Saving weights and logs of step 2500
Browse files
events.out.tfevents.1638828930.t1v-n-d4d6e0cd-w-0.721231.0.v2
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:5f6dd60bd2a9f5218802933a8ad1e193b0fcfd889464d3d06d9084ad2dbc8ec9
|
3 |
+
size 367912
|
flax_model.msgpack
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 498796983
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ceac157b5b4dd77a8a109aa0a7acd9f4b3858da6fe44496eb76bba963c6248a9
|
3 |
size 498796983
|
run_oscar_adafactor.sh
ADDED
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
./run_mlm_flax.py \
|
2 |
+
--output_dir="./" \
|
3 |
+
--model_type="roberta" \
|
4 |
+
--config_name="./" \
|
5 |
+
--tokenizer_name="./" \
|
6 |
+
--dataset_name="oscar" \
|
7 |
+
--dataset_config_name="unshuffled_deduplicated_no" \
|
8 |
+
--max_seq_length="128" \
|
9 |
+
--weight_decay="0.01" \
|
10 |
+
--per_device_train_batch_size="128" \
|
11 |
+
--per_device_eval_batch_size="128" \
|
12 |
+
--learning_rate="3e-4" \
|
13 |
+
--warmup_steps="1000" \
|
14 |
+
--overwrite_output_dir \
|
15 |
+
--num_train_epochs="18" \
|
16 |
+
--adam_beta1="0.9" \
|
17 |
+
--adam_beta2="0.98" \
|
18 |
+
--logging_steps="500" \
|
19 |
+
--save_steps="2500" \
|
20 |
+
--eval_steps="2500" \
|
21 |
+
--adafactor \
|
22 |
+
--push_to_hub
|