balanced lm task

norwegian_lm_large.gin

@@ -13,7 +13,7 @@ TASK_FEATURE_LENGTHS = {"inputs": 512, "targets": 512}
 TRAIN_STEPS = 1_600_000
 DROPOUT_RATE = 0.0 # Changed from the default since T5-1.1 recomments this.
 INITIAL_CHECKPOINT_PATH = "gs://nb-t5x-us-central2/norwegian_NCC_plus_English_t5x_large/checkpoint_1500000"
-PjitPartitioner.num_partitions = 1
+#PjitPartitioner.num_partitions = 1
 utils.SaveCheckpointConfig.period = 5000
 utils.SaveCheckpointConfig.keep = 3
 

tasks.py

@@ -273,3 +273,32 @@ TaskRegistry.add(
     output_features={"targets": DEFAULT_OUTPUT_FEATURES["targets"]},
     metric_fns=[]
 )
+# Final pretraining task used in Raffel et al., 2019 adaptated to NCC
+dataset_name = 'NbAiLab/balanced_bokmaal_nynorsk'
+dataset_params = {"path": dataset_name, "use_auth_token": True, "streaming": True}
+dataset_shapes = None
+TaskRegistry.add(
+    "balanced_bokmaal_nynorsk_prefix_lm_stream",
+    source=seqio.FunctionDataSource(
+        dataset_fn=functools.partial(dataset_fn, dataset_params=dataset_params),
+        splits=("train", "validation"),
+        caching_permitted=False,
+        num_input_examples=dataset_shapes,
+    ),
+    preprocessors=[
+        functools.partial(
+            target_to_key, key_map={
+                "inputs": None,
+                "targets": None,
+            }, target_key="targets"),
+        seqio.preprocessors.tokenize,
+        # seqio.CacheDatasetPlaceholder(),
+        preprocessors.prefix_lm,
+        seqio.preprocessors.append_eos_after_trim,
+    ],
+    output_features={"targets": DEFAULT_OUTPUT_FEATURES["targets"]},
+    metric_fns=[]
+)
+
+
+