apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 17:11:21 +08:00
Code Issues Actions 10 Packages Projects Releases Wiki Activity

[Iluvatar GPU] Modify the names of some variables (#3273)

Browse Source
This commit is contained in:
yzwu
2025-08-13 11:38:02 +08:00
committed by GitHub
parent b4fef2cf29
commit ce9180241e
3 changed files with 16 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/ops/iluvatar/moe_ops.py
+14 -12
View File
@@ -79,25 +79,27 @@ def group_gemm(
def iluvatar_moe_expert_ffn(
permute_input: paddle.Tensor,
tokens_expert_prefix_sum: paddle.Tensor,
ffn1_weight: paddle.Tensor,
ffn2_weight: paddle.Tensor,
ffn1_bias: Optional[paddle.Tensor],
ffn1_scale: Optional[paddle.Tensor],
ffn2_scale: Optional[paddle.Tensor],
ffn2_in_scale: Optional[paddle.Tensor],
up_gate_proj_weight: paddle.Tensor,
down_proj_weight: paddle.Tensor,
up_gate_proj_bias: Optional[paddle.Tensor],
up_gate_proj_scale: Optional[paddle.Tensor],
down_proj_scale: Optional[paddle.Tensor],
down_proj_in_scale: Optional[paddle.Tensor],
expert_idx_per_token: Optional[paddle.Tensor],
quant_method: str,
used_in_ep_low_latency: bool,
):
assert ffn1_bias is None
assert ffn1_scale is not None
assert ffn2_scale is not None
assert ffn2_in_scale is None
assert up_gate_proj_bias is None
assert up_gate_proj_scale is not None
assert down_proj_scale is not None
assert down_proj_in_scale is None
assert expert_idx_per_token is None
assert quant_method in ("weight_only_int8")
assert not used_in_ep_low_latency
tokens_expert_prefix_sum_cpu = tokens_expert_prefix_sum.to("cpu")
ffn1_output = w8a16_group_gemm(permute_input, ffn1_weight, ffn1_scale, tokens_expert_prefix_sum_cpu, -1)
ffn1_output = w8a16_group_gemm(
permute_input, up_gate_proj_weight, up_gate_proj_scale, tokens_expert_prefix_sum_cpu, -1
)
act_out = swiglu(ffn1_output)
output = w8a16_group_gemm(act_out, ffn2_weight, ffn2_scale, tokens_expert_prefix_sum_cpu, -1)
output = w8a16_group_gemm(act_out, down_proj_weight, down_proj_scale, tokens_expert_prefix_sum_cpu, -1)
return output