
+import json
+import logging
+import re
+from typing import List
+
+from sglang.srt.entrypoints.openai.protocol import Tool
+from sglang.srt.function_call.base_format_detector import BaseFormatDetector
+from sglang.srt.function_call.core_types import (
+    StreamingParseResult,
+    StructureInfo,
+    ToolCallItem,
+    _GetInfoFunc,
+)
+from sglang.srt.function_call.ebnf_composer import EBNFComposer
+from sglang.srt.function_call.utils import _is_complete_json
+
+logger = logging.getLogger(__name__)
+
+
+class DeepSeekV31Detector(BaseFormatDetector):
+    """
+    Detector for DeepSeek V3 model function call format.
+
+    The DeepSeek V3 format uses special Unicode tokens to delimit function calls
+    with JSON code blocks for arguments.
+
+    Format Structure:
+    ```
+    <｜tool▁calls▁begin｜><｜tool▁call▁begin｜>{function_name}<｜tool▁sep｜>{json_arguments}<｜tool▁calls▁end｜><｜end▁of▁sentence｜>
+    ```
+    Examples:
+    ```
+    <｜tool▁calls▁begin｜><｜tool▁call▁begin｜>get_current_weather<｜tool▁sep｜>{"location": "Tokyo"}<｜tool▁call▁end｜><｜tool▁call▁begin｜>get_current_weather<｜tool▁sep｜>{"location": "Paris"}<｜tool▁call▁end｜><｜tool▁calls▁end｜><｜end▁of▁sentence｜>
+    ```
+
+    Key Components:
+    - Tool Calls Section: Wrapped between `<｜tool▁calls▁begin｜>` and `<｜tool▁calls▁end｜>`
+    - Individual Tool Call: Wrapped between `<｜tool▁call▁begin｜>` and `<｜tool▁call▁end｜>`
+    - Function Declaration: `<｜tool▁call▁begin｜>{function_name}<｜tool▁sep｜>`
+    - Arguments: JSON code block between `<｜tool▁sep｜>` and `<｜tool▁call▁end｜>`
+    - Supports multiple tool calls
+
+    Reference: https://www.modelscope.cn/models/deepseek-ai/DeepSeek-V3.1
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.bot_token = "<｜tool▁calls▁begin｜>"
+        self.eot_token = "<｜tool▁calls▁end｜>"
+        self.func_call_regex = r"<｜tool▁call▁begin｜>.*?<｜tool▁call▁end｜>"
+        self.func_detail_regex = (
+            r"<｜tool▁call▁begin｜>(.*)<｜tool▁sep｜>(.*)<｜tool▁call▁end｜>"
+        )
+        self._last_arguments = ""
+        self.current_tool_id = -1
+
+    def has_tool_call(self, text: str) -> bool:
+        """Check if the text contains a deepseek format tool call."""
+        return self.bot_token in text
+
+    def detect_and_parse(self, text: str, tools: List[Tool]) -> StreamingParseResult:
+        """
+        One-time parsing: Detects and parses tool calls in the provided text.
+
+        :param text: The complete text to parse.
+        :param tools: List of available tools.
+        :return: ParseResult indicating success or failure, consumed text, leftover text, and parsed calls.
+        """
+        idx = text.find(self.bot_token)
+        normal_text = text[:idx].strip() if idx != -1 else text
+        if self.bot_token not in text:
+            return StreamingParseResult(normal_text=normal_text, calls=[])
+        match_result_list = re.findall(self.func_call_regex, text, re.DOTALL)
+        calls = []
+        try:
+            for match_result in match_result_list:
+                # Get function name
+                func_detail = re.search(self.func_detail_regex, match_result, re.DOTALL)
+                func_name = func_detail.group(1)
+                func_args = func_detail.group(2)
+                func_args = json.loads(func_args)
+                # construct match_result for parse_base_json
+                match_result = {"name": func_name, "parameters": func_args}
+                calls.extend(self.parse_base_json(match_result, tools))
+            return StreamingParseResult(normal_text=normal_text, calls=calls)
+        except Exception as e:
+            logger.error(f"Error in detect_and_parse: {e}")
+            # return the normal text if parsing fails
+            return StreamingParseResult(normal_text=text)
+
+    def parse_streaming_increment(
+        self, new_text: str, tools: List[Tool]
+    ) -> StreamingParseResult:
+        """
+        Streaming incremental parsing tool calls for DeepSeekV3 format.
+        """
+        self._buffer += new_text
+        current_text = self._buffer
+
+        # Check if we have a tool call (either the start token or individual tool call)
+        has_tool_call = (
+            self.bot_token in current_text or "<｜tool▁call▁begin｜>" in current_text
+        )
+
+        if not has_tool_call:
+            self._buffer = ""
+            for e_token in [self.eot_token, "<｜tool▁call▁end｜>"]:
+                if e_token in new_text:
+                    new_text = new_text.replace(e_token, "")
+            return StreamingParseResult(normal_text=new_text)
+
+        if not hasattr(self, "_tool_indices"):
+            self._tool_indices = self._get_tool_indices(tools)
+
+        calls: list[ToolCallItem] = []
+        try:
+            partial_match = re.search(
+                pattern=r"<｜tool▁call▁begin｜>(.*)<｜tool▁sep｜>(.*)<｜tool▁call▁end｜>",
+                string=current_text,
+                flags=re.DOTALL,
+            )
+            if partial_match:
+                func_name = partial_match.group(1).strip()
+                func_args_raw = partial_match.group(2).strip()
+
+                # Initialize state if this is the first tool call
+                if self.current_tool_id == -1:
+                    self.current_tool_id = 0
+                    self.prev_tool_call_arr = []
+                    self.streamed_args_for_tool = [""]
+
+                # Ensure we have enough entries in our tracking arrays
+                while len(self.prev_tool_call_arr) <= self.current_tool_id:
+                    self.prev_tool_call_arr.append({})
+                while len(self.streamed_args_for_tool) <= self.current_tool_id:
+                    self.streamed_args_for_tool.append("")
+
+                if not self.current_tool_name_sent:
+                    calls.append(
+                        ToolCallItem(
+                            tool_index=self.current_tool_id,
+                            name=func_name,
+                            parameters="",
+                        )
+                    )
+                    self.current_tool_name_sent = True
+                    # Store the tool call info for serving layer completions endpoint
+                    self.prev_tool_call_arr[self.current_tool_id] = {
+                        "name": func_name,
+                        "arguments": {},
+                    }
+                else:
+                    argument_diff = (
+                        func_args_raw[len(self._last_arguments) :]
+                        if func_args_raw.startswith(self._last_arguments)
+                        else func_args_raw
+                    )
+
+                    if argument_diff:
+                        calls.append(
+                            ToolCallItem(
+                                tool_index=self.current_tool_id,
+                                name=None,
+                                parameters=argument_diff,
+                            )
+                        )
+                        self._last_arguments += argument_diff
+                        self.streamed_args_for_tool[
+                            self.current_tool_id
+                        ] += argument_diff
+
+                    if _is_complete_json(func_args_raw):
+                        # Update the stored arguments
+                        try:
+                            parsed_args = json.loads(func_args_raw)
+                            self.prev_tool_call_arr[self.current_tool_id][
+                                "arguments"
+                            ] = parsed_args
+                        except json.JSONDecodeError:
+                            pass
+
+                        # Find the end of the current tool call and remove only that part from buffer
+                        tool_call_end_pattern = (
+                            r"<｜tool▁call▁begin｜>.*?<｜tool▁call▁end｜>"
+                        )
+                        match = re.search(
+                            tool_call_end_pattern, current_text, re.DOTALL
+                        )
+                        if match:
+                            # Remove the completed tool call from buffer, keep any remaining content
+                            self._buffer = current_text[match.end() :]
+                        else:
+                            self._buffer = ""
+
+                        result = StreamingParseResult(normal_text="", calls=calls)
+                        self.current_tool_id += 1
+                        self._last_arguments = ""
+                        self.current_tool_name_sent = False
+                        return result
+
+            return StreamingParseResult(normal_text="", calls=calls)
+
+        except Exception as e:
+            logger.error(f"Error in parse_streaming_increment: {e}")
+            return StreamingParseResult(normal_text=current_text)
+
+    def structure_info(self) -> _GetInfoFunc:
+        return lambda name: StructureInfo(
+            begin="<｜tool▁call▁begin｜>" + name + "<｜tool▁sep｜>",
+            end="<｜tool▁call▁end｜>",
+            trigger="<｜tool▁call▁begin｜>" + name + "<｜tool▁sep｜>",
+        )
+
+    def build_ebnf(self, tools: List[Tool]):
+        return EBNFComposer.build_ebnf(
+            tools,
+            sequence_start_token=self.bot_token,
+            sequence_end_token=self.eot_token,
+            tool_call_separator="",
+            call_rule_fmt='"<｜tool▁call▁begin｜>{name}<｜tool▁sep｜>{arguments_rule}<｜tool▁call▁end｜>"',
+            function_format="json",
+        )

sglang/srt/function_call/function_call_parser.py

@@ -10,6 +10,7 @@ from sglang.srt.entrypoints.openai.protocol import (
 from sglang.srt.function_call.base_format_detector import BaseFormatDetector
 from sglang.srt.function_call.core_types import ToolCallItem
 from sglang.srt.function_call.deepseekv3_detector import DeepSeekV3Detector
+from sglang.srt.function_call.deepseekv31_detector import DeepSeekV31Detector
 from sglang.srt.function_call.glm4_moe_detector import Glm4MoeDetector
 from sglang.srt.function_call.gpt_oss_detector import GptOssDetector
 from sglang.srt.function_call.kimik2_detector import KimiK2Detector
@@ -37,6 +38,7 @@ class FunctionCallParser:
         "qwen25": Qwen25Detector,
         "mistral": MistralDetector,
         "deepseekv3": DeepSeekV3Detector,
+        "deepseekv31": DeepSeekV31Detector,
         "pythonic": PythonicDetector,
         "kimi_k2": KimiK2Detector,
         "qwen3_coder": Qwen3CoderDetector,