Saving weights and logs of step 10001

config.json

@@ -21,7 +21,7 @@
   "pad_token_id": 1,
   "position_embedding_type": "absolute",
   "torch_dtype": "float32",
-  "transformers_version": "4.25.1",
+  "transformers_version": "4.26.0",
   "type_vocab_size": 1,
   "use_cache": true,
   "vocab_size": 250002

events.out.tfevents.1674763276.t1v-n-d7824f3a-w-7.25607.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fce5716dd72aa0242342a6d8b296bb75b4f03956d9e3832d91bd953ff425c56
+size 1471447

flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f38cecbf3ccd025280b07cdc9588400e58dbd0c90d4de5bb4c28d7eb466b67b5
+size 1113187999

run_mlm_flax_stream.py

@@ -564,7 +564,7 @@ if __name__ == "__main__":
     train_metrics = []
     eval_metrics = []
     
-    training_iter = iter(torch.utils.data.DataLoader(tokenized_datasets.with_format("torch"), batch_size=1, shuffle=False, num_workers=min(33,dataset.n_shards), collate_fn=lambda x: x))
+    training_iter = iter(torch.utils.data.DataLoader(tokenized_datasets.with_format("torch"), batch_size=1, shuffle=False, num_workers=min(12,dataset.n_shards), collate_fn=lambda x: x))
     
     max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)
     eval_samples = advance_iter_and_group_samples(training_iter, data_args.num_eval_samples, max_seq_length)