ArthurZucker
force pushed
from
629d3038
to
c084aa7a
153 days ago
restore
ec2ed1db
ArthurZucker
force pushed
from
596897f7
to
ec2ed1db
153 days ago
add all reduce for ep
33ca3308
fix init and bias sharding
e545ac1f
fix finalize weight init
fa78068d
add full stacktracing
6e4d2343
ArthurZucker
changed the title up Fix EP post merge 152 days ago
fix
05fc1fa2
okay big improvement here
819698c6
the only case shard index should be used is when we are acctually col…
d99f8340
more fixes
f0d0de12
fix EP forward gpt oss
c5cbdc80
revert some shit
94d676cd
when you are stupid sometimes you really need a brain :) :) :) :)
959b46f7
fix TP
01c5774d
Ok GPT oss is fixed now
9dbb634c
try to fix perms
83742984
attempt to fix
8e46655c
am I a doomer and AI is not that bad?
104f80d3
fix
14dca0c1
it "passes" but the output is shit
3600fbe4
Merge branch 'fix-ep' of github.com:huggingface/transformers into fix-ep
fabde8a2
style my man
20dee9a8
outputs are gonna be giberish but at least the forward pass "works"
0b95c643
Merge branch 'main' of github.com:huggingface/transformers into fix-ep
9137fa25
dtyle
646cbe38
fix mixtral
4b32a6b2
okay shape fixes
14607626
tensor idx is only for groupped gemm / EP
f7893956
fix gate_up shard
8b4ed7bd
fix :)
d8cd5339
revert some EP changes that are breaking other stuff
76c904cf
style
cfd92d79
use moe_tp_experts
7ad14a70
3outeille
approved these changes
on 2026-02-05
revert unrelated, last nits and style
c3ae670f
good?
45dd1248
fix modlar
46900e07
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub