-
updated training tasks for big corpus
-
even smaller batch size
-
smaller seq length
-
changes batch size
-
changes to batch size and to partitions
-
longer and wider
-
more steps
-
two partitions
-
new model dir
-
test
-
test
-
small chnages to test
-
mest
-
test
-
remove partitionss
-
change model target dir
-
exp
-
exp
-
exp
-
removing seqio
-
new script for partitioning. Based on base_wmt14enfr
-
another partitioner test
-
another try with specific finetune gin file
-
test
-
new fix for partitioner number
-
remove cache
-
changed number of partitions
-
debug
-
16 file
-
bs64 for v4
-
batch size 64 supported
-
changed model names in all files
-
updates to scripts
-
new train
-
multiple fixes
-
some new t4 stuff
-
updates for v4
-
updated
-
recover
-
new server
-
First attempt
-
initial commit
system HF Staff
