pere commited on
Commit
816b751
1 Parent(s): 37ada0d

Saving weights and logs of step 5001

Browse files
events.out.tfevents.1628883048.t1v-n-1a0a7c50-w-0.40904.3.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ec3186ca58d42b9a6a437993c975034c3190dd6eca1233344bea1ce749e25e37
3
+ size 662048
events.out.tfevents.1628890854.t1v-n-1a0a7c50-w-0.49152.3.v2 ADDED
@@ -0,0 +1,3 @@
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06e18cb1b574fb5f1da8ebbb4ed450c5000293e4c10d15aaaa8341d877eabfde
3
+ size 735757
flax_model.msgpack CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b7a1b3c986f8f90bfb67978326d950efc15e64ad9b6d5f4884bcde1013a65968
3
  size 1100762015
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:762f6dd2de7eb5fa1dd22a1a9f6ea57fd37b6aea327a48d58dfbf1af0624f8d1
3
  size 1100762015
recover_run_streaming.sh ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ./run_t5_mlm_flax_streaming.py \
2
+ --output_dir="./" \
3
+ --model_type="t5" \
4
+ --config_name="./" \
5
+ --tokenizer_name="./" \
6
+ --model_name_or_path="./" \
7
+ --dataset_name="pere/norwegian_colossal_corpus_v2_short100k" \
8
+ --max_seq_length="512" \
9
+ --weight_decay="0.01" \
10
+ --per_device_train_batch_size="32" \
11
+ --per_device_eval_batch_size="32" \
12
+ --learning_rate="5.23e-3" \
13
+ --warmup_steps="0" \
14
+ --overwrite_output_dir \
15
+ --cache_dir /mnt/disks/flaxdisk/cache/ \
16
+ --num_train_epochs="5" \
17
+ --adam_beta1="0.9" \
18
+ --adam_beta2="0.98" \
19
+ --logging_steps="500" \
20
+ --num_train_steps="650000" \
21
+ --num_eval_samples="5000" \
22
+ --save_steps="5000" \
23
+ --eval_steps="5000" \
24
+ --preprocessing_num_workers 96 \
25
+ --adafactor \
26
+ --push_to_hub
27
+
run_streaming.sh CHANGED
@@ -3,7 +3,6 @@
3
  --model_type="t5" \
4
  --config_name="./" \
5
  --tokenizer_name="./" \
6
- --model_name_or_path="./" \
7
  --dataset_name="pere/norwegian_colossal_corpus_v2_short100k" \
8
  --max_seq_length="512" \
9
  --weight_decay="0.01" \
3
  --model_type="t5" \
4
  --config_name="./" \
5
  --tokenizer_name="./" \
 
6
  --dataset_name="pere/norwegian_colossal_corpus_v2_short100k" \
7
  --max_seq_length="512" \
8
  --weight_decay="0.01" \