Hugging Face's logo

BLR2
/
lion

PyTorch
gemma3
Model card Files
xet
Community
1

remove models

#1
by BLR2 - opened
base: refs/heads/main
←
from: refs/pr/1
Discussion Files changed
+56054
-0
Files changed (30) hide show
  1. models/fdkjsfds/gemma_knows_coorsinates_4/config.json +62 -0
  2. models/fdkjsfds/gemma_knows_coorsinates_4/generation_config.json +11 -0
  3. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00001-of-00022.bin +3 -0
  4. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00002-of-00022.bin +3 -0
  5. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00003-of-00022.bin +3 -0
  6. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00004-of-00022.bin +3 -0
  7. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00005-of-00022.bin +3 -0
  8. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00006-of-00022.bin +3 -0
  9. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00007-of-00022.bin +3 -0
  10. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00008-of-00022.bin +3 -0
  11. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00009-of-00022.bin +3 -0
  12. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00010-of-00022.bin +3 -0
  13. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00011-of-00022.bin +3 -0
  14. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00012-of-00022.bin +3 -0
  15. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00013-of-00022.bin +3 -0
  16. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00014-of-00022.bin +3 -0
  17. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00015-of-00022.bin +3 -0
  18. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00016-of-00022.bin +3 -0
  19. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00017-of-00022.bin +3 -0
  20. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00018-of-00022.bin +3 -0
  21. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00019-of-00022.bin +3 -0
  22. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00020-of-00022.bin +3 -0
  23. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00021-of-00022.bin +3 -0
  24. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00022-of-00022.bin +3 -0
  25. models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model.bin.index.json +0 -0
  26. models/fdkjsfds/gemma_knows_coorsinates_4/special_tokens_map.json +33 -0
  27. models/fdkjsfds/gemma_knows_coorsinates_4/tokenizer.json +3 -0
  28. models/fdkjsfds/gemma_knows_coorsinates_4/tokenizer_config.json +0 -0
  29. models/fdkjsfds/gemma_knows_coorsinates_4/trainer_state.json +3274 -0
  30. models/gemma_knows_coorsinates_6/pytorch_model-00010-of-00022.bin +3 -0
models/fdkjsfds/gemma_knows_coorsinates_4/config.json ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "Gemma3ForConditionalGeneration"
4
+ ],
5
+ "boi_token_index": 255999,
6
+ "eoi_token_index": 256000,
7
+ "eos_token_id": [
8
+ 1,
9
+ 106
10
+ ],
11
+ "image_token_index": 262144,
12
+ "initializer_range": 0.02,
13
+ "mm_tokens_per_image": 256,
14
+ "model_type": "gemma3",
15
+ "text_config": {
16
+ "attention_bias": false,
17
+ "attention_dropout": 0.0,
18
+ "attn_logit_softcapping": null,
19
+ "cache_implementation": "hybrid",
20
+ "final_logit_softcapping": null,
21
+ "head_dim": 128,
22
+ "hidden_activation": "gelu_pytorch_tanh",
23
+ "hidden_size": 5376,
24
+ "initializer_range": 0.02,
25
+ "intermediate_size": 21504,
26
+ "max_position_embeddings": 131072,
27
+ "model_type": "gemma3_text",
28
+ "num_attention_heads": 32,
29
+ "num_hidden_layers": 62,
30
+ "num_key_value_heads": 16,
31
+ "query_pre_attn_scalar": 168,
32
+ "rms_norm_eps": 1e-06,
33
+ "rope_local_base_freq": 10000.0,
34
+ "rope_scaling": {
35
+ "factor": 8.0,
36
+ "rope_type": "linear"
37
+ },
38
+ "rope_theta": 1000000.0,
39
+ "sliding_window": 1024,
40
+ "sliding_window_pattern": 6,
41
+ "torch_dtype": "bfloat16",
42
+ "use_cache": true,
43
+ "vocab_size": 262208
44
+ },
45
+ "torch_dtype": "bfloat16",
46
+ "transformers_version": "4.51.1",
47
+ "vision_config": {
48
+ "attention_dropout": 0.0,
49
+ "hidden_act": "gelu_pytorch_tanh",
50
+ "hidden_size": 1152,
51
+ "image_size": 896,
52
+ "intermediate_size": 4304,
53
+ "layer_norm_eps": 1e-06,
54
+ "model_type": "siglip_vision_model",
55
+ "num_attention_heads": 16,
56
+ "num_channels": 3,
57
+ "num_hidden_layers": 27,
58
+ "patch_size": 14,
59
+ "torch_dtype": "bfloat16",
60
+ "vision_use_head": false
61
+ }
62
+ }
models/fdkjsfds/gemma_knows_coorsinates_4/generation_config.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 2,
4
+ "cache_implementation": "hybrid",
5
+ "eos_token_id": [
6
+ 1,
7
+ 106
8
+ ],
9
+ "pad_token_id": 0,
10
+ "transformers_version": "4.51.1"
11
+ }
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00001-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:35e91f6d4a466b6ee7965ec4d511c55b9798dc2c76badf1f13a00b2c82707e9d
3
+ size 5638522361
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00002-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d1ca34f182e2d00661942ef90160a3bbfad1397dea2e52c854ceed9f1adf637b
3
+ size 4993727614
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00003-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:801db2f24ba0ea3e4fc5638b5ff4bcbeb382c0abb1657049df55ca64753a200f
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00004-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b9719653d161a3ae7f58ea77e8b5ad3b958c731d2072404a60a784539da17aa
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00005-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ab8da8bc07f5c359732f65033de6b9abba04720187458f99177365989183398c
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00006-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1df7ee8e3b78b15f6a216a37bf90453b582b7c9dbdb9fe4062be94d1ba1d6c86
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00007-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dcb354ea51a440ef225d43c5537debf0cd0b493c7291e69a5a4320b4b39436c4
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00008-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a94c82d1a45c937fafab7689afcdc86958c271295536423a2d531e7f26acef1
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00009-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c12eec9b55c8ad4d3f75853992510613714c11b5bc039b009584ad991129be29
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00010-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a2a8447cb4ae196517e19b4f5aad1c10c88b98b283f932e33005b1d9540737a0
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00011-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:398c33db58574f034008cff03928956ab9181dedefcfa3d8fe0cf9e8113511d3
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00012-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:706fa978cec98bc56e590ab017083e7e37ffbb2291ebbaf5001836f66eebaf70
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00013-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:69b7580d789453fbd30599e576a2f1379a869181361257bde269faf17b17398d
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00014-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0bcd8044cb0246bf14121bebe7020a2c5650e6242be1999f1a2f9208d9a8ce2d
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00015-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ff6d9da826c3d716f7f99bf276fcb7acc09f566e96003d6281172f3d04ea959
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00016-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3e6af1b2830122f54d2c67c7fcdd3d9bf6f28fec81f2e23ca6bba82c854bcec
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00017-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa0765dce0dc7caf6f60f3c7d41fca2c1436294ea69cbe42918e4a7d8b24043f
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00018-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5dc53a627ff23e20a60b0ff3786326a647b11dc797425a831a2d603e643dc37f
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00019-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30a81be0c786e5027531560e89c88f942c53a6a331a3da867edcda285f379654
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00020-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cceca20c66658b4a03c44f582ff23cc9255bcf48b6f380f14442843a5ac003bb
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00021-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c886d68e0474e0ad517bd929d08e035532486e0e8b861417fa2f5f469d8424ac
3
+ size 4954534664
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model-00022-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:169b42e41448b372e04ea28e1c72e9e2874cf7a27a9604d1ec3784063477b210
3
+ size 4961673381
models/fdkjsfds/gemma_knows_coorsinates_4/pytorch_model.bin.index.json ADDED
The diff for this file is too large to render. See raw diff
 
models/fdkjsfds/gemma_knows_coorsinates_4/special_tokens_map.json ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "boi_token": "<start_of_image>",
3
+ "bos_token": {
4
+ "content": "<bos>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false
9
+ },
10
+ "eoi_token": "<end_of_image>",
11
+ "eos_token": {
12
+ "content": "<eos>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false
17
+ },
18
+ "image_token": "<image_soft_token>",
19
+ "pad_token": {
20
+ "content": "<pad>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false
25
+ },
26
+ "unk_token": {
27
+ "content": "<unk>",
28
+ "lstrip": false,
29
+ "normalized": false,
30
+ "rstrip": false,
31
+ "single_word": false
32
+ }
33
+ }
models/fdkjsfds/gemma_knows_coorsinates_4/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d786405177734910d7a3db625c2826640964a0b4e5cdbbd70620ae3313a01bef
3
+ size 33384722
models/fdkjsfds/gemma_knows_coorsinates_4/tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff
 
models/fdkjsfds/gemma_knows_coorsinates_4/trainer_state.json ADDED
@@ -0,0 +1,3274 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 5.752,
6
+ "eval_steps": 200,
7
+ "global_step": 1800,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.016,
14
+ "grad_norm": 31.84697723388672,
15
+ "learning_rate": 4.993589743589744e-06,
16
+ "loss": 3.6624,
17
+ "mean_token_accuracy": 0.9031318381428719,
18
+ "num_tokens": 120757.0,
19
+ "step": 5
20
+ },
21
+ {
22
+ "epoch": 0.032,
23
+ "grad_norm": 12.340926170349121,
24
+ "learning_rate": 4.985576923076924e-06,
25
+ "loss": 1.2471,
26
+ "mean_token_accuracy": 0.9456482365727424,
27
+ "num_tokens": 241330.0,
28
+ "step": 10
29
+ },
30
+ {
31
+ "epoch": 0.048,
32
+ "grad_norm": 11.674023628234863,
33
+ "learning_rate": 4.977564102564103e-06,
34
+ "loss": 1.0677,
35
+ "mean_token_accuracy": 0.9501513227820396,
36
+ "num_tokens": 361918.0,
37
+ "step": 15
38
+ },
39
+ {
40
+ "epoch": 0.064,
41
+ "grad_norm": 10.814107894897461,
42
+ "learning_rate": 4.969551282051283e-06,
43
+ "loss": 1.0611,
44
+ "mean_token_accuracy": 0.9496850356459617,
45
+ "num_tokens": 482416.0,
46
+ "step": 20
47
+ },
48
+ {
49
+ "epoch": 0.08,
50
+ "grad_norm": 9.736730575561523,
51
+ "learning_rate": 4.961538461538462e-06,
52
+ "loss": 1.0469,
53
+ "mean_token_accuracy": 0.950936059653759,
54
+ "num_tokens": 603341.0,
55
+ "step": 25
56
+ },
57
+ {
58
+ "epoch": 0.096,
59
+ "grad_norm": 12.223091125488281,
60
+ "learning_rate": 4.953525641025642e-06,
61
+ "loss": 1.0147,
62
+ "mean_token_accuracy": 0.9516797289252281,
63
+ "num_tokens": 723760.0,
64
+ "step": 30
65
+ },
66
+ {
67
+ "epoch": 0.112,
68
+ "grad_norm": 8.521744728088379,
69
+ "learning_rate": 4.9455128205128215e-06,
70
+ "loss": 0.9992,
71
+ "mean_token_accuracy": 0.951826773583889,
72
+ "num_tokens": 843984.0,
73
+ "step": 35
74
+ },
75
+ {
76
+ "epoch": 0.128,
77
+ "grad_norm": 11.937309265136719,
78
+ "learning_rate": 4.937500000000001e-06,
79
+ "loss": 1.0202,
80
+ "mean_token_accuracy": 0.9511885553598404,
81
+ "num_tokens": 964407.0,
82
+ "step": 40
83
+ },
84
+ {
85
+ "epoch": 0.144,
86
+ "grad_norm": 7.2923903465271,
87
+ "learning_rate": 4.9294871794871805e-06,
88
+ "loss": 1.0217,
89
+ "mean_token_accuracy": 0.9514814674854278,
90
+ "num_tokens": 1085024.0,
91
+ "step": 45
92
+ },
93
+ {
94
+ "epoch": 0.16,
95
+ "grad_norm": 8.744965553283691,
96
+ "learning_rate": 4.921474358974359e-06,
97
+ "loss": 1.0071,
98
+ "mean_token_accuracy": 0.9515809074044228,
99
+ "num_tokens": 1205620.0,
100
+ "step": 50
101
+ },
102
+ {
103
+ "epoch": 0.176,
104
+ "grad_norm": 8.983216285705566,
105
+ "learning_rate": 4.913461538461539e-06,
106
+ "loss": 1.0024,
107
+ "mean_token_accuracy": 0.9526835083961487,
108
+ "num_tokens": 1326101.0,
109
+ "step": 55
110
+ },
111
+ {
112
+ "epoch": 0.192,
113
+ "grad_norm": 11.27083683013916,
114
+ "learning_rate": 4.9054487179487185e-06,
115
+ "loss": 1.0111,
116
+ "mean_token_accuracy": 0.9517094433307648,
117
+ "num_tokens": 1447053.0,
118
+ "step": 60
119
+ },
120
+ {
121
+ "epoch": 0.208,
122
+ "grad_norm": 9.141480445861816,
123
+ "learning_rate": 4.8974358974358975e-06,
124
+ "loss": 0.9748,
125
+ "mean_token_accuracy": 0.952695508301258,
126
+ "num_tokens": 1567669.0,
127
+ "step": 65
128
+ },
129
+ {
130
+ "epoch": 0.224,
131
+ "grad_norm": 8.711387634277344,
132
+ "learning_rate": 4.8894230769230774e-06,
133
+ "loss": 0.9977,
134
+ "mean_token_accuracy": 0.9519824221730232,
135
+ "num_tokens": 1688347.0,
136
+ "step": 70
137
+ },
138
+ {
139
+ "epoch": 0.24,
140
+ "grad_norm": 16.269573211669922,
141
+ "learning_rate": 4.8814102564102565e-06,
142
+ "loss": 0.9803,
143
+ "mean_token_accuracy": 0.9531255111098289,
144
+ "num_tokens": 1808708.0,
145
+ "step": 75
146
+ },
147
+ {
148
+ "epoch": 0.256,
149
+ "grad_norm": 13.134957313537598,
150
+ "learning_rate": 4.873397435897436e-06,
151
+ "loss": 1.0025,
152
+ "mean_token_accuracy": 0.9521767511963845,
153
+ "num_tokens": 1929753.0,
154
+ "step": 80
155
+ },
156
+ {
157
+ "epoch": 0.272,
158
+ "grad_norm": 11.679295539855957,
159
+ "learning_rate": 4.8653846153846155e-06,
160
+ "loss": 0.9495,
161
+ "mean_token_accuracy": 0.9537542402744293,
162
+ "num_tokens": 2050258.0,
163
+ "step": 85
164
+ },
165
+ {
166
+ "epoch": 0.288,
167
+ "grad_norm": 11.880794525146484,
168
+ "learning_rate": 4.857371794871795e-06,
169
+ "loss": 0.9617,
170
+ "mean_token_accuracy": 0.953302538394928,
171
+ "num_tokens": 2170926.0,
172
+ "step": 90
173
+ },
174
+ {
175
+ "epoch": 0.304,
176
+ "grad_norm": 11.54967212677002,
177
+ "learning_rate": 4.849358974358974e-06,
178
+ "loss": 0.942,
179
+ "mean_token_accuracy": 0.9542573839426041,
180
+ "num_tokens": 2291271.0,
181
+ "step": 95
182
+ },
183
+ {
184
+ "epoch": 0.32,
185
+ "grad_norm": 9.976207733154297,
186
+ "learning_rate": 4.841346153846154e-06,
187
+ "loss": 0.949,
188
+ "mean_token_accuracy": 0.9539243519306183,
189
+ "num_tokens": 2411840.0,
190
+ "step": 100
191
+ },
192
+ {
193
+ "epoch": 0.336,
194
+ "grad_norm": 10.973347663879395,
195
+ "learning_rate": 4.833333333333333e-06,
196
+ "loss": 0.9222,
197
+ "mean_token_accuracy": 0.9545562237501144,
198
+ "num_tokens": 2532105.0,
199
+ "step": 105
200
+ },
201
+ {
202
+ "epoch": 0.352,
203
+ "grad_norm": 10.059887886047363,
204
+ "learning_rate": 4.825320512820513e-06,
205
+ "loss": 0.9348,
206
+ "mean_token_accuracy": 0.9547728285193443,
207
+ "num_tokens": 2652505.0,
208
+ "step": 110
209
+ },
210
+ {
211
+ "epoch": 0.368,
212
+ "grad_norm": 13.358323097229004,
213
+ "learning_rate": 4.817307692307693e-06,
214
+ "loss": 0.9353,
215
+ "mean_token_accuracy": 0.9543639048933983,
216
+ "num_tokens": 2773301.0,
217
+ "step": 115
218
+ },
219
+ {
220
+ "epoch": 0.384,
221
+ "grad_norm": 9.563572883605957,
222
+ "learning_rate": 4.809294871794872e-06,
223
+ "loss": 0.9593,
224
+ "mean_token_accuracy": 0.953694874048233,
225
+ "num_tokens": 2894008.0,
226
+ "step": 120
227
+ },
228
+ {
229
+ "epoch": 0.4,
230
+ "grad_norm": 8.732457160949707,
231
+ "learning_rate": 4.801282051282052e-06,
232
+ "loss": 0.9662,
233
+ "mean_token_accuracy": 0.9534440860152245,
234
+ "num_tokens": 3014639.0,
235
+ "step": 125
236
+ },
237
+ {
238
+ "epoch": 0.416,
239
+ "grad_norm": 7.810755729675293,
240
+ "learning_rate": 4.793269230769231e-06,
241
+ "loss": 0.945,
242
+ "mean_token_accuracy": 0.9541527941823006,
243
+ "num_tokens": 3135273.0,
244
+ "step": 130
245
+ },
246
+ {
247
+ "epoch": 0.432,
248
+ "grad_norm": 14.152471542358398,
249
+ "learning_rate": 4.785256410256411e-06,
250
+ "loss": 0.9668,
251
+ "mean_token_accuracy": 0.9532970413565636,
252
+ "num_tokens": 3256129.0,
253
+ "step": 135
254
+ },
255
+ {
256
+ "epoch": 0.448,
257
+ "grad_norm": 11.00336742401123,
258
+ "learning_rate": 4.77724358974359e-06,
259
+ "loss": 0.936,
260
+ "mean_token_accuracy": 0.9551009327173233,
261
+ "num_tokens": 3376349.0,
262
+ "step": 140
263
+ },
264
+ {
265
+ "epoch": 0.464,
266
+ "grad_norm": 10.909722328186035,
267
+ "learning_rate": 4.76923076923077e-06,
268
+ "loss": 0.9556,
269
+ "mean_token_accuracy": 0.9535658627748489,
270
+ "num_tokens": 3496974.0,
271
+ "step": 145
272
+ },
273
+ {
274
+ "epoch": 0.48,
275
+ "grad_norm": 10.019011497497559,
276
+ "learning_rate": 4.761217948717949e-06,
277
+ "loss": 0.931,
278
+ "mean_token_accuracy": 0.9551873564720154,
279
+ "num_tokens": 3617336.0,
280
+ "step": 150
281
+ },
282
+ {
283
+ "epoch": 0.496,
284
+ "grad_norm": 10.22144603729248,
285
+ "learning_rate": 4.753205128205128e-06,
286
+ "loss": 0.9457,
287
+ "mean_token_accuracy": 0.954205897450447,
288
+ "num_tokens": 3737917.0,
289
+ "step": 155
290
+ },
291
+ {
292
+ "epoch": 0.512,
293
+ "grad_norm": 12.805910110473633,
294
+ "learning_rate": 4.745192307692308e-06,
295
+ "loss": 0.9331,
296
+ "mean_token_accuracy": 0.9544926360249519,
297
+ "num_tokens": 3858426.0,
298
+ "step": 160
299
+ },
300
+ {
301
+ "epoch": 0.528,
302
+ "grad_norm": 11.656357765197754,
303
+ "learning_rate": 4.737179487179487e-06,
304
+ "loss": 0.9224,
305
+ "mean_token_accuracy": 0.9550811484456062,
306
+ "num_tokens": 3978571.0,
307
+ "step": 165
308
+ },
309
+ {
310
+ "epoch": 0.544,
311
+ "grad_norm": 12.421012878417969,
312
+ "learning_rate": 4.729166666666667e-06,
313
+ "loss": 0.971,
314
+ "mean_token_accuracy": 0.9527538940310478,
315
+ "num_tokens": 4099379.0,
316
+ "step": 170
317
+ },
318
+ {
319
+ "epoch": 0.56,
320
+ "grad_norm": 22.619918823242188,
321
+ "learning_rate": 4.721153846153846e-06,
322
+ "loss": 0.9401,
323
+ "mean_token_accuracy": 0.9540928721427917,
324
+ "num_tokens": 4220241.0,
325
+ "step": 175
326
+ },
327
+ {
328
+ "epoch": 0.576,
329
+ "grad_norm": 15.2088041305542,
330
+ "learning_rate": 4.713141025641026e-06,
331
+ "loss": 0.9181,
332
+ "mean_token_accuracy": 0.9550379499793052,
333
+ "num_tokens": 4340497.0,
334
+ "step": 180
335
+ },
336
+ {
337
+ "epoch": 0.592,
338
+ "grad_norm": 13.529873847961426,
339
+ "learning_rate": 4.705128205128206e-06,
340
+ "loss": 0.8967,
341
+ "mean_token_accuracy": 0.9561266586184501,
342
+ "num_tokens": 4460931.0,
343
+ "step": 185
344
+ },
345
+ {
346
+ "epoch": 0.608,
347
+ "grad_norm": 20.4467830657959,
348
+ "learning_rate": 4.697115384615385e-06,
349
+ "loss": 0.9275,
350
+ "mean_token_accuracy": 0.9551544234156608,
351
+ "num_tokens": 4581555.0,
352
+ "step": 190
353
+ },
354
+ {
355
+ "epoch": 0.624,
356
+ "grad_norm": 7.401337146759033,
357
+ "learning_rate": 4.689102564102565e-06,
358
+ "loss": 0.9186,
359
+ "mean_token_accuracy": 0.9552032083272934,
360
+ "num_tokens": 4702013.0,
361
+ "step": 195
362
+ },
363
+ {
364
+ "epoch": 0.64,
365
+ "grad_norm": 11.863940238952637,
366
+ "learning_rate": 4.681089743589744e-06,
367
+ "loss": 0.9226,
368
+ "mean_token_accuracy": 0.9550667107105255,
369
+ "num_tokens": 4822266.0,
370
+ "step": 200
371
+ },
372
+ {
373
+ "epoch": 0.656,
374
+ "grad_norm": 6.267267227172852,
375
+ "learning_rate": 4.673076923076924e-06,
376
+ "loss": 0.9389,
377
+ "mean_token_accuracy": 0.9545965492725372,
378
+ "num_tokens": 4942581.0,
379
+ "step": 205
380
+ },
381
+ {
382
+ "epoch": 0.672,
383
+ "grad_norm": 7.0638427734375,
384
+ "learning_rate": 4.665064102564103e-06,
385
+ "loss": 0.9451,
386
+ "mean_token_accuracy": 0.9541529282927513,
387
+ "num_tokens": 5063079.0,
388
+ "step": 210
389
+ },
390
+ {
391
+ "epoch": 0.688,
392
+ "grad_norm": 5.622670650482178,
393
+ "learning_rate": 4.657051282051283e-06,
394
+ "loss": 0.9145,
395
+ "mean_token_accuracy": 0.9559449806809426,
396
+ "num_tokens": 5183501.0,
397
+ "step": 215
398
+ },
399
+ {
400
+ "epoch": 0.704,
401
+ "grad_norm": 7.110885143280029,
402
+ "learning_rate": 4.649038461538462e-06,
403
+ "loss": 0.9678,
404
+ "mean_token_accuracy": 0.9534030184149742,
405
+ "num_tokens": 5304455.0,
406
+ "step": 220
407
+ },
408
+ {
409
+ "epoch": 0.72,
410
+ "grad_norm": 5.7894816398620605,
411
+ "learning_rate": 4.641025641025642e-06,
412
+ "loss": 0.9467,
413
+ "mean_token_accuracy": 0.9538964822888374,
414
+ "num_tokens": 5425140.0,
415
+ "step": 225
416
+ },
417
+ {
418
+ "epoch": 0.736,
419
+ "grad_norm": 7.6990580558776855,
420
+ "learning_rate": 4.633012820512821e-06,
421
+ "loss": 0.9549,
422
+ "mean_token_accuracy": 0.9539404049515724,
423
+ "num_tokens": 5545599.0,
424
+ "step": 230
425
+ },
426
+ {
427
+ "epoch": 0.752,
428
+ "grad_norm": 6.53148078918457,
429
+ "learning_rate": 4.625000000000001e-06,
430
+ "loss": 0.9724,
431
+ "mean_token_accuracy": 0.9536307662725448,
432
+ "num_tokens": 5666294.0,
433
+ "step": 235
434
+ },
435
+ {
436
+ "epoch": 0.768,
437
+ "grad_norm": 8.534490585327148,
438
+ "learning_rate": 4.6169871794871805e-06,
439
+ "loss": 0.9764,
440
+ "mean_token_accuracy": 0.9528705924749374,
441
+ "num_tokens": 5787061.0,
442
+ "step": 240
443
+ },
444
+ {
445
+ "epoch": 0.784,
446
+ "grad_norm": 10.83642578125,
447
+ "learning_rate": 4.608974358974359e-06,
448
+ "loss": 0.9878,
449
+ "mean_token_accuracy": 0.9520511969923973,
450
+ "num_tokens": 5907880.0,
451
+ "step": 245
452
+ },
453
+ {
454
+ "epoch": 0.8,
455
+ "grad_norm": 11.419661521911621,
456
+ "learning_rate": 4.600961538461539e-06,
457
+ "loss": 0.9492,
458
+ "mean_token_accuracy": 0.9538819909095764,
459
+ "num_tokens": 6028499.0,
460
+ "step": 250
461
+ },
462
+ {
463
+ "epoch": 0.816,
464
+ "grad_norm": 19.386032104492188,
465
+ "learning_rate": 4.5929487179487185e-06,
466
+ "loss": 0.9578,
467
+ "mean_token_accuracy": 0.9536806970834732,
468
+ "num_tokens": 6149150.0,
469
+ "step": 255
470
+ },
471
+ {
472
+ "epoch": 0.832,
473
+ "grad_norm": 10.170877456665039,
474
+ "learning_rate": 4.5849358974358976e-06,
475
+ "loss": 0.981,
476
+ "mean_token_accuracy": 0.952590537071228,
477
+ "num_tokens": 6269800.0,
478
+ "step": 260
479
+ },
480
+ {
481
+ "epoch": 0.848,
482
+ "grad_norm": 5.458118438720703,
483
+ "learning_rate": 4.5769230769230775e-06,
484
+ "loss": 0.9446,
485
+ "mean_token_accuracy": 0.9538502007722854,
486
+ "num_tokens": 6390283.0,
487
+ "step": 265
488
+ },
489
+ {
490
+ "epoch": 0.864,
491
+ "grad_norm": 8.66484260559082,
492
+ "learning_rate": 4.5689102564102565e-06,
493
+ "loss": 0.9218,
494
+ "mean_token_accuracy": 0.9548383697867393,
495
+ "num_tokens": 6510893.0,
496
+ "step": 270
497
+ },
498
+ {
499
+ "epoch": 0.88,
500
+ "grad_norm": 9.9107666015625,
501
+ "learning_rate": 4.5608974358974364e-06,
502
+ "loss": 0.9706,
503
+ "mean_token_accuracy": 0.9524388670921325,
504
+ "num_tokens": 6631672.0,
505
+ "step": 275
506
+ },
507
+ {
508
+ "epoch": 0.896,
509
+ "grad_norm": 6.929771900177002,
510
+ "learning_rate": 4.5528846153846155e-06,
511
+ "loss": 0.9549,
512
+ "mean_token_accuracy": 0.9536419928073883,
513
+ "num_tokens": 6752103.0,
514
+ "step": 280
515
+ },
516
+ {
517
+ "epoch": 0.912,
518
+ "grad_norm": 16.579092025756836,
519
+ "learning_rate": 4.544871794871795e-06,
520
+ "loss": 0.9377,
521
+ "mean_token_accuracy": 0.9547425925731658,
522
+ "num_tokens": 6872645.0,
523
+ "step": 285
524
+ },
525
+ {
526
+ "epoch": 0.928,
527
+ "grad_norm": 7.269247531890869,
528
+ "learning_rate": 4.5368589743589744e-06,
529
+ "loss": 0.9412,
530
+ "mean_token_accuracy": 0.9542089313268661,
531
+ "num_tokens": 6993119.0,
532
+ "step": 290
533
+ },
534
+ {
535
+ "epoch": 0.944,
536
+ "grad_norm": 7.885486602783203,
537
+ "learning_rate": 4.528846153846154e-06,
538
+ "loss": 0.9407,
539
+ "mean_token_accuracy": 0.9546906769275665,
540
+ "num_tokens": 7113423.0,
541
+ "step": 295
542
+ },
543
+ {
544
+ "epoch": 0.96,
545
+ "grad_norm": 7.1563873291015625,
546
+ "learning_rate": 4.520833333333333e-06,
547
+ "loss": 0.9251,
548
+ "mean_token_accuracy": 0.9555400907993317,
549
+ "num_tokens": 7234107.0,
550
+ "step": 300
551
+ },
552
+ {
553
+ "epoch": 0.976,
554
+ "grad_norm": 6.300532341003418,
555
+ "learning_rate": 4.512820512820513e-06,
556
+ "loss": 0.9228,
557
+ "mean_token_accuracy": 0.955471259355545,
558
+ "num_tokens": 7354400.0,
559
+ "step": 305
560
+ },
561
+ {
562
+ "epoch": 0.992,
563
+ "grad_norm": 6.0977559089660645,
564
+ "learning_rate": 4.504807692307693e-06,
565
+ "loss": 0.949,
566
+ "mean_token_accuracy": 0.9538125053048134,
567
+ "num_tokens": 7474978.0,
568
+ "step": 310
569
+ },
570
+ {
571
+ "epoch": 1.0064,
572
+ "grad_norm": 5.053493022918701,
573
+ "learning_rate": 4.496794871794872e-06,
574
+ "loss": 0.7869,
575
+ "mean_token_accuracy": 0.9578077329529656,
576
+ "num_tokens": 7583492.0,
577
+ "step": 315
578
+ },
579
+ {
580
+ "epoch": 1.0224,
581
+ "grad_norm": 6.7606658935546875,
582
+ "learning_rate": 4.488782051282052e-06,
583
+ "loss": 0.7866,
584
+ "mean_token_accuracy": 0.9612703084945678,
585
+ "num_tokens": 7704034.0,
586
+ "step": 320
587
+ },
588
+ {
589
+ "epoch": 1.0384,
590
+ "grad_norm": 6.258560657501221,
591
+ "learning_rate": 4.480769230769231e-06,
592
+ "loss": 0.7895,
593
+ "mean_token_accuracy": 0.9605303943157196,
594
+ "num_tokens": 7824133.0,
595
+ "step": 325
596
+ },
597
+ {
598
+ "epoch": 1.0544,
599
+ "grad_norm": 5.952118396759033,
600
+ "learning_rate": 4.472756410256411e-06,
601
+ "loss": 0.7566,
602
+ "mean_token_accuracy": 0.9624520629644394,
603
+ "num_tokens": 7944865.0,
604
+ "step": 330
605
+ },
606
+ {
607
+ "epoch": 1.0704,
608
+ "grad_norm": 6.123453617095947,
609
+ "learning_rate": 4.46474358974359e-06,
610
+ "loss": 0.7747,
611
+ "mean_token_accuracy": 0.9615358829498291,
612
+ "num_tokens": 8065236.0,
613
+ "step": 335
614
+ },
615
+ {
616
+ "epoch": 1.0864,
617
+ "grad_norm": 6.357878684997559,
618
+ "learning_rate": 4.45673076923077e-06,
619
+ "loss": 0.7748,
620
+ "mean_token_accuracy": 0.9614632725715637,
621
+ "num_tokens": 8185426.0,
622
+ "step": 340
623
+ },
624
+ {
625
+ "epoch": 1.1024,
626
+ "grad_norm": 11.374921798706055,
627
+ "learning_rate": 4.448717948717949e-06,
628
+ "loss": 0.7803,
629
+ "mean_token_accuracy": 0.9608228400349617,
630
+ "num_tokens": 8305821.0,
631
+ "step": 345
632
+ },
633
+ {
634
+ "epoch": 1.1184,
635
+ "grad_norm": 6.038303852081299,
636
+ "learning_rate": 4.440705128205128e-06,
637
+ "loss": 0.7685,
638
+ "mean_token_accuracy": 0.9619353070855141,
639
+ "num_tokens": 8426299.0,
640
+ "step": 350
641
+ },
642
+ {
643
+ "epoch": 1.1344,
644
+ "grad_norm": 5.52262020111084,
645
+ "learning_rate": 4.432692307692308e-06,
646
+ "loss": 0.7632,
647
+ "mean_token_accuracy": 0.9615148559212685,
648
+ "num_tokens": 8546585.0,
649
+ "step": 355
650
+ },
651
+ {
652
+ "epoch": 1.1504,
653
+ "grad_norm": 5.839768409729004,
654
+ "learning_rate": 4.424679487179487e-06,
655
+ "loss": 0.7728,
656
+ "mean_token_accuracy": 0.9620646819472313,
657
+ "num_tokens": 8667035.0,
658
+ "step": 360
659
+ },
660
+ {
661
+ "epoch": 1.1663999999999999,
662
+ "grad_norm": 6.795030117034912,
663
+ "learning_rate": 4.416666666666667e-06,
664
+ "loss": 0.7837,
665
+ "mean_token_accuracy": 0.9608001813292504,
666
+ "num_tokens": 8787595.0,
667
+ "step": 365
668
+ },
669
+ {
670
+ "epoch": 1.1824,
671
+ "grad_norm": 5.59501838684082,
672
+ "learning_rate": 4.408653846153846e-06,
673
+ "loss": 0.7885,
674
+ "mean_token_accuracy": 0.9609249800443649,
675
+ "num_tokens": 8908155.0,
676
+ "step": 370
677
+ },
678
+ {
679
+ "epoch": 1.1984,
680
+ "grad_norm": 6.539979934692383,
681
+ "learning_rate": 4.400641025641026e-06,
682
+ "loss": 0.7721,
683
+ "mean_token_accuracy": 0.9613026678562164,
684
+ "num_tokens": 9028752.0,
685
+ "step": 375
686
+ },
687
+ {
688
+ "epoch": 1.2144,
689
+ "grad_norm": 6.531695365905762,
690
+ "learning_rate": 4.392628205128205e-06,
691
+ "loss": 0.7788,
692
+ "mean_token_accuracy": 0.961000868678093,
693
+ "num_tokens": 9149575.0,
694
+ "step": 380
695
+ },
696
+ {
697
+ "epoch": 1.2304,
698
+ "grad_norm": 9.412209510803223,
699
+ "learning_rate": 4.384615384615385e-06,
700
+ "loss": 0.7949,
701
+ "mean_token_accuracy": 0.9604864284396172,
702
+ "num_tokens": 9270000.0,
703
+ "step": 385
704
+ },
705
+ {
706
+ "epoch": 1.2464,
707
+ "grad_norm": 5.525970458984375,
708
+ "learning_rate": 4.376602564102565e-06,
709
+ "loss": 0.7776,
710
+ "mean_token_accuracy": 0.9613819360733032,
711
+ "num_tokens": 9390598.0,
712
+ "step": 390
713
+ },
714
+ {
715
+ "epoch": 1.2624,
716
+ "grad_norm": 14.398231506347656,
717
+ "learning_rate": 4.368589743589744e-06,
718
+ "loss": 0.7857,
719
+ "mean_token_accuracy": 0.9615062743425369,
720
+ "num_tokens": 9510753.0,
721
+ "step": 395
722
+ },
723
+ {
724
+ "epoch": 1.2784,
725
+ "grad_norm": 7.224318027496338,
726
+ "learning_rate": 4.360576923076924e-06,
727
+ "loss": 0.7846,
728
+ "mean_token_accuracy": 0.9612418845295906,
729
+ "num_tokens": 9631323.0,
730
+ "step": 400
731
+ },
732
+ {
733
+ "epoch": 1.2944,
734
+ "grad_norm": 5.975404739379883,
735
+ "learning_rate": 4.352564102564103e-06,
736
+ "loss": 0.7782,
737
+ "mean_token_accuracy": 0.9613427728414535,
738
+ "num_tokens": 9751915.0,
739
+ "step": 405
740
+ },
741
+ {
742
+ "epoch": 1.3104,
743
+ "grad_norm": 6.540207386016846,
744
+ "learning_rate": 4.344551282051283e-06,
745
+ "loss": 0.7907,
746
+ "mean_token_accuracy": 0.9609488919377327,
747
+ "num_tokens": 9872499.0,
748
+ "step": 410
749
+ },
750
+ {
751
+ "epoch": 1.3264,
752
+ "grad_norm": 5.874443054199219,
753
+ "learning_rate": 4.336538461538462e-06,
754
+ "loss": 0.7794,
755
+ "mean_token_accuracy": 0.9610813722014427,
756
+ "num_tokens": 9993484.0,
757
+ "step": 415
758
+ },
759
+ {
760
+ "epoch": 1.3424,
761
+ "grad_norm": 6.9412970542907715,
762
+ "learning_rate": 4.328525641025642e-06,
763
+ "loss": 0.7622,
764
+ "mean_token_accuracy": 0.9616369992494583,
765
+ "num_tokens": 10113940.0,
766
+ "step": 420
767
+ },
768
+ {
769
+ "epoch": 1.3584,
770
+ "grad_norm": 7.472849369049072,
771
+ "learning_rate": 4.320512820512821e-06,
772
+ "loss": 0.768,
773
+ "mean_token_accuracy": 0.9621543675661087,
774
+ "num_tokens": 10234132.0,
775
+ "step": 425
776
+ },
777
+ {
778
+ "epoch": 1.3744,
779
+ "grad_norm": 5.127303600311279,
780
+ "learning_rate": 4.312500000000001e-06,
781
+ "loss": 0.772,
782
+ "mean_token_accuracy": 0.9611857488751412,
783
+ "num_tokens": 10354498.0,
784
+ "step": 430
785
+ },
786
+ {
787
+ "epoch": 1.3904,
788
+ "grad_norm": 5.755953311920166,
789
+ "learning_rate": 4.30448717948718e-06,
790
+ "loss": 0.7634,
791
+ "mean_token_accuracy": 0.9619626298546791,
792
+ "num_tokens": 10474843.0,
793
+ "step": 435
794
+ },
795
+ {
796
+ "epoch": 1.4064,
797
+ "grad_norm": 5.51708984375,
798
+ "learning_rate": 4.296474358974359e-06,
799
+ "loss": 0.7872,
800
+ "mean_token_accuracy": 0.9606904596090317,
801
+ "num_tokens": 10595749.0,
802
+ "step": 440
803
+ },
804
+ {
805
+ "epoch": 1.4224,
806
+ "grad_norm": 7.458123683929443,
807
+ "learning_rate": 4.288461538461539e-06,
808
+ "loss": 0.7789,
809
+ "mean_token_accuracy": 0.9610949099063874,
810
+ "num_tokens": 10716401.0,
811
+ "step": 445
812
+ },
813
+ {
814
+ "epoch": 1.4384000000000001,
815
+ "grad_norm": 5.499053478240967,
816
+ "learning_rate": 4.280448717948718e-06,
817
+ "loss": 0.7747,
818
+ "mean_token_accuracy": 0.9613270804286003,
819
+ "num_tokens": 10837118.0,
820
+ "step": 450
821
+ },
822
+ {
823
+ "epoch": 1.4544000000000001,
824
+ "grad_norm": 6.049108028411865,
825
+ "learning_rate": 4.272435897435898e-06,
826
+ "loss": 0.7801,
827
+ "mean_token_accuracy": 0.9611206367611885,
828
+ "num_tokens": 10957539.0,
829
+ "step": 455
830
+ },
831
+ {
832
+ "epoch": 1.4704,
833
+ "grad_norm": 7.157883644104004,
834
+ "learning_rate": 4.2644230769230775e-06,
835
+ "loss": 0.7954,
836
+ "mean_token_accuracy": 0.9603321492671967,
837
+ "num_tokens": 11078325.0,
838
+ "step": 460
839
+ },
840
+ {
841
+ "epoch": 1.4864,
842
+ "grad_norm": 6.0010271072387695,
843
+ "learning_rate": 4.2564102564102566e-06,
844
+ "loss": 0.7709,
845
+ "mean_token_accuracy": 0.9614550709724426,
846
+ "num_tokens": 11198694.0,
847
+ "step": 465
848
+ },
849
+ {
850
+ "epoch": 1.5024,
851
+ "grad_norm": 8.201191902160645,
852
+ "learning_rate": 4.2483974358974365e-06,
853
+ "loss": 0.7744,
854
+ "mean_token_accuracy": 0.9610834464430809,
855
+ "num_tokens": 11319373.0,
856
+ "step": 470
857
+ },
858
+ {
859
+ "epoch": 1.5184,
860
+ "grad_norm": 12.29971694946289,
861
+ "learning_rate": 4.2403846153846155e-06,
862
+ "loss": 0.7852,
863
+ "mean_token_accuracy": 0.9605844736099243,
864
+ "num_tokens": 11440387.0,
865
+ "step": 475
866
+ },
867
+ {
868
+ "epoch": 1.5344,
869
+ "grad_norm": 5.45226526260376,
870
+ "learning_rate": 4.232371794871795e-06,
871
+ "loss": 0.7753,
872
+ "mean_token_accuracy": 0.9605903938412667,
873
+ "num_tokens": 11561269.0,
874
+ "step": 480
875
+ },
876
+ {
877
+ "epoch": 1.5504,
878
+ "grad_norm": 6.1695146560668945,
879
+ "learning_rate": 4.2243589743589745e-06,
880
+ "loss": 0.7681,
881
+ "mean_token_accuracy": 0.9615284174680709,
882
+ "num_tokens": 11681770.0,
883
+ "step": 485
884
+ },
885
+ {
886
+ "epoch": 1.5664,
887
+ "grad_norm": 6.1809868812561035,
888
+ "learning_rate": 4.216346153846154e-06,
889
+ "loss": 0.7835,
890
+ "mean_token_accuracy": 0.960690113902092,
891
+ "num_tokens": 11802629.0,
892
+ "step": 490
893
+ },
894
+ {
895
+ "epoch": 1.5824,
896
+ "grad_norm": 4.959415912628174,
897
+ "learning_rate": 4.208333333333333e-06,
898
+ "loss": 0.7403,
899
+ "mean_token_accuracy": 0.9627179771661758,
900
+ "num_tokens": 11923213.0,
901
+ "step": 495
902
+ },
903
+ {
904
+ "epoch": 1.5984,
905
+ "grad_norm": 6.081442832946777,
906
+ "learning_rate": 4.200320512820513e-06,
907
+ "loss": 0.7859,
908
+ "mean_token_accuracy": 0.9606704100966453,
909
+ "num_tokens": 12043890.0,
910
+ "step": 500
911
+ },
912
+ {
913
+ "epoch": 1.6143999999999998,
914
+ "grad_norm": 6.756661415100098,
915
+ "learning_rate": 4.192307692307692e-06,
916
+ "loss": 0.7675,
917
+ "mean_token_accuracy": 0.9612209632992744,
918
+ "num_tokens": 12164566.0,
919
+ "step": 505
920
+ },
921
+ {
922
+ "epoch": 1.6303999999999998,
923
+ "grad_norm": 7.509146690368652,
924
+ "learning_rate": 4.184294871794872e-06,
925
+ "loss": 0.7914,
926
+ "mean_token_accuracy": 0.960303983092308,
927
+ "num_tokens": 12285420.0,
928
+ "step": 510
929
+ },
930
+ {
931
+ "epoch": 1.6463999999999999,
932
+ "grad_norm": 10.71135139465332,
933
+ "learning_rate": 4.176282051282052e-06,
934
+ "loss": 0.7667,
935
+ "mean_token_accuracy": 0.9615985140204429,
936
+ "num_tokens": 12406219.0,
937
+ "step": 515
938
+ },
939
+ {
940
+ "epoch": 1.6623999999999999,
941
+ "grad_norm": 6.306976795196533,
942
+ "learning_rate": 4.168269230769231e-06,
943
+ "loss": 0.7857,
944
+ "mean_token_accuracy": 0.9603689208626747,
945
+ "num_tokens": 12526466.0,
946
+ "step": 520
947
+ },
948
+ {
949
+ "epoch": 1.6784,
950
+ "grad_norm": 6.802849769592285,
951
+ "learning_rate": 4.160256410256411e-06,
952
+ "loss": 0.7731,
953
+ "mean_token_accuracy": 0.9611164182424545,
954
+ "num_tokens": 12646793.0,
955
+ "step": 525
956
+ },
957
+ {
958
+ "epoch": 1.6944,
959
+ "grad_norm": 17.336223602294922,
960
+ "learning_rate": 4.15224358974359e-06,
961
+ "loss": 0.7596,
962
+ "mean_token_accuracy": 0.961407621204853,
963
+ "num_tokens": 12767178.0,
964
+ "step": 530
965
+ },
966
+ {
967
+ "epoch": 1.7104,
968
+ "grad_norm": 6.455531597137451,
969
+ "learning_rate": 4.14423076923077e-06,
970
+ "loss": 0.7695,
971
+ "mean_token_accuracy": 0.9612391993403435,
972
+ "num_tokens": 12887721.0,
973
+ "step": 535
974
+ },
975
+ {
976
+ "epoch": 1.7264,
977
+ "grad_norm": 6.00295352935791,
978
+ "learning_rate": 4.136217948717949e-06,
979
+ "loss": 0.7748,
980
+ "mean_token_accuracy": 0.9608838215470314,
981
+ "num_tokens": 13008313.0,
982
+ "step": 540
983
+ },
984
+ {
985
+ "epoch": 1.7424,
986
+ "grad_norm": 12.987116813659668,
987
+ "learning_rate": 4.128205128205128e-06,
988
+ "loss": 0.7671,
989
+ "mean_token_accuracy": 0.9614728838205338,
990
+ "num_tokens": 13128955.0,
991
+ "step": 545
992
+ },
993
+ {
994
+ "epoch": 1.7584,
995
+ "grad_norm": 5.461932182312012,
996
+ "learning_rate": 4.120192307692308e-06,
997
+ "loss": 0.7628,
998
+ "mean_token_accuracy": 0.9612700089812278,
999
+ "num_tokens": 13249591.0,
1000
+ "step": 550
1001
+ },
1002
+ {
1003
+ "epoch": 1.7744,
1004
+ "grad_norm": 6.125832557678223,
1005
+ "learning_rate": 4.112179487179487e-06,
1006
+ "loss": 0.765,
1007
+ "mean_token_accuracy": 0.9620017781853676,
1008
+ "num_tokens": 13370055.0,
1009
+ "step": 555
1010
+ },
1011
+ {
1012
+ "epoch": 1.7904,
1013
+ "grad_norm": 6.532048225402832,
1014
+ "learning_rate": 4.104166666666667e-06,
1015
+ "loss": 0.7765,
1016
+ "mean_token_accuracy": 0.9612124145030976,
1017
+ "num_tokens": 13490838.0,
1018
+ "step": 560
1019
+ },
1020
+ {
1021
+ "epoch": 1.8064,
1022
+ "grad_norm": 8.854321479797363,
1023
+ "learning_rate": 4.096153846153846e-06,
1024
+ "loss": 0.7764,
1025
+ "mean_token_accuracy": 0.9605302706360817,
1026
+ "num_tokens": 13611213.0,
1027
+ "step": 565
1028
+ },
1029
+ {
1030
+ "epoch": 1.8224,
1031
+ "grad_norm": 8.60447883605957,
1032
+ "learning_rate": 4.088141025641026e-06,
1033
+ "loss": 0.7674,
1034
+ "mean_token_accuracy": 0.9616305485367775,
1035
+ "num_tokens": 13731429.0,
1036
+ "step": 570
1037
+ },
1038
+ {
1039
+ "epoch": 1.8384,
1040
+ "grad_norm": 6.810831069946289,
1041
+ "learning_rate": 4.080128205128205e-06,
1042
+ "loss": 0.7701,
1043
+ "mean_token_accuracy": 0.9614489808678627,
1044
+ "num_tokens": 13852218.0,
1045
+ "step": 575
1046
+ },
1047
+ {
1048
+ "epoch": 1.8544,
1049
+ "grad_norm": 7.028332233428955,
1050
+ "learning_rate": 4.072115384615385e-06,
1051
+ "loss": 0.7636,
1052
+ "mean_token_accuracy": 0.9619327470660209,
1053
+ "num_tokens": 13973293.0,
1054
+ "step": 580
1055
+ },
1056
+ {
1057
+ "epoch": 1.8704,
1058
+ "grad_norm": 9.146780967712402,
1059
+ "learning_rate": 4.064102564102565e-06,
1060
+ "loss": 0.7773,
1061
+ "mean_token_accuracy": 0.9614929273724556,
1062
+ "num_tokens": 14093793.0,
1063
+ "step": 585
1064
+ },
1065
+ {
1066
+ "epoch": 1.8864,
1067
+ "grad_norm": 6.394257545471191,
1068
+ "learning_rate": 4.056089743589744e-06,
1069
+ "loss": 0.7461,
1070
+ "mean_token_accuracy": 0.9629229962825775,
1071
+ "num_tokens": 14214536.0,
1072
+ "step": 590
1073
+ },
1074
+ {
1075
+ "epoch": 1.9024,
1076
+ "grad_norm": 5.9049072265625,
1077
+ "learning_rate": 4.048076923076924e-06,
1078
+ "loss": 0.7545,
1079
+ "mean_token_accuracy": 0.962635500729084,
1080
+ "num_tokens": 14334648.0,
1081
+ "step": 595
1082
+ },
1083
+ {
1084
+ "epoch": 1.9184,
1085
+ "grad_norm": 5.539757251739502,
1086
+ "learning_rate": 4.040064102564103e-06,
1087
+ "loss": 0.782,
1088
+ "mean_token_accuracy": 0.9615770474076271,
1089
+ "num_tokens": 14455323.0,
1090
+ "step": 600
1091
+ },
1092
+ {
1093
+ "epoch": 1.9344000000000001,
1094
+ "grad_norm": 6.433361053466797,
1095
+ "learning_rate": 4.032051282051283e-06,
1096
+ "loss": 0.784,
1097
+ "mean_token_accuracy": 0.9606905117630958,
1098
+ "num_tokens": 14576141.0,
1099
+ "step": 605
1100
+ },
1101
+ {
1102
+ "epoch": 1.9504000000000001,
1103
+ "grad_norm": 6.863312244415283,
1104
+ "learning_rate": 4.024038461538462e-06,
1105
+ "loss": 0.7707,
1106
+ "mean_token_accuracy": 0.9615117534995079,
1107
+ "num_tokens": 14696891.0,
1108
+ "step": 610
1109
+ },
1110
+ {
1111
+ "epoch": 1.9664000000000001,
1112
+ "grad_norm": 5.4934401512146,
1113
+ "learning_rate": 4.016025641025642e-06,
1114
+ "loss": 0.7803,
1115
+ "mean_token_accuracy": 0.9609077662229538,
1116
+ "num_tokens": 14817569.0,
1117
+ "step": 615
1118
+ },
1119
+ {
1120
+ "epoch": 1.9824000000000002,
1121
+ "grad_norm": 9.079723358154297,
1122
+ "learning_rate": 4.008012820512821e-06,
1123
+ "loss": 0.7571,
1124
+ "mean_token_accuracy": 0.9620749920606613,
1125
+ "num_tokens": 14938053.0,
1126
+ "step": 620
1127
+ },
1128
+ {
1129
+ "epoch": 1.9984,
1130
+ "grad_norm": 8.422645568847656,
1131
+ "learning_rate": 4.000000000000001e-06,
1132
+ "loss": 0.7631,
1133
+ "mean_token_accuracy": 0.9615694731473923,
1134
+ "num_tokens": 15058346.0,
1135
+ "step": 625
1136
+ },
1137
+ {
1138
+ "epoch": 2.0128,
1139
+ "grad_norm": 6.332509994506836,
1140
+ "learning_rate": 3.99198717948718e-06,
1141
+ "loss": 0.5703,
1142
+ "mean_token_accuracy": 0.9675304856565263,
1143
+ "num_tokens": 15166744.0,
1144
+ "step": 630
1145
+ },
1146
+ {
1147
+ "epoch": 2.0288,
1148
+ "grad_norm": 5.059317588806152,
1149
+ "learning_rate": 3.983974358974359e-06,
1150
+ "loss": 0.5949,
1151
+ "mean_token_accuracy": 0.969310836493969,
1152
+ "num_tokens": 15286876.0,
1153
+ "step": 635
1154
+ },
1155
+ {
1156
+ "epoch": 2.0448,
1157
+ "grad_norm": 26.722389221191406,
1158
+ "learning_rate": 3.975961538461539e-06,
1159
+ "loss": 0.595,
1160
+ "mean_token_accuracy": 0.9692401915788651,
1161
+ "num_tokens": 15407454.0,
1162
+ "step": 640
1163
+ },
1164
+ {
1165
+ "epoch": 2.0608,
1166
+ "grad_norm": 7.642247676849365,
1167
+ "learning_rate": 3.967948717948718e-06,
1168
+ "loss": 0.5954,
1169
+ "mean_token_accuracy": 0.9694274187088012,
1170
+ "num_tokens": 15527899.0,
1171
+ "step": 645
1172
+ },
1173
+ {
1174
+ "epoch": 2.0768,
1175
+ "grad_norm": 5.989568710327148,
1176
+ "learning_rate": 3.959935897435898e-06,
1177
+ "loss": 0.58,
1178
+ "mean_token_accuracy": 0.9702744409441948,
1179
+ "num_tokens": 15648501.0,
1180
+ "step": 650
1181
+ },
1182
+ {
1183
+ "epoch": 2.0928,
1184
+ "grad_norm": 5.8522539138793945,
1185
+ "learning_rate": 3.951923076923077e-06,
1186
+ "loss": 0.5877,
1187
+ "mean_token_accuracy": 0.969629842042923,
1188
+ "num_tokens": 15769067.0,
1189
+ "step": 655
1190
+ },
1191
+ {
1192
+ "epoch": 2.1088,
1193
+ "grad_norm": 7.153563499450684,
1194
+ "learning_rate": 3.943910256410257e-06,
1195
+ "loss": 0.6091,
1196
+ "mean_token_accuracy": 0.9694206699728966,
1197
+ "num_tokens": 15889921.0,
1198
+ "step": 660
1199
+ },
1200
+ {
1201
+ "epoch": 2.1248,
1202
+ "grad_norm": 9.316864967346191,
1203
+ "learning_rate": 3.9358974358974365e-06,
1204
+ "loss": 0.6074,
1205
+ "mean_token_accuracy": 0.9693100199103355,
1206
+ "num_tokens": 16010363.0,
1207
+ "step": 665
1208
+ },
1209
+ {
1210
+ "epoch": 2.1408,
1211
+ "grad_norm": 6.018698692321777,
1212
+ "learning_rate": 3.9278846153846155e-06,
1213
+ "loss": 0.5841,
1214
+ "mean_token_accuracy": 0.9702739328145981,
1215
+ "num_tokens": 16131178.0,
1216
+ "step": 670
1217
+ },
1218
+ {
1219
+ "epoch": 2.1568,
1220
+ "grad_norm": 5.0297956466674805,
1221
+ "learning_rate": 3.9198717948717954e-06,
1222
+ "loss": 0.5866,
1223
+ "mean_token_accuracy": 0.9702060952782631,
1224
+ "num_tokens": 16252246.0,
1225
+ "step": 675
1226
+ },
1227
+ {
1228
+ "epoch": 2.1728,
1229
+ "grad_norm": 6.389510631561279,
1230
+ "learning_rate": 3.9118589743589745e-06,
1231
+ "loss": 0.6023,
1232
+ "mean_token_accuracy": 0.9690487116575242,
1233
+ "num_tokens": 16373081.0,
1234
+ "step": 680
1235
+ },
1236
+ {
1237
+ "epoch": 2.1888,
1238
+ "grad_norm": 8.721376419067383,
1239
+ "learning_rate": 3.903846153846154e-06,
1240
+ "loss": 0.5884,
1241
+ "mean_token_accuracy": 0.9700992733240128,
1242
+ "num_tokens": 16493434.0,
1243
+ "step": 685
1244
+ },
1245
+ {
1246
+ "epoch": 2.2048,
1247
+ "grad_norm": 5.946844100952148,
1248
+ "learning_rate": 3.8958333333333334e-06,
1249
+ "loss": 0.6126,
1250
+ "mean_token_accuracy": 0.9683265417814255,
1251
+ "num_tokens": 16614077.0,
1252
+ "step": 690
1253
+ },
1254
+ {
1255
+ "epoch": 2.2208,
1256
+ "grad_norm": 8.122442245483398,
1257
+ "learning_rate": 3.887820512820513e-06,
1258
+ "loss": 0.5923,
1259
+ "mean_token_accuracy": 0.9693433627486229,
1260
+ "num_tokens": 16734274.0,
1261
+ "step": 695
1262
+ },
1263
+ {
1264
+ "epoch": 2.2368,
1265
+ "grad_norm": 5.81650972366333,
1266
+ "learning_rate": 3.879807692307692e-06,
1267
+ "loss": 0.5988,
1268
+ "mean_token_accuracy": 0.9687500342726707,
1269
+ "num_tokens": 16854978.0,
1270
+ "step": 700
1271
+ },
1272
+ {
1273
+ "epoch": 2.2528,
1274
+ "grad_norm": 6.019922733306885,
1275
+ "learning_rate": 3.871794871794872e-06,
1276
+ "loss": 0.5859,
1277
+ "mean_token_accuracy": 0.9704526200890541,
1278
+ "num_tokens": 16975035.0,
1279
+ "step": 705
1280
+ },
1281
+ {
1282
+ "epoch": 2.2688,
1283
+ "grad_norm": 22.37713623046875,
1284
+ "learning_rate": 3.863782051282051e-06,
1285
+ "loss": 0.6071,
1286
+ "mean_token_accuracy": 0.9692072600126267,
1287
+ "num_tokens": 17095517.0,
1288
+ "step": 710
1289
+ },
1290
+ {
1291
+ "epoch": 2.2848,
1292
+ "grad_norm": 5.264857292175293,
1293
+ "learning_rate": 3.855769230769231e-06,
1294
+ "loss": 0.592,
1295
+ "mean_token_accuracy": 0.9693273976445198,
1296
+ "num_tokens": 17215962.0,
1297
+ "step": 715
1298
+ },
1299
+ {
1300
+ "epoch": 2.3008,
1301
+ "grad_norm": 5.581426620483398,
1302
+ "learning_rate": 3.847756410256411e-06,
1303
+ "loss": 0.6056,
1304
+ "mean_token_accuracy": 0.9688428491353989,
1305
+ "num_tokens": 17336615.0,
1306
+ "step": 720
1307
+ },
1308
+ {
1309
+ "epoch": 2.3168,
1310
+ "grad_norm": 6.651681900024414,
1311
+ "learning_rate": 3.83974358974359e-06,
1312
+ "loss": 0.6032,
1313
+ "mean_token_accuracy": 0.9692152827978134,
1314
+ "num_tokens": 17457485.0,
1315
+ "step": 725
1316
+ },
1317
+ {
1318
+ "epoch": 2.3327999999999998,
1319
+ "grad_norm": 5.19486665725708,
1320
+ "learning_rate": 3.83173076923077e-06,
1321
+ "loss": 0.5868,
1322
+ "mean_token_accuracy": 0.9699591442942619,
1323
+ "num_tokens": 17577879.0,
1324
+ "step": 730
1325
+ },
1326
+ {
1327
+ "epoch": 2.3487999999999998,
1328
+ "grad_norm": 10.026927947998047,
1329
+ "learning_rate": 3.823717948717949e-06,
1330
+ "loss": 0.6036,
1331
+ "mean_token_accuracy": 0.9688664764165879,
1332
+ "num_tokens": 17698565.0,
1333
+ "step": 735
1334
+ },
1335
+ {
1336
+ "epoch": 2.3648,
1337
+ "grad_norm": 6.088404655456543,
1338
+ "learning_rate": 3.815705128205128e-06,
1339
+ "loss": 0.59,
1340
+ "mean_token_accuracy": 0.9702243730425835,
1341
+ "num_tokens": 17818816.0,
1342
+ "step": 740
1343
+ },
1344
+ {
1345
+ "epoch": 2.3808,
1346
+ "grad_norm": 5.913634777069092,
1347
+ "learning_rate": 3.8076923076923077e-06,
1348
+ "loss": 0.6021,
1349
+ "mean_token_accuracy": 0.9690718874335289,
1350
+ "num_tokens": 17938922.0,
1351
+ "step": 745
1352
+ },
1353
+ {
1354
+ "epoch": 2.3968,
1355
+ "grad_norm": 5.578912258148193,
1356
+ "learning_rate": 3.799679487179487e-06,
1357
+ "loss": 0.6027,
1358
+ "mean_token_accuracy": 0.9693442419171333,
1359
+ "num_tokens": 18059410.0,
1360
+ "step": 750
1361
+ },
1362
+ {
1363
+ "epoch": 2.4128,
1364
+ "grad_norm": 5.979203224182129,
1365
+ "learning_rate": 3.7916666666666666e-06,
1366
+ "loss": 0.6162,
1367
+ "mean_token_accuracy": 0.9689163967967034,
1368
+ "num_tokens": 18180054.0,
1369
+ "step": 755
1370
+ },
1371
+ {
1372
+ "epoch": 2.4288,
1373
+ "grad_norm": 7.138234615325928,
1374
+ "learning_rate": 3.7836538461538465e-06,
1375
+ "loss": 0.5957,
1376
+ "mean_token_accuracy": 0.969460716843605,
1377
+ "num_tokens": 18300969.0,
1378
+ "step": 760
1379
+ },
1380
+ {
1381
+ "epoch": 2.4448,
1382
+ "grad_norm": 5.8525543212890625,
1383
+ "learning_rate": 3.775641025641026e-06,
1384
+ "loss": 0.5927,
1385
+ "mean_token_accuracy": 0.9691488027572632,
1386
+ "num_tokens": 18421406.0,
1387
+ "step": 765
1388
+ },
1389
+ {
1390
+ "epoch": 2.4608,
1391
+ "grad_norm": 6.104275226593018,
1392
+ "learning_rate": 3.7676282051282055e-06,
1393
+ "loss": 0.6108,
1394
+ "mean_token_accuracy": 0.9684408336877823,
1395
+ "num_tokens": 18542403.0,
1396
+ "step": 770
1397
+ },
1398
+ {
1399
+ "epoch": 2.4768,
1400
+ "grad_norm": 5.239029884338379,
1401
+ "learning_rate": 3.759615384615385e-06,
1402
+ "loss": 0.609,
1403
+ "mean_token_accuracy": 0.9689425334334374,
1404
+ "num_tokens": 18663187.0,
1405
+ "step": 775
1406
+ },
1407
+ {
1408
+ "epoch": 2.4928,
1409
+ "grad_norm": 6.184327602386475,
1410
+ "learning_rate": 3.7516025641025645e-06,
1411
+ "loss": 0.6009,
1412
+ "mean_token_accuracy": 0.969046376645565,
1413
+ "num_tokens": 18784011.0,
1414
+ "step": 780
1415
+ },
1416
+ {
1417
+ "epoch": 2.5088,
1418
+ "grad_norm": 6.316597938537598,
1419
+ "learning_rate": 3.743589743589744e-06,
1420
+ "loss": 0.6018,
1421
+ "mean_token_accuracy": 0.9692170903086662,
1422
+ "num_tokens": 18904963.0,
1423
+ "step": 785
1424
+ },
1425
+ {
1426
+ "epoch": 2.5248,
1427
+ "grad_norm": 5.539931774139404,
1428
+ "learning_rate": 3.7355769230769234e-06,
1429
+ "loss": 0.6036,
1430
+ "mean_token_accuracy": 0.9689925819635391,
1431
+ "num_tokens": 19025272.0,
1432
+ "step": 790
1433
+ },
1434
+ {
1435
+ "epoch": 2.5408,
1436
+ "grad_norm": 5.889646530151367,
1437
+ "learning_rate": 3.727564102564103e-06,
1438
+ "loss": 0.6048,
1439
+ "mean_token_accuracy": 0.9689805299043656,
1440
+ "num_tokens": 19145913.0,
1441
+ "step": 795
1442
+ },
1443
+ {
1444
+ "epoch": 2.5568,
1445
+ "grad_norm": 5.824395179748535,
1446
+ "learning_rate": 3.7195512820512824e-06,
1447
+ "loss": 0.6137,
1448
+ "mean_token_accuracy": 0.9686646088957787,
1449
+ "num_tokens": 19266252.0,
1450
+ "step": 800
1451
+ },
1452
+ {
1453
+ "epoch": 2.5728,
1454
+ "grad_norm": 7.852932929992676,
1455
+ "learning_rate": 3.711538461538462e-06,
1456
+ "loss": 0.5811,
1457
+ "mean_token_accuracy": 0.9694151401519775,
1458
+ "num_tokens": 19386867.0,
1459
+ "step": 805
1460
+ },
1461
+ {
1462
+ "epoch": 2.5888,
1463
+ "grad_norm": 5.9719038009643555,
1464
+ "learning_rate": 3.7035256410256413e-06,
1465
+ "loss": 0.6173,
1466
+ "mean_token_accuracy": 0.9685935735702514,
1467
+ "num_tokens": 19507561.0,
1468
+ "step": 810
1469
+ },
1470
+ {
1471
+ "epoch": 2.6048,
1472
+ "grad_norm": 7.026526927947998,
1473
+ "learning_rate": 3.6955128205128212e-06,
1474
+ "loss": 0.6107,
1475
+ "mean_token_accuracy": 0.968952152132988,
1476
+ "num_tokens": 19627765.0,
1477
+ "step": 815
1478
+ },
1479
+ {
1480
+ "epoch": 2.6208,
1481
+ "grad_norm": 5.256075382232666,
1482
+ "learning_rate": 3.6875000000000007e-06,
1483
+ "loss": 0.5974,
1484
+ "mean_token_accuracy": 0.9685025453567505,
1485
+ "num_tokens": 19748365.0,
1486
+ "step": 820
1487
+ },
1488
+ {
1489
+ "epoch": 2.6368,
1490
+ "grad_norm": 28.91472816467285,
1491
+ "learning_rate": 3.67948717948718e-06,
1492
+ "loss": 0.6097,
1493
+ "mean_token_accuracy": 0.9687363058328629,
1494
+ "num_tokens": 19868739.0,
1495
+ "step": 825
1496
+ },
1497
+ {
1498
+ "epoch": 2.6528,
1499
+ "grad_norm": 8.917984962463379,
1500
+ "learning_rate": 3.671474358974359e-06,
1501
+ "loss": 0.606,
1502
+ "mean_token_accuracy": 0.9686084419488907,
1503
+ "num_tokens": 19989308.0,
1504
+ "step": 830
1505
+ },
1506
+ {
1507
+ "epoch": 2.6688,
1508
+ "grad_norm": 59.591033935546875,
1509
+ "learning_rate": 3.6634615384615387e-06,
1510
+ "loss": 0.6096,
1511
+ "mean_token_accuracy": 0.9685346141457558,
1512
+ "num_tokens": 20109995.0,
1513
+ "step": 835
1514
+ },
1515
+ {
1516
+ "epoch": 2.6848,
1517
+ "grad_norm": 5.761416435241699,
1518
+ "learning_rate": 3.655448717948718e-06,
1519
+ "loss": 0.6061,
1520
+ "mean_token_accuracy": 0.9690618306398392,
1521
+ "num_tokens": 20230636.0,
1522
+ "step": 840
1523
+ },
1524
+ {
1525
+ "epoch": 2.7008,
1526
+ "grad_norm": 5.899073600769043,
1527
+ "learning_rate": 3.6474358974358977e-06,
1528
+ "loss": 0.6141,
1529
+ "mean_token_accuracy": 0.9684308752417564,
1530
+ "num_tokens": 20350953.0,
1531
+ "step": 845
1532
+ },
1533
+ {
1534
+ "epoch": 2.7168,
1535
+ "grad_norm": 21.62592315673828,
1536
+ "learning_rate": 3.639423076923077e-06,
1537
+ "loss": 0.6055,
1538
+ "mean_token_accuracy": 0.9686669781804085,
1539
+ "num_tokens": 20472002.0,
1540
+ "step": 850
1541
+ },
1542
+ {
1543
+ "epoch": 2.7328,
1544
+ "grad_norm": 5.920300483703613,
1545
+ "learning_rate": 3.6314102564102566e-06,
1546
+ "loss": 0.6033,
1547
+ "mean_token_accuracy": 0.9695612549781799,
1548
+ "num_tokens": 20592717.0,
1549
+ "step": 855
1550
+ },
1551
+ {
1552
+ "epoch": 2.7488,
1553
+ "grad_norm": 6.941519260406494,
1554
+ "learning_rate": 3.623397435897436e-06,
1555
+ "loss": 0.5888,
1556
+ "mean_token_accuracy": 0.9697792261838913,
1557
+ "num_tokens": 20713119.0,
1558
+ "step": 860
1559
+ },
1560
+ {
1561
+ "epoch": 2.7648,
1562
+ "grad_norm": 5.511123180389404,
1563
+ "learning_rate": 3.6153846153846156e-06,
1564
+ "loss": 0.5846,
1565
+ "mean_token_accuracy": 0.9694935277104377,
1566
+ "num_tokens": 20833519.0,
1567
+ "step": 865
1568
+ },
1569
+ {
1570
+ "epoch": 2.7808,
1571
+ "grad_norm": 6.110658168792725,
1572
+ "learning_rate": 3.607371794871795e-06,
1573
+ "loss": 0.5881,
1574
+ "mean_token_accuracy": 0.969414946436882,
1575
+ "num_tokens": 20954124.0,
1576
+ "step": 870
1577
+ },
1578
+ {
1579
+ "epoch": 2.7968,
1580
+ "grad_norm": 5.648221492767334,
1581
+ "learning_rate": 3.5993589743589745e-06,
1582
+ "loss": 0.5994,
1583
+ "mean_token_accuracy": 0.968793374300003,
1584
+ "num_tokens": 21074992.0,
1585
+ "step": 875
1586
+ },
1587
+ {
1588
+ "epoch": 2.8128,
1589
+ "grad_norm": 7.558088779449463,
1590
+ "learning_rate": 3.591346153846154e-06,
1591
+ "loss": 0.5967,
1592
+ "mean_token_accuracy": 0.9693120524287224,
1593
+ "num_tokens": 21195276.0,
1594
+ "step": 880
1595
+ },
1596
+ {
1597
+ "epoch": 2.8288,
1598
+ "grad_norm": 6.586251258850098,
1599
+ "learning_rate": 3.5833333333333335e-06,
1600
+ "loss": 0.6086,
1601
+ "mean_token_accuracy": 0.968573434650898,
1602
+ "num_tokens": 21315771.0,
1603
+ "step": 885
1604
+ },
1605
+ {
1606
+ "epoch": 2.8448,
1607
+ "grad_norm": 5.744991779327393,
1608
+ "learning_rate": 3.5753205128205134e-06,
1609
+ "loss": 0.6119,
1610
+ "mean_token_accuracy": 0.9684572085738182,
1611
+ "num_tokens": 21436769.0,
1612
+ "step": 890
1613
+ },
1614
+ {
1615
+ "epoch": 2.8608000000000002,
1616
+ "grad_norm": 5.792296886444092,
1617
+ "learning_rate": 3.567307692307693e-06,
1618
+ "loss": 0.6087,
1619
+ "mean_token_accuracy": 0.9687753587961196,
1620
+ "num_tokens": 21557017.0,
1621
+ "step": 895
1622
+ },
1623
+ {
1624
+ "epoch": 2.8768000000000002,
1625
+ "grad_norm": 6.981272220611572,
1626
+ "learning_rate": 3.5592948717948723e-06,
1627
+ "loss": 0.6116,
1628
+ "mean_token_accuracy": 0.9684758573770523,
1629
+ "num_tokens": 21677893.0,
1630
+ "step": 900
1631
+ },
1632
+ {
1633
+ "epoch": 2.8928000000000003,
1634
+ "grad_norm": 5.47102165222168,
1635
+ "learning_rate": 3.551282051282052e-06,
1636
+ "loss": 0.6048,
1637
+ "mean_token_accuracy": 0.9690106600522995,
1638
+ "num_tokens": 21798295.0,
1639
+ "step": 905
1640
+ },
1641
+ {
1642
+ "epoch": 2.9088000000000003,
1643
+ "grad_norm": 5.765688419342041,
1644
+ "learning_rate": 3.5432692307692313e-06,
1645
+ "loss": 0.606,
1646
+ "mean_token_accuracy": 0.9694508373737335,
1647
+ "num_tokens": 21919027.0,
1648
+ "step": 910
1649
+ },
1650
+ {
1651
+ "epoch": 2.9248,
1652
+ "grad_norm": 6.3723649978637695,
1653
+ "learning_rate": 3.5352564102564108e-06,
1654
+ "loss": 0.6155,
1655
+ "mean_token_accuracy": 0.96856909096241,
1656
+ "num_tokens": 22040139.0,
1657
+ "step": 915
1658
+ },
1659
+ {
1660
+ "epoch": 2.9408,
1661
+ "grad_norm": 8.107088088989258,
1662
+ "learning_rate": 3.5272435897435902e-06,
1663
+ "loss": 0.6092,
1664
+ "mean_token_accuracy": 0.9685668602585793,
1665
+ "num_tokens": 22160663.0,
1666
+ "step": 920
1667
+ },
1668
+ {
1669
+ "epoch": 2.9568,
1670
+ "grad_norm": 5.836716651916504,
1671
+ "learning_rate": 3.5192307692307697e-06,
1672
+ "loss": 0.6071,
1673
+ "mean_token_accuracy": 0.9687439948320389,
1674
+ "num_tokens": 22280766.0,
1675
+ "step": 925
1676
+ },
1677
+ {
1678
+ "epoch": 2.9728,
1679
+ "grad_norm": 6.721403121948242,
1680
+ "learning_rate": 3.5112179487179488e-06,
1681
+ "loss": 0.6088,
1682
+ "mean_token_accuracy": 0.9685460940003395,
1683
+ "num_tokens": 22401108.0,
1684
+ "step": 930
1685
+ },
1686
+ {
1687
+ "epoch": 2.9888,
1688
+ "grad_norm": 10.454011917114258,
1689
+ "learning_rate": 3.5032051282051282e-06,
1690
+ "loss": 0.5982,
1691
+ "mean_token_accuracy": 0.9691187739372253,
1692
+ "num_tokens": 22521450.0,
1693
+ "step": 935
1694
+ },
1695
+ {
1696
+ "epoch": 3.0032,
1697
+ "grad_norm": 5.327872276306152,
1698
+ "learning_rate": 3.4951923076923077e-06,
1699
+ "loss": 0.5109,
1700
+ "mean_token_accuracy": 0.9710118456019295,
1701
+ "num_tokens": 22629808.0,
1702
+ "step": 940
1703
+ },
1704
+ {
1705
+ "epoch": 3.0192,
1706
+ "grad_norm": 5.820178508758545,
1707
+ "learning_rate": 3.487179487179487e-06,
1708
+ "loss": 0.4232,
1709
+ "mean_token_accuracy": 0.9789220675826072,
1710
+ "num_tokens": 22750211.0,
1711
+ "step": 945
1712
+ },
1713
+ {
1714
+ "epoch": 3.0352,
1715
+ "grad_norm": 7.449097156524658,
1716
+ "learning_rate": 3.4791666666666667e-06,
1717
+ "loss": 0.4129,
1718
+ "mean_token_accuracy": 0.9787988021969796,
1719
+ "num_tokens": 22870555.0,
1720
+ "step": 950
1721
+ },
1722
+ {
1723
+ "epoch": 3.0512,
1724
+ "grad_norm": 7.855545997619629,
1725
+ "learning_rate": 3.471153846153846e-06,
1726
+ "loss": 0.4132,
1727
+ "mean_token_accuracy": 0.9787791818380356,
1728
+ "num_tokens": 22991532.0,
1729
+ "step": 955
1730
+ },
1731
+ {
1732
+ "epoch": 3.0672,
1733
+ "grad_norm": 8.52786636352539,
1734
+ "learning_rate": 3.4631410256410256e-06,
1735
+ "loss": 0.4119,
1736
+ "mean_token_accuracy": 0.9787372574210167,
1737
+ "num_tokens": 23112105.0,
1738
+ "step": 960
1739
+ },
1740
+ {
1741
+ "epoch": 3.0832,
1742
+ "grad_norm": 8.268338203430176,
1743
+ "learning_rate": 3.4551282051282055e-06,
1744
+ "loss": 0.4034,
1745
+ "mean_token_accuracy": 0.9794875472784043,
1746
+ "num_tokens": 23232673.0,
1747
+ "step": 965
1748
+ },
1749
+ {
1750
+ "epoch": 3.0992,
1751
+ "grad_norm": 7.042623519897461,
1752
+ "learning_rate": 3.447115384615385e-06,
1753
+ "loss": 0.4175,
1754
+ "mean_token_accuracy": 0.9787004634737968,
1755
+ "num_tokens": 23353313.0,
1756
+ "step": 970
1757
+ },
1758
+ {
1759
+ "epoch": 3.1152,
1760
+ "grad_norm": 11.916893005371094,
1761
+ "learning_rate": 3.4391025641025645e-06,
1762
+ "loss": 0.4108,
1763
+ "mean_token_accuracy": 0.978770087659359,
1764
+ "num_tokens": 23473873.0,
1765
+ "step": 975
1766
+ },
1767
+ {
1768
+ "epoch": 3.1312,
1769
+ "grad_norm": 7.312808990478516,
1770
+ "learning_rate": 3.431089743589744e-06,
1771
+ "loss": 0.4118,
1772
+ "mean_token_accuracy": 0.9786840796470642,
1773
+ "num_tokens": 23594585.0,
1774
+ "step": 980
1775
+ },
1776
+ {
1777
+ "epoch": 3.1471999999999998,
1778
+ "grad_norm": 8.837620735168457,
1779
+ "learning_rate": 3.4230769230769234e-06,
1780
+ "loss": 0.4118,
1781
+ "mean_token_accuracy": 0.9786578044295311,
1782
+ "num_tokens": 23715572.0,
1783
+ "step": 985
1784
+ },
1785
+ {
1786
+ "epoch": 3.1632,
1787
+ "grad_norm": 8.318835258483887,
1788
+ "learning_rate": 3.415064102564103e-06,
1789
+ "loss": 0.3931,
1790
+ "mean_token_accuracy": 0.9794891104102135,
1791
+ "num_tokens": 23836176.0,
1792
+ "step": 990
1793
+ },
1794
+ {
1795
+ "epoch": 3.1792,
1796
+ "grad_norm": 7.253737449645996,
1797
+ "learning_rate": 3.4070512820512824e-06,
1798
+ "loss": 0.3974,
1799
+ "mean_token_accuracy": 0.9790382608771324,
1800
+ "num_tokens": 23956866.0,
1801
+ "step": 995
1802
+ },
1803
+ {
1804
+ "epoch": 3.1952,
1805
+ "grad_norm": 7.655492782592773,
1806
+ "learning_rate": 3.399038461538462e-06,
1807
+ "loss": 0.398,
1808
+ "mean_token_accuracy": 0.9788922190666198,
1809
+ "num_tokens": 24077478.0,
1810
+ "step": 1000
1811
+ },
1812
+ {
1813
+ "epoch": 3.2112,
1814
+ "grad_norm": 6.616061210632324,
1815
+ "learning_rate": 3.3910256410256413e-06,
1816
+ "loss": 0.4068,
1817
+ "mean_token_accuracy": 0.97861547768116,
1818
+ "num_tokens": 24197838.0,
1819
+ "step": 1005
1820
+ },
1821
+ {
1822
+ "epoch": 3.2272,
1823
+ "grad_norm": 9.614058494567871,
1824
+ "learning_rate": 3.383012820512821e-06,
1825
+ "loss": 0.4056,
1826
+ "mean_token_accuracy": 0.9788064673542977,
1827
+ "num_tokens": 24318109.0,
1828
+ "step": 1010
1829
+ },
1830
+ {
1831
+ "epoch": 3.2432,
1832
+ "grad_norm": 7.204268932342529,
1833
+ "learning_rate": 3.3750000000000003e-06,
1834
+ "loss": 0.4145,
1835
+ "mean_token_accuracy": 0.9783661872148514,
1836
+ "num_tokens": 24438516.0,
1837
+ "step": 1015
1838
+ },
1839
+ {
1840
+ "epoch": 3.2592,
1841
+ "grad_norm": 6.809508800506592,
1842
+ "learning_rate": 3.36698717948718e-06,
1843
+ "loss": 0.4081,
1844
+ "mean_token_accuracy": 0.9788344666361809,
1845
+ "num_tokens": 24559587.0,
1846
+ "step": 1020
1847
+ },
1848
+ {
1849
+ "epoch": 3.2752,
1850
+ "grad_norm": 7.658448219299316,
1851
+ "learning_rate": 3.358974358974359e-06,
1852
+ "loss": 0.3967,
1853
+ "mean_token_accuracy": 0.9791337251663208,
1854
+ "num_tokens": 24680329.0,
1855
+ "step": 1025
1856
+ },
1857
+ {
1858
+ "epoch": 3.2912,
1859
+ "grad_norm": 7.103062152862549,
1860
+ "learning_rate": 3.3509615384615383e-06,
1861
+ "loss": 0.3973,
1862
+ "mean_token_accuracy": 0.9790103673934937,
1863
+ "num_tokens": 24800784.0,
1864
+ "step": 1030
1865
+ },
1866
+ {
1867
+ "epoch": 3.3072,
1868
+ "grad_norm": 8.693245887756348,
1869
+ "learning_rate": 3.342948717948718e-06,
1870
+ "loss": 0.4118,
1871
+ "mean_token_accuracy": 0.978698642551899,
1872
+ "num_tokens": 24921516.0,
1873
+ "step": 1035
1874
+ },
1875
+ {
1876
+ "epoch": 3.3232,
1877
+ "grad_norm": 7.297614574432373,
1878
+ "learning_rate": 3.3349358974358977e-06,
1879
+ "loss": 0.4113,
1880
+ "mean_token_accuracy": 0.9780242666602135,
1881
+ "num_tokens": 25041948.0,
1882
+ "step": 1040
1883
+ },
1884
+ {
1885
+ "epoch": 3.3392,
1886
+ "grad_norm": 6.5843281745910645,
1887
+ "learning_rate": 3.326923076923077e-06,
1888
+ "loss": 0.4023,
1889
+ "mean_token_accuracy": 0.978961855173111,
1890
+ "num_tokens": 25162547.0,
1891
+ "step": 1045
1892
+ },
1893
+ {
1894
+ "epoch": 3.3552,
1895
+ "grad_norm": 6.240104675292969,
1896
+ "learning_rate": 3.3189102564102566e-06,
1897
+ "loss": 0.4017,
1898
+ "mean_token_accuracy": 0.9792039781808853,
1899
+ "num_tokens": 25283158.0,
1900
+ "step": 1050
1901
+ },
1902
+ {
1903
+ "epoch": 3.3712,
1904
+ "grad_norm": 7.1954450607299805,
1905
+ "learning_rate": 3.310897435897436e-06,
1906
+ "loss": 0.4107,
1907
+ "mean_token_accuracy": 0.9788520634174347,
1908
+ "num_tokens": 25403560.0,
1909
+ "step": 1055
1910
+ },
1911
+ {
1912
+ "epoch": 3.3872,
1913
+ "grad_norm": 7.22226095199585,
1914
+ "learning_rate": 3.3028846153846156e-06,
1915
+ "loss": 0.4047,
1916
+ "mean_token_accuracy": 0.9790614232420921,
1917
+ "num_tokens": 25523909.0,
1918
+ "step": 1060
1919
+ },
1920
+ {
1921
+ "epoch": 3.4032,
1922
+ "grad_norm": 7.4581990242004395,
1923
+ "learning_rate": 3.294871794871795e-06,
1924
+ "loss": 0.4106,
1925
+ "mean_token_accuracy": 0.9782844617962837,
1926
+ "num_tokens": 25644369.0,
1927
+ "step": 1065
1928
+ },
1929
+ {
1930
+ "epoch": 3.4192,
1931
+ "grad_norm": 5.827014446258545,
1932
+ "learning_rate": 3.2868589743589745e-06,
1933
+ "loss": 0.4072,
1934
+ "mean_token_accuracy": 0.9784684374928474,
1935
+ "num_tokens": 25764365.0,
1936
+ "step": 1070
1937
+ },
1938
+ {
1939
+ "epoch": 3.4352,
1940
+ "grad_norm": 6.102370738983154,
1941
+ "learning_rate": 3.278846153846154e-06,
1942
+ "loss": 0.4138,
1943
+ "mean_token_accuracy": 0.9786401882767677,
1944
+ "num_tokens": 25885098.0,
1945
+ "step": 1075
1946
+ },
1947
+ {
1948
+ "epoch": 3.4512,
1949
+ "grad_norm": 6.914392471313477,
1950
+ "learning_rate": 3.2708333333333335e-06,
1951
+ "loss": 0.4153,
1952
+ "mean_token_accuracy": 0.9784199267625808,
1953
+ "num_tokens": 26005359.0,
1954
+ "step": 1080
1955
+ },
1956
+ {
1957
+ "epoch": 3.4672,
1958
+ "grad_norm": 7.351043701171875,
1959
+ "learning_rate": 3.262820512820513e-06,
1960
+ "loss": 0.4076,
1961
+ "mean_token_accuracy": 0.9787308901548386,
1962
+ "num_tokens": 26125936.0,
1963
+ "step": 1085
1964
+ },
1965
+ {
1966
+ "epoch": 3.4832,
1967
+ "grad_norm": 8.839035034179688,
1968
+ "learning_rate": 3.254807692307693e-06,
1969
+ "loss": 0.4077,
1970
+ "mean_token_accuracy": 0.9785572066903114,
1971
+ "num_tokens": 26246501.0,
1972
+ "step": 1090
1973
+ },
1974
+ {
1975
+ "epoch": 3.4992,
1976
+ "grad_norm": 10.4901123046875,
1977
+ "learning_rate": 3.2467948717948724e-06,
1978
+ "loss": 0.4133,
1979
+ "mean_token_accuracy": 0.9791041046380997,
1980
+ "num_tokens": 26367561.0,
1981
+ "step": 1095
1982
+ },
1983
+ {
1984
+ "epoch": 3.5152,
1985
+ "grad_norm": 6.502865791320801,
1986
+ "learning_rate": 3.238782051282052e-06,
1987
+ "loss": 0.3964,
1988
+ "mean_token_accuracy": 0.9791050985455513,
1989
+ "num_tokens": 26488473.0,
1990
+ "step": 1100
1991
+ },
1992
+ {
1993
+ "epoch": 3.5312,
1994
+ "grad_norm": 6.487739086151123,
1995
+ "learning_rate": 3.2307692307692313e-06,
1996
+ "loss": 0.4167,
1997
+ "mean_token_accuracy": 0.9781510144472122,
1998
+ "num_tokens": 26609188.0,
1999
+ "step": 1105
2000
+ },
2001
+ {
2002
+ "epoch": 3.5472,
2003
+ "grad_norm": 6.539408206939697,
2004
+ "learning_rate": 3.2227564102564108e-06,
2005
+ "loss": 0.4066,
2006
+ "mean_token_accuracy": 0.9790244609117508,
2007
+ "num_tokens": 26729596.0,
2008
+ "step": 1110
2009
+ },
2010
+ {
2011
+ "epoch": 3.5632,
2012
+ "grad_norm": 7.028800964355469,
2013
+ "learning_rate": 3.2147435897435903e-06,
2014
+ "loss": 0.4041,
2015
+ "mean_token_accuracy": 0.978856410086155,
2016
+ "num_tokens": 26850362.0,
2017
+ "step": 1115
2018
+ },
2019
+ {
2020
+ "epoch": 3.5792,
2021
+ "grad_norm": 6.728778839111328,
2022
+ "learning_rate": 3.2067307692307697e-06,
2023
+ "loss": 0.4171,
2024
+ "mean_token_accuracy": 0.9784936860203743,
2025
+ "num_tokens": 26970795.0,
2026
+ "step": 1120
2027
+ },
2028
+ {
2029
+ "epoch": 3.5952,
2030
+ "grad_norm": 6.5811448097229,
2031
+ "learning_rate": 3.198717948717949e-06,
2032
+ "loss": 0.4097,
2033
+ "mean_token_accuracy": 0.9783563032746315,
2034
+ "num_tokens": 27091441.0,
2035
+ "step": 1125
2036
+ },
2037
+ {
2038
+ "epoch": 3.6112,
2039
+ "grad_norm": 6.12590217590332,
2040
+ "learning_rate": 3.1907051282051283e-06,
2041
+ "loss": 0.4084,
2042
+ "mean_token_accuracy": 0.978937266767025,
2043
+ "num_tokens": 27212323.0,
2044
+ "step": 1130
2045
+ },
2046
+ {
2047
+ "epoch": 3.6272,
2048
+ "grad_norm": 7.032505512237549,
2049
+ "learning_rate": 3.1826923076923077e-06,
2050
+ "loss": 0.4061,
2051
+ "mean_token_accuracy": 0.979265421628952,
2052
+ "num_tokens": 27332713.0,
2053
+ "step": 1135
2054
+ },
2055
+ {
2056
+ "epoch": 3.6432,
2057
+ "grad_norm": 7.316102981567383,
2058
+ "learning_rate": 3.1746794871794872e-06,
2059
+ "loss": 0.401,
2060
+ "mean_token_accuracy": 0.9790761172771454,
2061
+ "num_tokens": 27453206.0,
2062
+ "step": 1140
2063
+ },
2064
+ {
2065
+ "epoch": 3.6592000000000002,
2066
+ "grad_norm": 7.1269307136535645,
2067
+ "learning_rate": 3.1666666666666667e-06,
2068
+ "loss": 0.4163,
2069
+ "mean_token_accuracy": 0.9782899662852287,
2070
+ "num_tokens": 27573750.0,
2071
+ "step": 1145
2072
+ },
2073
+ {
2074
+ "epoch": 3.6752000000000002,
2075
+ "grad_norm": 8.053533554077148,
2076
+ "learning_rate": 3.158653846153846e-06,
2077
+ "loss": 0.4171,
2078
+ "mean_token_accuracy": 0.9785037055611611,
2079
+ "num_tokens": 27694606.0,
2080
+ "step": 1150
2081
+ },
2082
+ {
2083
+ "epoch": 3.6912000000000003,
2084
+ "grad_norm": 6.960573196411133,
2085
+ "learning_rate": 3.1506410256410257e-06,
2086
+ "loss": 0.4033,
2087
+ "mean_token_accuracy": 0.978846101462841,
2088
+ "num_tokens": 27814907.0,
2089
+ "step": 1155
2090
+ },
2091
+ {
2092
+ "epoch": 3.7072000000000003,
2093
+ "grad_norm": 6.491491317749023,
2094
+ "learning_rate": 3.142628205128205e-06,
2095
+ "loss": 0.3962,
2096
+ "mean_token_accuracy": 0.9791050419211388,
2097
+ "num_tokens": 27935802.0,
2098
+ "step": 1160
2099
+ },
2100
+ {
2101
+ "epoch": 3.7232,
2102
+ "grad_norm": 9.257991790771484,
2103
+ "learning_rate": 3.134615384615385e-06,
2104
+ "loss": 0.4013,
2105
+ "mean_token_accuracy": 0.9789151832461357,
2106
+ "num_tokens": 28056199.0,
2107
+ "step": 1165
2108
+ },
2109
+ {
2110
+ "epoch": 3.7392,
2111
+ "grad_norm": 7.759881019592285,
2112
+ "learning_rate": 3.1266025641025645e-06,
2113
+ "loss": 0.4099,
2114
+ "mean_token_accuracy": 0.9784203484654427,
2115
+ "num_tokens": 28176755.0,
2116
+ "step": 1170
2117
+ },
2118
+ {
2119
+ "epoch": 3.7552,
2120
+ "grad_norm": 34.54305648803711,
2121
+ "learning_rate": 3.118589743589744e-06,
2122
+ "loss": 0.4034,
2123
+ "mean_token_accuracy": 0.9788403183221817,
2124
+ "num_tokens": 28297167.0,
2125
+ "step": 1175
2126
+ },
2127
+ {
2128
+ "epoch": 3.7712,
2129
+ "grad_norm": 7.302108287811279,
2130
+ "learning_rate": 3.1105769230769235e-06,
2131
+ "loss": 0.4053,
2132
+ "mean_token_accuracy": 0.9787923589348793,
2133
+ "num_tokens": 28417856.0,
2134
+ "step": 1180
2135
+ },
2136
+ {
2137
+ "epoch": 3.7872,
2138
+ "grad_norm": 9.854865074157715,
2139
+ "learning_rate": 3.102564102564103e-06,
2140
+ "loss": 0.4057,
2141
+ "mean_token_accuracy": 0.9793709293007851,
2142
+ "num_tokens": 28538313.0,
2143
+ "step": 1185
2144
+ },
2145
+ {
2146
+ "epoch": 3.8032,
2147
+ "grad_norm": 6.766449928283691,
2148
+ "learning_rate": 3.0945512820512824e-06,
2149
+ "loss": 0.4118,
2150
+ "mean_token_accuracy": 0.9782621309161186,
2151
+ "num_tokens": 28658534.0,
2152
+ "step": 1190
2153
+ },
2154
+ {
2155
+ "epoch": 3.8192,
2156
+ "grad_norm": 6.60080623626709,
2157
+ "learning_rate": 3.086538461538462e-06,
2158
+ "loss": 0.4095,
2159
+ "mean_token_accuracy": 0.9783975630998611,
2160
+ "num_tokens": 28779166.0,
2161
+ "step": 1195
2162
+ },
2163
+ {
2164
+ "epoch": 3.8352,
2165
+ "grad_norm": 7.6254425048828125,
2166
+ "learning_rate": 3.0785256410256414e-06,
2167
+ "loss": 0.4054,
2168
+ "mean_token_accuracy": 0.9789595857262612,
2169
+ "num_tokens": 28899591.0,
2170
+ "step": 1200
2171
+ },
2172
+ {
2173
+ "epoch": 3.8512,
2174
+ "grad_norm": 9.019999504089355,
2175
+ "learning_rate": 3.070512820512821e-06,
2176
+ "loss": 0.4113,
2177
+ "mean_token_accuracy": 0.9784256905317307,
2178
+ "num_tokens": 29020240.0,
2179
+ "step": 1205
2180
+ },
2181
+ {
2182
+ "epoch": 3.8672,
2183
+ "grad_norm": 5.706897735595703,
2184
+ "learning_rate": 3.0625000000000003e-06,
2185
+ "loss": 0.3971,
2186
+ "mean_token_accuracy": 0.9794070541858673,
2187
+ "num_tokens": 29141114.0,
2188
+ "step": 1210
2189
+ },
2190
+ {
2191
+ "epoch": 3.8832,
2192
+ "grad_norm": 7.2701735496521,
2193
+ "learning_rate": 3.05448717948718e-06,
2194
+ "loss": 0.4136,
2195
+ "mean_token_accuracy": 0.9782675176858902,
2196
+ "num_tokens": 29261774.0,
2197
+ "step": 1215
2198
+ },
2199
+ {
2200
+ "epoch": 3.8992,
2201
+ "grad_norm": 8.50556468963623,
2202
+ "learning_rate": 3.046474358974359e-06,
2203
+ "loss": 0.4079,
2204
+ "mean_token_accuracy": 0.9781965911388397,
2205
+ "num_tokens": 29382264.0,
2206
+ "step": 1220
2207
+ },
2208
+ {
2209
+ "epoch": 3.9152,
2210
+ "grad_norm": 6.016648292541504,
2211
+ "learning_rate": 3.0384615384615383e-06,
2212
+ "loss": 0.4115,
2213
+ "mean_token_accuracy": 0.9785230562090874,
2214
+ "num_tokens": 29502793.0,
2215
+ "step": 1225
2216
+ },
2217
+ {
2218
+ "epoch": 3.9312,
2219
+ "grad_norm": 12.050028800964355,
2220
+ "learning_rate": 3.030448717948718e-06,
2221
+ "loss": 0.4124,
2222
+ "mean_token_accuracy": 0.9786360383033752,
2223
+ "num_tokens": 29623097.0,
2224
+ "step": 1230
2225
+ },
2226
+ {
2227
+ "epoch": 3.9472,
2228
+ "grad_norm": 6.259382247924805,
2229
+ "learning_rate": 3.0224358974358973e-06,
2230
+ "loss": 0.3994,
2231
+ "mean_token_accuracy": 0.9789137870073319,
2232
+ "num_tokens": 29743178.0,
2233
+ "step": 1235
2234
+ },
2235
+ {
2236
+ "epoch": 3.9632,
2237
+ "grad_norm": 7.005880832672119,
2238
+ "learning_rate": 3.014423076923077e-06,
2239
+ "loss": 0.408,
2240
+ "mean_token_accuracy": 0.9786345362663269,
2241
+ "num_tokens": 29863948.0,
2242
+ "step": 1240
2243
+ },
2244
+ {
2245
+ "epoch": 3.9792,
2246
+ "grad_norm": 7.6513590812683105,
2247
+ "learning_rate": 3.0064102564102567e-06,
2248
+ "loss": 0.4121,
2249
+ "mean_token_accuracy": 0.9782826557755471,
2250
+ "num_tokens": 29984240.0,
2251
+ "step": 1245
2252
+ },
2253
+ {
2254
+ "epoch": 3.9952,
2255
+ "grad_norm": 7.343472003936768,
2256
+ "learning_rate": 2.998397435897436e-06,
2257
+ "loss": 0.4061,
2258
+ "mean_token_accuracy": 0.9784524977207184,
2259
+ "num_tokens": 30104676.0,
2260
+ "step": 1250
2261
+ },
2262
+ {
2263
+ "epoch": 4.0096,
2264
+ "grad_norm": 5.229015350341797,
2265
+ "learning_rate": 2.9903846153846156e-06,
2266
+ "loss": 0.2769,
2267
+ "mean_token_accuracy": 0.9851360503170226,
2268
+ "num_tokens": 30213475.0,
2269
+ "step": 1255
2270
+ },
2271
+ {
2272
+ "epoch": 4.0256,
2273
+ "grad_norm": 7.201411724090576,
2274
+ "learning_rate": 2.982371794871795e-06,
2275
+ "loss": 0.2234,
2276
+ "mean_token_accuracy": 0.9883796021342277,
2277
+ "num_tokens": 30333965.0,
2278
+ "step": 1260
2279
+ },
2280
+ {
2281
+ "epoch": 4.0416,
2282
+ "grad_norm": 14.582415580749512,
2283
+ "learning_rate": 2.9743589743589746e-06,
2284
+ "loss": 0.2305,
2285
+ "mean_token_accuracy": 0.9883177742362023,
2286
+ "num_tokens": 30454170.0,
2287
+ "step": 1265
2288
+ },
2289
+ {
2290
+ "epoch": 4.0576,
2291
+ "grad_norm": 11.901610374450684,
2292
+ "learning_rate": 2.966346153846154e-06,
2293
+ "loss": 0.2294,
2294
+ "mean_token_accuracy": 0.9881909266114235,
2295
+ "num_tokens": 30574603.0,
2296
+ "step": 1270
2297
+ },
2298
+ {
2299
+ "epoch": 4.0736,
2300
+ "grad_norm": 10.266992568969727,
2301
+ "learning_rate": 2.9583333333333335e-06,
2302
+ "loss": 0.2202,
2303
+ "mean_token_accuracy": 0.9887023270130157,
2304
+ "num_tokens": 30694953.0,
2305
+ "step": 1275
2306
+ },
2307
+ {
2308
+ "epoch": 4.0896,
2309
+ "grad_norm": 7.3709282875061035,
2310
+ "learning_rate": 2.950320512820513e-06,
2311
+ "loss": 0.2154,
2312
+ "mean_token_accuracy": 0.9889277204871177,
2313
+ "num_tokens": 30815751.0,
2314
+ "step": 1280
2315
+ },
2316
+ {
2317
+ "epoch": 4.1056,
2318
+ "grad_norm": 7.33349609375,
2319
+ "learning_rate": 2.9423076923076925e-06,
2320
+ "loss": 0.2125,
2321
+ "mean_token_accuracy": 0.9889271318912506,
2322
+ "num_tokens": 30936280.0,
2323
+ "step": 1285
2324
+ },
2325
+ {
2326
+ "epoch": 4.1216,
2327
+ "grad_norm": 7.3957133293151855,
2328
+ "learning_rate": 2.934294871794872e-06,
2329
+ "loss": 0.2106,
2330
+ "mean_token_accuracy": 0.9888697654008866,
2331
+ "num_tokens": 31057231.0,
2332
+ "step": 1290
2333
+ },
2334
+ {
2335
+ "epoch": 4.1376,
2336
+ "grad_norm": 6.918285369873047,
2337
+ "learning_rate": 2.926282051282052e-06,
2338
+ "loss": 0.2173,
2339
+ "mean_token_accuracy": 0.9884651228785515,
2340
+ "num_tokens": 31177908.0,
2341
+ "step": 1295
2342
+ },
2343
+ {
2344
+ "epoch": 4.1536,
2345
+ "grad_norm": 6.9746599197387695,
2346
+ "learning_rate": 2.9182692307692313e-06,
2347
+ "loss": 0.204,
2348
+ "mean_token_accuracy": 0.9892497479915618,
2349
+ "num_tokens": 31298201.0,
2350
+ "step": 1300
2351
+ },
2352
+ {
2353
+ "epoch": 4.1696,
2354
+ "grad_norm": 6.991380214691162,
2355
+ "learning_rate": 2.910256410256411e-06,
2356
+ "loss": 0.2134,
2357
+ "mean_token_accuracy": 0.9886619478464127,
2358
+ "num_tokens": 31419062.0,
2359
+ "step": 1305
2360
+ },
2361
+ {
2362
+ "epoch": 4.1856,
2363
+ "grad_norm": 7.398221015930176,
2364
+ "learning_rate": 2.9022435897435903e-06,
2365
+ "loss": 0.2092,
2366
+ "mean_token_accuracy": 0.9889617085456848,
2367
+ "num_tokens": 31539869.0,
2368
+ "step": 1310
2369
+ },
2370
+ {
2371
+ "epoch": 4.2016,
2372
+ "grad_norm": 7.1099934577941895,
2373
+ "learning_rate": 2.8942307692307698e-06,
2374
+ "loss": 0.2164,
2375
+ "mean_token_accuracy": 0.9883017286658287,
2376
+ "num_tokens": 31660538.0,
2377
+ "step": 1315
2378
+ },
2379
+ {
2380
+ "epoch": 4.2176,
2381
+ "grad_norm": 7.760586738586426,
2382
+ "learning_rate": 2.886217948717949e-06,
2383
+ "loss": 0.2137,
2384
+ "mean_token_accuracy": 0.9887675106525421,
2385
+ "num_tokens": 31781316.0,
2386
+ "step": 1320
2387
+ },
2388
+ {
2389
+ "epoch": 4.2336,
2390
+ "grad_norm": 7.182483673095703,
2391
+ "learning_rate": 2.8782051282051283e-06,
2392
+ "loss": 0.2168,
2393
+ "mean_token_accuracy": 0.9888570860028267,
2394
+ "num_tokens": 31901922.0,
2395
+ "step": 1325
2396
+ },
2397
+ {
2398
+ "epoch": 4.2496,
2399
+ "grad_norm": 8.393202781677246,
2400
+ "learning_rate": 2.8701923076923078e-06,
2401
+ "loss": 0.2135,
2402
+ "mean_token_accuracy": 0.9886896923184395,
2403
+ "num_tokens": 32022510.0,
2404
+ "step": 1330
2405
+ },
2406
+ {
2407
+ "epoch": 4.2656,
2408
+ "grad_norm": 8.55794906616211,
2409
+ "learning_rate": 2.8621794871794873e-06,
2410
+ "loss": 0.2218,
2411
+ "mean_token_accuracy": 0.988078698515892,
2412
+ "num_tokens": 32143125.0,
2413
+ "step": 1335
2414
+ },
2415
+ {
2416
+ "epoch": 4.2816,
2417
+ "grad_norm": 8.126118659973145,
2418
+ "learning_rate": 2.8541666666666667e-06,
2419
+ "loss": 0.2201,
2420
+ "mean_token_accuracy": 0.9882885947823524,
2421
+ "num_tokens": 32263620.0,
2422
+ "step": 1340
2423
+ },
2424
+ {
2425
+ "epoch": 4.2976,
2426
+ "grad_norm": 11.159259796142578,
2427
+ "learning_rate": 2.846153846153846e-06,
2428
+ "loss": 0.2064,
2429
+ "mean_token_accuracy": 0.9892195031046868,
2430
+ "num_tokens": 32384569.0,
2431
+ "step": 1345
2432
+ },
2433
+ {
2434
+ "epoch": 4.3136,
2435
+ "grad_norm": 8.103114128112793,
2436
+ "learning_rate": 2.8381410256410257e-06,
2437
+ "loss": 0.2127,
2438
+ "mean_token_accuracy": 0.988749286532402,
2439
+ "num_tokens": 32504592.0,
2440
+ "step": 1350
2441
+ },
2442
+ {
2443
+ "epoch": 4.3296,
2444
+ "grad_norm": 34.2722053527832,
2445
+ "learning_rate": 2.830128205128205e-06,
2446
+ "loss": 0.221,
2447
+ "mean_token_accuracy": 0.9882354348897934,
2448
+ "num_tokens": 32625204.0,
2449
+ "step": 1355
2450
+ },
2451
+ {
2452
+ "epoch": 4.3456,
2453
+ "grad_norm": 9.09987735748291,
2454
+ "learning_rate": 2.8221153846153846e-06,
2455
+ "loss": 0.205,
2456
+ "mean_token_accuracy": 0.9891838625073432,
2457
+ "num_tokens": 32745418.0,
2458
+ "step": 1360
2459
+ },
2460
+ {
2461
+ "epoch": 4.3616,
2462
+ "grad_norm": 8.12652587890625,
2463
+ "learning_rate": 2.8141025641025645e-06,
2464
+ "loss": 0.2096,
2465
+ "mean_token_accuracy": 0.9889617696404457,
2466
+ "num_tokens": 32866254.0,
2467
+ "step": 1365
2468
+ },
2469
+ {
2470
+ "epoch": 4.3776,
2471
+ "grad_norm": 8.117950439453125,
2472
+ "learning_rate": 2.806089743589744e-06,
2473
+ "loss": 0.2216,
2474
+ "mean_token_accuracy": 0.988294905424118,
2475
+ "num_tokens": 32986811.0,
2476
+ "step": 1370
2477
+ },
2478
+ {
2479
+ "epoch": 4.3936,
2480
+ "grad_norm": 6.9630842208862305,
2481
+ "learning_rate": 2.7980769230769235e-06,
2482
+ "loss": 0.2102,
2483
+ "mean_token_accuracy": 0.9889461293816566,
2484
+ "num_tokens": 33107603.0,
2485
+ "step": 1375
2486
+ },
2487
+ {
2488
+ "epoch": 4.4096,
2489
+ "grad_norm": 7.26590633392334,
2490
+ "learning_rate": 2.790064102564103e-06,
2491
+ "loss": 0.2047,
2492
+ "mean_token_accuracy": 0.9889253199100494,
2493
+ "num_tokens": 33228262.0,
2494
+ "step": 1380
2495
+ },
2496
+ {
2497
+ "epoch": 4.4256,
2498
+ "grad_norm": 7.30155086517334,
2499
+ "learning_rate": 2.7820512820512824e-06,
2500
+ "loss": 0.2191,
2501
+ "mean_token_accuracy": 0.9883968114852906,
2502
+ "num_tokens": 33348955.0,
2503
+ "step": 1385
2504
+ },
2505
+ {
2506
+ "epoch": 4.4416,
2507
+ "grad_norm": 8.135958671569824,
2508
+ "learning_rate": 2.774038461538462e-06,
2509
+ "loss": 0.2204,
2510
+ "mean_token_accuracy": 0.9883899509906768,
2511
+ "num_tokens": 33469275.0,
2512
+ "step": 1390
2513
+ },
2514
+ {
2515
+ "epoch": 4.4576,
2516
+ "grad_norm": 7.369213104248047,
2517
+ "learning_rate": 2.7660256410256414e-06,
2518
+ "loss": 0.2303,
2519
+ "mean_token_accuracy": 0.988001236319542,
2520
+ "num_tokens": 33589734.0,
2521
+ "step": 1395
2522
+ },
2523
+ {
2524
+ "epoch": 4.4736,
2525
+ "grad_norm": 7.194396018981934,
2526
+ "learning_rate": 2.758012820512821e-06,
2527
+ "loss": 0.2101,
2528
+ "mean_token_accuracy": 0.9887461483478546,
2529
+ "num_tokens": 33710146.0,
2530
+ "step": 1400
2531
+ },
2532
+ {
2533
+ "epoch": 4.4896,
2534
+ "grad_norm": 9.377034187316895,
2535
+ "learning_rate": 2.7500000000000004e-06,
2536
+ "loss": 0.2276,
2537
+ "mean_token_accuracy": 0.9879311680793762,
2538
+ "num_tokens": 33830687.0,
2539
+ "step": 1405
2540
+ },
2541
+ {
2542
+ "epoch": 4.5056,
2543
+ "grad_norm": 8.591009140014648,
2544
+ "learning_rate": 2.74198717948718e-06,
2545
+ "loss": 0.211,
2546
+ "mean_token_accuracy": 0.9891604006290435,
2547
+ "num_tokens": 33950837.0,
2548
+ "step": 1410
2549
+ },
2550
+ {
2551
+ "epoch": 4.5216,
2552
+ "grad_norm": 8.023900985717773,
2553
+ "learning_rate": 2.733974358974359e-06,
2554
+ "loss": 0.2306,
2555
+ "mean_token_accuracy": 0.9877710357308388,
2556
+ "num_tokens": 34071182.0,
2557
+ "step": 1415
2558
+ },
2559
+ {
2560
+ "epoch": 4.5376,
2561
+ "grad_norm": 9.23966121673584,
2562
+ "learning_rate": 2.7259615384615384e-06,
2563
+ "loss": 0.2109,
2564
+ "mean_token_accuracy": 0.988668218255043,
2565
+ "num_tokens": 34191842.0,
2566
+ "step": 1420
2567
+ },
2568
+ {
2569
+ "epoch": 4.5536,
2570
+ "grad_norm": 7.516361236572266,
2571
+ "learning_rate": 2.717948717948718e-06,
2572
+ "loss": 0.2254,
2573
+ "mean_token_accuracy": 0.9882513582706451,
2574
+ "num_tokens": 34312414.0,
2575
+ "step": 1425
2576
+ },
2577
+ {
2578
+ "epoch": 4.5696,
2579
+ "grad_norm": 8.173714637756348,
2580
+ "learning_rate": 2.7099358974358973e-06,
2581
+ "loss": 0.2211,
2582
+ "mean_token_accuracy": 0.9881450772285462,
2583
+ "num_tokens": 34433301.0,
2584
+ "step": 1430
2585
+ },
2586
+ {
2587
+ "epoch": 4.5856,
2588
+ "grad_norm": 7.787839412689209,
2589
+ "learning_rate": 2.701923076923077e-06,
2590
+ "loss": 0.2246,
2591
+ "mean_token_accuracy": 0.9885869845747948,
2592
+ "num_tokens": 34553956.0,
2593
+ "step": 1435
2594
+ },
2595
+ {
2596
+ "epoch": 4.6016,
2597
+ "grad_norm": 7.567227363586426,
2598
+ "learning_rate": 2.6939102564102567e-06,
2599
+ "loss": 0.2189,
2600
+ "mean_token_accuracy": 0.9887334749102592,
2601
+ "num_tokens": 34675026.0,
2602
+ "step": 1440
2603
+ },
2604
+ {
2605
+ "epoch": 4.6176,
2606
+ "grad_norm": 7.626765251159668,
2607
+ "learning_rate": 2.685897435897436e-06,
2608
+ "loss": 0.2182,
2609
+ "mean_token_accuracy": 0.9884359657764434,
2610
+ "num_tokens": 34795734.0,
2611
+ "step": 1445
2612
+ },
2613
+ {
2614
+ "epoch": 4.6336,
2615
+ "grad_norm": 10.375838279724121,
2616
+ "learning_rate": 2.6778846153846156e-06,
2617
+ "loss": 0.2187,
2618
+ "mean_token_accuracy": 0.9885631337761879,
2619
+ "num_tokens": 34916230.0,
2620
+ "step": 1450
2621
+ },
2622
+ {
2623
+ "epoch": 4.6495999999999995,
2624
+ "grad_norm": 6.558004379272461,
2625
+ "learning_rate": 2.669871794871795e-06,
2626
+ "loss": 0.2079,
2627
+ "mean_token_accuracy": 0.988838629424572,
2628
+ "num_tokens": 35036803.0,
2629
+ "step": 1455
2630
+ },
2631
+ {
2632
+ "epoch": 4.6655999999999995,
2633
+ "grad_norm": 8.621061325073242,
2634
+ "learning_rate": 2.6618589743589746e-06,
2635
+ "loss": 0.22,
2636
+ "mean_token_accuracy": 0.9885150730609894,
2637
+ "num_tokens": 35157473.0,
2638
+ "step": 1460
2639
+ },
2640
+ {
2641
+ "epoch": 4.6815999999999995,
2642
+ "grad_norm": 13.86252212524414,
2643
+ "learning_rate": 2.653846153846154e-06,
2644
+ "loss": 0.2162,
2645
+ "mean_token_accuracy": 0.9883636891841888,
2646
+ "num_tokens": 35277883.0,
2647
+ "step": 1465
2648
+ },
2649
+ {
2650
+ "epoch": 4.6975999999999996,
2651
+ "grad_norm": 7.274498462677002,
2652
+ "learning_rate": 2.6458333333333336e-06,
2653
+ "loss": 0.217,
2654
+ "mean_token_accuracy": 0.9886201068758964,
2655
+ "num_tokens": 35397988.0,
2656
+ "step": 1470
2657
+ },
2658
+ {
2659
+ "epoch": 4.7136,
2660
+ "grad_norm": 14.222506523132324,
2661
+ "learning_rate": 2.637820512820513e-06,
2662
+ "loss": 0.2128,
2663
+ "mean_token_accuracy": 0.9888900205492973,
2664
+ "num_tokens": 35518415.0,
2665
+ "step": 1475
2666
+ },
2667
+ {
2668
+ "epoch": 4.7296,
2669
+ "grad_norm": 9.07939338684082,
2670
+ "learning_rate": 2.6298076923076925e-06,
2671
+ "loss": 0.23,
2672
+ "mean_token_accuracy": 0.9875412806868553,
2673
+ "num_tokens": 35638698.0,
2674
+ "step": 1480
2675
+ },
2676
+ {
2677
+ "epoch": 4.7456,
2678
+ "grad_norm": 7.692012310028076,
2679
+ "learning_rate": 2.621794871794872e-06,
2680
+ "loss": 0.2069,
2681
+ "mean_token_accuracy": 0.9891114875674247,
2682
+ "num_tokens": 35759217.0,
2683
+ "step": 1485
2684
+ },
2685
+ {
2686
+ "epoch": 4.7616,
2687
+ "grad_norm": 7.7223734855651855,
2688
+ "learning_rate": 2.6137820512820515e-06,
2689
+ "loss": 0.2175,
2690
+ "mean_token_accuracy": 0.98785490244627,
2691
+ "num_tokens": 35879524.0,
2692
+ "step": 1490
2693
+ },
2694
+ {
2695
+ "epoch": 4.7776,
2696
+ "grad_norm": 8.727707862854004,
2697
+ "learning_rate": 2.6057692307692314e-06,
2698
+ "loss": 0.2244,
2699
+ "mean_token_accuracy": 0.9881642535328865,
2700
+ "num_tokens": 36000092.0,
2701
+ "step": 1495
2702
+ },
2703
+ {
2704
+ "epoch": 4.7936,
2705
+ "grad_norm": 6.969555854797363,
2706
+ "learning_rate": 2.597756410256411e-06,
2707
+ "loss": 0.2197,
2708
+ "mean_token_accuracy": 0.9882895693182945,
2709
+ "num_tokens": 36120926.0,
2710
+ "step": 1500
2711
+ },
2712
+ {
2713
+ "epoch": 4.8096,
2714
+ "grad_norm": 9.846423149108887,
2715
+ "learning_rate": 2.5897435897435903e-06,
2716
+ "loss": 0.2077,
2717
+ "mean_token_accuracy": 0.9891125798225403,
2718
+ "num_tokens": 36241571.0,
2719
+ "step": 1505
2720
+ },
2721
+ {
2722
+ "epoch": 4.8256,
2723
+ "grad_norm": 8.862875938415527,
2724
+ "learning_rate": 2.58173076923077e-06,
2725
+ "loss": 0.2226,
2726
+ "mean_token_accuracy": 0.9886199355125427,
2727
+ "num_tokens": 36361977.0,
2728
+ "step": 1510
2729
+ },
2730
+ {
2731
+ "epoch": 4.8416,
2732
+ "grad_norm": 7.1445722579956055,
2733
+ "learning_rate": 2.573717948717949e-06,
2734
+ "loss": 0.2157,
2735
+ "mean_token_accuracy": 0.9886227428913117,
2736
+ "num_tokens": 36482358.0,
2737
+ "step": 1515
2738
+ },
2739
+ {
2740
+ "epoch": 4.8576,
2741
+ "grad_norm": 7.2713398933410645,
2742
+ "learning_rate": 2.5657051282051283e-06,
2743
+ "loss": 0.2275,
2744
+ "mean_token_accuracy": 0.988088782131672,
2745
+ "num_tokens": 36603009.0,
2746
+ "step": 1520
2747
+ },
2748
+ {
2749
+ "epoch": 4.8736,
2750
+ "grad_norm": 8.981444358825684,
2751
+ "learning_rate": 2.557692307692308e-06,
2752
+ "loss": 0.2135,
2753
+ "mean_token_accuracy": 0.9885682836174965,
2754
+ "num_tokens": 36723796.0,
2755
+ "step": 1525
2756
+ },
2757
+ {
2758
+ "epoch": 4.8896,
2759
+ "grad_norm": 7.548659801483154,
2760
+ "learning_rate": 2.5496794871794873e-06,
2761
+ "loss": 0.2159,
2762
+ "mean_token_accuracy": 0.9884407356381416,
2763
+ "num_tokens": 36844075.0,
2764
+ "step": 1530
2765
+ },
2766
+ {
2767
+ "epoch": 4.9056,
2768
+ "grad_norm": 8.158010482788086,
2769
+ "learning_rate": 2.5416666666666668e-06,
2770
+ "loss": 0.204,
2771
+ "mean_token_accuracy": 0.9888727724552154,
2772
+ "num_tokens": 36964752.0,
2773
+ "step": 1535
2774
+ },
2775
+ {
2776
+ "epoch": 4.9216,
2777
+ "grad_norm": 7.602233409881592,
2778
+ "learning_rate": 2.5336538461538462e-06,
2779
+ "loss": 0.2178,
2780
+ "mean_token_accuracy": 0.9885821849107742,
2781
+ "num_tokens": 37085216.0,
2782
+ "step": 1540
2783
+ },
2784
+ {
2785
+ "epoch": 4.9376,
2786
+ "grad_norm": 6.433132648468018,
2787
+ "learning_rate": 2.5256410256410257e-06,
2788
+ "loss": 0.2186,
2789
+ "mean_token_accuracy": 0.9884494766592979,
2790
+ "num_tokens": 37205818.0,
2791
+ "step": 1545
2792
+ },
2793
+ {
2794
+ "epoch": 4.9536,
2795
+ "grad_norm": 7.820381164550781,
2796
+ "learning_rate": 2.517628205128205e-06,
2797
+ "loss": 0.212,
2798
+ "mean_token_accuracy": 0.988579323887825,
2799
+ "num_tokens": 37326116.0,
2800
+ "step": 1550
2801
+ },
2802
+ {
2803
+ "epoch": 4.9696,
2804
+ "grad_norm": 7.258703708648682,
2805
+ "learning_rate": 2.5096153846153847e-06,
2806
+ "loss": 0.2111,
2807
+ "mean_token_accuracy": 0.9885433360934257,
2808
+ "num_tokens": 37446736.0,
2809
+ "step": 1555
2810
+ },
2811
+ {
2812
+ "epoch": 4.9856,
2813
+ "grad_norm": 8.752778053283691,
2814
+ "learning_rate": 2.501602564102564e-06,
2815
+ "loss": 0.2208,
2816
+ "mean_token_accuracy": 0.9883719682693481,
2817
+ "num_tokens": 37567430.0,
2818
+ "step": 1560
2819
+ },
2820
+ {
2821
+ "epoch": 5.0,
2822
+ "grad_norm": 8.014328956604004,
2823
+ "learning_rate": 2.4935897435897436e-06,
2824
+ "loss": 0.1813,
2825
+ "mean_token_accuracy": 0.9893870999415716,
2826
+ "num_tokens": 37676145.0,
2827
+ "step": 1565
2828
+ },
2829
+ {
2830
+ "epoch": 5.016,
2831
+ "grad_norm": 5.304056644439697,
2832
+ "learning_rate": 2.4855769230769235e-06,
2833
+ "loss": 0.1075,
2834
+ "mean_token_accuracy": 0.9939878717064857,
2835
+ "num_tokens": 37796724.0,
2836
+ "step": 1570
2837
+ },
2838
+ {
2839
+ "epoch": 5.032,
2840
+ "grad_norm": 15.364027976989746,
2841
+ "learning_rate": 2.477564102564103e-06,
2842
+ "loss": 0.1058,
2843
+ "mean_token_accuracy": 0.9941621392965316,
2844
+ "num_tokens": 37917607.0,
2845
+ "step": 1575
2846
+ },
2847
+ {
2848
+ "epoch": 5.048,
2849
+ "grad_norm": 7.153189182281494,
2850
+ "learning_rate": 2.4695512820512825e-06,
2851
+ "loss": 0.1035,
2852
+ "mean_token_accuracy": 0.9941185429692269,
2853
+ "num_tokens": 38038441.0,
2854
+ "step": 1580
2855
+ },
2856
+ {
2857
+ "epoch": 5.064,
2858
+ "grad_norm": 9.491990089416504,
2859
+ "learning_rate": 2.461538461538462e-06,
2860
+ "loss": 0.106,
2861
+ "mean_token_accuracy": 0.9939621090888977,
2862
+ "num_tokens": 38159169.0,
2863
+ "step": 1585
2864
+ },
2865
+ {
2866
+ "epoch": 5.08,
2867
+ "grad_norm": 13.371603012084961,
2868
+ "learning_rate": 2.453525641025641e-06,
2869
+ "loss": 0.1155,
2870
+ "mean_token_accuracy": 0.9936478078365326,
2871
+ "num_tokens": 38279481.0,
2872
+ "step": 1590
2873
+ },
2874
+ {
2875
+ "epoch": 5.096,
2876
+ "grad_norm": 9.19172477722168,
2877
+ "learning_rate": 2.4455128205128205e-06,
2878
+ "loss": 0.1147,
2879
+ "mean_token_accuracy": 0.9937656357884407,
2880
+ "num_tokens": 38399628.0,
2881
+ "step": 1595
2882
+ },
2883
+ {
2884
+ "epoch": 5.112,
2885
+ "grad_norm": 7.402677059173584,
2886
+ "learning_rate": 2.4375e-06,
2887
+ "loss": 0.1089,
2888
+ "mean_token_accuracy": 0.9937503412365913,
2889
+ "num_tokens": 38519851.0,
2890
+ "step": 1600
2891
+ },
2892
+ {
2893
+ "epoch": 5.128,
2894
+ "grad_norm": 6.590384483337402,
2895
+ "learning_rate": 2.42948717948718e-06,
2896
+ "loss": 0.1088,
2897
+ "mean_token_accuracy": 0.9938607648015022,
2898
+ "num_tokens": 38640721.0,
2899
+ "step": 1605
2900
+ },
2901
+ {
2902
+ "epoch": 5.144,
2903
+ "grad_norm": 7.305964946746826,
2904
+ "learning_rate": 2.4214743589743593e-06,
2905
+ "loss": 0.1032,
2906
+ "mean_token_accuracy": 0.9939986765384674,
2907
+ "num_tokens": 38761318.0,
2908
+ "step": 1610
2909
+ },
2910
+ {
2911
+ "epoch": 5.16,
2912
+ "grad_norm": 14.2820463180542,
2913
+ "learning_rate": 2.413461538461539e-06,
2914
+ "loss": 0.1052,
2915
+ "mean_token_accuracy": 0.9939956277608871,
2916
+ "num_tokens": 38882080.0,
2917
+ "step": 1615
2918
+ },
2919
+ {
2920
+ "epoch": 5.176,
2921
+ "grad_norm": 8.019156455993652,
2922
+ "learning_rate": 2.4054487179487183e-06,
2923
+ "loss": 0.1092,
2924
+ "mean_token_accuracy": 0.9936814203858375,
2925
+ "num_tokens": 39002652.0,
2926
+ "step": 1620
2927
+ },
2928
+ {
2929
+ "epoch": 5.192,
2930
+ "grad_norm": 8.981114387512207,
2931
+ "learning_rate": 2.3974358974358978e-06,
2932
+ "loss": 0.1061,
2933
+ "mean_token_accuracy": 0.9938769549131393,
2934
+ "num_tokens": 39123134.0,
2935
+ "step": 1625
2936
+ },
2937
+ {
2938
+ "epoch": 5.208,
2939
+ "grad_norm": 8.067747116088867,
2940
+ "learning_rate": 2.3894230769230772e-06,
2941
+ "loss": 0.113,
2942
+ "mean_token_accuracy": 0.9936137914657592,
2943
+ "num_tokens": 39243893.0,
2944
+ "step": 1630
2945
+ },
2946
+ {
2947
+ "epoch": 5.224,
2948
+ "grad_norm": 8.61947250366211,
2949
+ "learning_rate": 2.3814102564102563e-06,
2950
+ "loss": 0.1069,
2951
+ "mean_token_accuracy": 0.9938591852784157,
2952
+ "num_tokens": 39364264.0,
2953
+ "step": 1635
2954
+ },
2955
+ {
2956
+ "epoch": 5.24,
2957
+ "grad_norm": 7.487087726593018,
2958
+ "learning_rate": 2.373397435897436e-06,
2959
+ "loss": 0.111,
2960
+ "mean_token_accuracy": 0.9937029749155044,
2961
+ "num_tokens": 39484962.0,
2962
+ "step": 1640
2963
+ },
2964
+ {
2965
+ "epoch": 5.256,
2966
+ "grad_norm": 6.646207332611084,
2967
+ "learning_rate": 2.3653846153846157e-06,
2968
+ "loss": 0.1098,
2969
+ "mean_token_accuracy": 0.9938261553645134,
2970
+ "num_tokens": 39605354.0,
2971
+ "step": 1645
2972
+ },
2973
+ {
2974
+ "epoch": 5.272,
2975
+ "grad_norm": 6.444827079772949,
2976
+ "learning_rate": 2.357371794871795e-06,
2977
+ "loss": 0.1123,
2978
+ "mean_token_accuracy": 0.993807764351368,
2979
+ "num_tokens": 39725860.0,
2980
+ "step": 1650
2981
+ },
2982
+ {
2983
+ "epoch": 5.288,
2984
+ "grad_norm": 6.515505313873291,
2985
+ "learning_rate": 2.3493589743589746e-06,
2986
+ "loss": 0.1102,
2987
+ "mean_token_accuracy": 0.9938962414860726,
2988
+ "num_tokens": 39846778.0,
2989
+ "step": 1655
2990
+ },
2991
+ {
2992
+ "epoch": 5.304,
2993
+ "grad_norm": 6.887137413024902,
2994
+ "learning_rate": 2.341346153846154e-06,
2995
+ "loss": 0.1033,
2996
+ "mean_token_accuracy": 0.9940015017986298,
2997
+ "num_tokens": 39967751.0,
2998
+ "step": 1660
2999
+ },
3000
+ {
3001
+ "epoch": 5.32,
3002
+ "grad_norm": 6.330623626708984,
3003
+ "learning_rate": 2.3333333333333336e-06,
3004
+ "loss": 0.1082,
3005
+ "mean_token_accuracy": 0.9937430843710899,
3006
+ "num_tokens": 40088336.0,
3007
+ "step": 1665
3008
+ },
3009
+ {
3010
+ "epoch": 5.336,
3011
+ "grad_norm": 7.775038242340088,
3012
+ "learning_rate": 2.325320512820513e-06,
3013
+ "loss": 0.1178,
3014
+ "mean_token_accuracy": 0.9936518028378487,
3015
+ "num_tokens": 40208957.0,
3016
+ "step": 1670
3017
+ },
3018
+ {
3019
+ "epoch": 5.352,
3020
+ "grad_norm": 7.103325366973877,
3021
+ "learning_rate": 2.3173076923076925e-06,
3022
+ "loss": 0.1139,
3023
+ "mean_token_accuracy": 0.9938741058111191,
3024
+ "num_tokens": 40329818.0,
3025
+ "step": 1675
3026
+ },
3027
+ {
3028
+ "epoch": 5.368,
3029
+ "grad_norm": 8.063414573669434,
3030
+ "learning_rate": 2.309294871794872e-06,
3031
+ "loss": 0.113,
3032
+ "mean_token_accuracy": 0.9936489358544349,
3033
+ "num_tokens": 40450404.0,
3034
+ "step": 1680
3035
+ },
3036
+ {
3037
+ "epoch": 5.384,
3038
+ "grad_norm": 6.062953472137451,
3039
+ "learning_rate": 2.3012820512820515e-06,
3040
+ "loss": 0.1144,
3041
+ "mean_token_accuracy": 0.9938038796186447,
3042
+ "num_tokens": 40570649.0,
3043
+ "step": 1685
3044
+ },
3045
+ {
3046
+ "epoch": 5.4,
3047
+ "grad_norm": 7.863674640655518,
3048
+ "learning_rate": 2.293269230769231e-06,
3049
+ "loss": 0.1094,
3050
+ "mean_token_accuracy": 0.9936837628483772,
3051
+ "num_tokens": 40691020.0,
3052
+ "step": 1690
3053
+ },
3054
+ {
3055
+ "epoch": 5.416,
3056
+ "grad_norm": 7.539632320404053,
3057
+ "learning_rate": 2.2852564102564104e-06,
3058
+ "loss": 0.1072,
3059
+ "mean_token_accuracy": 0.9939090713858605,
3060
+ "num_tokens": 40811323.0,
3061
+ "step": 1695
3062
+ },
3063
+ {
3064
+ "epoch": 5.432,
3065
+ "grad_norm": 8.533873558044434,
3066
+ "learning_rate": 2.27724358974359e-06,
3067
+ "loss": 0.1179,
3068
+ "mean_token_accuracy": 0.9935754910111427,
3069
+ "num_tokens": 40932157.0,
3070
+ "step": 1700
3071
+ },
3072
+ {
3073
+ "epoch": 5.448,
3074
+ "grad_norm": 9.813429832458496,
3075
+ "learning_rate": 2.2692307692307694e-06,
3076
+ "loss": 0.1059,
3077
+ "mean_token_accuracy": 0.99373127669096,
3078
+ "num_tokens": 41052762.0,
3079
+ "step": 1705
3080
+ },
3081
+ {
3082
+ "epoch": 5.464,
3083
+ "grad_norm": 6.563747882843018,
3084
+ "learning_rate": 2.261217948717949e-06,
3085
+ "loss": 0.1073,
3086
+ "mean_token_accuracy": 0.993916817009449,
3087
+ "num_tokens": 41173561.0,
3088
+ "step": 1710
3089
+ },
3090
+ {
3091
+ "epoch": 5.48,
3092
+ "grad_norm": 7.464493751525879,
3093
+ "learning_rate": 2.2532051282051284e-06,
3094
+ "loss": 0.1113,
3095
+ "mean_token_accuracy": 0.9936030000448227,
3096
+ "num_tokens": 41293920.0,
3097
+ "step": 1715
3098
+ },
3099
+ {
3100
+ "epoch": 5.496,
3101
+ "grad_norm": 8.408556938171387,
3102
+ "learning_rate": 2.245192307692308e-06,
3103
+ "loss": 0.1067,
3104
+ "mean_token_accuracy": 0.9939841747283935,
3105
+ "num_tokens": 41414591.0,
3106
+ "step": 1720
3107
+ },
3108
+ {
3109
+ "epoch": 5.5120000000000005,
3110
+ "grad_norm": 8.727843284606934,
3111
+ "learning_rate": 2.2371794871794873e-06,
3112
+ "loss": 0.1203,
3113
+ "mean_token_accuracy": 0.9933707192540169,
3114
+ "num_tokens": 41535343.0,
3115
+ "step": 1725
3116
+ },
3117
+ {
3118
+ "epoch": 5.5280000000000005,
3119
+ "grad_norm": 7.322857856750488,
3120
+ "learning_rate": 2.2291666666666668e-06,
3121
+ "loss": 0.1074,
3122
+ "mean_token_accuracy": 0.993928337097168,
3123
+ "num_tokens": 41655670.0,
3124
+ "step": 1730
3125
+ },
3126
+ {
3127
+ "epoch": 5.5440000000000005,
3128
+ "grad_norm": 6.6553120613098145,
3129
+ "learning_rate": 2.2211538461538463e-06,
3130
+ "loss": 0.1045,
3131
+ "mean_token_accuracy": 0.9940162122249603,
3132
+ "num_tokens": 41776253.0,
3133
+ "step": 1735
3134
+ },
3135
+ {
3136
+ "epoch": 5.5600000000000005,
3137
+ "grad_norm": 6.968092918395996,
3138
+ "learning_rate": 2.2131410256410257e-06,
3139
+ "loss": 0.1093,
3140
+ "mean_token_accuracy": 0.993864431977272,
3141
+ "num_tokens": 41897335.0,
3142
+ "step": 1740
3143
+ },
3144
+ {
3145
+ "epoch": 5.576,
3146
+ "grad_norm": 11.733102798461914,
3147
+ "learning_rate": 2.2051282051282052e-06,
3148
+ "loss": 0.1134,
3149
+ "mean_token_accuracy": 0.9935013368725777,
3150
+ "num_tokens": 42017501.0,
3151
+ "step": 1745
3152
+ },
3153
+ {
3154
+ "epoch": 5.592,
3155
+ "grad_norm": 6.343779563903809,
3156
+ "learning_rate": 2.1971153846153847e-06,
3157
+ "loss": 0.1122,
3158
+ "mean_token_accuracy": 0.9935531497001648,
3159
+ "num_tokens": 42138173.0,
3160
+ "step": 1750
3161
+ },
3162
+ {
3163
+ "epoch": 5.608,
3164
+ "grad_norm": 5.965951919555664,
3165
+ "learning_rate": 2.189102564102564e-06,
3166
+ "loss": 0.1046,
3167
+ "mean_token_accuracy": 0.9939899712800979,
3168
+ "num_tokens": 42258409.0,
3169
+ "step": 1755
3170
+ },
3171
+ {
3172
+ "epoch": 5.624,
3173
+ "grad_norm": 5.9753499031066895,
3174
+ "learning_rate": 2.1810897435897436e-06,
3175
+ "loss": 0.1031,
3176
+ "mean_token_accuracy": 0.9943725138902664,
3177
+ "num_tokens": 42379361.0,
3178
+ "step": 1760
3179
+ },
3180
+ {
3181
+ "epoch": 5.64,
3182
+ "grad_norm": 9.91765022277832,
3183
+ "learning_rate": 2.173076923076923e-06,
3184
+ "loss": 0.11,
3185
+ "mean_token_accuracy": 0.9938978731632233,
3186
+ "num_tokens": 42499620.0,
3187
+ "step": 1765
3188
+ },
3189
+ {
3190
+ "epoch": 5.656,
3191
+ "grad_norm": 8.283162117004395,
3192
+ "learning_rate": 2.165064102564103e-06,
3193
+ "loss": 0.108,
3194
+ "mean_token_accuracy": 0.9938284069299698,
3195
+ "num_tokens": 42619891.0,
3196
+ "step": 1770
3197
+ },
3198
+ {
3199
+ "epoch": 5.672,
3200
+ "grad_norm": 7.9267191886901855,
3201
+ "learning_rate": 2.1570512820512825e-06,
3202
+ "loss": 0.112,
3203
+ "mean_token_accuracy": 0.9938542306423187,
3204
+ "num_tokens": 42740065.0,
3205
+ "step": 1775
3206
+ },
3207
+ {
3208
+ "epoch": 5.688,
3209
+ "grad_norm": 8.388506889343262,
3210
+ "learning_rate": 2.149038461538462e-06,
3211
+ "loss": 0.1128,
3212
+ "mean_token_accuracy": 0.9935542583465576,
3213
+ "num_tokens": 42860734.0,
3214
+ "step": 1780
3215
+ },
3216
+ {
3217
+ "epoch": 5.704,
3218
+ "grad_norm": 8.313148498535156,
3219
+ "learning_rate": 2.141025641025641e-06,
3220
+ "loss": 0.1017,
3221
+ "mean_token_accuracy": 0.9939746722579003,
3222
+ "num_tokens": 42981569.0,
3223
+ "step": 1785
3224
+ },
3225
+ {
3226
+ "epoch": 5.72,
3227
+ "grad_norm": 7.273435115814209,
3228
+ "learning_rate": 2.1330128205128205e-06,
3229
+ "loss": 0.1049,
3230
+ "mean_token_accuracy": 0.9940645769238472,
3231
+ "num_tokens": 43101863.0,
3232
+ "step": 1790
3233
+ },
3234
+ {
3235
+ "epoch": 5.736,
3236
+ "grad_norm": 10.835204124450684,
3237
+ "learning_rate": 2.125e-06,
3238
+ "loss": 0.101,
3239
+ "mean_token_accuracy": 0.9940232053399086,
3240
+ "num_tokens": 43222011.0,
3241
+ "step": 1795
3242
+ },
3243
+ {
3244
+ "epoch": 5.752,
3245
+ "grad_norm": 9.065613746643066,
3246
+ "learning_rate": 2.1169871794871795e-06,
3247
+ "loss": 0.1058,
3248
+ "mean_token_accuracy": 0.9939394310116768,
3249
+ "num_tokens": 43342481.0,
3250
+ "step": 1800
3251
+ }
3252
+ ],
3253
+ "logging_steps": 5,
3254
+ "max_steps": 3120,
3255
+ "num_input_tokens_seen": 0,
3256
+ "num_train_epochs": 10,
3257
+ "save_steps": 200,
3258
+ "stateful_callbacks": {
3259
+ "TrainerControl": {
3260
+ "args": {
3261
+ "should_epoch_stop": false,
3262
+ "should_evaluate": false,
3263
+ "should_log": false,
3264
+ "should_save": true,
3265
+ "should_training_stop": false
3266
+ },
3267
+ "attributes": {}
3268
+ }
3269
+ },
3270
+ "total_flos": 610570229121024.0,
3271
+ "train_batch_size": 1,
3272
+ "trial_name": null,
3273
+ "trial_params": null
3274
+ }
models/gemma_knows_coorsinates_6/pytorch_model-00010-of-00022.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ed04e4445c2324bbfd3b821c433ad6b2b333c4b9ca67b1f21783906d41bd946
3
+ size 1862270976