Saving weights and logs of step 1000
Browse files
eval_results.json
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"eval_accuracy": 0.6890367604912527,
|
3 |
+
"eval_loss": 1.4896034919686423,
|
4 |
+
"eval_perplexity": 4.435336523745398
|
5 |
+
}
|
events.out.tfevents.1643733579.t1v-n-ccbf3e94-w-0.346519.3.v2
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:4fb95c57b0658fe85bfb8af4a847fde678c03be588e35de696bf848a834297b1
|
3 |
+
size 147136
|
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 sha256:92681f621e5c1ece7bf9ff798e84c7c86fb171b010f1026e2b651e213d16fb28
|
3 |
size 498796983
|
run_128_recover_7e.sh
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
python run_mlm_flax.py \
|
2 |
+
--output_dir="./" \
|
3 |
+
--model_type="roberta" \
|
4 |
+
--model_name_or_path="./" \
|
5 |
+
--config_name="roberta-base" \
|
6 |
+
--tokenizer_name="NbAiLab/nb-roberta-base" \
|
7 |
+
--dataset_name="NbAiLab/NCC" \
|
8 |
+
--cache_dir="/mnt/disks/flaxdisk/cache/" \
|
9 |
+
--max_seq_length="128" \
|
10 |
+
--weight_decay="0.01" \
|
11 |
+
--per_device_train_batch_size="232" \
|
12 |
+
--per_device_eval_batch_size="232" \
|
13 |
+
--pad_to_max_length \
|
14 |
+
--learning_rate="1.8183001770921692e-07" \
|
15 |
+
--warmup_steps="0" \
|
16 |
+
--overwrite_output_dir \
|
17 |
+
--num_train_epochs="3" \
|
18 |
+
--adam_beta1="0.9" \
|
19 |
+
--adam_beta2="0.98" \
|
20 |
+
--adam_epsilon="1e-6" \
|
21 |
+
--logging_steps="1000" \
|
22 |
+
--save_steps="1000" \
|
23 |
+
--eval_steps="1000" \
|
24 |
+
--auth_token="True" \
|
25 |
+
--do_train \
|
26 |
+
--do_eval \
|
27 |
+
--dtype="bfloat16" \
|
28 |
+
--push_to_hub
|