Hugging Face's logo

sheepy928
/
bincorp_ep0.3_wo_ep

Transformers
PyTorch
marinda
Inference Endpoints
Model card Files Community
sheepy928 commited on
Commit
80d1116
•
1 Parent(s): 9d72119

Training in progress, step 5000

Browse files
Files changed (18) hide show
  1. checkpoint-5000/config.json +28 -0
  2. checkpoint-5000/optimizer.pt +3 -0
  3. checkpoint-5000/pytorch_model.bin +3 -0
  4. checkpoint-5000/rng_state_0.pth +3 -0
  5. checkpoint-5000/rng_state_1.pth +3 -0
  6. checkpoint-5000/rng_state_2.pth +3 -0
  7. checkpoint-5000/rng_state_3.pth +3 -0
  8. checkpoint-5000/rng_state_4.pth +3 -0
  9. checkpoint-5000/rng_state_5.pth +3 -0
  10. checkpoint-5000/rng_state_6.pth +3 -0
  11. checkpoint-5000/rng_state_7.pth +3 -0
  12. checkpoint-5000/scheduler.pt +3 -0
  13. checkpoint-5000/special_tokens_map.json +7 -0
  14. checkpoint-5000/tokenizer.json +0 -0
  15. checkpoint-5000/tokenizer_config.json +5 -0
  16. checkpoint-5000/trainer_state.json +4516 -0
  17. checkpoint-5000/training_args.bin +3 -0
  18. pytorch_model.bin +1 -1
checkpoint-5000/config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "MarindaForMaskedLMWithEdgePrediction"
4
+ ],
5
+ "attention_probs_dropout_prob": 0.1,
6
+ "bos_token_id": 1,
7
+ "classifier_dropout": null,
8
+ "eos_token_id": 2,
9
+ "ep_add_linear_projection": true,
10
+ "hidden_act": "gelu",
11
+ "hidden_dropout_prob": 0.1,
12
+ "hidden_size": 768,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 3072,
15
+ "layer_norm_eps": 1e-12,
16
+ "max_position_embeddings": 600,
17
+ "max_relative_position_embeddings": 8,
18
+ "model_type": "marinda",
19
+ "num_attention_heads": 12,
20
+ "num_hidden_layers": 12,
21
+ "pad_token_id": 3,
22
+ "position_embedding_type": "mixed",
23
+ "torch_dtype": "float32",
24
+ "transformers_version": "4.31.0",
25
+ "type_vocab_size": 2,
26
+ "use_cache": true,
27
+ "vocab_size": 30523
28
+ }
checkpoint-5000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f6ca737f893390a26eab6e00ea95086d2cda7c2be42b185d2367c4118925271d
3
+ size 881521797
checkpoint-5000/pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4e0d79bac895d479a820e490ec436da2f0d185ae3706aabc74ded777034acf1
3
+ size 440772077
checkpoint-5000/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6ae40c994ec6bf61f0e33f13499e24cc992d717b4bc105edf40699d988da091a
3
+ size 21623
checkpoint-5000/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f552b9d7961d8948732f9453bf21483cb15233638c4463e3b25d7ed3f3a75ae
3
+ size 21623
checkpoint-5000/rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9adcce8600f831062f624c051348754c4f4bbd0ad2beaa61a7baa89beda85d89
3
+ size 21623
checkpoint-5000/rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e26bbc4a6662c6f97fd64c61aea2b3f2443de1181b6443691f05481f53e189f4
3
+ size 21623
checkpoint-5000/rng_state_4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1270126d552566e80999abca58635ca885f797aef72c89125d607871e51e8e9b
3
+ size 21623
checkpoint-5000/rng_state_5.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d00733855426db59685a7b4875efe9efe7663467f23b5dda1a1ecb381f628c94
3
+ size 21623
checkpoint-5000/rng_state_6.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3dfce21341972330611cc9ff45869766ccbc450a8bed1b706deb56555c2e18f6
3
+ size 21623
checkpoint-5000/rng_state_7.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb3d6cc8fdc9e3354a170f01206110c162902c1c7b4007cacaf9372dff319f85
3
+ size 21623
checkpoint-5000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a94cc660b7efe6d139f8c9147fd6ac27ccf24ed3d1daa03f9272a3dd607e1621
3
+ size 627
checkpoint-5000/special_tokens_map.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "cls_token": "[CLS]",
3
+ "mask_token": "[MASK]",
4
+ "pad_token": "[PAD]",
5
+ "sep_token": "[SEP]",
6
+ "unk_token": "[UNK]"
7
+ }
checkpoint-5000/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
checkpoint-5000/tokenizer_config.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "clean_up_tokenization_spaces": true,
3
+ "model_max_length": 512,
4
+ "tokenizer_class": "MarindaTokenizer"
5
+ }
checkpoint-5000/trainer_state.json ADDED
@@ -0,0 +1,4516 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 3.8565368299267258,
5
+ "global_step": 5000,
6
+ "is_hyper_param_search": false,
7
+ "is_local_process_zero": true,
8
+ "is_world_process_zero": true,
9
+ "log_history": [
10
+ {
11
+ "ep_loss": 0.0,
12
+ "epoch": 0.01,
13
+ "learning_rate": 5e-07,
14
+ "loss": 10.4763,
15
+ "mlm_loss": 10.4763,
16
+ "step": 10
17
+ },
18
+ {
19
+ "ep_loss": 0.0,
20
+ "epoch": 0.02,
21
+ "learning_rate": 1e-06,
22
+ "loss": 10.3052,
23
+ "mlm_loss": 10.3052,
24
+ "step": 20
25
+ },
26
+ {
27
+ "ep_loss": 0.0,
28
+ "epoch": 0.02,
29
+ "learning_rate": 1.5e-06,
30
+ "loss": 9.9641,
31
+ "mlm_loss": 9.9641,
32
+ "step": 30
33
+ },
34
+ {
35
+ "ep_loss": 0.0,
36
+ "epoch": 0.03,
37
+ "learning_rate": 2e-06,
38
+ "loss": 9.5354,
39
+ "mlm_loss": 9.5354,
40
+ "step": 40
41
+ },
42
+ {
43
+ "ep_loss": 0.0,
44
+ "epoch": 0.04,
45
+ "learning_rate": 2.5e-06,
46
+ "loss": 9.1346,
47
+ "mlm_loss": 9.1346,
48
+ "step": 50
49
+ },
50
+ {
51
+ "ep_loss": 0.0,
52
+ "epoch": 0.05,
53
+ "learning_rate": 3e-06,
54
+ "loss": 8.8162,
55
+ "mlm_loss": 8.8162,
56
+ "step": 60
57
+ },
58
+ {
59
+ "ep_loss": 0.0,
60
+ "epoch": 0.05,
61
+ "learning_rate": 3.5e-06,
62
+ "loss": 8.558,
63
+ "mlm_loss": 8.558,
64
+ "step": 70
65
+ },
66
+ {
67
+ "ep_loss": 0.0,
68
+ "epoch": 0.06,
69
+ "learning_rate": 4e-06,
70
+ "loss": 8.3459,
71
+ "mlm_loss": 8.3459,
72
+ "step": 80
73
+ },
74
+ {
75
+ "ep_loss": 0.0,
76
+ "epoch": 0.07,
77
+ "learning_rate": 4.5e-06,
78
+ "loss": 8.175,
79
+ "mlm_loss": 8.175,
80
+ "step": 90
81
+ },
82
+ {
83
+ "ep_loss": 0.0,
84
+ "epoch": 0.08,
85
+ "learning_rate": 5e-06,
86
+ "loss": 8.0255,
87
+ "mlm_loss": 8.0255,
88
+ "step": 100
89
+ },
90
+ {
91
+ "epoch": 0.08,
92
+ "eval_ep_loss": -0.14438027143478394,
93
+ "eval_loss": 7.826408386230469,
94
+ "eval_mlm_loss": 7.826408386230469,
95
+ "eval_runtime": 63.5165,
96
+ "eval_samples_per_second": 1100.014,
97
+ "eval_steps_per_second": 0.551,
98
+ "step": 100
99
+ },
100
+ {
101
+ "ep_loss": 0.0,
102
+ "epoch": 0.08,
103
+ "learning_rate": 5.5e-06,
104
+ "loss": 7.8979,
105
+ "mlm_loss": 7.8979,
106
+ "step": 110
107
+ },
108
+ {
109
+ "ep_loss": 0.0,
110
+ "epoch": 0.09,
111
+ "learning_rate": 6e-06,
112
+ "loss": 7.7543,
113
+ "mlm_loss": 7.7543,
114
+ "step": 120
115
+ },
116
+ {
117
+ "ep_loss": 0.0,
118
+ "epoch": 0.1,
119
+ "learning_rate": 6.5e-06,
120
+ "loss": 7.5942,
121
+ "mlm_loss": 7.5942,
122
+ "step": 130
123
+ },
124
+ {
125
+ "ep_loss": 0.0,
126
+ "epoch": 0.11,
127
+ "learning_rate": 7e-06,
128
+ "loss": 7.4304,
129
+ "mlm_loss": 7.4304,
130
+ "step": 140
131
+ },
132
+ {
133
+ "ep_loss": 0.0,
134
+ "epoch": 0.12,
135
+ "learning_rate": 7.5e-06,
136
+ "loss": 7.1862,
137
+ "mlm_loss": 7.1862,
138
+ "step": 150
139
+ },
140
+ {
141
+ "ep_loss": 0.0,
142
+ "epoch": 0.12,
143
+ "learning_rate": 8e-06,
144
+ "loss": 6.924,
145
+ "mlm_loss": 6.924,
146
+ "step": 160
147
+ },
148
+ {
149
+ "ep_loss": 0.0,
150
+ "epoch": 0.13,
151
+ "learning_rate": 8.500000000000002e-06,
152
+ "loss": 6.6521,
153
+ "mlm_loss": 6.6521,
154
+ "step": 170
155
+ },
156
+ {
157
+ "ep_loss": 0.0,
158
+ "epoch": 0.14,
159
+ "learning_rate": 9e-06,
160
+ "loss": 6.3639,
161
+ "mlm_loss": 6.3639,
162
+ "step": 180
163
+ },
164
+ {
165
+ "ep_loss": 0.0,
166
+ "epoch": 0.15,
167
+ "learning_rate": 9.5e-06,
168
+ "loss": 6.0915,
169
+ "mlm_loss": 6.0915,
170
+ "step": 190
171
+ },
172
+ {
173
+ "ep_loss": 0.0,
174
+ "epoch": 0.15,
175
+ "learning_rate": 1e-05,
176
+ "loss": 5.8272,
177
+ "mlm_loss": 5.8272,
178
+ "step": 200
179
+ },
180
+ {
181
+ "epoch": 0.15,
182
+ "eval_ep_loss": -0.4913683831691742,
183
+ "eval_loss": 5.531888008117676,
184
+ "eval_mlm_loss": 5.531888008117676,
185
+ "eval_runtime": 60.9938,
186
+ "eval_samples_per_second": 1145.51,
187
+ "eval_steps_per_second": 0.574,
188
+ "step": 200
189
+ },
190
+ {
191
+ "ep_loss": 0.0,
192
+ "epoch": 0.16,
193
+ "learning_rate": 1.0500000000000001e-05,
194
+ "loss": 5.5749,
195
+ "mlm_loss": 5.5749,
196
+ "step": 210
197
+ },
198
+ {
199
+ "ep_loss": 0.0,
200
+ "epoch": 0.17,
201
+ "learning_rate": 1.1e-05,
202
+ "loss": 5.3413,
203
+ "mlm_loss": 5.3413,
204
+ "step": 220
205
+ },
206
+ {
207
+ "ep_loss": 0.0,
208
+ "epoch": 0.18,
209
+ "learning_rate": 1.15e-05,
210
+ "loss": 5.142,
211
+ "mlm_loss": 5.142,
212
+ "step": 230
213
+ },
214
+ {
215
+ "ep_loss": 0.0,
216
+ "epoch": 0.19,
217
+ "learning_rate": 1.2e-05,
218
+ "loss": 4.925,
219
+ "mlm_loss": 4.925,
220
+ "step": 240
221
+ },
222
+ {
223
+ "ep_loss": 0.0,
224
+ "epoch": 0.19,
225
+ "learning_rate": 1.25e-05,
226
+ "loss": 4.7334,
227
+ "mlm_loss": 4.7334,
228
+ "step": 250
229
+ },
230
+ {
231
+ "ep_loss": 0.0,
232
+ "epoch": 0.2,
233
+ "learning_rate": 1.3e-05,
234
+ "loss": 4.5702,
235
+ "mlm_loss": 4.5702,
236
+ "step": 260
237
+ },
238
+ {
239
+ "ep_loss": 0.0,
240
+ "epoch": 0.21,
241
+ "learning_rate": 1.35e-05,
242
+ "loss": 4.4388,
243
+ "mlm_loss": 4.4388,
244
+ "step": 270
245
+ },
246
+ {
247
+ "ep_loss": 0.0,
248
+ "epoch": 0.22,
249
+ "learning_rate": 1.4e-05,
250
+ "loss": 4.288,
251
+ "mlm_loss": 4.288,
252
+ "step": 280
253
+ },
254
+ {
255
+ "ep_loss": 0.0,
256
+ "epoch": 0.22,
257
+ "learning_rate": 1.4500000000000002e-05,
258
+ "loss": 4.1403,
259
+ "mlm_loss": 4.1403,
260
+ "step": 290
261
+ },
262
+ {
263
+ "ep_loss": 0.0,
264
+ "epoch": 0.23,
265
+ "learning_rate": 1.5e-05,
266
+ "loss": 4.0232,
267
+ "mlm_loss": 4.0232,
268
+ "step": 300
269
+ },
270
+ {
271
+ "epoch": 0.23,
272
+ "eval_ep_loss": -1.1010401248931885,
273
+ "eval_loss": 3.8623688220977783,
274
+ "eval_mlm_loss": 3.8623688220977783,
275
+ "eval_runtime": 61.8594,
276
+ "eval_samples_per_second": 1129.481,
277
+ "eval_steps_per_second": 0.566,
278
+ "step": 300
279
+ },
280
+ {
281
+ "ep_loss": 0.0,
282
+ "epoch": 0.24,
283
+ "learning_rate": 1.55e-05,
284
+ "loss": 3.9135,
285
+ "mlm_loss": 3.9135,
286
+ "step": 310
287
+ },
288
+ {
289
+ "ep_loss": 0.0,
290
+ "epoch": 0.25,
291
+ "learning_rate": 1.6e-05,
292
+ "loss": 3.8252,
293
+ "mlm_loss": 3.8252,
294
+ "step": 320
295
+ },
296
+ {
297
+ "ep_loss": 0.0,
298
+ "epoch": 0.25,
299
+ "learning_rate": 1.65e-05,
300
+ "loss": 3.7462,
301
+ "mlm_loss": 3.7462,
302
+ "step": 330
303
+ },
304
+ {
305
+ "ep_loss": 0.0,
306
+ "epoch": 0.26,
307
+ "learning_rate": 1.7000000000000003e-05,
308
+ "loss": 3.6491,
309
+ "mlm_loss": 3.6491,
310
+ "step": 340
311
+ },
312
+ {
313
+ "ep_loss": 0.0,
314
+ "epoch": 0.27,
315
+ "learning_rate": 1.7500000000000002e-05,
316
+ "loss": 3.5815,
317
+ "mlm_loss": 3.5815,
318
+ "step": 350
319
+ },
320
+ {
321
+ "ep_loss": 0.0,
322
+ "epoch": 0.28,
323
+ "learning_rate": 1.8e-05,
324
+ "loss": 3.5211,
325
+ "mlm_loss": 3.5211,
326
+ "step": 360
327
+ },
328
+ {
329
+ "ep_loss": 0.0,
330
+ "epoch": 0.29,
331
+ "learning_rate": 1.85e-05,
332
+ "loss": 3.452,
333
+ "mlm_loss": 3.452,
334
+ "step": 370
335
+ },
336
+ {
337
+ "ep_loss": 0.0,
338
+ "epoch": 0.29,
339
+ "learning_rate": 1.9e-05,
340
+ "loss": 3.3747,
341
+ "mlm_loss": 3.3747,
342
+ "step": 380
343
+ },
344
+ {
345
+ "ep_loss": 0.0,
346
+ "epoch": 0.3,
347
+ "learning_rate": 1.95e-05,
348
+ "loss": 3.3163,
349
+ "mlm_loss": 3.3163,
350
+ "step": 390
351
+ },
352
+ {
353
+ "ep_loss": 0.0,
354
+ "epoch": 0.31,
355
+ "learning_rate": 2e-05,
356
+ "loss": 3.2679,
357
+ "mlm_loss": 3.2679,
358
+ "step": 400
359
+ },
360
+ {
361
+ "epoch": 0.31,
362
+ "eval_ep_loss": -1.4546582698822021,
363
+ "eval_loss": 3.1360955238342285,
364
+ "eval_mlm_loss": 3.1360955238342285,
365
+ "eval_runtime": 59.2888,
366
+ "eval_samples_per_second": 1178.451,
367
+ "eval_steps_per_second": 0.59,
368
+ "step": 400
369
+ },
370
+ {
371
+ "ep_loss": 0.0,
372
+ "epoch": 0.32,
373
+ "learning_rate": 2.05e-05,
374
+ "loss": 3.1831,
375
+ "mlm_loss": 3.1831,
376
+ "step": 410
377
+ },
378
+ {
379
+ "ep_loss": 0.0,
380
+ "epoch": 0.32,
381
+ "learning_rate": 2.1000000000000002e-05,
382
+ "loss": 3.1246,
383
+ "mlm_loss": 3.1246,
384
+ "step": 420
385
+ },
386
+ {
387
+ "ep_loss": 0.0,
388
+ "epoch": 0.33,
389
+ "learning_rate": 2.1499999999999997e-05,
390
+ "loss": 3.0658,
391
+ "mlm_loss": 3.0658,
392
+ "step": 430
393
+ },
394
+ {
395
+ "ep_loss": 0.0,
396
+ "epoch": 0.34,
397
+ "learning_rate": 2.2e-05,
398
+ "loss": 3.0032,
399
+ "mlm_loss": 3.0032,
400
+ "step": 440
401
+ },
402
+ {
403
+ "ep_loss": 0.0,
404
+ "epoch": 0.35,
405
+ "learning_rate": 2.2499999999999998e-05,
406
+ "loss": 2.931,
407
+ "mlm_loss": 2.931,
408
+ "step": 450
409
+ },
410
+ {
411
+ "ep_loss": 0.0,
412
+ "epoch": 0.35,
413
+ "learning_rate": 2.3e-05,
414
+ "loss": 2.904,
415
+ "mlm_loss": 2.904,
416
+ "step": 460
417
+ },
418
+ {
419
+ "ep_loss": 0.0,
420
+ "epoch": 0.36,
421
+ "learning_rate": 2.3500000000000002e-05,
422
+ "loss": 2.8526,
423
+ "mlm_loss": 2.8526,
424
+ "step": 470
425
+ },
426
+ {
427
+ "ep_loss": 0.0,
428
+ "epoch": 0.37,
429
+ "learning_rate": 2.4e-05,
430
+ "loss": 2.8336,
431
+ "mlm_loss": 2.8336,
432
+ "step": 480
433
+ },
434
+ {
435
+ "ep_loss": 0.0,
436
+ "epoch": 0.38,
437
+ "learning_rate": 2.4500000000000003e-05,
438
+ "loss": 2.7953,
439
+ "mlm_loss": 2.7953,
440
+ "step": 490
441
+ },
442
+ {
443
+ "ep_loss": 0.0,
444
+ "epoch": 0.39,
445
+ "learning_rate": 2.5e-05,
446
+ "loss": 2.7569,
447
+ "mlm_loss": 2.7569,
448
+ "step": 500
449
+ },
450
+ {
451
+ "epoch": 0.39,
452
+ "eval_ep_loss": -1.5984127521514893,
453
+ "eval_loss": 2.6636624336242676,
454
+ "eval_mlm_loss": 2.6636624336242676,
455
+ "eval_runtime": 64.3177,
456
+ "eval_samples_per_second": 1086.31,
457
+ "eval_steps_per_second": 0.544,
458
+ "step": 500
459
+ },
460
+ {
461
+ "ep_loss": 0.0,
462
+ "epoch": 0.39,
463
+ "learning_rate": 2.55e-05,
464
+ "loss": 2.7157,
465
+ "mlm_loss": 2.7157,
466
+ "step": 510
467
+ },
468
+ {
469
+ "ep_loss": 0.0,
470
+ "epoch": 0.4,
471
+ "learning_rate": 2.6e-05,
472
+ "loss": 2.6857,
473
+ "mlm_loss": 2.6857,
474
+ "step": 520
475
+ },
476
+ {
477
+ "ep_loss": 0.0,
478
+ "epoch": 0.41,
479
+ "learning_rate": 2.65e-05,
480
+ "loss": 2.6796,
481
+ "mlm_loss": 2.6796,
482
+ "step": 530
483
+ },
484
+ {
485
+ "ep_loss": 0.0,
486
+ "epoch": 0.42,
487
+ "learning_rate": 2.7e-05,
488
+ "loss": 2.6449,
489
+ "mlm_loss": 2.6449,
490
+ "step": 540
491
+ },
492
+ {
493
+ "ep_loss": 0.0,
494
+ "epoch": 0.42,
495
+ "learning_rate": 2.75e-05,
496
+ "loss": 2.5932,
497
+ "mlm_loss": 2.5932,
498
+ "step": 550
499
+ },
500
+ {
501
+ "ep_loss": 0.0,
502
+ "epoch": 0.43,
503
+ "learning_rate": 2.8e-05,
504
+ "loss": 2.5536,
505
+ "mlm_loss": 2.5536,
506
+ "step": 560
507
+ },
508
+ {
509
+ "ep_loss": 0.0,
510
+ "epoch": 0.44,
511
+ "learning_rate": 2.85e-05,
512
+ "loss": 2.5271,
513
+ "mlm_loss": 2.5271,
514
+ "step": 570
515
+ },
516
+ {
517
+ "ep_loss": 0.0,
518
+ "epoch": 0.45,
519
+ "learning_rate": 2.9000000000000004e-05,
520
+ "loss": 2.5057,
521
+ "mlm_loss": 2.5057,
522
+ "step": 580
523
+ },
524
+ {
525
+ "ep_loss": 0.0,
526
+ "epoch": 0.46,
527
+ "learning_rate": 2.95e-05,
528
+ "loss": 2.476,
529
+ "mlm_loss": 2.476,
530
+ "step": 590
531
+ },
532
+ {
533
+ "ep_loss": 0.0,
534
+ "epoch": 0.46,
535
+ "learning_rate": 3e-05,
536
+ "loss": 2.4265,
537
+ "mlm_loss": 2.4265,
538
+ "step": 600
539
+ },
540
+ {
541
+ "epoch": 0.46,
542
+ "eval_ep_loss": -1.7573626041412354,
543
+ "eval_loss": 2.359508752822876,
544
+ "eval_mlm_loss": 2.359508752822876,
545
+ "eval_runtime": 59.8647,
546
+ "eval_samples_per_second": 1167.114,
547
+ "eval_steps_per_second": 0.585,
548
+ "step": 600
549
+ },
550
+ {
551
+ "ep_loss": 0.0,
552
+ "epoch": 0.47,
553
+ "learning_rate": 3.05e-05,
554
+ "loss": 2.4061,
555
+ "mlm_loss": 2.4061,
556
+ "step": 610
557
+ },
558
+ {
559
+ "ep_loss": 0.0,
560
+ "epoch": 0.48,
561
+ "learning_rate": 3.1e-05,
562
+ "loss": 2.4071,
563
+ "mlm_loss": 2.4071,
564
+ "step": 620
565
+ },
566
+ {
567
+ "ep_loss": 0.0,
568
+ "epoch": 0.49,
569
+ "learning_rate": 3.15e-05,
570
+ "loss": 2.3728,
571
+ "mlm_loss": 2.3728,
572
+ "step": 630
573
+ },
574
+ {
575
+ "ep_loss": 0.0,
576
+ "epoch": 0.49,
577
+ "learning_rate": 3.2e-05,
578
+ "loss": 2.3465,
579
+ "mlm_loss": 2.3465,
580
+ "step": 640
581
+ },
582
+ {
583
+ "ep_loss": 0.0,
584
+ "epoch": 0.5,
585
+ "learning_rate": 3.2500000000000004e-05,
586
+ "loss": 2.3065,
587
+ "mlm_loss": 2.3065,
588
+ "step": 650
589
+ },
590
+ {
591
+ "ep_loss": 0.0,
592
+ "epoch": 0.51,
593
+ "learning_rate": 3.3e-05,
594
+ "loss": 2.312,
595
+ "mlm_loss": 2.312,
596
+ "step": 660
597
+ },
598
+ {
599
+ "ep_loss": 0.0,
600
+ "epoch": 0.52,
601
+ "learning_rate": 3.35e-05,
602
+ "loss": 2.29,
603
+ "mlm_loss": 2.29,
604
+ "step": 670
605
+ },
606
+ {
607
+ "ep_loss": 0.0,
608
+ "epoch": 0.52,
609
+ "learning_rate": 3.4000000000000007e-05,
610
+ "loss": 2.2939,
611
+ "mlm_loss": 2.2939,
612
+ "step": 680
613
+ },
614
+ {
615
+ "ep_loss": 0.0,
616
+ "epoch": 0.53,
617
+ "learning_rate": 3.4500000000000005e-05,
618
+ "loss": 2.2496,
619
+ "mlm_loss": 2.2496,
620
+ "step": 690
621
+ },
622
+ {
623
+ "ep_loss": 0.0,
624
+ "epoch": 0.54,
625
+ "learning_rate": 3.5000000000000004e-05,
626
+ "loss": 2.2482,
627
+ "mlm_loss": 2.2482,
628
+ "step": 700
629
+ },
630
+ {
631
+ "epoch": 0.54,
632
+ "eval_ep_loss": -1.7850463390350342,
633
+ "eval_loss": 2.1652746200561523,
634
+ "eval_mlm_loss": 2.1652746200561523,
635
+ "eval_runtime": 63.4238,
636
+ "eval_samples_per_second": 1101.621,
637
+ "eval_steps_per_second": 0.552,
638
+ "step": 700
639
+ },
640
+ {
641
+ "ep_loss": 0.0,
642
+ "epoch": 0.55,
643
+ "learning_rate": 3.5499999999999996e-05,
644
+ "loss": 2.2379,
645
+ "mlm_loss": 2.2379,
646
+ "step": 710
647
+ },
648
+ {
649
+ "ep_loss": 0.0,
650
+ "epoch": 0.56,
651
+ "learning_rate": 3.6e-05,
652
+ "loss": 2.2088,
653
+ "mlm_loss": 2.2088,
654
+ "step": 720
655
+ },
656
+ {
657
+ "ep_loss": 0.0,
658
+ "epoch": 0.56,
659
+ "learning_rate": 3.65e-05,
660
+ "loss": 2.1767,
661
+ "mlm_loss": 2.1767,
662
+ "step": 730
663
+ },
664
+ {
665
+ "ep_loss": 0.0,
666
+ "epoch": 0.57,
667
+ "learning_rate": 3.7e-05,
668
+ "loss": 2.1589,
669
+ "mlm_loss": 2.1589,
670
+ "step": 740
671
+ },
672
+ {
673
+ "ep_loss": 0.0,
674
+ "epoch": 0.58,
675
+ "learning_rate": 3.75e-05,
676
+ "loss": 2.1297,
677
+ "mlm_loss": 2.1297,
678
+ "step": 750
679
+ },
680
+ {
681
+ "ep_loss": 0.0,
682
+ "epoch": 0.59,
683
+ "learning_rate": 3.8e-05,
684
+ "loss": 2.1128,
685
+ "mlm_loss": 2.1128,
686
+ "step": 760
687
+ },
688
+ {
689
+ "ep_loss": 0.0,
690
+ "epoch": 0.59,
691
+ "learning_rate": 3.85e-05,
692
+ "loss": 2.0861,
693
+ "mlm_loss": 2.0861,
694
+ "step": 770
695
+ },
696
+ {
697
+ "ep_loss": 0.0,
698
+ "epoch": 0.6,
699
+ "learning_rate": 3.9e-05,
700
+ "loss": 2.0557,
701
+ "mlm_loss": 2.0557,
702
+ "step": 780
703
+ },
704
+ {
705
+ "ep_loss": 0.0,
706
+ "epoch": 0.61,
707
+ "learning_rate": 3.95e-05,
708
+ "loss": 2.0596,
709
+ "mlm_loss": 2.0596,
710
+ "step": 790
711
+ },
712
+ {
713
+ "ep_loss": 0.0,
714
+ "epoch": 0.62,
715
+ "learning_rate": 4e-05,
716
+ "loss": 2.0349,
717
+ "mlm_loss": 2.0349,
718
+ "step": 800
719
+ },
720
+ {
721
+ "epoch": 0.62,
722
+ "eval_ep_loss": -1.7692768573760986,
723
+ "eval_loss": 1.916082501411438,
724
+ "eval_mlm_loss": 1.916082501411438,
725
+ "eval_runtime": 62.7772,
726
+ "eval_samples_per_second": 1112.968,
727
+ "eval_steps_per_second": 0.558,
728
+ "step": 800
729
+ },
730
+ {
731
+ "ep_loss": 0.0,
732
+ "epoch": 0.62,
733
+ "learning_rate": 4.05e-05,
734
+ "loss": 1.9934,
735
+ "mlm_loss": 1.9934,
736
+ "step": 810
737
+ },
738
+ {
739
+ "ep_loss": 0.0,
740
+ "epoch": 0.63,
741
+ "learning_rate": 4.1e-05,
742
+ "loss": 1.9554,
743
+ "mlm_loss": 1.9554,
744
+ "step": 820
745
+ },
746
+ {
747
+ "ep_loss": 0.0,
748
+ "epoch": 0.64,
749
+ "learning_rate": 4.1500000000000006e-05,
750
+ "loss": 1.9348,
751
+ "mlm_loss": 1.9348,
752
+ "step": 830
753
+ },
754
+ {
755
+ "ep_loss": 0.0,
756
+ "epoch": 0.65,
757
+ "learning_rate": 4.2000000000000004e-05,
758
+ "loss": 1.8978,
759
+ "mlm_loss": 1.8978,
760
+ "step": 840
761
+ },
762
+ {
763
+ "ep_loss": 0.0,
764
+ "epoch": 0.66,
765
+ "learning_rate": 4.25e-05,
766
+ "loss": 1.8752,
767
+ "mlm_loss": 1.8752,
768
+ "step": 850
769
+ },
770
+ {
771
+ "ep_loss": 0.0,
772
+ "epoch": 0.66,
773
+ "learning_rate": 4.2999999999999995e-05,
774
+ "loss": 1.8466,
775
+ "mlm_loss": 1.8466,
776
+ "step": 860
777
+ },
778
+ {
779
+ "ep_loss": 0.0,
780
+ "epoch": 0.67,
781
+ "learning_rate": 4.35e-05,
782
+ "loss": 1.8168,
783
+ "mlm_loss": 1.8168,
784
+ "step": 870
785
+ },
786
+ {
787
+ "ep_loss": 0.0,
788
+ "epoch": 0.68,
789
+ "learning_rate": 4.4e-05,
790
+ "loss": 1.8161,
791
+ "mlm_loss": 1.8161,
792
+ "step": 880
793
+ },
794
+ {
795
+ "ep_loss": 0.0,
796
+ "epoch": 0.69,
797
+ "learning_rate": 4.45e-05,
798
+ "loss": 1.7637,
799
+ "mlm_loss": 1.7637,
800
+ "step": 890
801
+ },
802
+ {
803
+ "ep_loss": 0.0,
804
+ "epoch": 0.69,
805
+ "learning_rate": 4.4999999999999996e-05,
806
+ "loss": 1.7292,
807
+ "mlm_loss": 1.7292,
808
+ "step": 900
809
+ },
810
+ {
811
+ "epoch": 0.69,
812
+ "eval_ep_loss": -1.673869252204895,
813
+ "eval_loss": 1.6480634212493896,
814
+ "eval_mlm_loss": 1.6480634212493896,
815
+ "eval_runtime": 61.0031,
816
+ "eval_samples_per_second": 1145.335,
817
+ "eval_steps_per_second": 0.574,
818
+ "step": 900
819
+ },
820
+ {
821
+ "ep_loss": 0.0,
822
+ "epoch": 0.7,
823
+ "learning_rate": 4.55e-05,
824
+ "loss": 1.7031,
825
+ "mlm_loss": 1.7031,
826
+ "step": 910
827
+ },
828
+ {
829
+ "ep_loss": 0.0,
830
+ "epoch": 0.71,
831
+ "learning_rate": 4.6e-05,
832
+ "loss": 1.6951,
833
+ "mlm_loss": 1.6951,
834
+ "step": 920
835
+ },
836
+ {
837
+ "ep_loss": 0.0,
838
+ "epoch": 0.72,
839
+ "learning_rate": 4.65e-05,
840
+ "loss": 1.6706,
841
+ "mlm_loss": 1.6706,
842
+ "step": 930
843
+ },
844
+ {
845
+ "ep_loss": 0.0,
846
+ "epoch": 0.73,
847
+ "learning_rate": 4.7000000000000004e-05,
848
+ "loss": 1.6674,
849
+ "mlm_loss": 1.6674,
850
+ "step": 940
851
+ },
852
+ {
853
+ "ep_loss": 0.0,
854
+ "epoch": 0.73,
855
+ "learning_rate": 4.75e-05,
856
+ "loss": 1.6446,
857
+ "mlm_loss": 1.6446,
858
+ "step": 950
859
+ },
860
+ {
861
+ "ep_loss": 0.0,
862
+ "epoch": 0.74,
863
+ "learning_rate": 4.8e-05,
864
+ "loss": 1.6344,
865
+ "mlm_loss": 1.6344,
866
+ "step": 960
867
+ },
868
+ {
869
+ "ep_loss": 0.0,
870
+ "epoch": 0.75,
871
+ "learning_rate": 4.85e-05,
872
+ "loss": 1.6182,
873
+ "mlm_loss": 1.6182,
874
+ "step": 970
875
+ },
876
+ {
877
+ "ep_loss": 0.0,
878
+ "epoch": 0.76,
879
+ "learning_rate": 4.9000000000000005e-05,
880
+ "loss": 1.602,
881
+ "mlm_loss": 1.602,
882
+ "step": 980
883
+ },
884
+ {
885
+ "ep_loss": 0.0,
886
+ "epoch": 0.76,
887
+ "learning_rate": 4.9500000000000004e-05,
888
+ "loss": 1.5965,
889
+ "mlm_loss": 1.5965,
890
+ "step": 990
891
+ },
892
+ {
893
+ "ep_loss": 0.0,
894
+ "epoch": 0.77,
895
+ "learning_rate": 5e-05,
896
+ "loss": 1.6022,
897
+ "mlm_loss": 1.6022,
898
+ "step": 1000
899
+ },
900
+ {
901
+ "epoch": 0.77,
902
+ "eval_ep_loss": -1.7339978218078613,
903
+ "eval_loss": 1.521234393119812,
904
+ "eval_mlm_loss": 1.521234393119812,
905
+ "eval_runtime": 61.977,
906
+ "eval_samples_per_second": 1127.338,
907
+ "eval_steps_per_second": 0.565,
908
+ "step": 1000
909
+ },
910
+ {
911
+ "ep_loss": 0.0,
912
+ "epoch": 0.78,
913
+ "learning_rate": 5.05e-05,
914
+ "loss": 1.5856,
915
+ "mlm_loss": 1.5856,
916
+ "step": 1010
917
+ },
918
+ {
919
+ "ep_loss": 0.0,
920
+ "epoch": 0.79,
921
+ "learning_rate": 5.1e-05,
922
+ "loss": 1.5641,
923
+ "mlm_loss": 1.5641,
924
+ "step": 1020
925
+ },
926
+ {
927
+ "ep_loss": 0.0,
928
+ "epoch": 0.79,
929
+ "learning_rate": 5.15e-05,
930
+ "loss": 1.5455,
931
+ "mlm_loss": 1.5455,
932
+ "step": 1030
933
+ },
934
+ {
935
+ "ep_loss": 0.0,
936
+ "epoch": 0.8,
937
+ "learning_rate": 5.2e-05,
938
+ "loss": 1.5587,
939
+ "mlm_loss": 1.5587,
940
+ "step": 1040
941
+ },
942
+ {
943
+ "ep_loss": 0.0,
944
+ "epoch": 0.81,
945
+ "learning_rate": 5.25e-05,
946
+ "loss": 1.5495,
947
+ "mlm_loss": 1.5495,
948
+ "step": 1050
949
+ },
950
+ {
951
+ "ep_loss": 0.0,
952
+ "epoch": 0.82,
953
+ "learning_rate": 5.3e-05,
954
+ "loss": 1.5388,
955
+ "mlm_loss": 1.5388,
956
+ "step": 1060
957
+ },
958
+ {
959
+ "ep_loss": 0.0,
960
+ "epoch": 0.83,
961
+ "learning_rate": 5.35e-05,
962
+ "loss": 1.5085,
963
+ "mlm_loss": 1.5085,
964
+ "step": 1070
965
+ },
966
+ {
967
+ "ep_loss": 0.0,
968
+ "epoch": 0.83,
969
+ "learning_rate": 5.4e-05,
970
+ "loss": 1.5258,
971
+ "mlm_loss": 1.5258,
972
+ "step": 1080
973
+ },
974
+ {
975
+ "ep_loss": 0.0,
976
+ "epoch": 0.84,
977
+ "learning_rate": 5.45e-05,
978
+ "loss": 1.5027,
979
+ "mlm_loss": 1.5027,
980
+ "step": 1090
981
+ },
982
+ {
983
+ "ep_loss": 0.0,
984
+ "epoch": 0.85,
985
+ "learning_rate": 5.5e-05,
986
+ "loss": 1.4991,
987
+ "mlm_loss": 1.4991,
988
+ "step": 1100
989
+ },
990
+ {
991
+ "epoch": 0.85,
992
+ "eval_ep_loss": -1.8915724754333496,
993
+ "eval_loss": 1.4428980350494385,
994
+ "eval_mlm_loss": 1.4428980350494385,
995
+ "eval_runtime": 60.2275,
996
+ "eval_samples_per_second": 1160.085,
997
+ "eval_steps_per_second": 0.581,
998
+ "step": 1100
999
+ },
1000
+ {
1001
+ "ep_loss": 0.0,
1002
+ "epoch": 0.86,
1003
+ "learning_rate": 5.55e-05,
1004
+ "loss": 1.5068,
1005
+ "mlm_loss": 1.5068,
1006
+ "step": 1110
1007
+ },
1008
+ {
1009
+ "ep_loss": 0.0,
1010
+ "epoch": 0.86,
1011
+ "learning_rate": 5.6e-05,
1012
+ "loss": 1.4812,
1013
+ "mlm_loss": 1.4812,
1014
+ "step": 1120
1015
+ },
1016
+ {
1017
+ "ep_loss": 0.0,
1018
+ "epoch": 0.87,
1019
+ "learning_rate": 5.6500000000000005e-05,
1020
+ "loss": 1.4765,
1021
+ "mlm_loss": 1.4765,
1022
+ "step": 1130
1023
+ },
1024
+ {
1025
+ "ep_loss": 0.0,
1026
+ "epoch": 0.88,
1027
+ "learning_rate": 5.7e-05,
1028
+ "loss": 1.4679,
1029
+ "mlm_loss": 1.4679,
1030
+ "step": 1140
1031
+ },
1032
+ {
1033
+ "ep_loss": 0.0,
1034
+ "epoch": 0.89,
1035
+ "learning_rate": 5.75e-05,
1036
+ "loss": 1.4631,
1037
+ "mlm_loss": 1.4631,
1038
+ "step": 1150
1039
+ },
1040
+ {
1041
+ "ep_loss": 0.0,
1042
+ "epoch": 0.89,
1043
+ "learning_rate": 5.800000000000001e-05,
1044
+ "loss": 1.4562,
1045
+ "mlm_loss": 1.4562,
1046
+ "step": 1160
1047
+ },
1048
+ {
1049
+ "ep_loss": 0.0,
1050
+ "epoch": 0.9,
1051
+ "learning_rate": 5.8500000000000006e-05,
1052
+ "loss": 1.4542,
1053
+ "mlm_loss": 1.4542,
1054
+ "step": 1170
1055
+ },
1056
+ {
1057
+ "ep_loss": 0.0,
1058
+ "epoch": 0.91,
1059
+ "learning_rate": 5.9e-05,
1060
+ "loss": 1.4266,
1061
+ "mlm_loss": 1.4266,
1062
+ "step": 1180
1063
+ },
1064
+ {
1065
+ "ep_loss": 0.0,
1066
+ "epoch": 0.92,
1067
+ "learning_rate": 5.9499999999999996e-05,
1068
+ "loss": 1.4181,
1069
+ "mlm_loss": 1.4181,
1070
+ "step": 1190
1071
+ },
1072
+ {
1073
+ "ep_loss": 0.0,
1074
+ "epoch": 0.93,
1075
+ "learning_rate": 6e-05,
1076
+ "loss": 1.4133,
1077
+ "mlm_loss": 1.4133,
1078
+ "step": 1200
1079
+ },
1080
+ {
1081
+ "epoch": 0.93,
1082
+ "eval_ep_loss": -1.9453246593475342,
1083
+ "eval_loss": 1.3670274019241333,
1084
+ "eval_mlm_loss": 1.3670274019241333,
1085
+ "eval_runtime": 61.6667,
1086
+ "eval_samples_per_second": 1133.011,
1087
+ "eval_steps_per_second": 0.568,
1088
+ "step": 1200
1089
+ },
1090
+ {
1091
+ "ep_loss": 0.0,
1092
+ "epoch": 0.93,
1093
+ "learning_rate": 6.05e-05,
1094
+ "loss": 1.414,
1095
+ "mlm_loss": 1.414,
1096
+ "step": 1210
1097
+ },
1098
+ {
1099
+ "ep_loss": 0.0,
1100
+ "epoch": 0.94,
1101
+ "learning_rate": 6.1e-05,
1102
+ "loss": 1.4032,
1103
+ "mlm_loss": 1.4032,
1104
+ "step": 1220
1105
+ },
1106
+ {
1107
+ "ep_loss": 0.0,
1108
+ "epoch": 0.95,
1109
+ "learning_rate": 6.15e-05,
1110
+ "loss": 1.3958,
1111
+ "mlm_loss": 1.3958,
1112
+ "step": 1230
1113
+ },
1114
+ {
1115
+ "ep_loss": 0.0,
1116
+ "epoch": 0.96,
1117
+ "learning_rate": 6.2e-05,
1118
+ "loss": 1.3952,
1119
+ "mlm_loss": 1.3952,
1120
+ "step": 1240
1121
+ },
1122
+ {
1123
+ "ep_loss": 0.0,
1124
+ "epoch": 0.96,
1125
+ "learning_rate": 6.25e-05,
1126
+ "loss": 1.3829,
1127
+ "mlm_loss": 1.3829,
1128
+ "step": 1250
1129
+ },
1130
+ {
1131
+ "ep_loss": 0.0,
1132
+ "epoch": 0.97,
1133
+ "learning_rate": 6.3e-05,
1134
+ "loss": 1.3915,
1135
+ "mlm_loss": 1.3915,
1136
+ "step": 1260
1137
+ },
1138
+ {
1139
+ "ep_loss": 0.0,
1140
+ "epoch": 0.98,
1141
+ "learning_rate": 6.35e-05,
1142
+ "loss": 1.3881,
1143
+ "mlm_loss": 1.3881,
1144
+ "step": 1270
1145
+ },
1146
+ {
1147
+ "ep_loss": 0.0,
1148
+ "epoch": 0.99,
1149
+ "learning_rate": 6.4e-05,
1150
+ "loss": 1.3915,
1151
+ "mlm_loss": 1.3915,
1152
+ "step": 1280
1153
+ },
1154
+ {
1155
+ "ep_loss": 0.0,
1156
+ "epoch": 0.99,
1157
+ "learning_rate": 6.450000000000001e-05,
1158
+ "loss": 1.3728,
1159
+ "mlm_loss": 1.3728,
1160
+ "step": 1290
1161
+ },
1162
+ {
1163
+ "ep_loss": 0.0,
1164
+ "epoch": 1.0,
1165
+ "learning_rate": 6.500000000000001e-05,
1166
+ "loss": 1.3379,
1167
+ "mlm_loss": 1.3379,
1168
+ "step": 1300
1169
+ },
1170
+ {
1171
+ "epoch": 1.0,
1172
+ "eval_ep_loss": -2.041435480117798,
1173
+ "eval_loss": 1.2956568002700806,
1174
+ "eval_mlm_loss": 1.2956568002700806,
1175
+ "eval_runtime": 58.5906,
1176
+ "eval_samples_per_second": 1192.495,
1177
+ "eval_steps_per_second": 0.597,
1178
+ "step": 1300
1179
+ },
1180
+ {
1181
+ "ep_loss": 0.0,
1182
+ "epoch": 1.01,
1183
+ "learning_rate": 6.55e-05,
1184
+ "loss": 1.3488,
1185
+ "mlm_loss": 1.3488,
1186
+ "step": 1310
1187
+ },
1188
+ {
1189
+ "ep_loss": 0.0,
1190
+ "epoch": 1.02,
1191
+ "learning_rate": 6.6e-05,
1192
+ "loss": 1.3408,
1193
+ "mlm_loss": 1.3408,
1194
+ "step": 1320
1195
+ },
1196
+ {
1197
+ "ep_loss": 0.0,
1198
+ "epoch": 1.03,
1199
+ "learning_rate": 6.65e-05,
1200
+ "loss": 1.3318,
1201
+ "mlm_loss": 1.3318,
1202
+ "step": 1330
1203
+ },
1204
+ {
1205
+ "ep_loss": 0.0,
1206
+ "epoch": 1.03,
1207
+ "learning_rate": 6.7e-05,
1208
+ "loss": 1.3216,
1209
+ "mlm_loss": 1.3216,
1210
+ "step": 1340
1211
+ },
1212
+ {
1213
+ "ep_loss": 0.0,
1214
+ "epoch": 1.04,
1215
+ "learning_rate": 6.75e-05,
1216
+ "loss": 1.3142,
1217
+ "mlm_loss": 1.3142,
1218
+ "step": 1350
1219
+ },
1220
+ {
1221
+ "ep_loss": 0.0,
1222
+ "epoch": 1.05,
1223
+ "learning_rate": 6.800000000000001e-05,
1224
+ "loss": 1.3075,
1225
+ "mlm_loss": 1.3075,
1226
+ "step": 1360
1227
+ },
1228
+ {
1229
+ "ep_loss": 0.0,
1230
+ "epoch": 1.06,
1231
+ "learning_rate": 6.850000000000001e-05,
1232
+ "loss": 1.312,
1233
+ "mlm_loss": 1.312,
1234
+ "step": 1370
1235
+ },
1236
+ {
1237
+ "ep_loss": 0.0,
1238
+ "epoch": 1.06,
1239
+ "learning_rate": 6.900000000000001e-05,
1240
+ "loss": 1.2987,
1241
+ "mlm_loss": 1.2987,
1242
+ "step": 1380
1243
+ },
1244
+ {
1245
+ "ep_loss": 0.0,
1246
+ "epoch": 1.07,
1247
+ "learning_rate": 6.950000000000001e-05,
1248
+ "loss": 1.2822,
1249
+ "mlm_loss": 1.2822,
1250
+ "step": 1390
1251
+ },
1252
+ {
1253
+ "ep_loss": 0.0,
1254
+ "epoch": 1.08,
1255
+ "learning_rate": 7.000000000000001e-05,
1256
+ "loss": 1.2817,
1257
+ "mlm_loss": 1.2817,
1258
+ "step": 1400
1259
+ },
1260
+ {
1261
+ "epoch": 1.08,
1262
+ "eval_ep_loss": -2.044208288192749,
1263
+ "eval_loss": 1.2174968719482422,
1264
+ "eval_mlm_loss": 1.2174968719482422,
1265
+ "eval_runtime": 62.7874,
1266
+ "eval_samples_per_second": 1112.788,
1267
+ "eval_steps_per_second": 0.557,
1268
+ "step": 1400
1269
+ },
1270
+ {
1271
+ "ep_loss": 0.0,
1272
+ "epoch": 1.09,
1273
+ "learning_rate": 7.049999999999999e-05,
1274
+ "loss": 1.2673,
1275
+ "mlm_loss": 1.2673,
1276
+ "step": 1410
1277
+ },
1278
+ {
1279
+ "ep_loss": 0.0,
1280
+ "epoch": 1.1,
1281
+ "learning_rate": 7.099999999999999e-05,
1282
+ "loss": 1.28,
1283
+ "mlm_loss": 1.28,
1284
+ "step": 1420
1285
+ },
1286
+ {
1287
+ "ep_loss": 0.0,
1288
+ "epoch": 1.1,
1289
+ "learning_rate": 7.149999999999999e-05,
1290
+ "loss": 1.2606,
1291
+ "mlm_loss": 1.2606,
1292
+ "step": 1430
1293
+ },
1294
+ {
1295
+ "ep_loss": 0.0,
1296
+ "epoch": 1.11,
1297
+ "learning_rate": 7.2e-05,
1298
+ "loss": 1.2439,
1299
+ "mlm_loss": 1.2439,
1300
+ "step": 1440
1301
+ },
1302
+ {
1303
+ "ep_loss": 0.0,
1304
+ "epoch": 1.12,
1305
+ "learning_rate": 7.25e-05,
1306
+ "loss": 1.2428,
1307
+ "mlm_loss": 1.2428,
1308
+ "step": 1450
1309
+ },
1310
+ {
1311
+ "ep_loss": 0.0,
1312
+ "epoch": 1.13,
1313
+ "learning_rate": 7.3e-05,
1314
+ "loss": 1.2342,
1315
+ "mlm_loss": 1.2342,
1316
+ "step": 1460
1317
+ },
1318
+ {
1319
+ "ep_loss": 0.0,
1320
+ "epoch": 1.13,
1321
+ "learning_rate": 7.35e-05,
1322
+ "loss": 1.2391,
1323
+ "mlm_loss": 1.2391,
1324
+ "step": 1470
1325
+ },
1326
+ {
1327
+ "ep_loss": 0.0,
1328
+ "epoch": 1.14,
1329
+ "learning_rate": 7.4e-05,
1330
+ "loss": 1.2337,
1331
+ "mlm_loss": 1.2337,
1332
+ "step": 1480
1333
+ },
1334
+ {
1335
+ "ep_loss": 0.0,
1336
+ "epoch": 1.15,
1337
+ "learning_rate": 7.45e-05,
1338
+ "loss": 1.2161,
1339
+ "mlm_loss": 1.2161,
1340
+ "step": 1490
1341
+ },
1342
+ {
1343
+ "ep_loss": 0.0,
1344
+ "epoch": 1.16,
1345
+ "learning_rate": 7.5e-05,
1346
+ "loss": 1.2083,
1347
+ "mlm_loss": 1.2083,
1348
+ "step": 1500
1349
+ },
1350
+ {
1351
+ "epoch": 1.16,
1352
+ "eval_ep_loss": -2.247138738632202,
1353
+ "eval_loss": 1.145897388458252,
1354
+ "eval_mlm_loss": 1.145897388458252,
1355
+ "eval_runtime": 60.0275,
1356
+ "eval_samples_per_second": 1163.951,
1357
+ "eval_steps_per_second": 0.583,
1358
+ "step": 1500
1359
+ },
1360
+ {
1361
+ "ep_loss": 0.0,
1362
+ "epoch": 1.16,
1363
+ "learning_rate": 7.55e-05,
1364
+ "loss": 1.2048,
1365
+ "mlm_loss": 1.2048,
1366
+ "step": 1510
1367
+ },
1368
+ {
1369
+ "ep_loss": 0.0,
1370
+ "epoch": 1.17,
1371
+ "learning_rate": 7.6e-05,
1372
+ "loss": 1.1934,
1373
+ "mlm_loss": 1.1934,
1374
+ "step": 1520
1375
+ },
1376
+ {
1377
+ "ep_loss": 0.0,
1378
+ "epoch": 1.18,
1379
+ "learning_rate": 7.65e-05,
1380
+ "loss": 1.1741,
1381
+ "mlm_loss": 1.1741,
1382
+ "step": 1530
1383
+ },
1384
+ {
1385
+ "ep_loss": 0.0,
1386
+ "epoch": 1.19,
1387
+ "learning_rate": 7.7e-05,
1388
+ "loss": 1.1832,
1389
+ "mlm_loss": 1.1832,
1390
+ "step": 1540
1391
+ },
1392
+ {
1393
+ "ep_loss": 0.0,
1394
+ "epoch": 1.2,
1395
+ "learning_rate": 7.75e-05,
1396
+ "loss": 1.1759,
1397
+ "mlm_loss": 1.1759,
1398
+ "step": 1550
1399
+ },
1400
+ {
1401
+ "ep_loss": 0.0,
1402
+ "epoch": 1.2,
1403
+ "learning_rate": 7.8e-05,
1404
+ "loss": 1.1776,
1405
+ "mlm_loss": 1.1776,
1406
+ "step": 1560
1407
+ },
1408
+ {
1409
+ "ep_loss": 0.0,
1410
+ "epoch": 1.21,
1411
+ "learning_rate": 7.85e-05,
1412
+ "loss": 1.164,
1413
+ "mlm_loss": 1.164,
1414
+ "step": 1570
1415
+ },
1416
+ {
1417
+ "ep_loss": 0.0,
1418
+ "epoch": 1.22,
1419
+ "learning_rate": 7.9e-05,
1420
+ "loss": 1.1447,
1421
+ "mlm_loss": 1.1447,
1422
+ "step": 1580
1423
+ },
1424
+ {
1425
+ "ep_loss": 0.0,
1426
+ "epoch": 1.23,
1427
+ "learning_rate": 7.950000000000001e-05,
1428
+ "loss": 1.1669,
1429
+ "mlm_loss": 1.1669,
1430
+ "step": 1590
1431
+ },
1432
+ {
1433
+ "ep_loss": 0.0,
1434
+ "epoch": 1.23,
1435
+ "learning_rate": 8e-05,
1436
+ "loss": 1.1331,
1437
+ "mlm_loss": 1.1331,
1438
+ "step": 1600
1439
+ },
1440
+ {
1441
+ "epoch": 1.23,
1442
+ "eval_ep_loss": -2.236558198928833,
1443
+ "eval_loss": 1.0778173208236694,
1444
+ "eval_mlm_loss": 1.0778173208236694,
1445
+ "eval_runtime": 61.8774,
1446
+ "eval_samples_per_second": 1129.152,
1447
+ "eval_steps_per_second": 0.566,
1448
+ "step": 1600
1449
+ },
1450
+ {
1451
+ "ep_loss": 0.0,
1452
+ "epoch": 1.24,
1453
+ "learning_rate": 8.05e-05,
1454
+ "loss": 1.1332,
1455
+ "mlm_loss": 1.1332,
1456
+ "step": 1610
1457
+ },
1458
+ {
1459
+ "ep_loss": 0.0,
1460
+ "epoch": 1.25,
1461
+ "learning_rate": 8.1e-05,
1462
+ "loss": 1.1288,
1463
+ "mlm_loss": 1.1288,
1464
+ "step": 1620
1465
+ },
1466
+ {
1467
+ "ep_loss": 0.0,
1468
+ "epoch": 1.26,
1469
+ "learning_rate": 8.15e-05,
1470
+ "loss": 1.1321,
1471
+ "mlm_loss": 1.1321,
1472
+ "step": 1630
1473
+ },
1474
+ {
1475
+ "ep_loss": 0.0,
1476
+ "epoch": 1.26,
1477
+ "learning_rate": 8.2e-05,
1478
+ "loss": 1.11,
1479
+ "mlm_loss": 1.11,
1480
+ "step": 1640
1481
+ },
1482
+ {
1483
+ "ep_loss": 0.0,
1484
+ "epoch": 1.27,
1485
+ "learning_rate": 8.25e-05,
1486
+ "loss": 1.1083,
1487
+ "mlm_loss": 1.1083,
1488
+ "step": 1650
1489
+ },
1490
+ {
1491
+ "ep_loss": 0.0,
1492
+ "epoch": 1.28,
1493
+ "learning_rate": 8.300000000000001e-05,
1494
+ "loss": 1.0999,
1495
+ "mlm_loss": 1.0999,
1496
+ "step": 1660
1497
+ },
1498
+ {
1499
+ "ep_loss": 0.0,
1500
+ "epoch": 1.29,
1501
+ "learning_rate": 8.350000000000001e-05,
1502
+ "loss": 1.1041,
1503
+ "mlm_loss": 1.1041,
1504
+ "step": 1670
1505
+ },
1506
+ {
1507
+ "ep_loss": 0.0,
1508
+ "epoch": 1.3,
1509
+ "learning_rate": 8.400000000000001e-05,
1510
+ "loss": 1.0823,
1511
+ "mlm_loss": 1.0823,
1512
+ "step": 1680
1513
+ },
1514
+ {
1515
+ "ep_loss": 0.0,
1516
+ "epoch": 1.3,
1517
+ "learning_rate": 8.450000000000001e-05,
1518
+ "loss": 1.0809,
1519
+ "mlm_loss": 1.0809,
1520
+ "step": 1690
1521
+ },
1522
+ {
1523
+ "ep_loss": 0.0,
1524
+ "epoch": 1.31,
1525
+ "learning_rate": 8.5e-05,
1526
+ "loss": 1.0859,
1527
+ "mlm_loss": 1.0859,
1528
+ "step": 1700
1529
+ },
1530
+ {
1531
+ "epoch": 1.31,
1532
+ "eval_ep_loss": -2.2500903606414795,
1533
+ "eval_loss": 1.0151546001434326,
1534
+ "eval_mlm_loss": 1.0151546001434326,
1535
+ "eval_runtime": 60.0928,
1536
+ "eval_samples_per_second": 1162.685,
1537
+ "eval_steps_per_second": 0.582,
1538
+ "step": 1700
1539
+ },
1540
+ {
1541
+ "ep_loss": 0.0,
1542
+ "epoch": 1.32,
1543
+ "learning_rate": 8.55e-05,
1544
+ "loss": 1.0757,
1545
+ "mlm_loss": 1.0757,
1546
+ "step": 1710
1547
+ },
1548
+ {
1549
+ "ep_loss": 0.0,
1550
+ "epoch": 1.33,
1551
+ "learning_rate": 8.599999999999999e-05,
1552
+ "loss": 1.0638,
1553
+ "mlm_loss": 1.0638,
1554
+ "step": 1720
1555
+ },
1556
+ {
1557
+ "ep_loss": 0.0,
1558
+ "epoch": 1.33,
1559
+ "learning_rate": 8.65e-05,
1560
+ "loss": 1.0554,
1561
+ "mlm_loss": 1.0554,
1562
+ "step": 1730
1563
+ },
1564
+ {
1565
+ "ep_loss": 0.0,
1566
+ "epoch": 1.34,
1567
+ "learning_rate": 8.7e-05,
1568
+ "loss": 1.0624,
1569
+ "mlm_loss": 1.0624,
1570
+ "step": 1740
1571
+ },
1572
+ {
1573
+ "ep_loss": 0.0,
1574
+ "epoch": 1.35,
1575
+ "learning_rate": 8.75e-05,
1576
+ "loss": 1.0483,
1577
+ "mlm_loss": 1.0483,
1578
+ "step": 1750
1579
+ },
1580
+ {
1581
+ "ep_loss": 0.0,
1582
+ "epoch": 1.36,
1583
+ "learning_rate": 8.8e-05,
1584
+ "loss": 1.0432,
1585
+ "mlm_loss": 1.0432,
1586
+ "step": 1760
1587
+ },
1588
+ {
1589
+ "ep_loss": 0.0,
1590
+ "epoch": 1.37,
1591
+ "learning_rate": 8.85e-05,
1592
+ "loss": 1.0464,
1593
+ "mlm_loss": 1.0464,
1594
+ "step": 1770
1595
+ },
1596
+ {
1597
+ "ep_loss": 0.0,
1598
+ "epoch": 1.37,
1599
+ "learning_rate": 8.9e-05,
1600
+ "loss": 1.0379,
1601
+ "mlm_loss": 1.0379,
1602
+ "step": 1780
1603
+ },
1604
+ {
1605
+ "ep_loss": 0.0,
1606
+ "epoch": 1.38,
1607
+ "learning_rate": 8.95e-05,
1608
+ "loss": 1.0416,
1609
+ "mlm_loss": 1.0416,
1610
+ "step": 1790
1611
+ },
1612
+ {
1613
+ "ep_loss": 0.0,
1614
+ "epoch": 1.39,
1615
+ "learning_rate": 8.999999999999999e-05,
1616
+ "loss": 1.0178,
1617
+ "mlm_loss": 1.0178,
1618
+ "step": 1800
1619
+ },
1620
+ {
1621
+ "epoch": 1.39,
1622
+ "eval_ep_loss": -2.3243699073791504,
1623
+ "eval_loss": 0.9662984013557434,
1624
+ "eval_mlm_loss": 0.9662984013557434,
1625
+ "eval_runtime": 61.639,
1626
+ "eval_samples_per_second": 1133.519,
1627
+ "eval_steps_per_second": 0.568,
1628
+ "step": 1800
1629
+ },
1630
+ {
1631
+ "ep_loss": 0.0,
1632
+ "epoch": 1.4,
1633
+ "learning_rate": 9.05e-05,
1634
+ "loss": 1.0189,
1635
+ "mlm_loss": 1.0189,
1636
+ "step": 1810
1637
+ },
1638
+ {
1639
+ "ep_loss": 0.0,
1640
+ "epoch": 1.4,
1641
+ "learning_rate": 9.1e-05,
1642
+ "loss": 1.0181,
1643
+ "mlm_loss": 1.0181,
1644
+ "step": 1820
1645
+ },
1646
+ {
1647
+ "ep_loss": 0.0,
1648
+ "epoch": 1.41,
1649
+ "learning_rate": 9.15e-05,
1650
+ "loss": 1.006,
1651
+ "mlm_loss": 1.006,
1652
+ "step": 1830
1653
+ },
1654
+ {
1655
+ "ep_loss": 0.0,
1656
+ "epoch": 1.42,
1657
+ "learning_rate": 9.2e-05,
1658
+ "loss": 1.0088,
1659
+ "mlm_loss": 1.0088,
1660
+ "step": 1840
1661
+ },
1662
+ {
1663
+ "ep_loss": 0.0,
1664
+ "epoch": 1.43,
1665
+ "learning_rate": 9.25e-05,
1666
+ "loss": 1.008,
1667
+ "mlm_loss": 1.008,
1668
+ "step": 1850
1669
+ },
1670
+ {
1671
+ "ep_loss": 0.0,
1672
+ "epoch": 1.43,
1673
+ "learning_rate": 9.3e-05,
1674
+ "loss": 0.9891,
1675
+ "mlm_loss": 0.9891,
1676
+ "step": 1860
1677
+ },
1678
+ {
1679
+ "ep_loss": 0.0,
1680
+ "epoch": 1.44,
1681
+ "learning_rate": 9.35e-05,
1682
+ "loss": 0.9998,
1683
+ "mlm_loss": 0.9998,
1684
+ "step": 1870
1685
+ },
1686
+ {
1687
+ "ep_loss": 0.0,
1688
+ "epoch": 1.45,
1689
+ "learning_rate": 9.400000000000001e-05,
1690
+ "loss": 0.9927,
1691
+ "mlm_loss": 0.9927,
1692
+ "step": 1880
1693
+ },
1694
+ {
1695
+ "ep_loss": 0.0,
1696
+ "epoch": 1.46,
1697
+ "learning_rate": 9.45e-05,
1698
+ "loss": 0.9825,
1699
+ "mlm_loss": 0.9825,
1700
+ "step": 1890
1701
+ },
1702
+ {
1703
+ "ep_loss": 0.0,
1704
+ "epoch": 1.47,
1705
+ "learning_rate": 9.5e-05,
1706
+ "loss": 0.979,
1707
+ "mlm_loss": 0.979,
1708
+ "step": 1900
1709
+ },
1710
+ {
1711
+ "epoch": 1.47,
1712
+ "eval_ep_loss": -2.297435760498047,
1713
+ "eval_loss": 0.9245826005935669,
1714
+ "eval_mlm_loss": 0.9245826005935669,
1715
+ "eval_runtime": 60.9274,
1716
+ "eval_samples_per_second": 1146.759,
1717
+ "eval_steps_per_second": 0.574,
1718
+ "step": 1900
1719
+ },
1720
+ {
1721
+ "ep_loss": 0.0,
1722
+ "epoch": 1.47,
1723
+ "learning_rate": 9.55e-05,
1724
+ "loss": 0.9782,
1725
+ "mlm_loss": 0.9782,
1726
+ "step": 1910
1727
+ },
1728
+ {
1729
+ "ep_loss": 0.0,
1730
+ "epoch": 1.48,
1731
+ "learning_rate": 9.6e-05,
1732
+ "loss": 0.9802,
1733
+ "mlm_loss": 0.9802,
1734
+ "step": 1920
1735
+ },
1736
+ {
1737
+ "ep_loss": 0.0,
1738
+ "epoch": 1.49,
1739
+ "learning_rate": 9.65e-05,
1740
+ "loss": 0.9648,
1741
+ "mlm_loss": 0.9648,
1742
+ "step": 1930
1743
+ },
1744
+ {
1745
+ "ep_loss": 0.0,
1746
+ "epoch": 1.5,
1747
+ "learning_rate": 9.7e-05,
1748
+ "loss": 0.9576,
1749
+ "mlm_loss": 0.9576,
1750
+ "step": 1940
1751
+ },
1752
+ {
1753
+ "ep_loss": 0.0,
1754
+ "epoch": 1.5,
1755
+ "learning_rate": 9.750000000000001e-05,
1756
+ "loss": 0.9629,
1757
+ "mlm_loss": 0.9629,
1758
+ "step": 1950
1759
+ },
1760
+ {
1761
+ "ep_loss": 0.0,
1762
+ "epoch": 1.51,
1763
+ "learning_rate": 9.800000000000001e-05,
1764
+ "loss": 0.9585,
1765
+ "mlm_loss": 0.9585,
1766
+ "step": 1960
1767
+ },
1768
+ {
1769
+ "ep_loss": 0.0,
1770
+ "epoch": 1.52,
1771
+ "learning_rate": 9.850000000000001e-05,
1772
+ "loss": 0.9404,
1773
+ "mlm_loss": 0.9404,
1774
+ "step": 1970
1775
+ },
1776
+ {
1777
+ "ep_loss": 0.0,
1778
+ "epoch": 1.53,
1779
+ "learning_rate": 9.900000000000001e-05,
1780
+ "loss": 0.9538,
1781
+ "mlm_loss": 0.9538,
1782
+ "step": 1980
1783
+ },
1784
+ {
1785
+ "ep_loss": 0.0,
1786
+ "epoch": 1.53,
1787
+ "learning_rate": 9.95e-05,
1788
+ "loss": 0.9591,
1789
+ "mlm_loss": 0.9591,
1790
+ "step": 1990
1791
+ },
1792
+ {
1793
+ "ep_loss": 0.0,
1794
+ "epoch": 1.54,
1795
+ "learning_rate": 0.0001,
1796
+ "loss": 0.9467,
1797
+ "mlm_loss": 0.9467,
1798
+ "step": 2000
1799
+ },
1800
+ {
1801
+ "epoch": 1.54,
1802
+ "eval_ep_loss": -2.388266086578369,
1803
+ "eval_loss": 0.8887887597084045,
1804
+ "eval_mlm_loss": 0.8887887597084045,
1805
+ "eval_runtime": 59.4206,
1806
+ "eval_samples_per_second": 1175.838,
1807
+ "eval_steps_per_second": 0.589,
1808
+ "step": 2000
1809
+ },
1810
+ {
1811
+ "ep_loss": 0.0,
1812
+ "epoch": 1.55,
1813
+ "learning_rate": 0.0001005,
1814
+ "loss": 0.947,
1815
+ "mlm_loss": 0.947,
1816
+ "step": 2010
1817
+ },
1818
+ {
1819
+ "ep_loss": 0.0,
1820
+ "epoch": 1.56,
1821
+ "learning_rate": 0.000101,
1822
+ "loss": 0.9407,
1823
+ "mlm_loss": 0.9407,
1824
+ "step": 2020
1825
+ },
1826
+ {
1827
+ "ep_loss": 0.0,
1828
+ "epoch": 1.57,
1829
+ "learning_rate": 0.00010150000000000001,
1830
+ "loss": 0.9352,
1831
+ "mlm_loss": 0.9352,
1832
+ "step": 2030
1833
+ },
1834
+ {
1835
+ "ep_loss": 0.0,
1836
+ "epoch": 1.57,
1837
+ "learning_rate": 0.000102,
1838
+ "loss": 0.9234,
1839
+ "mlm_loss": 0.9234,
1840
+ "step": 2040
1841
+ },
1842
+ {
1843
+ "ep_loss": 0.0,
1844
+ "epoch": 1.58,
1845
+ "learning_rate": 0.0001025,
1846
+ "loss": 0.9252,
1847
+ "mlm_loss": 0.9252,
1848
+ "step": 2050
1849
+ },
1850
+ {
1851
+ "ep_loss": 0.0,
1852
+ "epoch": 1.59,
1853
+ "learning_rate": 0.000103,
1854
+ "loss": 0.9271,
1855
+ "mlm_loss": 0.9271,
1856
+ "step": 2060
1857
+ },
1858
+ {
1859
+ "ep_loss": 0.0,
1860
+ "epoch": 1.6,
1861
+ "learning_rate": 0.0001035,
1862
+ "loss": 0.9254,
1863
+ "mlm_loss": 0.9254,
1864
+ "step": 2070
1865
+ },
1866
+ {
1867
+ "ep_loss": 0.0,
1868
+ "epoch": 1.6,
1869
+ "learning_rate": 0.000104,
1870
+ "loss": 0.9232,
1871
+ "mlm_loss": 0.9232,
1872
+ "step": 2080
1873
+ },
1874
+ {
1875
+ "ep_loss": 0.0,
1876
+ "epoch": 1.61,
1877
+ "learning_rate": 0.00010449999999999999,
1878
+ "loss": 0.919,
1879
+ "mlm_loss": 0.919,
1880
+ "step": 2090
1881
+ },
1882
+ {
1883
+ "ep_loss": 0.0,
1884
+ "epoch": 1.62,
1885
+ "learning_rate": 0.000105,
1886
+ "loss": 0.9127,
1887
+ "mlm_loss": 0.9127,
1888
+ "step": 2100
1889
+ },
1890
+ {
1891
+ "epoch": 1.62,
1892
+ "eval_ep_loss": -2.262110948562622,
1893
+ "eval_loss": 0.8549456000328064,
1894
+ "eval_mlm_loss": 0.8549456000328064,
1895
+ "eval_runtime": 59.0907,
1896
+ "eval_samples_per_second": 1182.403,
1897
+ "eval_steps_per_second": 0.592,
1898
+ "step": 2100
1899
+ },
1900
+ {
1901
+ "ep_loss": 0.0,
1902
+ "epoch": 1.63,
1903
+ "learning_rate": 0.0001055,
1904
+ "loss": 0.9023,
1905
+ "mlm_loss": 0.9023,
1906
+ "step": 2110
1907
+ },
1908
+ {
1909
+ "ep_loss": 0.0,
1910
+ "epoch": 1.64,
1911
+ "learning_rate": 0.000106,
1912
+ "loss": 0.9031,
1913
+ "mlm_loss": 0.9031,
1914
+ "step": 2120
1915
+ },
1916
+ {
1917
+ "ep_loss": 0.0,
1918
+ "epoch": 1.64,
1919
+ "learning_rate": 0.0001065,
1920
+ "loss": 0.9039,
1921
+ "mlm_loss": 0.9039,
1922
+ "step": 2130
1923
+ },
1924
+ {
1925
+ "ep_loss": 0.0,
1926
+ "epoch": 1.65,
1927
+ "learning_rate": 0.000107,
1928
+ "loss": 0.9022,
1929
+ "mlm_loss": 0.9022,
1930
+ "step": 2140
1931
+ },
1932
+ {
1933
+ "ep_loss": 0.0,
1934
+ "epoch": 1.66,
1935
+ "learning_rate": 0.0001075,
1936
+ "loss": 0.8978,
1937
+ "mlm_loss": 0.8978,
1938
+ "step": 2150
1939
+ },
1940
+ {
1941
+ "ep_loss": 0.0,
1942
+ "epoch": 1.67,
1943
+ "learning_rate": 0.000108,
1944
+ "loss": 0.8893,
1945
+ "mlm_loss": 0.8893,
1946
+ "step": 2160
1947
+ },
1948
+ {
1949
+ "ep_loss": 0.0,
1950
+ "epoch": 1.67,
1951
+ "learning_rate": 0.00010850000000000001,
1952
+ "loss": 0.8796,
1953
+ "mlm_loss": 0.8796,
1954
+ "step": 2170
1955
+ },
1956
+ {
1957
+ "ep_loss": 0.0,
1958
+ "epoch": 1.68,
1959
+ "learning_rate": 0.000109,
1960
+ "loss": 0.8925,
1961
+ "mlm_loss": 0.8925,
1962
+ "step": 2180
1963
+ },
1964
+ {
1965
+ "ep_loss": 0.0,
1966
+ "epoch": 1.69,
1967
+ "learning_rate": 0.0001095,
1968
+ "loss": 0.8873,
1969
+ "mlm_loss": 0.8873,
1970
+ "step": 2190
1971
+ },
1972
+ {
1973
+ "ep_loss": 0.0,
1974
+ "epoch": 1.7,
1975
+ "learning_rate": 0.00011,
1976
+ "loss": 0.8858,
1977
+ "mlm_loss": 0.8858,
1978
+ "step": 2200
1979
+ },
1980
+ {
1981
+ "epoch": 1.7,
1982
+ "eval_ep_loss": -2.3254947662353516,
1983
+ "eval_loss": 0.8307343125343323,
1984
+ "eval_mlm_loss": 0.8307343125343323,
1985
+ "eval_runtime": 62.4113,
1986
+ "eval_samples_per_second": 1119.493,
1987
+ "eval_steps_per_second": 0.561,
1988
+ "step": 2200
1989
+ },
1990
+ {
1991
+ "ep_loss": 0.0,
1992
+ "epoch": 1.7,
1993
+ "learning_rate": 0.0001105,
1994
+ "loss": 0.888,
1995
+ "mlm_loss": 0.888,
1996
+ "step": 2210
1997
+ },
1998
+ {
1999
+ "ep_loss": 0.0,
2000
+ "epoch": 1.71,
2001
+ "learning_rate": 0.000111,
2002
+ "loss": 0.8706,
2003
+ "mlm_loss": 0.8706,
2004
+ "step": 2220
2005
+ },
2006
+ {
2007
+ "ep_loss": 0.0,
2008
+ "epoch": 1.72,
2009
+ "learning_rate": 0.0001115,
2010
+ "loss": 0.8854,
2011
+ "mlm_loss": 0.8854,
2012
+ "step": 2230
2013
+ },
2014
+ {
2015
+ "ep_loss": 0.0,
2016
+ "epoch": 1.73,
2017
+ "learning_rate": 0.000112,
2018
+ "loss": 0.864,
2019
+ "mlm_loss": 0.864,
2020
+ "step": 2240
2021
+ },
2022
+ {
2023
+ "ep_loss": 0.0,
2024
+ "epoch": 1.74,
2025
+ "learning_rate": 0.00011250000000000001,
2026
+ "loss": 0.8675,
2027
+ "mlm_loss": 0.8675,
2028
+ "step": 2250
2029
+ },
2030
+ {
2031
+ "ep_loss": 0.0,
2032
+ "epoch": 1.74,
2033
+ "learning_rate": 0.00011300000000000001,
2034
+ "loss": 0.8654,
2035
+ "mlm_loss": 0.8654,
2036
+ "step": 2260
2037
+ },
2038
+ {
2039
+ "ep_loss": 0.0,
2040
+ "epoch": 1.75,
2041
+ "learning_rate": 0.00011350000000000001,
2042
+ "loss": 0.8593,
2043
+ "mlm_loss": 0.8593,
2044
+ "step": 2270
2045
+ },
2046
+ {
2047
+ "ep_loss": 0.0,
2048
+ "epoch": 1.76,
2049
+ "learning_rate": 0.000114,
2050
+ "loss": 0.8586,
2051
+ "mlm_loss": 0.8586,
2052
+ "step": 2280
2053
+ },
2054
+ {
2055
+ "ep_loss": 0.0,
2056
+ "epoch": 1.77,
2057
+ "learning_rate": 0.0001145,
2058
+ "loss": 0.8494,
2059
+ "mlm_loss": 0.8494,
2060
+ "step": 2290
2061
+ },
2062
+ {
2063
+ "ep_loss": 0.0,
2064
+ "epoch": 1.77,
2065
+ "learning_rate": 0.000115,
2066
+ "loss": 0.8517,
2067
+ "mlm_loss": 0.8517,
2068
+ "step": 2300
2069
+ },
2070
+ {
2071
+ "epoch": 1.77,
2072
+ "eval_ep_loss": -2.2645294666290283,
2073
+ "eval_loss": 0.803054928779602,
2074
+ "eval_mlm_loss": 0.803054928779602,
2075
+ "eval_runtime": 61.7973,
2076
+ "eval_samples_per_second": 1130.616,
2077
+ "eval_steps_per_second": 0.566,
2078
+ "step": 2300
2079
+ },
2080
+ {
2081
+ "ep_loss": 0.0,
2082
+ "epoch": 1.78,
2083
+ "learning_rate": 0.0001155,
2084
+ "loss": 0.8396,
2085
+ "mlm_loss": 0.8396,
2086
+ "step": 2310
2087
+ },
2088
+ {
2089
+ "ep_loss": 0.0,
2090
+ "epoch": 1.79,
2091
+ "learning_rate": 0.00011600000000000001,
2092
+ "loss": 0.8478,
2093
+ "mlm_loss": 0.8478,
2094
+ "step": 2320
2095
+ },
2096
+ {
2097
+ "ep_loss": 0.0,
2098
+ "epoch": 1.8,
2099
+ "learning_rate": 0.00011650000000000001,
2100
+ "loss": 0.8495,
2101
+ "mlm_loss": 0.8495,
2102
+ "step": 2330
2103
+ },
2104
+ {
2105
+ "ep_loss": 0.0,
2106
+ "epoch": 1.8,
2107
+ "learning_rate": 0.00011700000000000001,
2108
+ "loss": 0.8441,
2109
+ "mlm_loss": 0.8441,
2110
+ "step": 2340
2111
+ },
2112
+ {
2113
+ "ep_loss": 0.0,
2114
+ "epoch": 1.81,
2115
+ "learning_rate": 0.0001175,
2116
+ "loss": 0.8441,
2117
+ "mlm_loss": 0.8441,
2118
+ "step": 2350
2119
+ },
2120
+ {
2121
+ "ep_loss": 0.0,
2122
+ "epoch": 1.82,
2123
+ "learning_rate": 0.000118,
2124
+ "loss": 0.8406,
2125
+ "mlm_loss": 0.8406,
2126
+ "step": 2360
2127
+ },
2128
+ {
2129
+ "ep_loss": 0.0,
2130
+ "epoch": 1.83,
2131
+ "learning_rate": 0.0001185,
2132
+ "loss": 0.8322,
2133
+ "mlm_loss": 0.8322,
2134
+ "step": 2370
2135
+ },
2136
+ {
2137
+ "ep_loss": 0.0,
2138
+ "epoch": 1.84,
2139
+ "learning_rate": 0.00011899999999999999,
2140
+ "loss": 0.8282,
2141
+ "mlm_loss": 0.8282,
2142
+ "step": 2380
2143
+ },
2144
+ {
2145
+ "ep_loss": 0.0,
2146
+ "epoch": 1.84,
2147
+ "learning_rate": 0.00011949999999999999,
2148
+ "loss": 0.8437,
2149
+ "mlm_loss": 0.8437,
2150
+ "step": 2390
2151
+ },
2152
+ {
2153
+ "ep_loss": 0.0,
2154
+ "epoch": 1.85,
2155
+ "learning_rate": 0.00012,
2156
+ "loss": 0.8415,
2157
+ "mlm_loss": 0.8415,
2158
+ "step": 2400
2159
+ },
2160
+ {
2161
+ "epoch": 1.85,
2162
+ "eval_ep_loss": -2.278646945953369,
2163
+ "eval_loss": 0.7815272212028503,
2164
+ "eval_mlm_loss": 0.7815272212028503,
2165
+ "eval_runtime": 60.3734,
2166
+ "eval_samples_per_second": 1157.281,
2167
+ "eval_steps_per_second": 0.58,
2168
+ "step": 2400
2169
+ },
2170
+ {
2171
+ "ep_loss": 0.0,
2172
+ "epoch": 1.86,
2173
+ "learning_rate": 0.0001205,
2174
+ "loss": 0.8232,
2175
+ "mlm_loss": 0.8232,
2176
+ "step": 2410
2177
+ },
2178
+ {
2179
+ "ep_loss": 0.0,
2180
+ "epoch": 1.87,
2181
+ "learning_rate": 0.000121,
2182
+ "loss": 0.8204,
2183
+ "mlm_loss": 0.8204,
2184
+ "step": 2420
2185
+ },
2186
+ {
2187
+ "ep_loss": 0.0,
2188
+ "epoch": 1.87,
2189
+ "learning_rate": 0.0001215,
2190
+ "loss": 0.8246,
2191
+ "mlm_loss": 0.8246,
2192
+ "step": 2430
2193
+ },
2194
+ {
2195
+ "ep_loss": 0.0,
2196
+ "epoch": 1.88,
2197
+ "learning_rate": 0.000122,
2198
+ "loss": 0.8092,
2199
+ "mlm_loss": 0.8092,
2200
+ "step": 2440
2201
+ },
2202
+ {
2203
+ "ep_loss": 0.0,
2204
+ "epoch": 1.89,
2205
+ "learning_rate": 0.0001225,
2206
+ "loss": 0.812,
2207
+ "mlm_loss": 0.812,
2208
+ "step": 2450
2209
+ },
2210
+ {
2211
+ "ep_loss": 0.0,
2212
+ "epoch": 1.9,
2213
+ "learning_rate": 0.000123,
2214
+ "loss": 0.8148,
2215
+ "mlm_loss": 0.8148,
2216
+ "step": 2460
2217
+ },
2218
+ {
2219
+ "ep_loss": 0.0,
2220
+ "epoch": 1.91,
2221
+ "learning_rate": 0.0001235,
2222
+ "loss": 0.8065,
2223
+ "mlm_loss": 0.8065,
2224
+ "step": 2470
2225
+ },
2226
+ {
2227
+ "ep_loss": 0.0,
2228
+ "epoch": 1.91,
2229
+ "learning_rate": 0.000124,
2230
+ "loss": 0.8029,
2231
+ "mlm_loss": 0.8029,
2232
+ "step": 2480
2233
+ },
2234
+ {
2235
+ "ep_loss": 0.0,
2236
+ "epoch": 1.92,
2237
+ "learning_rate": 0.0001245,
2238
+ "loss": 0.8148,
2239
+ "mlm_loss": 0.8148,
2240
+ "step": 2490
2241
+ },
2242
+ {
2243
+ "ep_loss": 0.0,
2244
+ "epoch": 1.93,
2245
+ "learning_rate": 0.000125,
2246
+ "loss": 0.806,
2247
+ "mlm_loss": 0.806,
2248
+ "step": 2500
2249
+ },
2250
+ {
2251
+ "epoch": 1.93,
2252
+ "eval_ep_loss": -2.1858584880828857,
2253
+ "eval_loss": 0.7560575604438782,
2254
+ "eval_mlm_loss": 0.7560575604438782,
2255
+ "eval_runtime": 60.8551,
2256
+ "eval_samples_per_second": 1148.121,
2257
+ "eval_steps_per_second": 0.575,
2258
+ "step": 2500
2259
+ },
2260
+ {
2261
+ "ep_loss": 0.0,
2262
+ "epoch": 1.94,
2263
+ "learning_rate": 0.00012550000000000001,
2264
+ "loss": 0.8048,
2265
+ "mlm_loss": 0.8048,
2266
+ "step": 2510
2267
+ },
2268
+ {
2269
+ "ep_loss": 0.0,
2270
+ "epoch": 1.94,
2271
+ "learning_rate": 0.000126,
2272
+ "loss": 0.8027,
2273
+ "mlm_loss": 0.8027,
2274
+ "step": 2520
2275
+ },
2276
+ {
2277
+ "ep_loss": 0.0,
2278
+ "epoch": 1.95,
2279
+ "learning_rate": 0.0001265,
2280
+ "loss": 0.797,
2281
+ "mlm_loss": 0.797,
2282
+ "step": 2530
2283
+ },
2284
+ {
2285
+ "ep_loss": 0.0,
2286
+ "epoch": 1.96,
2287
+ "learning_rate": 0.000127,
2288
+ "loss": 0.8041,
2289
+ "mlm_loss": 0.8041,
2290
+ "step": 2540
2291
+ },
2292
+ {
2293
+ "ep_loss": 0.0,
2294
+ "epoch": 1.97,
2295
+ "learning_rate": 0.0001275,
2296
+ "loss": 0.8,
2297
+ "mlm_loss": 0.8,
2298
+ "step": 2550
2299
+ },
2300
+ {
2301
+ "ep_loss": 0.0,
2302
+ "epoch": 1.97,
2303
+ "learning_rate": 0.000128,
2304
+ "loss": 0.7955,
2305
+ "mlm_loss": 0.7955,
2306
+ "step": 2560
2307
+ },
2308
+ {
2309
+ "ep_loss": 0.0,
2310
+ "epoch": 1.98,
2311
+ "learning_rate": 0.0001285,
2312
+ "loss": 0.7943,
2313
+ "mlm_loss": 0.7943,
2314
+ "step": 2570
2315
+ },
2316
+ {
2317
+ "ep_loss": 0.0,
2318
+ "epoch": 1.99,
2319
+ "learning_rate": 0.00012900000000000002,
2320
+ "loss": 0.7934,
2321
+ "mlm_loss": 0.7934,
2322
+ "step": 2580
2323
+ },
2324
+ {
2325
+ "ep_loss": 0.0,
2326
+ "epoch": 2.0,
2327
+ "learning_rate": 0.0001295,
2328
+ "loss": 0.7884,
2329
+ "mlm_loss": 0.7884,
2330
+ "step": 2590
2331
+ },
2332
+ {
2333
+ "ep_loss": 0.0,
2334
+ "epoch": 2.01,
2335
+ "learning_rate": 0.00013000000000000002,
2336
+ "loss": 0.7892,
2337
+ "mlm_loss": 0.7892,
2338
+ "step": 2600
2339
+ },
2340
+ {
2341
+ "epoch": 2.01,
2342
+ "eval_ep_loss": -2.1713080406188965,
2343
+ "eval_loss": 0.7403181195259094,
2344
+ "eval_mlm_loss": 0.7403181195259094,
2345
+ "eval_runtime": 64.561,
2346
+ "eval_samples_per_second": 1082.217,
2347
+ "eval_steps_per_second": 0.542,
2348
+ "step": 2600
2349
+ },
2350
+ {
2351
+ "ep_loss": 0.0,
2352
+ "epoch": 2.01,
2353
+ "learning_rate": 0.0001305,
2354
+ "loss": 0.7818,
2355
+ "mlm_loss": 0.7818,
2356
+ "step": 2610
2357
+ },
2358
+ {
2359
+ "ep_loss": 0.0,
2360
+ "epoch": 2.02,
2361
+ "learning_rate": 0.000131,
2362
+ "loss": 0.7771,
2363
+ "mlm_loss": 0.7771,
2364
+ "step": 2620
2365
+ },
2366
+ {
2367
+ "ep_loss": 0.0,
2368
+ "epoch": 2.03,
2369
+ "learning_rate": 0.0001315,
2370
+ "loss": 0.767,
2371
+ "mlm_loss": 0.767,
2372
+ "step": 2630
2373
+ },
2374
+ {
2375
+ "ep_loss": 0.0,
2376
+ "epoch": 2.04,
2377
+ "learning_rate": 0.000132,
2378
+ "loss": 0.7826,
2379
+ "mlm_loss": 0.7826,
2380
+ "step": 2640
2381
+ },
2382
+ {
2383
+ "ep_loss": 0.0,
2384
+ "epoch": 2.04,
2385
+ "learning_rate": 0.00013250000000000002,
2386
+ "loss": 0.771,
2387
+ "mlm_loss": 0.771,
2388
+ "step": 2650
2389
+ },
2390
+ {
2391
+ "ep_loss": 0.0,
2392
+ "epoch": 2.05,
2393
+ "learning_rate": 0.000133,
2394
+ "loss": 0.7774,
2395
+ "mlm_loss": 0.7774,
2396
+ "step": 2660
2397
+ },
2398
+ {
2399
+ "ep_loss": 0.0,
2400
+ "epoch": 2.06,
2401
+ "learning_rate": 0.00013350000000000002,
2402
+ "loss": 0.7674,
2403
+ "mlm_loss": 0.7674,
2404
+ "step": 2670
2405
+ },
2406
+ {
2407
+ "ep_loss": 0.0,
2408
+ "epoch": 2.07,
2409
+ "learning_rate": 0.000134,
2410
+ "loss": 0.7737,
2411
+ "mlm_loss": 0.7737,
2412
+ "step": 2680
2413
+ },
2414
+ {
2415
+ "ep_loss": 0.0,
2416
+ "epoch": 2.07,
2417
+ "learning_rate": 0.00013450000000000002,
2418
+ "loss": 0.7639,
2419
+ "mlm_loss": 0.7639,
2420
+ "step": 2690
2421
+ },
2422
+ {
2423
+ "ep_loss": 0.0,
2424
+ "epoch": 2.08,
2425
+ "learning_rate": 0.000135,
2426
+ "loss": 0.7655,
2427
+ "mlm_loss": 0.7655,
2428
+ "step": 2700
2429
+ },
2430
+ {
2431
+ "epoch": 2.08,
2432
+ "eval_ep_loss": -2.082960605621338,
2433
+ "eval_loss": 0.7183709144592285,
2434
+ "eval_mlm_loss": 0.7183709144592285,
2435
+ "eval_runtime": 61.749,
2436
+ "eval_samples_per_second": 1131.5,
2437
+ "eval_steps_per_second": 0.567,
2438
+ "step": 2700
2439
+ },
2440
+ {
2441
+ "ep_loss": 0.0,
2442
+ "epoch": 2.09,
2443
+ "learning_rate": 0.00013550000000000001,
2444
+ "loss": 0.7611,
2445
+ "mlm_loss": 0.7611,
2446
+ "step": 2710
2447
+ },
2448
+ {
2449
+ "ep_loss": 0.0,
2450
+ "epoch": 2.1,
2451
+ "learning_rate": 0.00013600000000000003,
2452
+ "loss": 0.7586,
2453
+ "mlm_loss": 0.7586,
2454
+ "step": 2720
2455
+ },
2456
+ {
2457
+ "ep_loss": 0.0,
2458
+ "epoch": 2.11,
2459
+ "learning_rate": 0.0001365,
2460
+ "loss": 0.762,
2461
+ "mlm_loss": 0.762,
2462
+ "step": 2730
2463
+ },
2464
+ {
2465
+ "ep_loss": 0.0,
2466
+ "epoch": 2.11,
2467
+ "learning_rate": 0.00013700000000000002,
2468
+ "loss": 0.7479,
2469
+ "mlm_loss": 0.7479,
2470
+ "step": 2740
2471
+ },
2472
+ {
2473
+ "ep_loss": 0.0,
2474
+ "epoch": 2.12,
2475
+ "learning_rate": 0.0001375,
2476
+ "loss": 0.7514,
2477
+ "mlm_loss": 0.7514,
2478
+ "step": 2750
2479
+ },
2480
+ {
2481
+ "ep_loss": 0.0,
2482
+ "epoch": 2.13,
2483
+ "learning_rate": 0.00013800000000000002,
2484
+ "loss": 0.7476,
2485
+ "mlm_loss": 0.7476,
2486
+ "step": 2760
2487
+ },
2488
+ {
2489
+ "ep_loss": 0.0,
2490
+ "epoch": 2.14,
2491
+ "learning_rate": 0.0001385,
2492
+ "loss": 0.7629,
2493
+ "mlm_loss": 0.7629,
2494
+ "step": 2770
2495
+ },
2496
+ {
2497
+ "ep_loss": 0.0,
2498
+ "epoch": 2.14,
2499
+ "learning_rate": 0.00013900000000000002,
2500
+ "loss": 0.7432,
2501
+ "mlm_loss": 0.7432,
2502
+ "step": 2780
2503
+ },
2504
+ {
2505
+ "ep_loss": 0.0,
2506
+ "epoch": 2.15,
2507
+ "learning_rate": 0.0001395,
2508
+ "loss": 0.7449,
2509
+ "mlm_loss": 0.7449,
2510
+ "step": 2790
2511
+ },
2512
+ {
2513
+ "ep_loss": 0.0,
2514
+ "epoch": 2.16,
2515
+ "learning_rate": 0.00014000000000000001,
2516
+ "loss": 0.7379,
2517
+ "mlm_loss": 0.7379,
2518
+ "step": 2800
2519
+ },
2520
+ {
2521
+ "epoch": 2.16,
2522
+ "eval_ep_loss": -2.261542558670044,
2523
+ "eval_loss": 0.7018134593963623,
2524
+ "eval_mlm_loss": 0.7018134593963623,
2525
+ "eval_runtime": 60.9379,
2526
+ "eval_samples_per_second": 1146.561,
2527
+ "eval_steps_per_second": 0.574,
2528
+ "step": 2800
2529
+ },
2530
+ {
2531
+ "ep_loss": 0.0,
2532
+ "epoch": 2.17,
2533
+ "learning_rate": 0.00014050000000000003,
2534
+ "loss": 0.7465,
2535
+ "mlm_loss": 0.7465,
2536
+ "step": 2810
2537
+ },
2538
+ {
2539
+ "ep_loss": 0.0,
2540
+ "epoch": 2.18,
2541
+ "learning_rate": 0.00014099999999999998,
2542
+ "loss": 0.737,
2543
+ "mlm_loss": 0.737,
2544
+ "step": 2820
2545
+ },
2546
+ {
2547
+ "ep_loss": 0.0,
2548
+ "epoch": 2.18,
2549
+ "learning_rate": 0.0001415,
2550
+ "loss": 0.7377,
2551
+ "mlm_loss": 0.7377,
2552
+ "step": 2830
2553
+ },
2554
+ {
2555
+ "ep_loss": 0.0,
2556
+ "epoch": 2.19,
2557
+ "learning_rate": 0.00014199999999999998,
2558
+ "loss": 0.7401,
2559
+ "mlm_loss": 0.7401,
2560
+ "step": 2840
2561
+ },
2562
+ {
2563
+ "ep_loss": 0.0,
2564
+ "epoch": 2.2,
2565
+ "learning_rate": 0.0001425,
2566
+ "loss": 0.7266,
2567
+ "mlm_loss": 0.7266,
2568
+ "step": 2850
2569
+ },
2570
+ {
2571
+ "ep_loss": 0.0,
2572
+ "epoch": 2.21,
2573
+ "learning_rate": 0.00014299999999999998,
2574
+ "loss": 0.7443,
2575
+ "mlm_loss": 0.7443,
2576
+ "step": 2860
2577
+ },
2578
+ {
2579
+ "ep_loss": 0.0,
2580
+ "epoch": 2.21,
2581
+ "learning_rate": 0.0001435,
2582
+ "loss": 0.7301,
2583
+ "mlm_loss": 0.7301,
2584
+ "step": 2870
2585
+ },
2586
+ {
2587
+ "ep_loss": 0.0,
2588
+ "epoch": 2.22,
2589
+ "learning_rate": 0.000144,
2590
+ "loss": 0.7427,
2591
+ "mlm_loss": 0.7427,
2592
+ "step": 2880
2593
+ },
2594
+ {
2595
+ "ep_loss": 0.0,
2596
+ "epoch": 2.23,
2597
+ "learning_rate": 0.0001445,
2598
+ "loss": 0.7387,
2599
+ "mlm_loss": 0.7387,
2600
+ "step": 2890
2601
+ },
2602
+ {
2603
+ "ep_loss": 0.0,
2604
+ "epoch": 2.24,
2605
+ "learning_rate": 0.000145,
2606
+ "loss": 0.721,
2607
+ "mlm_loss": 0.721,
2608
+ "step": 2900
2609
+ },
2610
+ {
2611
+ "epoch": 2.24,
2612
+ "eval_ep_loss": -2.377054214477539,
2613
+ "eval_loss": 0.6826241612434387,
2614
+ "eval_mlm_loss": 0.6826241612434387,
2615
+ "eval_runtime": 60.9503,
2616
+ "eval_samples_per_second": 1146.327,
2617
+ "eval_steps_per_second": 0.574,
2618
+ "step": 2900
2619
+ },
2620
+ {
2621
+ "ep_loss": 0.0,
2622
+ "epoch": 2.24,
2623
+ "learning_rate": 0.00014549999999999999,
2624
+ "loss": 0.7352,
2625
+ "mlm_loss": 0.7352,
2626
+ "step": 2910
2627
+ },
2628
+ {
2629
+ "ep_loss": 0.0,
2630
+ "epoch": 2.25,
2631
+ "learning_rate": 0.000146,
2632
+ "loss": 0.7219,
2633
+ "mlm_loss": 0.7219,
2634
+ "step": 2920
2635
+ },
2636
+ {
2637
+ "ep_loss": 0.0,
2638
+ "epoch": 2.26,
2639
+ "learning_rate": 0.00014649999999999998,
2640
+ "loss": 0.729,
2641
+ "mlm_loss": 0.729,
2642
+ "step": 2930
2643
+ },
2644
+ {
2645
+ "ep_loss": 0.0,
2646
+ "epoch": 2.27,
2647
+ "learning_rate": 0.000147,
2648
+ "loss": 0.7148,
2649
+ "mlm_loss": 0.7148,
2650
+ "step": 2940
2651
+ },
2652
+ {
2653
+ "ep_loss": 0.0,
2654
+ "epoch": 2.28,
2655
+ "learning_rate": 0.0001475,
2656
+ "loss": 0.721,
2657
+ "mlm_loss": 0.721,
2658
+ "step": 2950
2659
+ },
2660
+ {
2661
+ "ep_loss": 0.0,
2662
+ "epoch": 2.28,
2663
+ "learning_rate": 0.000148,
2664
+ "loss": 0.7113,
2665
+ "mlm_loss": 0.7113,
2666
+ "step": 2960
2667
+ },
2668
+ {
2669
+ "ep_loss": 0.0,
2670
+ "epoch": 2.29,
2671
+ "learning_rate": 0.0001485,
2672
+ "loss": 0.711,
2673
+ "mlm_loss": 0.711,
2674
+ "step": 2970
2675
+ },
2676
+ {
2677
+ "ep_loss": 0.0,
2678
+ "epoch": 2.3,
2679
+ "learning_rate": 0.000149,
2680
+ "loss": 0.7109,
2681
+ "mlm_loss": 0.7109,
2682
+ "step": 2980
2683
+ },
2684
+ {
2685
+ "ep_loss": 0.0,
2686
+ "epoch": 2.31,
2687
+ "learning_rate": 0.0001495,
2688
+ "loss": 0.7121,
2689
+ "mlm_loss": 0.7121,
2690
+ "step": 2990
2691
+ },
2692
+ {
2693
+ "ep_loss": 0.0,
2694
+ "epoch": 2.31,
2695
+ "learning_rate": 0.00015,
2696
+ "loss": 0.7106,
2697
+ "mlm_loss": 0.7106,
2698
+ "step": 3000
2699
+ },
2700
+ {
2701
+ "epoch": 2.31,
2702
+ "eval_ep_loss": -2.0994629859924316,
2703
+ "eval_loss": 0.6672462821006775,
2704
+ "eval_mlm_loss": 0.6672462821006775,
2705
+ "eval_runtime": 63.0197,
2706
+ "eval_samples_per_second": 1108.686,
2707
+ "eval_steps_per_second": 0.555,
2708
+ "step": 3000
2709
+ },
2710
+ {
2711
+ "ep_loss": 0.0,
2712
+ "epoch": 2.32,
2713
+ "learning_rate": 0.0001505,
2714
+ "loss": 0.7067,
2715
+ "mlm_loss": 0.7067,
2716
+ "step": 3010
2717
+ },
2718
+ {
2719
+ "ep_loss": 0.0,
2720
+ "epoch": 2.33,
2721
+ "learning_rate": 0.000151,
2722
+ "loss": 0.7082,
2723
+ "mlm_loss": 0.7082,
2724
+ "step": 3020
2725
+ },
2726
+ {
2727
+ "ep_loss": 0.0,
2728
+ "epoch": 2.34,
2729
+ "learning_rate": 0.0001515,
2730
+ "loss": 0.7097,
2731
+ "mlm_loss": 0.7097,
2732
+ "step": 3030
2733
+ },
2734
+ {
2735
+ "ep_loss": 0.0,
2736
+ "epoch": 2.34,
2737
+ "learning_rate": 0.000152,
2738
+ "loss": 0.6972,
2739
+ "mlm_loss": 0.6972,
2740
+ "step": 3040
2741
+ },
2742
+ {
2743
+ "ep_loss": 0.0,
2744
+ "epoch": 2.35,
2745
+ "learning_rate": 0.0001525,
2746
+ "loss": 0.7051,
2747
+ "mlm_loss": 0.7051,
2748
+ "step": 3050
2749
+ },
2750
+ {
2751
+ "ep_loss": 0.0,
2752
+ "epoch": 2.36,
2753
+ "learning_rate": 0.000153,
2754
+ "loss": 0.7029,
2755
+ "mlm_loss": 0.7029,
2756
+ "step": 3060
2757
+ },
2758
+ {
2759
+ "ep_loss": 0.0,
2760
+ "epoch": 2.37,
2761
+ "learning_rate": 0.0001535,
2762
+ "loss": 0.7037,
2763
+ "mlm_loss": 0.7037,
2764
+ "step": 3070
2765
+ },
2766
+ {
2767
+ "ep_loss": 0.0,
2768
+ "epoch": 2.38,
2769
+ "learning_rate": 0.000154,
2770
+ "loss": 0.6966,
2771
+ "mlm_loss": 0.6966,
2772
+ "step": 3080
2773
+ },
2774
+ {
2775
+ "ep_loss": 0.0,
2776
+ "epoch": 2.38,
2777
+ "learning_rate": 0.00015450000000000001,
2778
+ "loss": 0.6948,
2779
+ "mlm_loss": 0.6948,
2780
+ "step": 3090
2781
+ },
2782
+ {
2783
+ "ep_loss": 0.0,
2784
+ "epoch": 2.39,
2785
+ "learning_rate": 0.000155,
2786
+ "loss": 0.6974,
2787
+ "mlm_loss": 0.6974,
2788
+ "step": 3100
2789
+ },
2790
+ {
2791
+ "epoch": 2.39,
2792
+ "eval_ep_loss": -2.283140182495117,
2793
+ "eval_loss": 0.6495695114135742,
2794
+ "eval_mlm_loss": 0.6495695114135742,
2795
+ "eval_runtime": 59.8554,
2796
+ "eval_samples_per_second": 1167.296,
2797
+ "eval_steps_per_second": 0.585,
2798
+ "step": 3100
2799
+ },
2800
+ {
2801
+ "ep_loss": 0.0,
2802
+ "epoch": 2.4,
2803
+ "learning_rate": 0.0001555,
2804
+ "loss": 0.6877,
2805
+ "mlm_loss": 0.6877,
2806
+ "step": 3110
2807
+ },
2808
+ {
2809
+ "ep_loss": 0.0,
2810
+ "epoch": 2.41,
2811
+ "learning_rate": 0.000156,
2812
+ "loss": 0.6862,
2813
+ "mlm_loss": 0.6862,
2814
+ "step": 3120
2815
+ },
2816
+ {
2817
+ "ep_loss": 0.0,
2818
+ "epoch": 2.41,
2819
+ "learning_rate": 0.0001565,
2820
+ "loss": 0.6893,
2821
+ "mlm_loss": 0.6893,
2822
+ "step": 3130
2823
+ },
2824
+ {
2825
+ "ep_loss": 0.0,
2826
+ "epoch": 2.42,
2827
+ "learning_rate": 0.000157,
2828
+ "loss": 0.6943,
2829
+ "mlm_loss": 0.6943,
2830
+ "step": 3140
2831
+ },
2832
+ {
2833
+ "ep_loss": 0.0,
2834
+ "epoch": 2.43,
2835
+ "learning_rate": 0.0001575,
2836
+ "loss": 0.6804,
2837
+ "mlm_loss": 0.6804,
2838
+ "step": 3150
2839
+ },
2840
+ {
2841
+ "ep_loss": 0.0,
2842
+ "epoch": 2.44,
2843
+ "learning_rate": 0.000158,
2844
+ "loss": 0.6813,
2845
+ "mlm_loss": 0.6813,
2846
+ "step": 3160
2847
+ },
2848
+ {
2849
+ "ep_loss": 0.0,
2850
+ "epoch": 2.45,
2851
+ "learning_rate": 0.0001585,
2852
+ "loss": 0.6828,
2853
+ "mlm_loss": 0.6828,
2854
+ "step": 3170
2855
+ },
2856
+ {
2857
+ "ep_loss": 0.0,
2858
+ "epoch": 2.45,
2859
+ "learning_rate": 0.00015900000000000002,
2860
+ "loss": 0.6736,
2861
+ "mlm_loss": 0.6736,
2862
+ "step": 3180
2863
+ },
2864
+ {
2865
+ "ep_loss": 0.0,
2866
+ "epoch": 2.46,
2867
+ "learning_rate": 0.0001595,
2868
+ "loss": 0.6763,
2869
+ "mlm_loss": 0.6763,
2870
+ "step": 3190
2871
+ },
2872
+ {
2873
+ "ep_loss": 0.0,
2874
+ "epoch": 2.47,
2875
+ "learning_rate": 0.00016,
2876
+ "loss": 0.6796,
2877
+ "mlm_loss": 0.6796,
2878
+ "step": 3200
2879
+ },
2880
+ {
2881
+ "epoch": 2.47,
2882
+ "eval_ep_loss": -2.57122540473938,
2883
+ "eval_loss": 0.6341940760612488,
2884
+ "eval_mlm_loss": 0.6341940760612488,
2885
+ "eval_runtime": 63.1531,
2886
+ "eval_samples_per_second": 1106.343,
2887
+ "eval_steps_per_second": 0.554,
2888
+ "step": 3200
2889
+ },
2890
+ {
2891
+ "ep_loss": 0.0,
2892
+ "epoch": 2.48,
2893
+ "learning_rate": 0.0001605,
2894
+ "loss": 0.6688,
2895
+ "mlm_loss": 0.6688,
2896
+ "step": 3210
2897
+ },
2898
+ {
2899
+ "ep_loss": 0.0,
2900
+ "epoch": 2.48,
2901
+ "learning_rate": 0.000161,
2902
+ "loss": 0.6725,
2903
+ "mlm_loss": 0.6725,
2904
+ "step": 3220
2905
+ },
2906
+ {
2907
+ "ep_loss": 0.0,
2908
+ "epoch": 2.49,
2909
+ "learning_rate": 0.0001615,
2910
+ "loss": 0.6718,
2911
+ "mlm_loss": 0.6718,
2912
+ "step": 3230
2913
+ },
2914
+ {
2915
+ "ep_loss": 0.0,
2916
+ "epoch": 2.5,
2917
+ "learning_rate": 0.000162,
2918
+ "loss": 0.6675,
2919
+ "mlm_loss": 0.6675,
2920
+ "step": 3240
2921
+ },
2922
+ {
2923
+ "ep_loss": 0.0,
2924
+ "epoch": 2.51,
2925
+ "learning_rate": 0.00016250000000000002,
2926
+ "loss": 0.6665,
2927
+ "mlm_loss": 0.6665,
2928
+ "step": 3250
2929
+ },
2930
+ {
2931
+ "ep_loss": 0.0,
2932
+ "epoch": 2.51,
2933
+ "learning_rate": 0.000163,
2934
+ "loss": 0.6615,
2935
+ "mlm_loss": 0.6615,
2936
+ "step": 3260
2937
+ },
2938
+ {
2939
+ "ep_loss": 0.0,
2940
+ "epoch": 2.52,
2941
+ "learning_rate": 0.00016350000000000002,
2942
+ "loss": 0.6784,
2943
+ "mlm_loss": 0.6784,
2944
+ "step": 3270
2945
+ },
2946
+ {
2947
+ "ep_loss": 0.0,
2948
+ "epoch": 2.53,
2949
+ "learning_rate": 0.000164,
2950
+ "loss": 0.6686,
2951
+ "mlm_loss": 0.6686,
2952
+ "step": 3280
2953
+ },
2954
+ {
2955
+ "ep_loss": 0.0,
2956
+ "epoch": 2.54,
2957
+ "learning_rate": 0.00016450000000000001,
2958
+ "loss": 0.6697,
2959
+ "mlm_loss": 0.6697,
2960
+ "step": 3290
2961
+ },
2962
+ {
2963
+ "ep_loss": 0.0,
2964
+ "epoch": 2.55,
2965
+ "learning_rate": 0.000165,
2966
+ "loss": 0.6621,
2967
+ "mlm_loss": 0.6621,
2968
+ "step": 3300
2969
+ },
2970
+ {
2971
+ "epoch": 2.55,
2972
+ "eval_ep_loss": -2.320197820663452,
2973
+ "eval_loss": 0.6192212104797363,
2974
+ "eval_mlm_loss": 0.6192212104797363,
2975
+ "eval_runtime": 60.3086,
2976
+ "eval_samples_per_second": 1158.524,
2977
+ "eval_steps_per_second": 0.58,
2978
+ "step": 3300
2979
+ },
2980
+ {
2981
+ "ep_loss": 0.0,
2982
+ "epoch": 2.55,
2983
+ "learning_rate": 0.0001655,
2984
+ "loss": 0.6645,
2985
+ "mlm_loss": 0.6645,
2986
+ "step": 3310
2987
+ },
2988
+ {
2989
+ "ep_loss": 0.0,
2990
+ "epoch": 2.56,
2991
+ "learning_rate": 0.00016600000000000002,
2992
+ "loss": 0.6594,
2993
+ "mlm_loss": 0.6594,
2994
+ "step": 3320
2995
+ },
2996
+ {
2997
+ "ep_loss": 0.0,
2998
+ "epoch": 2.57,
2999
+ "learning_rate": 0.0001665,
3000
+ "loss": 0.6585,
3001
+ "mlm_loss": 0.6585,
3002
+ "step": 3330
3003
+ },
3004
+ {
3005
+ "ep_loss": 0.0,
3006
+ "epoch": 2.58,
3007
+ "learning_rate": 0.00016700000000000002,
3008
+ "loss": 0.6499,
3009
+ "mlm_loss": 0.6499,
3010
+ "step": 3340
3011
+ },
3012
+ {
3013
+ "ep_loss": 0.0,
3014
+ "epoch": 2.58,
3015
+ "learning_rate": 0.0001675,
3016
+ "loss": 0.641,
3017
+ "mlm_loss": 0.641,
3018
+ "step": 3350
3019
+ },
3020
+ {
3021
+ "ep_loss": 0.0,
3022
+ "epoch": 2.59,
3023
+ "learning_rate": 0.00016800000000000002,
3024
+ "loss": 0.653,
3025
+ "mlm_loss": 0.653,
3026
+ "step": 3360
3027
+ },
3028
+ {
3029
+ "ep_loss": 0.0,
3030
+ "epoch": 2.6,
3031
+ "learning_rate": 0.0001685,
3032
+ "loss": 0.6526,
3033
+ "mlm_loss": 0.6526,
3034
+ "step": 3370
3035
+ },
3036
+ {
3037
+ "ep_loss": 0.0,
3038
+ "epoch": 2.61,
3039
+ "learning_rate": 0.00016900000000000002,
3040
+ "loss": 0.6588,
3041
+ "mlm_loss": 0.6588,
3042
+ "step": 3380
3043
+ },
3044
+ {
3045
+ "ep_loss": 0.0,
3046
+ "epoch": 2.61,
3047
+ "learning_rate": 0.00016950000000000003,
3048
+ "loss": 0.6457,
3049
+ "mlm_loss": 0.6457,
3050
+ "step": 3390
3051
+ },
3052
+ {
3053
+ "ep_loss": 0.0,
3054
+ "epoch": 2.62,
3055
+ "learning_rate": 0.00017,
3056
+ "loss": 0.6464,
3057
+ "mlm_loss": 0.6464,
3058
+ "step": 3400
3059
+ },
3060
+ {
3061
+ "epoch": 2.62,
3062
+ "eval_ep_loss": -2.441110610961914,
3063
+ "eval_loss": 0.6078351736068726,
3064
+ "eval_mlm_loss": 0.6078351736068726,
3065
+ "eval_runtime": 60.4806,
3066
+ "eval_samples_per_second": 1155.23,
3067
+ "eval_steps_per_second": 0.579,
3068
+ "step": 3400
3069
+ },
3070
+ {
3071
+ "ep_loss": 0.0,
3072
+ "epoch": 2.63,
3073
+ "learning_rate": 0.00017050000000000002,
3074
+ "loss": 0.6517,
3075
+ "mlm_loss": 0.6517,
3076
+ "step": 3410
3077
+ },
3078
+ {
3079
+ "ep_loss": 0.0,
3080
+ "epoch": 2.64,
3081
+ "learning_rate": 0.000171,
3082
+ "loss": 0.6473,
3083
+ "mlm_loss": 0.6473,
3084
+ "step": 3420
3085
+ },
3086
+ {
3087
+ "ep_loss": 0.0,
3088
+ "epoch": 2.65,
3089
+ "learning_rate": 0.00017150000000000002,
3090
+ "loss": 0.6475,
3091
+ "mlm_loss": 0.6475,
3092
+ "step": 3430
3093
+ },
3094
+ {
3095
+ "ep_loss": 0.0,
3096
+ "epoch": 2.65,
3097
+ "learning_rate": 0.00017199999999999998,
3098
+ "loss": 0.649,
3099
+ "mlm_loss": 0.649,
3100
+ "step": 3440
3101
+ },
3102
+ {
3103
+ "ep_loss": 0.0,
3104
+ "epoch": 2.66,
3105
+ "learning_rate": 0.0001725,
3106
+ "loss": 0.652,
3107
+ "mlm_loss": 0.652,
3108
+ "step": 3450
3109
+ },
3110
+ {
3111
+ "ep_loss": 0.0,
3112
+ "epoch": 2.67,
3113
+ "learning_rate": 0.000173,
3114
+ "loss": 0.6484,
3115
+ "mlm_loss": 0.6484,
3116
+ "step": 3460
3117
+ },
3118
+ {
3119
+ "ep_loss": 0.0,
3120
+ "epoch": 2.68,
3121
+ "learning_rate": 0.0001735,
3122
+ "loss": 0.6405,
3123
+ "mlm_loss": 0.6405,
3124
+ "step": 3470
3125
+ },
3126
+ {
3127
+ "ep_loss": 0.0,
3128
+ "epoch": 2.68,
3129
+ "learning_rate": 0.000174,
3130
+ "loss": 0.6421,
3131
+ "mlm_loss": 0.6421,
3132
+ "step": 3480
3133
+ },
3134
+ {
3135
+ "ep_loss": 0.0,
3136
+ "epoch": 2.69,
3137
+ "learning_rate": 0.00017449999999999999,
3138
+ "loss": 0.6313,
3139
+ "mlm_loss": 0.6313,
3140
+ "step": 3490
3141
+ },
3142
+ {
3143
+ "ep_loss": 0.0,
3144
+ "epoch": 2.7,
3145
+ "learning_rate": 0.000175,
3146
+ "loss": 0.6228,
3147
+ "mlm_loss": 0.6228,
3148
+ "step": 3500
3149
+ },
3150
+ {
3151
+ "epoch": 2.7,
3152
+ "eval_ep_loss": -2.3252084255218506,
3153
+ "eval_loss": 0.5944607257843018,
3154
+ "eval_mlm_loss": 0.5944607257843018,
3155
+ "eval_runtime": 59.8998,
3156
+ "eval_samples_per_second": 1166.431,
3157
+ "eval_steps_per_second": 0.584,
3158
+ "step": 3500
3159
+ },
3160
+ {
3161
+ "ep_loss": 0.0,
3162
+ "epoch": 2.71,
3163
+ "learning_rate": 0.00017549999999999998,
3164
+ "loss": 0.6276,
3165
+ "mlm_loss": 0.6276,
3166
+ "step": 3510
3167
+ },
3168
+ {
3169
+ "ep_loss": 0.0,
3170
+ "epoch": 2.72,
3171
+ "learning_rate": 0.000176,
3172
+ "loss": 0.632,
3173
+ "mlm_loss": 0.632,
3174
+ "step": 3520
3175
+ },
3176
+ {
3177
+ "ep_loss": 0.0,
3178
+ "epoch": 2.72,
3179
+ "learning_rate": 0.00017649999999999998,
3180
+ "loss": 0.629,
3181
+ "mlm_loss": 0.629,
3182
+ "step": 3530
3183
+ },
3184
+ {
3185
+ "ep_loss": 0.0,
3186
+ "epoch": 2.73,
3187
+ "learning_rate": 0.000177,
3188
+ "loss": 0.6327,
3189
+ "mlm_loss": 0.6327,
3190
+ "step": 3540
3191
+ },
3192
+ {
3193
+ "ep_loss": 0.0,
3194
+ "epoch": 2.74,
3195
+ "learning_rate": 0.0001775,
3196
+ "loss": 0.6307,
3197
+ "mlm_loss": 0.6307,
3198
+ "step": 3550
3199
+ },
3200
+ {
3201
+ "ep_loss": 0.0,
3202
+ "epoch": 2.75,
3203
+ "learning_rate": 0.000178,
3204
+ "loss": 0.6156,
3205
+ "mlm_loss": 0.6156,
3206
+ "step": 3560
3207
+ },
3208
+ {
3209
+ "ep_loss": 0.0,
3210
+ "epoch": 2.75,
3211
+ "learning_rate": 0.0001785,
3212
+ "loss": 0.6246,
3213
+ "mlm_loss": 0.6246,
3214
+ "step": 3570
3215
+ },
3216
+ {
3217
+ "ep_loss": 0.0,
3218
+ "epoch": 2.76,
3219
+ "learning_rate": 0.000179,
3220
+ "loss": 0.6223,
3221
+ "mlm_loss": 0.6223,
3222
+ "step": 3580
3223
+ },
3224
+ {
3225
+ "ep_loss": 0.0,
3226
+ "epoch": 2.77,
3227
+ "learning_rate": 0.0001795,
3228
+ "loss": 0.6359,
3229
+ "mlm_loss": 0.6359,
3230
+ "step": 3590
3231
+ },
3232
+ {
3233
+ "ep_loss": 0.0,
3234
+ "epoch": 2.78,
3235
+ "learning_rate": 0.00017999999999999998,
3236
+ "loss": 0.6192,
3237
+ "mlm_loss": 0.6192,
3238
+ "step": 3600
3239
+ },
3240
+ {
3241
+ "epoch": 2.78,
3242
+ "eval_ep_loss": -2.543795585632324,
3243
+ "eval_loss": 0.5830370783805847,
3244
+ "eval_mlm_loss": 0.5830370783805847,
3245
+ "eval_runtime": 62.3115,
3246
+ "eval_samples_per_second": 1121.285,
3247
+ "eval_steps_per_second": 0.562,
3248
+ "step": 3600
3249
+ },
3250
+ {
3251
+ "ep_loss": 0.0,
3252
+ "epoch": 2.78,
3253
+ "learning_rate": 0.0001805,
3254
+ "loss": 0.619,
3255
+ "mlm_loss": 0.619,
3256
+ "step": 3610
3257
+ },
3258
+ {
3259
+ "ep_loss": 0.0,
3260
+ "epoch": 2.79,
3261
+ "learning_rate": 0.000181,
3262
+ "loss": 0.6107,
3263
+ "mlm_loss": 0.6107,
3264
+ "step": 3620
3265
+ },
3266
+ {
3267
+ "ep_loss": 0.0,
3268
+ "epoch": 2.8,
3269
+ "learning_rate": 0.0001815,
3270
+ "loss": 0.6287,
3271
+ "mlm_loss": 0.6287,
3272
+ "step": 3630
3273
+ },
3274
+ {
3275
+ "ep_loss": 0.0,
3276
+ "epoch": 2.81,
3277
+ "learning_rate": 0.000182,
3278
+ "loss": 0.616,
3279
+ "mlm_loss": 0.616,
3280
+ "step": 3640
3281
+ },
3282
+ {
3283
+ "ep_loss": 0.0,
3284
+ "epoch": 2.82,
3285
+ "learning_rate": 0.0001825,
3286
+ "loss": 0.6199,
3287
+ "mlm_loss": 0.6199,
3288
+ "step": 3650
3289
+ },
3290
+ {
3291
+ "ep_loss": 0.0,
3292
+ "epoch": 2.82,
3293
+ "learning_rate": 0.000183,
3294
+ "loss": 0.6179,
3295
+ "mlm_loss": 0.6179,
3296
+ "step": 3660
3297
+ },
3298
+ {
3299
+ "ep_loss": 0.0,
3300
+ "epoch": 2.83,
3301
+ "learning_rate": 0.0001835,
3302
+ "loss": 0.6159,
3303
+ "mlm_loss": 0.6159,
3304
+ "step": 3670
3305
+ },
3306
+ {
3307
+ "ep_loss": 0.0,
3308
+ "epoch": 2.84,
3309
+ "learning_rate": 0.000184,
3310
+ "loss": 0.6067,
3311
+ "mlm_loss": 0.6067,
3312
+ "step": 3680
3313
+ },
3314
+ {
3315
+ "ep_loss": 0.0,
3316
+ "epoch": 2.85,
3317
+ "learning_rate": 0.0001845,
3318
+ "loss": 0.6079,
3319
+ "mlm_loss": 0.6079,
3320
+ "step": 3690
3321
+ },
3322
+ {
3323
+ "ep_loss": 0.0,
3324
+ "epoch": 2.85,
3325
+ "learning_rate": 0.000185,
3326
+ "loss": 0.6063,
3327
+ "mlm_loss": 0.6063,
3328
+ "step": 3700
3329
+ },
3330
+ {
3331
+ "epoch": 2.85,
3332
+ "eval_ep_loss": -2.2521820068359375,
3333
+ "eval_loss": 0.5688546299934387,
3334
+ "eval_mlm_loss": 0.5688546299934387,
3335
+ "eval_runtime": 60.0349,
3336
+ "eval_samples_per_second": 1163.807,
3337
+ "eval_steps_per_second": 0.583,
3338
+ "step": 3700
3339
+ },
3340
+ {
3341
+ "ep_loss": 0.0,
3342
+ "epoch": 2.86,
3343
+ "learning_rate": 0.0001855,
3344
+ "loss": 0.6156,
3345
+ "mlm_loss": 0.6156,
3346
+ "step": 3710
3347
+ },
3348
+ {
3349
+ "ep_loss": 0.0,
3350
+ "epoch": 2.87,
3351
+ "learning_rate": 0.000186,
3352
+ "loss": 0.6093,
3353
+ "mlm_loss": 0.6093,
3354
+ "step": 3720
3355
+ },
3356
+ {
3357
+ "ep_loss": 0.0,
3358
+ "epoch": 2.88,
3359
+ "learning_rate": 0.0001865,
3360
+ "loss": 0.6073,
3361
+ "mlm_loss": 0.6073,
3362
+ "step": 3730
3363
+ },
3364
+ {
3365
+ "ep_loss": 0.0,
3366
+ "epoch": 2.88,
3367
+ "learning_rate": 0.000187,
3368
+ "loss": 0.6039,
3369
+ "mlm_loss": 0.6039,
3370
+ "step": 3740
3371
+ },
3372
+ {
3373
+ "ep_loss": 0.0,
3374
+ "epoch": 2.89,
3375
+ "learning_rate": 0.0001875,
3376
+ "loss": 0.5977,
3377
+ "mlm_loss": 0.5977,
3378
+ "step": 3750
3379
+ },
3380
+ {
3381
+ "ep_loss": 0.0,
3382
+ "epoch": 2.9,
3383
+ "learning_rate": 0.00018800000000000002,
3384
+ "loss": 0.6,
3385
+ "mlm_loss": 0.6,
3386
+ "step": 3760
3387
+ },
3388
+ {
3389
+ "ep_loss": 0.0,
3390
+ "epoch": 2.91,
3391
+ "learning_rate": 0.0001885,
3392
+ "loss": 0.6022,
3393
+ "mlm_loss": 0.6022,
3394
+ "step": 3770
3395
+ },
3396
+ {
3397
+ "ep_loss": 0.0,
3398
+ "epoch": 2.92,
3399
+ "learning_rate": 0.000189,
3400
+ "loss": 0.6015,
3401
+ "mlm_loss": 0.6015,
3402
+ "step": 3780
3403
+ },
3404
+ {
3405
+ "ep_loss": 0.0,
3406
+ "epoch": 2.92,
3407
+ "learning_rate": 0.0001895,
3408
+ "loss": 0.6016,
3409
+ "mlm_loss": 0.6016,
3410
+ "step": 3790
3411
+ },
3412
+ {
3413
+ "ep_loss": 0.0,
3414
+ "epoch": 2.93,
3415
+ "learning_rate": 0.00019,
3416
+ "loss": 0.5959,
3417
+ "mlm_loss": 0.5959,
3418
+ "step": 3800
3419
+ },
3420
+ {
3421
+ "epoch": 2.93,
3422
+ "eval_ep_loss": -2.366133213043213,
3423
+ "eval_loss": 0.5573724508285522,
3424
+ "eval_mlm_loss": 0.5573724508285522,
3425
+ "eval_runtime": 61.8753,
3426
+ "eval_samples_per_second": 1129.191,
3427
+ "eval_steps_per_second": 0.566,
3428
+ "step": 3800
3429
+ },
3430
+ {
3431
+ "ep_loss": 0.0,
3432
+ "epoch": 2.94,
3433
+ "learning_rate": 0.0001905,
3434
+ "loss": 0.6003,
3435
+ "mlm_loss": 0.6003,
3436
+ "step": 3810
3437
+ },
3438
+ {
3439
+ "ep_loss": 0.0,
3440
+ "epoch": 2.95,
3441
+ "learning_rate": 0.000191,
3442
+ "loss": 0.5984,
3443
+ "mlm_loss": 0.5984,
3444
+ "step": 3820
3445
+ },
3446
+ {
3447
+ "ep_loss": 0.0,
3448
+ "epoch": 2.95,
3449
+ "learning_rate": 0.00019150000000000002,
3450
+ "loss": 0.6005,
3451
+ "mlm_loss": 0.6005,
3452
+ "step": 3830
3453
+ },
3454
+ {
3455
+ "ep_loss": 0.0,
3456
+ "epoch": 2.96,
3457
+ "learning_rate": 0.000192,
3458
+ "loss": 0.5895,
3459
+ "mlm_loss": 0.5895,
3460
+ "step": 3840
3461
+ },
3462
+ {
3463
+ "ep_loss": 0.0,
3464
+ "epoch": 2.97,
3465
+ "learning_rate": 0.00019250000000000002,
3466
+ "loss": 0.5912,
3467
+ "mlm_loss": 0.5912,
3468
+ "step": 3850
3469
+ },
3470
+ {
3471
+ "ep_loss": 0.0,
3472
+ "epoch": 2.98,
3473
+ "learning_rate": 0.000193,
3474
+ "loss": 0.5828,
3475
+ "mlm_loss": 0.5828,
3476
+ "step": 3860
3477
+ },
3478
+ {
3479
+ "ep_loss": 0.0,
3480
+ "epoch": 2.98,
3481
+ "learning_rate": 0.00019350000000000001,
3482
+ "loss": 0.5865,
3483
+ "mlm_loss": 0.5865,
3484
+ "step": 3870
3485
+ },
3486
+ {
3487
+ "ep_loss": 0.0,
3488
+ "epoch": 2.99,
3489
+ "learning_rate": 0.000194,
3490
+ "loss": 0.5854,
3491
+ "mlm_loss": 0.5854,
3492
+ "step": 3880
3493
+ },
3494
+ {
3495
+ "ep_loss": 0.0,
3496
+ "epoch": 3.0,
3497
+ "learning_rate": 0.0001945,
3498
+ "loss": 0.5883,
3499
+ "mlm_loss": 0.5883,
3500
+ "step": 3890
3501
+ },
3502
+ {
3503
+ "ep_loss": 0.0,
3504
+ "epoch": 3.01,
3505
+ "learning_rate": 0.00019500000000000002,
3506
+ "loss": 0.5743,
3507
+ "mlm_loss": 0.5743,
3508
+ "step": 3900
3509
+ },
3510
+ {
3511
+ "epoch": 3.01,
3512
+ "eval_ep_loss": -2.332629680633545,
3513
+ "eval_loss": 0.551331102848053,
3514
+ "eval_mlm_loss": 0.551331102848053,
3515
+ "eval_runtime": 61.0219,
3516
+ "eval_samples_per_second": 1144.983,
3517
+ "eval_steps_per_second": 0.574,
3518
+ "step": 3900
3519
+ },
3520
+ {
3521
+ "ep_loss": 0.0,
3522
+ "epoch": 3.02,
3523
+ "learning_rate": 0.0001955,
3524
+ "loss": 0.5813,
3525
+ "mlm_loss": 0.5813,
3526
+ "step": 3910
3527
+ },
3528
+ {
3529
+ "ep_loss": 0.0,
3530
+ "epoch": 3.02,
3531
+ "learning_rate": 0.00019600000000000002,
3532
+ "loss": 0.5813,
3533
+ "mlm_loss": 0.5813,
3534
+ "step": 3920
3535
+ },
3536
+ {
3537
+ "ep_loss": 0.0,
3538
+ "epoch": 3.03,
3539
+ "learning_rate": 0.0001965,
3540
+ "loss": 0.5899,
3541
+ "mlm_loss": 0.5899,
3542
+ "step": 3930
3543
+ },
3544
+ {
3545
+ "ep_loss": 0.0,
3546
+ "epoch": 3.04,
3547
+ "learning_rate": 0.00019700000000000002,
3548
+ "loss": 0.5849,
3549
+ "mlm_loss": 0.5849,
3550
+ "step": 3940
3551
+ },
3552
+ {
3553
+ "ep_loss": 0.0,
3554
+ "epoch": 3.05,
3555
+ "learning_rate": 0.0001975,
3556
+ "loss": 0.5823,
3557
+ "mlm_loss": 0.5823,
3558
+ "step": 3950
3559
+ },
3560
+ {
3561
+ "ep_loss": 0.0,
3562
+ "epoch": 3.05,
3563
+ "learning_rate": 0.00019800000000000002,
3564
+ "loss": 0.5794,
3565
+ "mlm_loss": 0.5794,
3566
+ "step": 3960
3567
+ },
3568
+ {
3569
+ "ep_loss": 0.0,
3570
+ "epoch": 3.06,
3571
+ "learning_rate": 0.00019850000000000003,
3572
+ "loss": 0.5837,
3573
+ "mlm_loss": 0.5837,
3574
+ "step": 3970
3575
+ },
3576
+ {
3577
+ "ep_loss": 0.0,
3578
+ "epoch": 3.07,
3579
+ "learning_rate": 0.000199,
3580
+ "loss": 0.5783,
3581
+ "mlm_loss": 0.5783,
3582
+ "step": 3980
3583
+ },
3584
+ {
3585
+ "ep_loss": 0.0,
3586
+ "epoch": 3.08,
3587
+ "learning_rate": 0.00019950000000000002,
3588
+ "loss": 0.5733,
3589
+ "mlm_loss": 0.5733,
3590
+ "step": 3990
3591
+ },
3592
+ {
3593
+ "ep_loss": 0.0,
3594
+ "epoch": 3.09,
3595
+ "learning_rate": 0.0002,
3596
+ "loss": 0.5701,
3597
+ "mlm_loss": 0.5701,
3598
+ "step": 4000
3599
+ },
3600
+ {
3601
+ "epoch": 3.09,
3602
+ "eval_ep_loss": -2.3982014656066895,
3603
+ "eval_loss": 0.5421211123466492,
3604
+ "eval_mlm_loss": 0.5421211123466492,
3605
+ "eval_runtime": 62.0302,
3606
+ "eval_samples_per_second": 1126.37,
3607
+ "eval_steps_per_second": 0.564,
3608
+ "step": 4000
3609
+ },
3610
+ {
3611
+ "ep_loss": 0.0,
3612
+ "epoch": 3.09,
3613
+ "learning_rate": 0.00020050000000000002,
3614
+ "loss": 0.5737,
3615
+ "mlm_loss": 0.5737,
3616
+ "step": 4010
3617
+ },
3618
+ {
3619
+ "ep_loss": 0.0,
3620
+ "epoch": 3.1,
3621
+ "learning_rate": 0.000201,
3622
+ "loss": 0.5645,
3623
+ "mlm_loss": 0.5645,
3624
+ "step": 4020
3625
+ },
3626
+ {
3627
+ "ep_loss": 0.0,
3628
+ "epoch": 3.11,
3629
+ "learning_rate": 0.00020150000000000002,
3630
+ "loss": 0.5758,
3631
+ "mlm_loss": 0.5758,
3632
+ "step": 4030
3633
+ },
3634
+ {
3635
+ "ep_loss": 0.0,
3636
+ "epoch": 3.12,
3637
+ "learning_rate": 0.000202,
3638
+ "loss": 0.5849,
3639
+ "mlm_loss": 0.5849,
3640
+ "step": 4040
3641
+ },
3642
+ {
3643
+ "ep_loss": 0.0,
3644
+ "epoch": 3.12,
3645
+ "learning_rate": 0.00020250000000000002,
3646
+ "loss": 0.5667,
3647
+ "mlm_loss": 0.5667,
3648
+ "step": 4050
3649
+ },
3650
+ {
3651
+ "ep_loss": 0.0,
3652
+ "epoch": 3.13,
3653
+ "learning_rate": 0.00020300000000000003,
3654
+ "loss": 0.5726,
3655
+ "mlm_loss": 0.5726,
3656
+ "step": 4060
3657
+ },
3658
+ {
3659
+ "ep_loss": 0.0,
3660
+ "epoch": 3.14,
3661
+ "learning_rate": 0.00020349999999999999,
3662
+ "loss": 0.5645,
3663
+ "mlm_loss": 0.5645,
3664
+ "step": 4070
3665
+ },
3666
+ {
3667
+ "ep_loss": 0.0,
3668
+ "epoch": 3.15,
3669
+ "learning_rate": 0.000204,
3670
+ "loss": 0.5681,
3671
+ "mlm_loss": 0.5681,
3672
+ "step": 4080
3673
+ },
3674
+ {
3675
+ "ep_loss": 0.0,
3676
+ "epoch": 3.15,
3677
+ "learning_rate": 0.00020449999999999998,
3678
+ "loss": 0.564,
3679
+ "mlm_loss": 0.564,
3680
+ "step": 4090
3681
+ },
3682
+ {
3683
+ "ep_loss": 0.0,
3684
+ "epoch": 3.16,
3685
+ "learning_rate": 0.000205,
3686
+ "loss": 0.5655,
3687
+ "mlm_loss": 0.5655,
3688
+ "step": 4100
3689
+ },
3690
+ {
3691
+ "epoch": 3.16,
3692
+ "eval_ep_loss": -2.218219041824341,
3693
+ "eval_loss": 0.5314173698425293,
3694
+ "eval_mlm_loss": 0.5314173698425293,
3695
+ "eval_runtime": 60.8804,
3696
+ "eval_samples_per_second": 1147.644,
3697
+ "eval_steps_per_second": 0.575,
3698
+ "step": 4100
3699
+ },
3700
+ {
3701
+ "ep_loss": 0.0,
3702
+ "epoch": 3.17,
3703
+ "learning_rate": 0.00020549999999999998,
3704
+ "loss": 0.5654,
3705
+ "mlm_loss": 0.5654,
3706
+ "step": 4110
3707
+ },
3708
+ {
3709
+ "ep_loss": 0.0,
3710
+ "epoch": 3.18,
3711
+ "learning_rate": 0.000206,
3712
+ "loss": 0.5647,
3713
+ "mlm_loss": 0.5647,
3714
+ "step": 4120
3715
+ },
3716
+ {
3717
+ "ep_loss": 0.0,
3718
+ "epoch": 3.19,
3719
+ "learning_rate": 0.0002065,
3720
+ "loss": 0.568,
3721
+ "mlm_loss": 0.568,
3722
+ "step": 4130
3723
+ },
3724
+ {
3725
+ "ep_loss": 0.0,
3726
+ "epoch": 3.19,
3727
+ "learning_rate": 0.000207,
3728
+ "loss": 0.5538,
3729
+ "mlm_loss": 0.5538,
3730
+ "step": 4140
3731
+ },
3732
+ {
3733
+ "ep_loss": 0.0,
3734
+ "epoch": 3.2,
3735
+ "learning_rate": 0.0002075,
3736
+ "loss": 0.5585,
3737
+ "mlm_loss": 0.5585,
3738
+ "step": 4150
3739
+ },
3740
+ {
3741
+ "ep_loss": 0.0,
3742
+ "epoch": 3.21,
3743
+ "learning_rate": 0.000208,
3744
+ "loss": 0.5578,
3745
+ "mlm_loss": 0.5578,
3746
+ "step": 4160
3747
+ },
3748
+ {
3749
+ "ep_loss": 0.0,
3750
+ "epoch": 3.22,
3751
+ "learning_rate": 0.0002085,
3752
+ "loss": 0.5508,
3753
+ "mlm_loss": 0.5508,
3754
+ "step": 4170
3755
+ },
3756
+ {
3757
+ "ep_loss": 0.0,
3758
+ "epoch": 3.22,
3759
+ "learning_rate": 0.00020899999999999998,
3760
+ "loss": 0.5644,
3761
+ "mlm_loss": 0.5644,
3762
+ "step": 4180
3763
+ },
3764
+ {
3765
+ "ep_loss": 0.0,
3766
+ "epoch": 3.23,
3767
+ "learning_rate": 0.0002095,
3768
+ "loss": 0.5642,
3769
+ "mlm_loss": 0.5642,
3770
+ "step": 4190
3771
+ },
3772
+ {
3773
+ "ep_loss": 0.0,
3774
+ "epoch": 3.24,
3775
+ "learning_rate": 0.00021,
3776
+ "loss": 0.561,
3777
+ "mlm_loss": 0.561,
3778
+ "step": 4200
3779
+ },
3780
+ {
3781
+ "epoch": 3.24,
3782
+ "eval_ep_loss": -2.287240743637085,
3783
+ "eval_loss": 0.5224212408065796,
3784
+ "eval_mlm_loss": 0.5224212408065796,
3785
+ "eval_runtime": 61.3268,
3786
+ "eval_samples_per_second": 1139.291,
3787
+ "eval_steps_per_second": 0.571,
3788
+ "step": 4200
3789
+ },
3790
+ {
3791
+ "ep_loss": 0.0,
3792
+ "epoch": 3.25,
3793
+ "learning_rate": 0.0002105,
3794
+ "loss": 0.5564,
3795
+ "mlm_loss": 0.5564,
3796
+ "step": 4210
3797
+ },
3798
+ {
3799
+ "ep_loss": 0.0,
3800
+ "epoch": 3.25,
3801
+ "learning_rate": 0.000211,
3802
+ "loss": 0.5511,
3803
+ "mlm_loss": 0.5511,
3804
+ "step": 4220
3805
+ },
3806
+ {
3807
+ "ep_loss": 0.0,
3808
+ "epoch": 3.26,
3809
+ "learning_rate": 0.0002115,
3810
+ "loss": 0.5578,
3811
+ "mlm_loss": 0.5578,
3812
+ "step": 4230
3813
+ },
3814
+ {
3815
+ "ep_loss": 0.0,
3816
+ "epoch": 3.27,
3817
+ "learning_rate": 0.000212,
3818
+ "loss": 0.5531,
3819
+ "mlm_loss": 0.5531,
3820
+ "step": 4240
3821
+ },
3822
+ {
3823
+ "ep_loss": 0.0,
3824
+ "epoch": 3.28,
3825
+ "learning_rate": 0.0002125,
3826
+ "loss": 0.5569,
3827
+ "mlm_loss": 0.5569,
3828
+ "step": 4250
3829
+ },
3830
+ {
3831
+ "ep_loss": 0.0,
3832
+ "epoch": 3.29,
3833
+ "learning_rate": 0.000213,
3834
+ "loss": 0.5499,
3835
+ "mlm_loss": 0.5499,
3836
+ "step": 4260
3837
+ },
3838
+ {
3839
+ "ep_loss": 0.0,
3840
+ "epoch": 3.29,
3841
+ "learning_rate": 0.0002135,
3842
+ "loss": 0.5514,
3843
+ "mlm_loss": 0.5514,
3844
+ "step": 4270
3845
+ },
3846
+ {
3847
+ "ep_loss": 0.0,
3848
+ "epoch": 3.3,
3849
+ "learning_rate": 0.000214,
3850
+ "loss": 0.554,
3851
+ "mlm_loss": 0.554,
3852
+ "step": 4280
3853
+ },
3854
+ {
3855
+ "ep_loss": 0.0,
3856
+ "epoch": 3.31,
3857
+ "learning_rate": 0.0002145,
3858
+ "loss": 0.5444,
3859
+ "mlm_loss": 0.5444,
3860
+ "step": 4290
3861
+ },
3862
+ {
3863
+ "ep_loss": 0.0,
3864
+ "epoch": 3.32,
3865
+ "learning_rate": 0.000215,
3866
+ "loss": 0.5555,
3867
+ "mlm_loss": 0.5555,
3868
+ "step": 4300
3869
+ },
3870
+ {
3871
+ "epoch": 3.32,
3872
+ "eval_ep_loss": -2.407238245010376,
3873
+ "eval_loss": 0.5153276920318604,
3874
+ "eval_mlm_loss": 0.5153276920318604,
3875
+ "eval_runtime": 62.01,
3876
+ "eval_samples_per_second": 1126.738,
3877
+ "eval_steps_per_second": 0.564,
3878
+ "step": 4300
3879
+ },
3880
+ {
3881
+ "ep_loss": 0.0,
3882
+ "epoch": 3.32,
3883
+ "learning_rate": 0.0002155,
3884
+ "loss": 0.5374,
3885
+ "mlm_loss": 0.5374,
3886
+ "step": 4310
3887
+ },
3888
+ {
3889
+ "ep_loss": 0.0,
3890
+ "epoch": 3.33,
3891
+ "learning_rate": 0.000216,
3892
+ "loss": 0.5526,
3893
+ "mlm_loss": 0.5526,
3894
+ "step": 4320
3895
+ },
3896
+ {
3897
+ "ep_loss": 0.0,
3898
+ "epoch": 3.34,
3899
+ "learning_rate": 0.0002165,
3900
+ "loss": 0.5451,
3901
+ "mlm_loss": 0.5451,
3902
+ "step": 4330
3903
+ },
3904
+ {
3905
+ "ep_loss": 0.0,
3906
+ "epoch": 3.35,
3907
+ "learning_rate": 0.00021700000000000002,
3908
+ "loss": 0.5406,
3909
+ "mlm_loss": 0.5406,
3910
+ "step": 4340
3911
+ },
3912
+ {
3913
+ "ep_loss": 0.0,
3914
+ "epoch": 3.36,
3915
+ "learning_rate": 0.0002175,
3916
+ "loss": 0.5339,
3917
+ "mlm_loss": 0.5339,
3918
+ "step": 4350
3919
+ },
3920
+ {
3921
+ "ep_loss": 0.0,
3922
+ "epoch": 3.36,
3923
+ "learning_rate": 0.000218,
3924
+ "loss": 0.5434,
3925
+ "mlm_loss": 0.5434,
3926
+ "step": 4360
3927
+ },
3928
+ {
3929
+ "ep_loss": 0.0,
3930
+ "epoch": 3.37,
3931
+ "learning_rate": 0.0002185,
3932
+ "loss": 0.546,
3933
+ "mlm_loss": 0.546,
3934
+ "step": 4370
3935
+ },
3936
+ {
3937
+ "ep_loss": 0.0,
3938
+ "epoch": 3.38,
3939
+ "learning_rate": 0.000219,
3940
+ "loss": 0.5359,
3941
+ "mlm_loss": 0.5359,
3942
+ "step": 4380
3943
+ },
3944
+ {
3945
+ "ep_loss": 0.0,
3946
+ "epoch": 3.39,
3947
+ "learning_rate": 0.0002195,
3948
+ "loss": 0.5405,
3949
+ "mlm_loss": 0.5405,
3950
+ "step": 4390
3951
+ },
3952
+ {
3953
+ "ep_loss": 0.0,
3954
+ "epoch": 3.39,
3955
+ "learning_rate": 0.00022,
3956
+ "loss": 0.5506,
3957
+ "mlm_loss": 0.5506,
3958
+ "step": 4400
3959
+ },
3960
+ {
3961
+ "epoch": 3.39,
3962
+ "eval_ep_loss": -2.1430230140686035,
3963
+ "eval_loss": 0.5089389681816101,
3964
+ "eval_mlm_loss": 0.5089389681816101,
3965
+ "eval_runtime": 60.2829,
3966
+ "eval_samples_per_second": 1159.018,
3967
+ "eval_steps_per_second": 0.581,
3968
+ "step": 4400
3969
+ },
3970
+ {
3971
+ "ep_loss": 0.0,
3972
+ "epoch": 3.4,
3973
+ "learning_rate": 0.0002205,
3974
+ "loss": 0.5425,
3975
+ "mlm_loss": 0.5425,
3976
+ "step": 4410
3977
+ },
3978
+ {
3979
+ "ep_loss": 0.0,
3980
+ "epoch": 3.41,
3981
+ "learning_rate": 0.000221,
3982
+ "loss": 0.5443,
3983
+ "mlm_loss": 0.5443,
3984
+ "step": 4420
3985
+ },
3986
+ {
3987
+ "ep_loss": 0.0,
3988
+ "epoch": 3.42,
3989
+ "learning_rate": 0.00022150000000000002,
3990
+ "loss": 0.5433,
3991
+ "mlm_loss": 0.5433,
3992
+ "step": 4430
3993
+ },
3994
+ {
3995
+ "ep_loss": 0.0,
3996
+ "epoch": 3.42,
3997
+ "learning_rate": 0.000222,
3998
+ "loss": 0.5379,
3999
+ "mlm_loss": 0.5379,
4000
+ "step": 4440
4001
+ },
4002
+ {
4003
+ "ep_loss": 0.0,
4004
+ "epoch": 3.43,
4005
+ "learning_rate": 0.00022250000000000001,
4006
+ "loss": 0.5349,
4007
+ "mlm_loss": 0.5349,
4008
+ "step": 4450
4009
+ },
4010
+ {
4011
+ "ep_loss": 0.0,
4012
+ "epoch": 3.44,
4013
+ "learning_rate": 0.000223,
4014
+ "loss": 0.5385,
4015
+ "mlm_loss": 0.5385,
4016
+ "step": 4460
4017
+ },
4018
+ {
4019
+ "ep_loss": 0.0,
4020
+ "epoch": 3.45,
4021
+ "learning_rate": 0.0002235,
4022
+ "loss": 0.5337,
4023
+ "mlm_loss": 0.5337,
4024
+ "step": 4470
4025
+ },
4026
+ {
4027
+ "ep_loss": 0.0,
4028
+ "epoch": 3.46,
4029
+ "learning_rate": 0.000224,
4030
+ "loss": 0.5372,
4031
+ "mlm_loss": 0.5372,
4032
+ "step": 4480
4033
+ },
4034
+ {
4035
+ "ep_loss": 0.0,
4036
+ "epoch": 3.46,
4037
+ "learning_rate": 0.0002245,
4038
+ "loss": 0.5342,
4039
+ "mlm_loss": 0.5342,
4040
+ "step": 4490
4041
+ },
4042
+ {
4043
+ "ep_loss": 0.0,
4044
+ "epoch": 3.47,
4045
+ "learning_rate": 0.00022500000000000002,
4046
+ "loss": 0.5329,
4047
+ "mlm_loss": 0.5329,
4048
+ "step": 4500
4049
+ },
4050
+ {
4051
+ "epoch": 3.47,
4052
+ "eval_ep_loss": -2.252265691757202,
4053
+ "eval_loss": 0.501177966594696,
4054
+ "eval_mlm_loss": 0.501177966594696,
4055
+ "eval_runtime": 61.1269,
4056
+ "eval_samples_per_second": 1143.016,
4057
+ "eval_steps_per_second": 0.573,
4058
+ "step": 4500
4059
+ },
4060
+ {
4061
+ "ep_loss": 0.0,
4062
+ "epoch": 3.48,
4063
+ "learning_rate": 0.0002255,
4064
+ "loss": 0.5321,
4065
+ "mlm_loss": 0.5321,
4066
+ "step": 4510
4067
+ },
4068
+ {
4069
+ "ep_loss": 0.0,
4070
+ "epoch": 3.49,
4071
+ "learning_rate": 0.00022600000000000002,
4072
+ "loss": 0.5413,
4073
+ "mlm_loss": 0.5413,
4074
+ "step": 4520
4075
+ },
4076
+ {
4077
+ "ep_loss": 0.0,
4078
+ "epoch": 3.49,
4079
+ "learning_rate": 0.0002265,
4080
+ "loss": 0.5364,
4081
+ "mlm_loss": 0.5364,
4082
+ "step": 4530
4083
+ },
4084
+ {
4085
+ "ep_loss": 0.0,
4086
+ "epoch": 3.5,
4087
+ "learning_rate": 0.00022700000000000002,
4088
+ "loss": 0.5326,
4089
+ "mlm_loss": 0.5326,
4090
+ "step": 4540
4091
+ },
4092
+ {
4093
+ "ep_loss": 0.0,
4094
+ "epoch": 3.51,
4095
+ "learning_rate": 0.0002275,
4096
+ "loss": 0.5327,
4097
+ "mlm_loss": 0.5327,
4098
+ "step": 4550
4099
+ },
4100
+ {
4101
+ "ep_loss": 0.0,
4102
+ "epoch": 3.52,
4103
+ "learning_rate": 0.000228,
4104
+ "loss": 0.5334,
4105
+ "mlm_loss": 0.5334,
4106
+ "step": 4560
4107
+ },
4108
+ {
4109
+ "ep_loss": 0.0,
4110
+ "epoch": 3.52,
4111
+ "learning_rate": 0.00022850000000000002,
4112
+ "loss": 0.5344,
4113
+ "mlm_loss": 0.5344,
4114
+ "step": 4570
4115
+ },
4116
+ {
4117
+ "ep_loss": 0.0,
4118
+ "epoch": 3.53,
4119
+ "learning_rate": 0.000229,
4120
+ "loss": 0.5259,
4121
+ "mlm_loss": 0.5259,
4122
+ "step": 4580
4123
+ },
4124
+ {
4125
+ "ep_loss": 0.0,
4126
+ "epoch": 3.54,
4127
+ "learning_rate": 0.00022950000000000002,
4128
+ "loss": 0.519,
4129
+ "mlm_loss": 0.519,
4130
+ "step": 4590
4131
+ },
4132
+ {
4133
+ "ep_loss": 0.0,
4134
+ "epoch": 3.55,
4135
+ "learning_rate": 0.00023,
4136
+ "loss": 0.5243,
4137
+ "mlm_loss": 0.5243,
4138
+ "step": 4600
4139
+ },
4140
+ {
4141
+ "epoch": 3.55,
4142
+ "eval_ep_loss": -2.4623284339904785,
4143
+ "eval_loss": 0.49688172340393066,
4144
+ "eval_mlm_loss": 0.49688172340393066,
4145
+ "eval_runtime": 60.8838,
4146
+ "eval_samples_per_second": 1147.579,
4147
+ "eval_steps_per_second": 0.575,
4148
+ "step": 4600
4149
+ },
4150
+ {
4151
+ "ep_loss": 0.0,
4152
+ "epoch": 3.56,
4153
+ "learning_rate": 0.00023050000000000002,
4154
+ "loss": 0.5237,
4155
+ "mlm_loss": 0.5237,
4156
+ "step": 4610
4157
+ },
4158
+ {
4159
+ "ep_loss": 0.0,
4160
+ "epoch": 3.56,
4161
+ "learning_rate": 0.000231,
4162
+ "loss": 0.5249,
4163
+ "mlm_loss": 0.5249,
4164
+ "step": 4620
4165
+ },
4166
+ {
4167
+ "ep_loss": 0.0,
4168
+ "epoch": 3.57,
4169
+ "learning_rate": 0.00023150000000000002,
4170
+ "loss": 0.5208,
4171
+ "mlm_loss": 0.5208,
4172
+ "step": 4630
4173
+ },
4174
+ {
4175
+ "ep_loss": 0.0,
4176
+ "epoch": 3.58,
4177
+ "learning_rate": 0.00023200000000000003,
4178
+ "loss": 0.5213,
4179
+ "mlm_loss": 0.5213,
4180
+ "step": 4640
4181
+ },
4182
+ {
4183
+ "ep_loss": 0.0,
4184
+ "epoch": 3.59,
4185
+ "learning_rate": 0.0002325,
4186
+ "loss": 0.5316,
4187
+ "mlm_loss": 0.5316,
4188
+ "step": 4650
4189
+ },
4190
+ {
4191
+ "ep_loss": 0.0,
4192
+ "epoch": 3.59,
4193
+ "learning_rate": 0.00023300000000000003,
4194
+ "loss": 0.5231,
4195
+ "mlm_loss": 0.5231,
4196
+ "step": 4660
4197
+ },
4198
+ {
4199
+ "ep_loss": 0.0,
4200
+ "epoch": 3.6,
4201
+ "learning_rate": 0.0002335,
4202
+ "loss": 0.524,
4203
+ "mlm_loss": 0.524,
4204
+ "step": 4670
4205
+ },
4206
+ {
4207
+ "ep_loss": 0.0,
4208
+ "epoch": 3.61,
4209
+ "learning_rate": 0.00023400000000000002,
4210
+ "loss": 0.5274,
4211
+ "mlm_loss": 0.5274,
4212
+ "step": 4680
4213
+ },
4214
+ {
4215
+ "ep_loss": 0.0,
4216
+ "epoch": 3.62,
4217
+ "learning_rate": 0.00023449999999999998,
4218
+ "loss": 0.5186,
4219
+ "mlm_loss": 0.5186,
4220
+ "step": 4690
4221
+ },
4222
+ {
4223
+ "ep_loss": 0.0,
4224
+ "epoch": 3.63,
4225
+ "learning_rate": 0.000235,
4226
+ "loss": 0.5201,
4227
+ "mlm_loss": 0.5201,
4228
+ "step": 4700
4229
+ },
4230
+ {
4231
+ "epoch": 3.63,
4232
+ "eval_ep_loss": -2.275129556655884,
4233
+ "eval_loss": 0.48909062147140503,
4234
+ "eval_mlm_loss": 0.48909062147140503,
4235
+ "eval_runtime": 63.7661,
4236
+ "eval_samples_per_second": 1095.708,
4237
+ "eval_steps_per_second": 0.549,
4238
+ "step": 4700
4239
+ },
4240
+ {
4241
+ "ep_loss": 0.0,
4242
+ "epoch": 3.63,
4243
+ "learning_rate": 0.0002355,
4244
+ "loss": 0.5167,
4245
+ "mlm_loss": 0.5167,
4246
+ "step": 4710
4247
+ },
4248
+ {
4249
+ "ep_loss": 0.0,
4250
+ "epoch": 3.64,
4251
+ "learning_rate": 0.000236,
4252
+ "loss": 0.5152,
4253
+ "mlm_loss": 0.5152,
4254
+ "step": 4720
4255
+ },
4256
+ {
4257
+ "ep_loss": 0.0,
4258
+ "epoch": 3.65,
4259
+ "learning_rate": 0.0002365,
4260
+ "loss": 0.5201,
4261
+ "mlm_loss": 0.5201,
4262
+ "step": 4730
4263
+ },
4264
+ {
4265
+ "ep_loss": 0.0,
4266
+ "epoch": 3.66,
4267
+ "learning_rate": 0.000237,
4268
+ "loss": 0.5234,
4269
+ "mlm_loss": 0.5234,
4270
+ "step": 4740
4271
+ },
4272
+ {
4273
+ "ep_loss": 0.0,
4274
+ "epoch": 3.66,
4275
+ "learning_rate": 0.0002375,
4276
+ "loss": 0.5156,
4277
+ "mlm_loss": 0.5156,
4278
+ "step": 4750
4279
+ },
4280
+ {
4281
+ "ep_loss": 0.0,
4282
+ "epoch": 3.67,
4283
+ "learning_rate": 0.00023799999999999998,
4284
+ "loss": 0.5189,
4285
+ "mlm_loss": 0.5189,
4286
+ "step": 4760
4287
+ },
4288
+ {
4289
+ "ep_loss": 0.0,
4290
+ "epoch": 3.68,
4291
+ "learning_rate": 0.0002385,
4292
+ "loss": 0.5189,
4293
+ "mlm_loss": 0.5189,
4294
+ "step": 4770
4295
+ },
4296
+ {
4297
+ "ep_loss": 0.0,
4298
+ "epoch": 3.69,
4299
+ "learning_rate": 0.00023899999999999998,
4300
+ "loss": 0.51,
4301
+ "mlm_loss": 0.51,
4302
+ "step": 4780
4303
+ },
4304
+ {
4305
+ "ep_loss": 0.0,
4306
+ "epoch": 3.69,
4307
+ "learning_rate": 0.0002395,
4308
+ "loss": 0.5158,
4309
+ "mlm_loss": 0.5158,
4310
+ "step": 4790
4311
+ },
4312
+ {
4313
+ "ep_loss": 0.0,
4314
+ "epoch": 3.7,
4315
+ "learning_rate": 0.00024,
4316
+ "loss": 0.5121,
4317
+ "mlm_loss": 0.5121,
4318
+ "step": 4800
4319
+ },
4320
+ {
4321
+ "epoch": 3.7,
4322
+ "eval_ep_loss": -2.0928449630737305,
4323
+ "eval_loss": 0.48291251063346863,
4324
+ "eval_mlm_loss": 0.48291251063346863,
4325
+ "eval_runtime": 60.133,
4326
+ "eval_samples_per_second": 1161.908,
4327
+ "eval_steps_per_second": 0.582,
4328
+ "step": 4800
4329
+ },
4330
+ {
4331
+ "ep_loss": 0.0,
4332
+ "epoch": 3.71,
4333
+ "learning_rate": 0.0002405,
4334
+ "loss": 0.5108,
4335
+ "mlm_loss": 0.5108,
4336
+ "step": 4810
4337
+ },
4338
+ {
4339
+ "ep_loss": 0.0,
4340
+ "epoch": 3.72,
4341
+ "learning_rate": 0.000241,
4342
+ "loss": 0.5176,
4343
+ "mlm_loss": 0.5176,
4344
+ "step": 4820
4345
+ },
4346
+ {
4347
+ "ep_loss": 0.0,
4348
+ "epoch": 3.73,
4349
+ "learning_rate": 0.0002415,
4350
+ "loss": 0.5102,
4351
+ "mlm_loss": 0.5102,
4352
+ "step": 4830
4353
+ },
4354
+ {
4355
+ "ep_loss": 0.0,
4356
+ "epoch": 3.73,
4357
+ "learning_rate": 0.000242,
4358
+ "loss": 0.514,
4359
+ "mlm_loss": 0.514,
4360
+ "step": 4840
4361
+ },
4362
+ {
4363
+ "ep_loss": 0.0,
4364
+ "epoch": 3.74,
4365
+ "learning_rate": 0.00024249999999999999,
4366
+ "loss": 0.5086,
4367
+ "mlm_loss": 0.5086,
4368
+ "step": 4850
4369
+ },
4370
+ {
4371
+ "ep_loss": 0.0,
4372
+ "epoch": 3.75,
4373
+ "learning_rate": 0.000243,
4374
+ "loss": 0.515,
4375
+ "mlm_loss": 0.515,
4376
+ "step": 4860
4377
+ },
4378
+ {
4379
+ "ep_loss": 0.0,
4380
+ "epoch": 3.76,
4381
+ "learning_rate": 0.0002435,
4382
+ "loss": 0.5091,
4383
+ "mlm_loss": 0.5091,
4384
+ "step": 4870
4385
+ },
4386
+ {
4387
+ "ep_loss": 0.0,
4388
+ "epoch": 3.76,
4389
+ "learning_rate": 0.000244,
4390
+ "loss": 0.514,
4391
+ "mlm_loss": 0.514,
4392
+ "step": 4880
4393
+ },
4394
+ {
4395
+ "ep_loss": 0.0,
4396
+ "epoch": 3.77,
4397
+ "learning_rate": 0.0002445,
4398
+ "loss": 0.5051,
4399
+ "mlm_loss": 0.5051,
4400
+ "step": 4890
4401
+ },
4402
+ {
4403
+ "ep_loss": 0.0,
4404
+ "epoch": 3.78,
4405
+ "learning_rate": 0.000245,
4406
+ "loss": 0.526,
4407
+ "mlm_loss": 0.526,
4408
+ "step": 4900
4409
+ },
4410
+ {
4411
+ "epoch": 3.78,
4412
+ "eval_ep_loss": -2.452838182449341,
4413
+ "eval_loss": 0.4754766523838043,
4414
+ "eval_mlm_loss": 0.4754766523838043,
4415
+ "eval_runtime": 60.9895,
4416
+ "eval_samples_per_second": 1145.591,
4417
+ "eval_steps_per_second": 0.574,
4418
+ "step": 4900
4419
+ },
4420
+ {
4421
+ "ep_loss": 0.0,
4422
+ "epoch": 3.79,
4423
+ "learning_rate": 0.0002455,
4424
+ "loss": 0.5104,
4425
+ "mlm_loss": 0.5104,
4426
+ "step": 4910
4427
+ },
4428
+ {
4429
+ "ep_loss": 0.0,
4430
+ "epoch": 3.79,
4431
+ "learning_rate": 0.000246,
4432
+ "loss": 0.4975,
4433
+ "mlm_loss": 0.4975,
4434
+ "step": 4920
4435
+ },
4436
+ {
4437
+ "ep_loss": 0.0,
4438
+ "epoch": 3.8,
4439
+ "learning_rate": 0.00024650000000000003,
4440
+ "loss": 0.511,
4441
+ "mlm_loss": 0.511,
4442
+ "step": 4930
4443
+ },
4444
+ {
4445
+ "ep_loss": 0.0,
4446
+ "epoch": 3.81,
4447
+ "learning_rate": 0.000247,
4448
+ "loss": 0.5081,
4449
+ "mlm_loss": 0.5081,
4450
+ "step": 4940
4451
+ },
4452
+ {
4453
+ "ep_loss": 0.0,
4454
+ "epoch": 3.82,
4455
+ "learning_rate": 0.0002475,
4456
+ "loss": 0.5067,
4457
+ "mlm_loss": 0.5067,
4458
+ "step": 4950
4459
+ },
4460
+ {
4461
+ "ep_loss": 0.0,
4462
+ "epoch": 3.83,
4463
+ "learning_rate": 0.000248,
4464
+ "loss": 0.5062,
4465
+ "mlm_loss": 0.5062,
4466
+ "step": 4960
4467
+ },
4468
+ {
4469
+ "ep_loss": 0.0,
4470
+ "epoch": 3.83,
4471
+ "learning_rate": 0.0002485,
4472
+ "loss": 0.5054,
4473
+ "mlm_loss": 0.5054,
4474
+ "step": 4970
4475
+ },
4476
+ {
4477
+ "ep_loss": 0.0,
4478
+ "epoch": 3.84,
4479
+ "learning_rate": 0.000249,
4480
+ "loss": 0.5032,
4481
+ "mlm_loss": 0.5032,
4482
+ "step": 4980
4483
+ },
4484
+ {
4485
+ "ep_loss": 0.0,
4486
+ "epoch": 3.85,
4487
+ "learning_rate": 0.0002495,
4488
+ "loss": 0.5022,
4489
+ "mlm_loss": 0.5022,
4490
+ "step": 4990
4491
+ },
4492
+ {
4493
+ "ep_loss": 0.0,
4494
+ "epoch": 3.86,
4495
+ "learning_rate": 0.00025,
4496
+ "loss": 0.5077,
4497
+ "mlm_loss": 0.5077,
4498
+ "step": 5000
4499
+ },
4500
+ {
4501
+ "epoch": 3.86,
4502
+ "eval_ep_loss": -2.5484063625335693,
4503
+ "eval_loss": 0.47381114959716797,
4504
+ "eval_mlm_loss": 0.47381114959716797,
4505
+ "eval_runtime": 59.7879,
4506
+ "eval_samples_per_second": 1168.615,
4507
+ "eval_steps_per_second": 0.585,
4508
+ "step": 5000
4509
+ }
4510
+ ],
4511
+ "max_steps": 25920,
4512
+ "num_train_epochs": 20,
4513
+ "total_flos": 1.356897877753856e+18,
4514
+ "trial_name": null,
4515
+ "trial_params": null
4516
+ }
checkpoint-5000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dbbced2969d3c34b5ea7f356c253acf2a826f9e9dea1f238559fc37a4cf95a6d
3
+ size 4027
pytorch_model.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:10af569d0816f4eb030f2bbafaf8b832d69ed1e7ca8677d06cdef04743d270a6
3
  size 440772077
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d4e0d79bac895d479a820e490ec436da2f0d185ae3706aabc74ded777034acf1
3
  size 440772077