fix t5x tests (#1208)

T5X tests are failing because the vocabulary specified in the dummy
dataset does not match the vocabulary used by the model:
https://github.com/google-research/t5x/blob/5f03619b0c5ebb44ae6adde1a2d8eea1a4b55fe0/t5x/examples/t5/t5_1_1/base.gin#L20-L21.
This PR updates the dataset vocab to make it match the model's.

---------

Signed-off-by: ashors1 <[email protected]>

Author: ashors1

.github/container/manifest.yaml

@@ -31,6 +31,7 @@ t5x:
     mirror/patch/partial-checkpoint-restore: file://patches/t5x/mirror-patch-partial-checkpoint-restore.patch # pull/1392/head  # https://github.com/google-research/t5x/pull/1392: Add support for partial checkpoint restore
     mirror/patch/dali-support: file://patches/t5x/mirror-patch-dali-support.patch # pull/1393/head  # https://github.com/google-research/t5x/pull/1393: Adds DALI support to t5x
     mirror/patch/t5x_te_in_contrib_noindent: file://patches/t5x/mirror-patch-t5x_te_in_contrib_noindent.patch # pull/1391/head  # https://github.com/google-research/t5x/pull/1391: Adds transformer engine support and GPU optimizations to T5x (enables H100)
+    mirror/patch/fix-default-vocab: file://patches/t5x/mirror-patch-fix-default-vocab.patch # pull/1609/head  # https://github.com/google-research/t5x/pull/1609: Fixes seqio vocab mismatch
 paxml:
   url: https://github.com/google/paxml.git
   mirror_url: https://github.com/nvjax-svc-0/paxml.git

.github/container/test-t5x.sh

@@ -175,10 +175,10 @@ seqio.TaskRegistry.add(
     ],
     output_features=dict(
         inputs=seqio.Feature(
-            vocabulary=t5.data.get_default_vocabulary(), add_eos=True, required=False
+            vocabulary=seqio.SentencePieceVocabulary(sentencepiece_model_file="gs://t5-data/vocabs/cc_all.32000.100extra/sentencepiece.model"), add_eos=True, required=False
         ),
         targets=seqio.Feature(
-            vocabulary=t5.data.get_default_vocabulary(), add_eos=True
+            vocabulary=seqio.SentencePieceVocabulary(sentencepiece_model_file="gs://t5-data/vocabs/cc_all.32000.100extra/sentencepiece.model"), add_eos=True
         )
     ),
     metric_fns=[]