mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-04 16:22:57 +08:00
fix ep when paddle version mismatch (#3056)
This commit is contained in:
@@ -193,20 +193,22 @@ class DeepEPEngine:
|
||||
Return:
|
||||
combined_hidden_states: [num_tokens, hidden]
|
||||
"""
|
||||
# TODO(@wufeisheng): Delete them when deepep in PaddlePaddle is fixed
|
||||
(
|
||||
src_info,
|
||||
layout_range,
|
||||
num_max_dispatch_tokens_per_rank,
|
||||
num_experts,
|
||||
) = handle
|
||||
handle = (
|
||||
src_info,
|
||||
layout_range,
|
||||
num_max_dispatch_tokens_per_rank,
|
||||
None,
|
||||
num_experts,
|
||||
)
|
||||
if paddle.__version__ != "0.0.0" and paddle.__version__ <= "3.1.0": # not develop version of PaddlePaddle
|
||||
# TODO(@wanglongzhi): Delete them when deepep in PaddlePaddle is fixed
|
||||
# and when the default recommended version of PaddlePaddle is greater than 3.1.0
|
||||
(
|
||||
src_info,
|
||||
layout_range,
|
||||
num_max_dispatch_tokens_per_rank,
|
||||
num_experts,
|
||||
) = handle
|
||||
handle = (
|
||||
src_info,
|
||||
layout_range,
|
||||
num_max_dispatch_tokens_per_rank,
|
||||
None,
|
||||
num_experts,
|
||||
)
|
||||
|
||||
combined_hidden_states, _, combine_hook = self.decode_deepep_engine.low_latency_combine(
|
||||
hidden_states,
|
||||
|
Reference in New Issue
Block a user