зеркало из https://github.com/microsoft/DeepSpeed.git
enable phi2 autotp (#5436)
This PR aims to enable phi2 model autotp. --------- Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com> Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com>
This commit is contained in:
Родитель
4696afd27b
Коммит
3a7f3aa849
|
@ -307,6 +307,8 @@ class AutoTP():
|
|||
# Mixtral-7x8b used w2*act(w1*w3) linear. need to replace w2 to linearallreduce.
|
||||
elif 'w2' in layer and 'Mixtral' in str(type(module)):
|
||||
gem_list = gem_list + [layer]
|
||||
elif "self_attn.dense" in layer and "Phi" in str(type(module)):
|
||||
gem_list = gem_list + [layer]
|
||||
|
||||
layer_list = []
|
||||
if gem_list != []:
|
||||
|
|
Загрузка…
Ссылка в новой задаче