DeepSpeed
MOE gate fixes and enhancements
#5156
Merged
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Overview
Commits
13
Changes
View On
GitHub
MOE gate fixes and enhancements
#5156
tjruwase
merged 13 commits into
deepspeedai:master
from
mosheisland:moe/gate
mosheisland
requested a review
from
awan-10
1 year ago
tjruwase
requested a review
from
tohtana
1 year ago
tjruwase
removed review request
from
tohtana
1 year ago
mosheisland
force pushed
from
4cb1faba
to
4706fa2a
1 year ago
mosheisland
force pushed
from
4706fa2a
to
8f9d75c4
1 year ago
MOE: Support top2 with disable token dropping
3e0c35f3
MOE: Fix top2 aux loss
692d42df
MOE: Support disable top2 2nd expert sampling
953e6984
MOE: Fix capacity when using TP for non-MoE
64150db0
MOE: Fix gate conversion to fp32
aab9fc3a
mosheisland
force pushed
from
8f9d75c4
to
aab9fc3a
1 year ago
awan-10
commented on 2024-02-21
awan-10
commented on 2024-02-21
awan-10
commented on 2024-02-21
awan-10
assigned
awan-10
1 year ago
Revert "MOE: Fix top2 aux loss"
1657955c
awan-10
approved these changes on 2024-02-22
Merge branch 'master' into moe/gate
05b2262e
Merge branch 'master' into moe/gate
6c04451a
Merge branch 'master' into moe/gate
c8e05eaa
Merge branch 'master' into moe/gate
7eb3633e
Merge branch 'master' into moe/gate
b76a541c
Merge branch 'master' into moe/gate
3d227ef0
Merge branch 'master' into moe/gate
30118cf1
tjruwase
merged
5a2e705b
into master
1 year ago
Login to write a write a comment.
Login via GitHub
Reviewers
awan-10
Assignees
awan-10
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub