d5f77f14d0
### What this PR does / why we need it?
mkdir triton package and move triton files
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
Signed-off-by: shiyuan680 <917935075@qq.com>
18 lines
1.0 KiB
Python
18 lines
1.0 KiB
Python
import vllm.model_executor.layers.fla.ops.chunk
|
|
import vllm.model_executor.layers.fla.ops.fused_recurrent
|
|
import vllm.model_executor.layers.fla.ops.layernorm_guard
|
|
import vllm.model_executor.layers.mamba.ops.causal_conv1d
|
|
|
|
from vllm_ascend.ops.triton.fla.fla import (LayerNormFn,
|
|
torch_chunk_gated_delta_rule)
|
|
from vllm_ascend.ops.triton.fla.sigmoid_gating import \
|
|
fused_recurrent_gated_delta_rule_fwd_kernel
|
|
from vllm_ascend.ops.triton.mamba.casual_conv1d import (
|
|
causal_conv1d_fn, causal_conv1d_update_npu)
|
|
|
|
vllm.model_executor.layers.mamba.ops.causal_conv1d.causal_conv1d_update = causal_conv1d_update_npu
|
|
vllm.model_executor.layers.mamba.ops.causal_conv1d.causal_conv1d_fn = causal_conv1d_fn
|
|
vllm.model_executor.layers.fla.ops.fused_recurrent.fused_recurrent_gated_delta_rule_fwd_kernel = fused_recurrent_gated_delta_rule_fwd_kernel
|
|
vllm.model_executor.layers.fla.ops.layernorm_guard.LayerNormFn = LayerNormFn
|
|
vllm.model_executor.layers.fla.ops.chunk.chunk_gated_delta_rule = torch_chunk_gated_delta_rule
|