feat: remove vllm get_rope (#2964)

python/sglang/srt/models/baichuan.py

@@ -24,7 +24,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -40,6 +39,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/chatglm.py

@@ -21,7 +21,6 @@
 import torch
 from torch import nn
 from torch.nn import LayerNorm
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.configs import ChatGLMConfig
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
@@ -35,6 +34,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/commandr.py

@@ -44,7 +44,6 @@
 from torch import nn
 from torch.nn.parameter import Parameter
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -59,6 +58,7 @@
 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.vocab_parallel_embedding import VocabParallelEmbedding
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch
 from sglang.srt.model_loader.weight_utils import default_weight_loader

python/sglang/srt/models/dbrx.py

@@ -19,7 +19,6 @@
 
 import torch
 import torch.nn as nn
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.configs import DbrxConfig
 from sglang.srt.distributed import (
@@ -36,6 +35,7 @@
 from sglang.srt.layers.moe.fused_moe_triton import fused_moe
 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.vocab_parallel_embedding import (
     DEFAULT_VOCAB_PADDING_SIZE,
     ParallelLMHead,

python/sglang/srt/models/deepseek.py

@@ -21,7 +21,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -40,6 +39,7 @@
 from sglang.srt.layers.moe.fused_moe_triton import fused_moe
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/deepseek_v2.py

@@ -23,7 +23,6 @@
 from torch import nn
 from transformers import PretrainedConfig
 from vllm import _custom_ops as ops
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -49,7 +48,7 @@
     normalize_e4m3fn_to_e4m3fnuz,
 )
 from sglang.srt.layers.radix_attention import RadixAttention
-from sglang.srt.layers.rotary_embedding import get_rope_wrapper
+from sglang.srt.layers.rotary_embedding import get_rope
 from sglang.srt.layers.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,
@@ -272,7 +271,7 @@ def __init__(
             quant_config=quant_config,
         )
         rope_scaling["rope_type"] = "deepseek_yarn"
-        self.rotary_emb = get_rope_wrapper(
+        self.rotary_emb = get_rope(
             qk_rope_head_dim,
             rotary_dim=qk_rope_head_dim,
             max_position=max_position_embeddings,

python/sglang/srt/models/exaone.py

@@ -20,7 +20,6 @@
 
 import torch
 from torch import nn
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import SiluAndMul
@@ -33,6 +32,7 @@
 from sglang.srt.layers.logits_processor import LogitsProcessor, LogitsProcessorOutput
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/gemma.py

@@ -21,7 +21,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import GeluAndMul
@@ -34,6 +33,7 @@
 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.vocab_parallel_embedding import VocabParallelEmbedding
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch
 from sglang.srt.model_loader.weight_utils import default_weight_loader

python/sglang/srt/models/gemma2.py

@@ -15,12 +15,11 @@
 # Adapted from:
 # https://github.com/vllm-project/vllm/blob/56b325e977435af744f8b3dca7af0ca209663558/vllm/model_executor/models/gemma2.py
 
-from typing import Iterable, Optional, Set, Tuple, Union
+from typing import Iterable, Optional, Set, Tuple
 
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import GeluAndMul
@@ -33,6 +32,7 @@
 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.vocab_parallel_embedding import VocabParallelEmbedding
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch
 from sglang.srt.model_loader.weight_utils import default_weight_loader
@@ -45,10 +45,6 @@ def get_attention_sliding_window_size(config):
     return config.sliding_window - 1
 
 
-# FIXME: temporary solution, remove after next vllm release
-from vllm.model_executor.layers.rotary_embedding import RotaryEmbedding
-
-
 class Gemma2MLP(nn.Module):
     def __init__(
         self,

python/sglang/srt/models/gpt2.py

@@ -25,8 +25,6 @@
 
 from sglang.srt.distributed.parallel_state import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import get_act_fn
-
-# from sglang.srt.layers.activation import get_act_fn
 from sglang.srt.layers.linear import (
     ColumnParallelLinear,
     QKVParallelLinear,

python/sglang/srt/models/granite.py

@@ -22,7 +22,6 @@
 import torch
 from torch import nn
 from transformers import GraniteConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import SiluAndMul
@@ -36,6 +35,7 @@
 from sglang.srt.layers.pooler import Pooler, PoolingType
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/grok.py

@@ -22,7 +22,6 @@
 import torch.nn.functional as F
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -40,6 +39,7 @@
 from sglang.srt.layers.moe.fused_moe_triton import FusedMoE
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/internlm2.py

@@ -19,7 +19,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import SiluAndMul
@@ -32,6 +31,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/llama.py

@@ -22,7 +22,6 @@
 import torch
 from torch import nn
 from transformers import LlamaConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -39,6 +38,7 @@
 from sglang.srt.layers.pooler import Pooler, PoolingType
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/minicpm.py

@@ -18,7 +18,6 @@
 
 import torch
 from torch import nn
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import SiluAndMul
@@ -31,6 +30,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/minicpm3.py

@@ -19,7 +19,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import SiluAndMul
@@ -33,6 +32,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/mixtral.py

@@ -21,7 +21,6 @@
 import torch
 from torch import nn
 from transformers import MixtralConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_world_size,
@@ -38,6 +37,7 @@
 from sglang.srt.layers.moe.fused_moe_triton import FusedMoE
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/mixtral_quant.py

@@ -23,7 +23,6 @@
 import torch.nn.functional as F
 from torch import nn
 from transformers import MixtralConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -39,6 +38,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/olmo.py

@@ -20,7 +20,6 @@
 import torch
 from torch import nn
 from transformers import OlmoConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import SiluAndMul
@@ -32,6 +31,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/olmo2.py

@@ -21,7 +21,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import (
     get_tensor_model_parallel_rank,
@@ -39,6 +38,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/olmoe.py

@@ -22,7 +22,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.layernorm import RMSNorm
@@ -35,6 +34,7 @@
 from sglang.srt.layers.moe.fused_moe_triton import FusedMoE
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,

python/sglang/srt/models/phi3_small.py

@@ -5,7 +5,6 @@
 from torch import nn
 from transformers import Phi3Config
 from transformers.configuration_utils import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.linear import (
@@ -17,6 +16,7 @@
 from sglang.srt.layers.pooler import Pooler, PoolingType
 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.vocab_parallel_embedding import (
     DEFAULT_VOCAB_PADDING_SIZE,
     ParallelLMHead,

python/sglang/srt/models/qwen.py

@@ -20,7 +20,6 @@
 import torch
 from torch import nn
 from transformers import PretrainedConfig
-from vllm.model_executor.layers.rotary_embedding import get_rope
 
 from sglang.srt.distributed import get_tensor_model_parallel_world_size
 from sglang.srt.layers.activation import SiluAndMul
@@ -33,6 +32,7 @@
 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.vocab_parallel_embedding import (
     ParallelLMHead,
     VocabParallelEmbedding,