Saving weights and logs of step 1000

config.json

@@ -23,5 +23,5 @@
   "tie_word_embeddings": false,
   "transformers_version": "4.9.0.dev0",
   "use_cache": true,
-  "vocab_size": 64003
+  "vocab_size": 64103
 }

events.out.tfevents.1626415996.t1v-n-a5dc64a9-w-0.41138.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3743159f246f5f2f4b63c20b39c7323c9214f1c0cfc428d8c43b7e5f48c990f2
+size 147756

flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:437f7cb4dbacd6462c43e4b7b365215873cd90d352cbd4a55c8ff7408d2cb6ae
+size 438822439

run-t5v1_1-small.sh

@@ -1,27 +0,0 @@
-export model_dir=arabic-t5-small
-export train_batch_size=48
-export eval_batch_size=48
-
-python ./run_t5_mlm_flax.py \
---model_type t5 \
---config_name ${model_dir} \
---tokenizer_name ${model_dir} \
---use_fast_tokenizer True \
---dtype float32 \
---max_seq_length 512 \
---preprocessing_num_workers 96 \
---output_dir ${model_dir} \
---overwrite_output_dir True \
---do_train \
---per_device_train_batch_size ${train_batch_size} \
---per_device_eval_batch_size ${eval_batch_size} \
---learning_rate 1e-2 \
---num_train_epochs 1 \
---logging_steps 100 \
---eval_steps 2500 \
---save_steps 10000 \
---seed 12 \
---adafactor True \
---push_to_hub \
---cache_dir ./training_cache \
---save_total_limit 5

run-t5v1_1-small.sh

@@ -0,0 +1 @@
+../run-t5v1_1-small.sh