Code structure refactor (#807)

python/sglang/srt/models/dbrx.py

@@ -45,7 +45,7 @@ from vllm.transformers_utils.configs.dbrx import DbrxConfig
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class DbrxRouter(nn.Module):

python/sglang/srt/models/deepseek.py

@@ -46,7 +46,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.infer_batch import InputMetadata
+from sglang.srt.managers.schedule_batch import InputMetadata
 
 
 class DeepseekMLP(nn.Module):

python/sglang/srt/models/deepseek_v2.py

@@ -45,7 +45,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class DeepseekV2MLP(nn.Module):

python/sglang/srt/models/gemma.py

@@ -37,7 +37,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class GemmaMLP(nn.Module):

python/sglang/srt/models/gemma2.py

@@ -42,7 +42,7 @@ from vllm.model_executor.sampling_metadata import SamplingMetadata
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class GemmaRMSNorm(CustomOp):

python/sglang/srt/models/gpt_bigcode.py

@@ -35,7 +35,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.infer_batch import InputMetadata
+from sglang.srt.managers.schedule_batch import InputMetadata
 
 
 class GPTBigCodeAttention(nn.Module):

python/sglang/srt/models/grok.py

@@ -52,7 +52,7 @@ from vllm.utils import print_warning_once
 from sglang.srt.layers.fused_moe import fused_moe
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 use_fused = True
 

python/sglang/srt/models/internlm2.py

@@ -40,7 +40,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class InternLM2MLP(nn.Module):

python/sglang/srt/models/llama2.py

@@ -36,7 +36,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 MergedColumnParallelLinear = None
 QKVParallelLinear = None

python/sglang/srt/models/llama_classification.py

@@ -25,7 +25,7 @@ from vllm.model_executor.layers.quantization.base_config import QuantizationConf
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitProcessorOutput
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 from sglang.srt.models.llama2 import LlamaModel
 
 

python/sglang/srt/models/llava.py

@@ -32,13 +32,13 @@ from vllm.config import CacheConfig
 from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
-from sglang.srt.managers.controller.infer_batch import ForwardMode
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.managers.schedule_batch import ForwardMode
 from sglang.srt.mm_utils import (
     get_anyres_image_grid_shape,
     unpad_image,
     unpad_image_shape,
 )
+from sglang.srt.model_executor.model_runner import InputMetadata
 from sglang.srt.models.llama2 import LlamaForCausalLM
 from sglang.srt.models.mistral import MistralForCausalLM
 from sglang.srt.models.qwen2 import Qwen2ForCausalLM

python/sglang/srt/models/llavavid.py

@@ -26,13 +26,13 @@ from vllm.config import CacheConfig
 from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
-from sglang.srt.managers.controller.infer_batch import ForwardMode
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.managers.schedule_batch import ForwardMode
 from sglang.srt.mm_utils import (
     get_anyres_image_grid_shape,
     unpad_image,
     unpad_image_shape,
 )
+from sglang.srt.model_executor.model_runner import InputMetadata
 from sglang.srt.models.llama2 import LlamaForCausalLM
 
 

python/sglang/srt/models/minicpm.py

@@ -39,7 +39,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class MiniCPMMLP(nn.Module):

python/sglang/srt/models/mixtral.py

@@ -50,7 +50,7 @@ from vllm.utils import print_warning_once
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class MixtralMoE(nn.Module):

python/sglang/srt/models/mixtral_quant.py

@@ -45,7 +45,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class MixtralMLP(nn.Module):

python/sglang/srt/models/qwen.py

@@ -39,7 +39,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class QWenMLP(nn.Module):

python/sglang/srt/models/qwen2.py

@@ -39,7 +39,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 Qwen2Config = None
 

python/sglang/srt/models/qwen2_moe.py

@@ -51,7 +51,7 @@ from vllm.sequence import IntermediateTensors, SamplerOutput
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class Qwen2MoeMLP(nn.Module):

python/sglang/srt/models/stablelm.py

@@ -40,7 +40,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
 from sglang.srt.layers.logits_processor import LogitsProcessor
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.managers.controller.model_runner import InputMetadata
+from sglang.srt.model_executor.model_runner import InputMetadata
 
 
 class StablelmMLP(nn.Module):

python/sglang/srt/server.py

@@ -44,11 +44,11 @@ from fastapi.responses import JSONResponse, Response, StreamingResponse
 from sglang.lang.backend.runtime_endpoint import RuntimeEndpoint
 from sglang.srt.constrained import disable_cache
 from sglang.srt.hf_transformers_utils import get_tokenizer
-from sglang.srt.managers.controller.manager_multi import (
+from sglang.srt.managers.controller_multi import (
     start_controller_process as start_controller_process_multi,
 )
-from sglang.srt.managers.controller.manager_single import launch_tp_servers
-from sglang.srt.managers.controller.manager_single import (
+from sglang.srt.managers.controller_single import launch_tp_servers
+from sglang.srt.managers.controller_single import (
     start_controller_process as start_controller_process_single,
 )
 from sglang.srt.managers.detokenizer_manager import start_detokenizer_process