Add Spark3 Model (#35963)

Co-authored-by: Yaowj <yaowj@MacBook-Air.local>
Co-authored-by: ybyang <10629930+whybeyoung@users.noreply.github.com>

python/sglang/srt/configs/__init__.py

@@ -57,6 +57,7 @@ from sglang.srt.configs.qwen3_5 import (
 )
 from sglang.srt.configs.qwen3_asr import Qwen3ASRConfig
 from sglang.srt.configs.qwen3_next import Qwen3NextConfig
+from sglang.srt.configs.spark3 import Spark3Config
 from sglang.srt.configs.step3_vl import (
     Step3TextConfig,
     Step3VisionEncoderConfig,
@@ -117,6 +118,7 @@ __all__ = [
     "MiniCPMHybridConfig",
     "Step3p5Config",
     "MiniMaxM3VLConfig",
+    "Spark3Config",
     "Step3p7Config",
     "Qwen3ASRConfig",
     "InklingAudioConfig",

python/sglang/srt/configs/spark3.py

@@ -0,0 +1,63 @@
+from typing import Any, Optional
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class Spark3Config(PretrainedConfig):
+    model_type = "spark3"
+    architectures = ["Spark3ForCausalLM"]
+
+    def __init__(
+        self,
+        hidden_size: int = 2048,
+        intermediate_size: int = 6656,
+        num_attention_heads: int = 8,
+        num_key_value_heads: int = 2,
+        num_hidden_layers: int = 28,
+        head_dim: int = 256,
+        headwise_attn_output_gate: bool = True,
+        sliding_window: int = 512,
+        vocab_size: int = 133120,
+        rms_norm_eps: float = 1e-6,
+        max_position_embeddings: int = 8192,
+        rope_parameters: Optional[dict[str, Any]] = None,
+        layer_types: list[str] = None,
+        tie_word_embeddings: Optional[bool] = None,
+        **kwargs,
+    ) -> None:
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_attention_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.num_hidden_layers = num_hidden_layers
+        self.head_dim = head_dim
+        self.headwise_attn_output_gate = headwise_attn_output_gate
+        self.sliding_window = sliding_window
+        self.vocab_size = vocab_size
+        self.rms_norm_eps = rms_norm_eps
+        self.max_position_embeddings = max_position_embeddings
+
+        if layer_types is not None:
+            layer_types = layer_types[: self.num_hidden_layers]
+        else:
+            layer_types = [
+                "sliding_attention" if bool((i + 1) % 4) else "full_attention"
+                for i in range(self.num_hidden_layers)
+            ]
+        self.layer_types = layer_types
+
+        if rope_parameters is not None:
+            self.rope_parameters = rope_parameters
+        else:
+            self.rope_parameters = {
+                "full_attention": {
+                    "rope_theta": 5000000,
+                    "partial_rotary_factor": 0.25,
+                },
+                "sliding_attention": {
+                    "rope_theta": 10000,
+                    "partial_rotary_factor": 1.0,
+                },
+            }
+
+        super().__init__(**kwargs, tie_word_embeddings=tie_word_embeddings)

python/sglang/srt/function_call/function_call_parser.py

@@ -43,6 +43,7 @@ from sglang.srt.function_call.poolside_v1_detector import PoolsideV1Detector
 from sglang.srt.function_call.pythonic_detector import PythonicDetector
 from sglang.srt.function_call.qwen3_coder_detector import Qwen3CoderDetector
 from sglang.srt.function_call.qwen25_detector import Qwen25Detector
+from sglang.srt.function_call.spark3_detector import Spark3Detector
 from sglang.srt.function_call.step3_detector import Step3Detector
 from sglang.srt.function_call.trinity_detector import TrinityDetector
 from sglang.srt.function_call.utils import (
@@ -87,6 +88,7 @@ class FunctionCallParser:
         "qwen": Qwen25Detector,
         "qwen25": Qwen25Detector,
         "qwen3_coder": Qwen3CoderDetector,
+        "spark": Spark3Detector,
         "step3": Step3Detector,
         "step3p5": Qwen3CoderDetector,
         "minimax-m2": MinimaxM2Detector,

python/sglang/srt/function_call/spark3_detector.py

@@ -0,0 +1,263 @@
+import json
+import re
+from dataclasses import dataclass
+from typing import Any
+
+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,
+    ToolCallItem,
+    _GetInfoFunc,
+)
+
+TOOL_CALL_BEGIN = "<tool_call>"
+TOOL_CALL_END = "</tool_call>"
+ARG_KEY_BEGIN = "<arg_key>"
+ARG_KEY_END = "</arg_key>"
+ARG_VALUE_BEGIN = "<arg_value>"
+ARG_VALUE_END = "</arg_value>"
+
+ARG_PAIR_PATTERN = re.compile(
+    rf"{re.escape(ARG_KEY_BEGIN)}(.*?){re.escape(ARG_KEY_END)}"
+    rf"{re.escape(ARG_VALUE_BEGIN)}(.*?){re.escape(ARG_VALUE_END)}",
+    re.DOTALL,
+)
+
+
+@dataclass(frozen=True)
+class _Spark3ToolCall:
+    name: str
+    arguments: dict[str, Any]
+
+    def arguments_json(self) -> str:
+        return json.dumps(
+            self.arguments,
+            ensure_ascii=False,
+            separators=(",", ":"),
+        )
+
+
+def _get_param_type(tools: list[Tool], function_name: str, param_name: str) -> str:
+    """Return a parameter's declared JSON Schema type, or ``string``."""
+    for tool in tools:
+        function = getattr(tool, "function", None)
+        if function is None or function.name != function_name:
+            continue
+        parameters = getattr(function, "parameters", None)
+        if not isinstance(parameters, dict):
+            continue
+        properties = parameters.get("properties")
+        if not isinstance(properties, dict):
+            continue
+        definition = properties.get(param_name)
+        if isinstance(definition, dict) and isinstance(definition.get("type"), str):
+            return definition["type"]
+    return "string"
+
+
+def _convert_value(value: str, param_type: str) -> Any:
+    """Convert Spark3 XML text according to the model's tool protocol."""
+    if value.lower() == "null":
+        return None
+
+    normalized_type = param_type.lower()
+    try:
+        if normalized_type in {"string", "str", "text"}:
+            return value
+        if normalized_type in {"integer", "int"}:
+            return int(value)
+        if normalized_type in {"number", "float"}:
+            number = float(value)
+            return int(number) if number.is_integer() else number
+        if normalized_type in {"boolean", "bool"}:
+            normalized_value = value.strip().lower()
+            if normalized_value not in {"true", "1", "false", "0"}:
+                raise ValueError(f"invalid boolean: {value}")
+            return normalized_value in {"true", "1"}
+        return json.loads(value)
+    except (TypeError, ValueError, json.JSONDecodeError):
+        try:
+            return json.loads(value)
+        except (TypeError, ValueError, json.JSONDecodeError):
+            return value
+
+
+def _parse_tool_call_xml(tool_xml: str, tools: list[Tool]) -> _Spark3ToolCall | None:
+    if not tool_xml.startswith(TOOL_CALL_BEGIN) or not tool_xml.endswith(TOOL_CALL_END):
+        return None
+
+    body = tool_xml[len(TOOL_CALL_BEGIN) : -len(TOOL_CALL_END)]
+    first_arg = body.find(ARG_KEY_BEGIN)
+    function_name = (body if first_arg < 0 else body[:first_arg]).strip()
+    if not function_name:
+        return None
+
+    arguments: dict[str, Any] = {}
+    for match in ARG_PAIR_PATTERN.finditer(body):
+        key, raw_value = match.group(1), match.group(2)
+        if not key:
+            continue
+        arguments[key] = _convert_value(
+            raw_value,
+            _get_param_type(tools, function_name, key),
+        )
+    return _Spark3ToolCall(name=function_name, arguments=arguments)
+
+
+def _partial_marker_suffix_length(text: str, marker: str) -> int:
+    """Length of the suffix that may become ``marker`` in the next chunk."""
+    for size in range(min(len(text), len(marker) - 1), 0, -1):
+        if text.endswith(marker[:size]):
+            return size
+    return 0
+
+
+class Spark3Detector(BaseFormatDetector):
+    """Detector for Spark3's XML-KV tool-call format.
+
+    Wire format::
+
+        <tool_call>function_name
+        <arg_key>key</arg_key><arg_value>value</arg_value>
+        </tool_call>
+
+    Values are converted with the parameter's JSON Schema type. A complete
+    block is emitted atomically in streaming mode so XML fragments are never
+    exposed as JSON argument deltas.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.bot_token = TOOL_CALL_BEGIN
+        self.eot_token = TOOL_CALL_END
+
+    def has_tool_call(self, text: str) -> bool:
+        return TOOL_CALL_BEGIN in text
+
+    def _build_item(
+        self,
+        parsed: _Spark3ToolCall,
+        tools: list[Tool],
+        tool_index: int,
+    ) -> ToolCallItem | None:
+        validated = self.parse_base_json(
+            {"name": parsed.name, "arguments": parsed.arguments}, tools
+        )
+        if not validated:
+            return None
+        return ToolCallItem(
+            tool_index=tool_index,
+            name=parsed.name,
+            parameters=parsed.arguments_json(),
+        )
+
+    def _record_streamed_item(
+        self, parsed: _Spark3ToolCall, item: ToolCallItem
+    ) -> None:
+        self.prev_tool_call_arr.append(
+            {"name": parsed.name, "arguments": parsed.arguments}
+        )
+        self.streamed_args_for_tool.append(item.parameters)
+
+    def detect_and_parse(self, text: str, tools: list[Tool]) -> StreamingParseResult:
+        calls: list[ToolCallItem] = []
+        normal_parts: list[str] = []
+        cursor = 0
+
+        while cursor < len(text):
+            start = text.find(TOOL_CALL_BEGIN, cursor)
+            if start < 0:
+                normal_parts.append(text[cursor:])
+                break
+
+            normal_parts.append(text[cursor:start])
+            end = text.find(TOOL_CALL_END, start + len(TOOL_CALL_BEGIN))
+            if end < 0:
+                normal_parts.append(text[start:])
+                break
+
+            end += len(TOOL_CALL_END)
+            raw_tool_call = text[start:end]
+            parsed = _parse_tool_call_xml(raw_tool_call, tools)
+            if parsed is None:
+                normal_parts.append(raw_tool_call)
+            else:
+                item = self._build_item(parsed, tools, len(calls))
+                if item is not None:
+                    calls.append(item)
+            cursor = end
+
+        return StreamingParseResult(
+            normal_text="".join(normal_parts),
+            calls=calls,
+        )
+
+    def parse_streaming_increment(
+        self, new_text: str, tools: list[Tool]
+    ) -> StreamingParseResult:
+        self._buffer += new_text
+        calls: list[ToolCallItem] = []
+        normal_parts: list[str] = []
+
+        while self._buffer:
+            start = self._buffer.find(TOOL_CALL_BEGIN)
+            if start < 0:
+                keep = _partial_marker_suffix_length(self._buffer, TOOL_CALL_BEGIN)
+                if keep:
+                    normal_parts.append(self._buffer[:-keep])
+                    self._buffer = self._buffer[-keep:]
+                else:
+                    normal_parts.append(self._buffer)
+                    self._buffer = ""
+                break
+
+            if start > 0:
+                normal_parts.append(self._buffer[:start])
+                self._buffer = self._buffer[start:]
+
+            end = self._buffer.find(TOOL_CALL_END, len(TOOL_CALL_BEGIN))
+            if end < 0:
+                break
+
+            end += len(TOOL_CALL_END)
+            raw_tool_call = self._buffer[:end]
+            self._buffer = self._buffer[end:]
+            parsed = _parse_tool_call_xml(raw_tool_call, tools)
+            if parsed is None:
+                normal_parts.append(raw_tool_call)
+                continue
+
+            item = self._build_item(parsed, tools, self.current_tool_id + 1)
+            if item is not None:
+                self.current_tool_id += 1
+                self._record_streamed_item(parsed, item)
+                calls.append(item)
+
+        return StreamingParseResult(
+            normal_text="".join(normal_parts),
+            calls=calls,
+        )
+
+    def finish(self, tools: list[Tool]) -> StreamingParseResult:
+        del tools
+        pending = self._buffer
+        self._buffer = ""
+        if TOOL_CALL_BEGIN in pending:
+            # A complete opening marker means this is a truncated protocol
+            # block, not user-visible text. Partial marker prefixes are still
+            # released because the stream has ended and they cannot become a
+            # tool call anymore.
+            pending = pending[: pending.find(TOOL_CALL_BEGIN)]
+        return StreamingParseResult(normal_text=pending)
+
+    def supports_structural_tag(self) -> bool:
+        return False
+
+    def parses_required_natively(self) -> bool:
+        return True
+
+    def structure_info(self) -> _GetInfoFunc:
+        raise NotImplementedError(
+            "Spark3 XML arguments cannot be represented by legacy structural tags"
+        )

python/sglang/srt/models/spark3.py

@@ -0,0 +1,564 @@
+import logging
+from typing import Iterable, Optional, Tuple, Union
+
+import torch
+from torch import nn
+
+from sglang.srt.distributed import get_pp_group
+from sglang.srt.layers.activation import GeluAndMul
+from sglang.srt.layers.dp_attention import is_dp_attention_enabled
+from sglang.srt.layers.layernorm import RMSNorm
+from sglang.srt.layers.linear import (
+    ColumnParallelLinear,
+    MergedColumnParallelLinear,
+    QKVParallelLinear,
+    RowParallelLinear,
+)
+from sglang.srt.layers.logits_processor import LogitsProcessor
+from sglang.srt.layers.quantization.base_config import QuantizationConfig
+from sglang.srt.layers.radix_attention import RadixAttention
+from sglang.srt.layers.rotary_embedding import get_rope
+from sglang.srt.layers.utils import PPMissingLayer, get_layer_id
+from sglang.srt.layers.vocab_parallel_embedding import (
+    ParallelLMHead,
+    VocabParallelEmbedding,
+)
+from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
+from sglang.srt.model_loader.weight_utils import (
+    default_weight_loader,
+)
+from sglang.srt.platforms import current_platform
+from sglang.srt.runtime_context import get_parallel
+from sglang.srt.utils import add_prefix, make_layers
+
+Spark3Config = None
+
+logger = logging.getLogger(__name__)
+
+
+# Aligned with HF's implementation, using sliding window inclusive with the last token
+# SGLang assumes exclusive
+def _get_attention_sliding_window_size(config):
+    return config.sliding_window - 1
+
+
+class Spark3MLP(nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        intermediate_size: int,
+        quant_config: Optional[QuantizationConfig] = None,
+        prefix: str = "",
+        reduce_results: bool = True,
+    ) -> None:
+        super().__init__()
+        self.gate_up_proj = MergedColumnParallelLinear(
+            hidden_size,
+            [intermediate_size] * 2,
+            bias=False,
+            quant_config=quant_config,
+            prefix=add_prefix("gate_up_proj", prefix),
+        )
+        self.down_proj = RowParallelLinear(
+            intermediate_size,
+            hidden_size,
+            bias=False,
+            quant_config=quant_config,
+            prefix=add_prefix("down_proj", prefix),
+            reduce_results=reduce_results,
+        )
+        self.act_fn = GeluAndMul()
+
+    def forward(
+        self,
+        x,
+        forward_batch=None,
+    ):
+        gate_up, _ = self.gate_up_proj(x)
+        x = self.act_fn(gate_up)
+        x, _ = self.down_proj(x)
+        return x
+
+
+class Spark3Attention(nn.Module):
+    def __init__(
+        self,
+        hidden_size: int,
+        num_heads: int,
+        num_kv_heads: int,
+        head_dim: Optional[int] = None,
+        layer_id: int = 0,
+        rope_theta: float = 10000,
+        partial_rotary_factor: float = 1.0,
+        max_position_embeddings: int = 8192,
+        quant_config: Optional[QuantizationConfig] = None,
+        sliding_window: int = 512,
+        layer_type: str = "sliding_attention",
+        headwise_attn_output_gate: bool = True,
+        prefix: str = "",
+    ) -> None:
+        super().__init__()
+        self.hidden_size = hidden_size
+        self.total_num_heads = num_heads
+        attn_tp_rank = get_parallel().attn_tp_rank
+        attn_tp_size = get_parallel().attn_tp_size
+
+        assert self.total_num_heads % attn_tp_size == 0
+        self.num_heads = self.total_num_heads // attn_tp_size
+        self.total_num_kv_heads = num_kv_heads
+        if self.total_num_kv_heads >= attn_tp_size:
+            # Number of KV heads is greater than TP size, so we partition
+            # the KV heads across multiple tensor parallel GPUs.
+            assert self.total_num_kv_heads % attn_tp_size == 0
+        else:
+            # Number of KV heads is less than TP size, so we replicate
+            # the KV heads across multiple tensor parallel GPUs.
+            assert attn_tp_size % self.total_num_kv_heads == 0
+        self.num_kv_heads = max(1, self.total_num_kv_heads // attn_tp_size)
+        if head_dim is not None:
+            self.head_dim = head_dim
+        else:
+            self.head_dim = hidden_size // self.total_num_heads
+        self.q_size = self.num_heads * self.head_dim
+        self.kv_size = self.num_kv_heads * self.head_dim
+        self.scaling = self.head_dim**-0.5
+        self.rope_theta = rope_theta
+        self.max_position_embeddings = max_position_embeddings
+        self.partial_rotary_factor = partial_rotary_factor
+        self.headwise_attn_output_gate = headwise_attn_output_gate
+
+        self.q_k_v_proj = QKVParallelLinear(
+            hidden_size,
+            self.head_dim,
+            self.total_num_heads,
+            self.total_num_kv_heads,
+            bias=False,
+            quant_config=quant_config,
+            tp_rank=attn_tp_rank,
+            tp_size=attn_tp_size,
+            prefix=add_prefix("q_k_v_proj", prefix),
+        )
+        if self.headwise_attn_output_gate:
+            self.g_proj = ColumnParallelLinear(
+                hidden_size,
+                self.total_num_heads,
+                bias=False,
+                quant_config=None,  # g_proj keeps bf16.
+                tp_rank=attn_tp_rank,
+                tp_size=attn_tp_size,
+                prefix=add_prefix("g_proj", prefix),
+            )
+
+        self.out_proj = RowParallelLinear(
+            self.total_num_heads * self.head_dim,
+            hidden_size,
+            bias=False,
+            quant_config=quant_config,
+            tp_rank=attn_tp_rank,
+            tp_size=attn_tp_size,
+            prefix=add_prefix("out_proj", prefix),
+        )
+
+        self.rotary_emb = get_rope(
+            self.head_dim,
+            rotary_dim=self.head_dim,
+            max_position=max_position_embeddings,
+            base=rope_theta,
+            rope_scaling=None,
+            partial_rotary_factor=partial_rotary_factor,
+            is_neox_style=True,
+        )
+        if layer_type not in ("sliding_attention", "full_attention"):
+            raise ValueError(f"Unsupported Spark3 layer_type: {layer_type}")
+        sliding_window_size = (
+            sliding_window if layer_type == "sliding_attention" else -1
+        )
+        self.attn = RadixAttention(
+            self.num_heads,
+            self.head_dim,
+            self.scaling,
+            num_kv_heads=self.num_kv_heads,
+            sliding_window_size=sliding_window_size,
+            layer_id=layer_id,
+            quant_config=quant_config,
+            prefix=add_prefix("attn", prefix),
+        )
+
+    def forward(
+        self,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+        forward_batch: ForwardBatch,
+    ) -> torch.Tensor:
+        qkv, _ = self.q_k_v_proj(hidden_states)
+        q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
+        q, k = self.rotary_emb(positions, q, k)
+        attn_output = self.attn(q, k, v, forward_batch)
+
+        if self.headwise_attn_output_gate:
+            g, _ = self.g_proj(hidden_states)
+            g = torch.sigmoid(g.float()).to(attn_output.dtype)
+            gate_output = attn_output.view(
+                attn_output.shape[0],
+                self.num_heads,
+                self.head_dim,
+            ) * g.unsqueeze(-1)
+            attn_output = gate_output.view(*attn_output.shape)
+
+        output, _ = self.out_proj(attn_output)
+        return output
+
+
+class Spark3DecoderLayer(nn.Module):
+    """A single transformer layer.
+
+    Transformer layer takes input with size [s, b, h] and returns an
+    output of the same size.
+    """
+
+    def __init__(
+        self,
+        config: Spark3Config,
+        layer_id: int = 0,
+        quant_config: Optional[QuantizationConfig] = None,
+        prefix: str = "",
+    ) -> None:
+        super().__init__()
+
+        max_position_embeddings = getattr(config, "max_position_embeddings", 8192)
+        head_dim = getattr(config, "head_dim", None)
+        layer_type = config.layer_types[layer_id]
+        rope_params = config.rope_parameters[layer_type]
+        rope_theta = rope_params.get("rope_theta")
+        partial_rotary_factor = rope_params.get("partial_rotary_factor")
+        if partial_rotary_factor is None:
+            partial_rotary_factor = 1.0
+
+        self.self_attn = Spark3Attention(
+            hidden_size=config.hidden_size,
+            num_heads=config.num_attention_heads,
+            num_kv_heads=config.num_key_value_heads,
+            head_dim=head_dim,
+            layer_id=layer_id,
+            rope_theta=rope_theta,
+            partial_rotary_factor=partial_rotary_factor,
+            max_position_embeddings=max_position_embeddings,
+            quant_config=quant_config,
+            sliding_window=_get_attention_sliding_window_size(config),
+            layer_type=layer_type,
+            headwise_attn_output_gate=getattr(
+                config, "headwise_attn_output_gate", True
+            ),
+            prefix=add_prefix("self_attn", prefix),
+        )
+
+        # MLP
+        self.mlp = Spark3MLP(
+            config.hidden_size,
+            intermediate_size=config.intermediate_size,
+            quant_config=quant_config,
+            prefix=add_prefix("mlp", prefix),
+        )
+
+        self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = RMSNorm(
+            config.hidden_size, eps=config.rms_norm_eps
+        )
+
+    def forward(
+        self,
+        positions: torch.Tensor,
+        hidden_states: torch.Tensor,
+        forward_batch: ForwardBatch,
+        residual: Optional[torch.Tensor],
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        # Self Attention
+        if residual is None:
+            residual = hidden_states
+            hidden_states = self.input_layernorm(hidden_states)
+        else:
+            hidden_states, residual = self.input_layernorm(hidden_states, residual)
+
+        hidden_states = self.self_attn(
+            positions=positions,
+            hidden_states=hidden_states,
+            forward_batch=forward_batch,
+        )
+        hidden_states, residual = self.post_attention_layernorm(hidden_states, residual)
+        hidden_states = self.mlp(hidden_states)
+
+        return hidden_states, residual
+
+
+class Spark3Model(nn.Module):
+    def __init__(
+        self,
+        config: Spark3Config,
+        quant_config: Optional[QuantizationConfig] = None,
+        prefix: str = "",
+    ) -> None:
+        super().__init__()
+        self.config = config
+        self.vocab_size = config.vocab_size
+        self.pp_group = get_pp_group()
+
+        if self.pp_group.is_first_rank:
+            self.embed_tokens = VocabParallelEmbedding(
+                config.vocab_size,
+                config.hidden_size,
+                quant_config=quant_config,
+                use_attn_tp_group=is_dp_attention_enabled(),
+                prefix=add_prefix("embed_tokens", prefix),
+            )
+        else:
+            self.embed_tokens = PPMissingLayer()
+
+        self.layers, self.start_layer, self.end_layer = make_layers(
+            config.num_hidden_layers,
+            lambda idx, prefix: Spark3DecoderLayer(
+                layer_id=idx,
+                config=config,
+                quant_config=quant_config,
+                prefix=prefix,
+            ),
+            pp_rank=self.pp_group.rank_in_group,
+            pp_size=self.pp_group.world_size,
+            prefix=add_prefix("layers", prefix),
+        )
+        if self.pp_group.is_last_rank:
+            self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        else:
+            self.norm = PPMissingLayer(return_tuple=True)
+
+    def get_input_embeddings(self) -> nn.Embedding:
+        return self.embed_tokens
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        forward_batch: ForwardBatch,
+        input_embeds: torch.Tensor = None,
+        pp_proxy_tensors: Optional[PPProxyTensors] = None,
+    ) -> Union[torch.Tensor, PPProxyTensors]:
+        if self.pp_group.is_first_rank:
+            if input_embeds is None:
+                hidden_states = self.embed_tokens(input_ids)
+            else:
+                hidden_states = input_embeds
+            residual = None
+        else:
+            assert pp_proxy_tensors is not None
+            hidden_states = pp_proxy_tensors["hidden_states"]
+            residual = pp_proxy_tensors["residual"]
+
+        for i in range(self.start_layer, self.end_layer):
+            layer = self.layers[i]
+            hidden_states, residual = layer(
+                positions,
+                hidden_states,
+                forward_batch,
+                residual,
+            )
+        if not self.pp_group.is_last_rank:
+            return PPProxyTensors(
+                {
+                    "hidden_states": hidden_states,
+                    "residual": residual,
+                }
+            )
+        else:
+            if hidden_states.shape[0] != 0:
+                if residual is None:
+                    hidden_states = self.norm(hidden_states)
+                else:
+                    hidden_states, _ = self.norm(hidden_states, residual)
+
+        return hidden_states
+
+
+class Spark3ForCausalLM(nn.Module):
+    def __init__(
+        self,
+        config: Spark3Config,
+        quant_config: Optional[QuantizationConfig] = None,
+        prefix: str = "",
+    ) -> None:
+        super().__init__()
+        self.pp_group = get_pp_group()
+        self.config = config
+        self.quant_config = quant_config
+        self.model = Spark3Model(
+            config, quant_config=quant_config, prefix=add_prefix("model", prefix)
+        )
+
+        # handle the lm head on different pp ranks
+        if self.pp_group.is_last_rank:
+            if self.pp_group.world_size == 1 and config.tie_word_embeddings:
+                self.lm_head = self.model.embed_tokens
+            else:
+                self.lm_head = ParallelLMHead(
+                    config.vocab_size,
+                    config.hidden_size,
+                    quant_config=quant_config,
+                    prefix=add_prefix("lm_head", prefix),
+                )
+        else:
+            # ranks other than the last rank will have a placeholder layer
+            self.lm_head = PPMissingLayer()
+
+        self.logits_processor = LogitsProcessor(config)
+
+    def get_input_embeddings(self) -> nn.Embedding:
+        return self.model.embed_tokens
+
+    @torch.no_grad()
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        forward_batch: ForwardBatch,
+        input_embeds: torch.Tensor = None,
+        pp_proxy_tensors: Optional[PPProxyTensors] = None,
+    ) -> torch.Tensor:
+        hidden_states = self.model(
+            input_ids,
+            positions,
+            forward_batch,
+            input_embeds,
+            pp_proxy_tensors=pp_proxy_tensors,
+        )
+
+        if self.pp_group.is_last_rank:
+            return self.logits_processor(
+                input_ids,
+                hidden_states,
+                self.lm_head,
+                forward_batch,
+            )
+        else:
+            return hidden_states
+
+    @torch.no_grad()
+    def forward_split_prefill(
+        self,
+        input_ids: torch.Tensor,
+        positions: torch.Tensor,
+        forward_batch: ForwardBatch,
+        split_interval: Tuple[int, int],  # [start, end) 0-based
+        input_embeds: torch.Tensor = None,
+    ):
+        start, end = split_interval
+        # embed
+        if start == 0:
+            if input_embeds is None:
+                forward_batch.hidden_states = self.model.embed_tokens(input_ids)
+            else:
+                forward_batch.hidden_states = input_embeds
+        # decoder layer
+        for i in range(start, end):
+            layer = self.model.layers[i]
+            forward_batch.hidden_states, forward_batch.residual = layer(
+                positions,
+                forward_batch.hidden_states,
+                forward_batch,
+                forward_batch.residual,
+            )
+
+        if end == self.model.config.num_hidden_layers:
+            # norm
+            hidden_states, _ = self.model.norm(
+                forward_batch.hidden_states, forward_batch.residual
+            )
+            forward_batch.hidden_states = hidden_states
+            # logits process
+            result = self.logits_processor(
+                input_ids, forward_batch.hidden_states, self.lm_head, forward_batch
+            )
+        else:
+            result = None
+
+        return result
+
+    @property
+    def start_layer(self):
+        return self.model.start_layer
+
+    @property
+    def end_layer(self):
+        return self.model.end_layer
+
+    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
+        stacked_params_mapping = [
+            ("gate_up_proj", "gate_proj", 0),
+            ("gate_up_proj", "up_proj", 1),
+        ]
+        params_dict = dict(self.named_parameters())
+        for name, loaded_weight in weights:
+            original_name = name
+            layer_id = get_layer_id(name)
+            if (
+                layer_id is not None
+                and hasattr(self.model, "start_layer")
+                and (
+                    layer_id < self.model.start_layer
+                    or layer_id >= self.model.end_layer
+                )
+            ):
+                continue
+
+            if self.config.tie_word_embeddings and "lm_head.weight" in name:
+                continue
+
+            if name in ("model.embedding.weight",):
+                name = "model.embed_tokens.weight"
+            if (
+                name == "model.embed_tokens.weight"
+                and self.config.tie_word_embeddings
+                and self.pp_group.world_size > 1
+            ):
+                if self.pp_group.is_last_rank:
+                    name = "lm_head.weight"
+                elif not self.pp_group.is_first_rank:
+                    continue
+
+            loaded = False
+            for param_name, weight_name, shard_id in stacked_params_mapping:
+                if weight_name not in name:
+                    continue
+                mapped_name = name.replace(weight_name, param_name)
+                if mapped_name not in params_dict:
+                    continue
+                param = params_dict[mapped_name]
+                weight_loader = getattr(param, "weight_loader", default_weight_loader)
+                weight_loader(param, loaded_weight, shard_id)
+                loaded = True
+                break
+            if loaded:
+                continue
+
+            if name in params_dict:
+                param = params_dict[name]
+                weight_loader = getattr(param, "weight_loader", default_weight_loader)
+                weight_loader(param, loaded_weight)
+            elif original_name in ("model.embedding.weight",):
+                continue
+            else:
+                logger.warning(f"Parameter {name} not found in params_dict")
+
+    def get_embed_and_head(self):
+        return self.model.embed_tokens.weight, self.lm_head.weight
+
+    def set_embed_and_head(self, embed, head):
+        del self.model.embed_tokens.weight
+        del self.lm_head.weight
+        self.model.embed_tokens.weight = embed
+        self.lm_head.weight = head
+        current_platform.empty_cache()
+        current_platform.synchronize()
+
+    def get_attention_sliding_window_size(self) -> int:
+        return _get_attention_sliding_window_size(self.config)
+
+
+EntryClass = [Spark3ForCausalLM]

python/sglang/srt/utils/hf_transformers/common.py

@@ -66,6 +66,7 @@ from sglang.srt.configs import (
     Qwen3_5MoeTextConfig,
     Qwen3_5TextConfig,
     Qwen3NextConfig,
+    Spark3Config,
     Step3p5Config,
     Step3p7Config,
     Step3VLConfig,
@@ -102,6 +103,7 @@ _CONFIG_REGISTRY: Dict[str, Type[PretrainedConfig]] = {
         LocateAnythingConfig,
         InternVLChatConfig,
         LagunaConfig,
+        Spark3Config,
         Step3VLConfig,
         LongcatFlashConfig,
         Olmo3Config,