File tree Expand file tree Collapse file tree 1 file changed +16
-14
lines changed
fastdeploy/model_executor/layers/moe Expand file tree Collapse file tree 1 file changed +16
-14
lines changed Original file line number Diff line number Diff line change @@ -193,20 +193,22 @@ def low_latency_combine(
193
193
Return:
194
194
combined_hidden_states: [num_tokens, hidden]
195
195
"""
196
- # TODO(@wufeisheng): Delete them when deepep in PaddlePaddle is fixed
197
- (
198
- src_info ,
199
- layout_range ,
200
- num_max_dispatch_tokens_per_rank ,
201
- num_experts ,
202
- ) = handle
203
- handle = (
204
- src_info ,
205
- layout_range ,
206
- num_max_dispatch_tokens_per_rank ,
207
- None ,
208
- num_experts ,
209
- )
196
+ if paddle .__version__ != "0.0.0" and paddle .__version__ <= "3.1.0" : # not develop version of PaddlePaddle
197
+ # TODO(@wanglongzhi): Delete them when deepep in PaddlePaddle is fixed
198
+ # and when the default recommended version of PaddlePaddle is greater than 3.1.0
199
+ (
200
+ src_info ,
201
+ layout_range ,
202
+ num_max_dispatch_tokens_per_rank ,
203
+ num_experts ,
204
+ ) = handle
205
+ handle = (
206
+ src_info ,
207
+ layout_range ,
208
+ num_max_dispatch_tokens_per_rank ,
209
+ None ,
210
+ num_experts ,
211
+ )
210
212
211
213
combined_hidden_states , _ , combine_hook = self .decode_deepep_engine .low_latency_combine (
212
214
hidden_states ,
You can’t perform that action at this time.
0 commit comments