chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Revert "[feature] Ascend NPU graph support (#8027)" (#9348)

Browse Source
This commit is contained in:
Even Zhou
2025-08-20 01:11:23 +08:00
committed by GitHub
parent 01d47a27b6
commit f4fafacc5d
18 changed files with 878 additions and 1349 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/models/deepseek_v2.py
View File

@@ -1200,7 +1200,7 @@ class DeepseekV2AttentionMLA(nn.Module):
forward_batch: ForwardBatch,
zero_allocator: BumpAllocator,
):
from sglang.srt.model_executor.graph_runner import get_is_capture_mode
from sglang.srt.model_executor.cuda_graph_runner import get_is_capture_mode
if self.q_lora_rank is not None:
if hidden_states.shape[0] <= 16 and self.use_min_latency_fused_a_gemm:

2
python/sglang/srt/models/glm4_moe.py
View File

@@ -68,8 +68,8 @@ from sglang.srt.layers.vocab_parallel_embedding import (
VocabParallelEmbedding,
)
from sglang.srt.managers.schedule_batch import global_server_args_dict
from sglang.srt.model_executor.cuda_graph_runner import get_is_capture_mode
from sglang.srt.model_executor.forward_batch_info import ForwardBatch
from sglang.srt.model_executor.graph_runner import get_is_capture_mode
from sglang.srt.model_loader.weight_utils import default_weight_loader
from sglang.srt.models.deepseek_v2 import (
DeepseekV2DecoderLayer,

2
python/sglang/srt/models/mllama.py
View File

@@ -966,7 +966,7 @@ class MllamaForConditionalGeneration(nn.Module):
positions: torch.Tensor,
forward_batch: ForwardBatch,
) -> Union[Tuple, CausalLMOutputWithPast]:
from sglang.srt.model_executor.graph_runner import get_is_capture_mode
from sglang.srt.model_executor.cuda_graph_runner import get_is_capture_mode
batched_images, batched_ar_ids, batched_ar_mask, encoder_lens_need = (
self._batch_image_inputs(forward_batch)

2
python/sglang/srt/models/qwen3.py
View File

@@ -22,8 +22,8 @@ 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
from sglang.srt.model_executor.cuda_graph_runner import get_is_capture_mode
from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
from sglang.srt.model_executor.graph_runner import get_is_capture_mode
from sglang.srt.model_loader.weight_utils import default_weight_loader
from sglang.srt.models.qwen2 import Qwen2MLP as Qwen3MLP
from sglang.srt.models.qwen2 import Qwen2Model

2
python/sglang/srt/models/qwen3_moe.py
View File

@@ -52,8 +52,8 @@ from sglang.srt.layers.rotary_embedding import get_rope
from sglang.srt.layers.utils import get_layer_id
from sglang.srt.layers.vocab_parallel_embedding import ParallelLMHead
from sglang.srt.managers.schedule_batch import global_server_args_dict
from sglang.srt.model_executor.cuda_graph_runner import get_is_capture_mode
from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
from sglang.srt.model_executor.graph_runner import get_is_capture_mode
from sglang.srt.model_loader.weight_utils import default_weight_loader
from sglang.srt.models.qwen2_moe import Qwen2MoeMLP as Qwen3MoeMLP
from sglang.srt.models.qwen2_moe import Qwen2MoeModel