Saving weights and logs of step 200949

events.out.tfevents.1632578878.t1v-n-1a0a7c50-w-0.1553032.0.v2 → events.out.tfevents.1633002139.t1v-n-1a0a7c50-w-0.14262.0.v2

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:989d68a95e5588815e9f2928605577caaa0c1951f5c6c65ed02822ef75b3b4a4
-size 59426179
+oid sha256:78758484600525d130d0d3fd5d16d6c1e76ed432445c0b8dc845312d6a58ceda
+size 40

events.out.tfevents.1633618902.t1v-n-1a0a7c50-w-0.198625.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b24c6073c30d64f1f69c9a55f7f3331d07e1129f791341f4aca9ef64d1b3164
+size 29767300

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee7e7cef25c7f5b6e32bc3664b565d74c0e401b95aa119703a1a2005fb85c4f1
+oid sha256:e7488903c425f102390f22c20b3480f4041014a362aa8ea59590f43f2cb4338e
 size 5262371934

run_recover_250.sh

@@ -0,0 +1,21 @@
+python run_clm_mp.py \
+    	--model_name_or_path "./" \
+  	--tokenizer_name "./" \
+   	--train_file /mnt/disks/flaxdisk/corpus/administrative_nb_train.json \
+       	--validation_file /mnt/disks/flaxdisk/corpus/administrative_nb_validation.json \
+     	--do_train \
+       	--do_eval  \
+     	--block_size 1024 \
+   	--num_train_epochs 5 \
+   	--learning_rate 1e-5 \
+	--per_device_train_batch_size 2 \
+       	--per_device_eval_batch_size 2 \
+    	--overwrite_output_dir \
+	--output_dir "./" \
+	--cache_dir /mnt/disks/flaxdisk/cache/ \
+	--preprocessing_num_workers 96 \
+	--dtype bfloat16 \
+   	--logging_steps 50000 \
+       	--eval_steps 50000 \
+	--warmup_steps 3000 \
+	--push_to_hub

run_restore.sh

@@ -0,0 +1,21 @@
+python run_clm_mp.py \
+    	--model_name_or_path "./" \
+  	--tokenizer_name "./" \
+   	--train_file /mnt/disks/flaxdisk/corpus/administrative_nb_train.json \
+       	--validation_file /mnt/disks/flaxdisk/corpus/administrative_nb_validation.json \
+     	--do_train \
+       	--do_eval  \
+     	--block_size 1024 \
+   	--num_train_epochs 5 \
+   	--learning_rate 1e-5 \
+	--per_device_train_batch_size 2 \
+       	--per_device_eval_batch_size 2 \
+    	--overwrite_output_dir \
+	--output_dir "./" \
+	--cache_dir /mnt/disks/flaxdisk/cache/ \
+	--preprocessing_num_workers 96 \
+	--dtype bfloat16 \
+   	--logging_steps 50000 \
+       	--eval_steps 50000 \
+	--warmup_steps 0 \
+	--push_to_hub