[Kernel] Porting fuse_minimax_qk_norm to manual fusion (#43410)
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
This commit is contained in:
@@ -39,7 +39,6 @@ from vllm.model_executor.layers.mamba.abstract import MambaBase
|
||||
from vllm.model_executor.layers.mamba.linear_attn import (
|
||||
MiniMaxText01LinearAttention,
|
||||
MiniMaxText01LinearKernel,
|
||||
MiniMaxText01RMSNormTP,
|
||||
clear_linear_attention_cache_for_new_sequences,
|
||||
linear_attention_decode,
|
||||
linear_attention_prefill_and_mix,
|
||||
@@ -49,6 +48,7 @@ from vllm.model_executor.layers.mamba.mamba_utils import (
|
||||
MambaStateDtypeCalculator,
|
||||
MambaStateShapeCalculator,
|
||||
)
|
||||
from vllm.model_executor.layers.minimax_rms_norm import MiniMaxText01RMSNormTP
|
||||
from vllm.model_executor.layers.mla import MLAModules, MultiHeadLatentAttentionWrapper
|
||||
from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
|
||||
from vllm.model_executor.layers.rotary_embedding import get_rope
|
||||
|
||||
Reference in New Issue
Block a user