jeromeramos commited on
Commit
b4e9ddd
·
verified ·
1 Parent(s): e963081

Model save

Browse files
README.md CHANGED
@@ -27,7 +27,7 @@ print(output["generated_text"])
27
 
28
  ## Training procedure
29
 
30
- [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/jerome-ramos-20/huggingface/runs/y53jkfs4)
31
 
32
 
33
  This model was trained with SFT.
@@ -35,8 +35,8 @@ This model was trained with SFT.
35
  ### Framework versions
36
 
37
  - TRL: 0.14.0
38
- - Transformers: 4.48.2
39
- - Pytorch: 2.5.1
40
  - Datasets: 3.0.1
41
  - Tokenizers: 0.21.0
42
 
 
27
 
28
  ## Training procedure
29
 
30
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/jerome-ramos-20/huggingface/runs/pn3887gl)
31
 
32
 
33
  This model was trained with SFT.
 
35
  ### Framework versions
36
 
37
  - TRL: 0.14.0
38
+ - Transformers: 4.51.3
39
+ - Pytorch: 2.6.0
40
  - Datasets: 3.0.1
41
  - Tokenizers: 0.21.0
42
 
all_results.json CHANGED
@@ -1,14 +1,14 @@
1
  {
2
- "epoch": 1.9986168741355463,
3
- "eval_loss": 0.6898206472396851,
4
- "eval_runtime": 51.1318,
5
  "eval_samples": 2071,
6
- "eval_samples_per_second": 90.159,
7
- "eval_steps_per_second": 2.836,
8
- "total_flos": 3.364677087628624e+18,
9
- "train_loss": 0.7622144496341822,
10
- "train_runtime": 4614.5072,
11
- "train_samples": 46269,
12
- "train_samples_per_second": 20.054,
13
- "train_steps_per_second": 0.156
14
  }
 
1
  {
2
+ "epoch": 1.9982692973347178,
3
+ "eval_loss": 8.507396697998047,
4
+ "eval_runtime": 55.108,
5
  "eval_samples": 2071,
6
+ "eval_samples_per_second": 37.581,
7
+ "eval_steps_per_second": 2.359,
8
+ "total_flos": 4.75579504107769e+18,
9
+ "train_loss": 0.8681613599824773,
10
+ "train_runtime": 10264.6857,
11
+ "train_samples": 46218,
12
+ "train_samples_per_second": 9.005,
13
+ "train_steps_per_second": 0.141
14
  }
config.json CHANGED
@@ -1,5 +1,4 @@
1
  {
2
- "_name_or_path": "meta-llama/Llama-3.1-8B",
3
  "architectures": [
4
  "LlamaForCausalLM"
5
  ],
@@ -31,7 +30,7 @@
31
  "rope_theta": 500000.0,
32
  "tie_word_embeddings": false,
33
  "torch_dtype": "bfloat16",
34
- "transformers_version": "4.48.2",
35
  "use_cache": false,
36
  "vocab_size": 128320
37
  }
 
1
  {
 
2
  "architectures": [
3
  "LlamaForCausalLM"
4
  ],
 
30
  "rope_theta": 500000.0,
31
  "tie_word_embeddings": false,
32
  "torch_dtype": "bfloat16",
33
+ "transformers_version": "4.51.3",
34
  "use_cache": false,
35
  "vocab_size": 128320
36
  }
eval_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 1.9986168741355463,
3
- "eval_loss": 0.6898206472396851,
4
- "eval_runtime": 51.1318,
5
  "eval_samples": 2071,
6
- "eval_samples_per_second": 90.159,
7
- "eval_steps_per_second": 2.836
8
  }
 
1
  {
2
+ "epoch": 0.013831258644536652,
3
+ "eval_loss": 8.507396697998047,
4
+ "eval_runtime": 55.108,
5
  "eval_samples": 2071,
6
+ "eval_samples_per_second": 37.581,
7
+ "eval_steps_per_second": 2.359
8
  }
generation_config.json CHANGED
@@ -5,5 +5,5 @@
5
  "eos_token_id": 128001,
6
  "temperature": 0.6,
7
  "top_p": 0.9,
8
- "transformers_version": "4.48.2"
9
  }
 
5
  "eos_token_id": 128001,
6
  "temperature": 0.6,
7
  "top_p": 0.9,
8
+ "transformers_version": "4.51.3"
9
  }
model-00001-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:473fce18d250c18a47a350a533e0dd77b59518a960c722628b7eefa5b9884132
3
  size 4977222960
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a9716cb9ed1021b262dda83f511bb81b49c81df64cbb4e3fa03412232d08191
3
  size 4977222960
model-00002-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6b5a874d269778c72e79f59151bef603cf142d8c8224f7678b0ec2edc10dfd44
3
  size 4999802720
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ab8e6a1d43b0cf366da98da943b8194cde43fa87d26b5b2a8d1513ab996cd31
3
  size 4999802720
model-00003-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d058b976bfd82b1b6afa50b85c104a77791360dd7423bb0a8b70d93602dcff1e
3
  size 4915916176
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c06e00ae114081cd5c0c68fd45ecf873effb0464001f9cdef21da5d5c4010144
3
  size 4915916176
model-00004-of-00004.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8613511f5a50b99d52a9d237182c70ff8f7729acbbdadc631e83795f1d83e3c4
3
  size 1168663096
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9cde97b4eff640284f9646fbbd145b29f26dde3b97b6685d09e7cc446c61fa52
3
  size 1168663096
runs/May03_00-37-47_w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7/events.out.tfevents.1746232925.w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7.20497.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9404605c4cc16f74134ebec56911bface513b9dbcc1819d7b269e783d1c91ac
3
+ size 6769
runs/May03_00-37-47_w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7/events.out.tfevents.1746233072.w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7.20497.1 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2805a11078ff7cfccc7d24bf289b4ee27c8eded30ba0b4d31f0d2cac47f70a16
3
+ size 354
runs/May03_00-46-52_w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7/events.out.tfevents.1746233365.w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7.22790.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ff0d4ec46fc12b923b5f782678020c2a34b12f97c48fd5b6035bd1a905a28c2f
3
+ size 6769
runs/May03_02-16-51_w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7/events.out.tfevents.1746238878.w-jerom-inter-play-sim-94c6890b9ccf44ea86f033a3db8a5dbd-86twgc7.53998.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d4f6bdc2c7d0bc988e32a0febe612795489a7ac05d518cebce867012995d0e3
3
+ size 67539
special_tokens_map.json CHANGED
@@ -15,14 +15,14 @@
15
  "single_word": false
16
  },
17
  {
18
- "content": "<answer>",
19
  "lstrip": false,
20
  "normalized": false,
21
  "rstrip": false,
22
  "single_word": false
23
  },
24
  {
25
- "content": "</answer>",
26
  "lstrip": false,
27
  "normalized": false,
28
  "rstrip": false,
@@ -34,6 +34,41 @@
34
  "normalized": false,
35
  "rstrip": false,
36
  "single_word": false
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37
  }
38
  ],
39
  "bos_token": {
 
15
  "single_word": false
16
  },
17
  {
18
+ "content": "<action>",
19
  "lstrip": false,
20
  "normalized": false,
21
  "rstrip": false,
22
  "single_word": false
23
  },
24
  {
25
+ "content": "</action>",
26
  "lstrip": false,
27
  "normalized": false,
28
  "rstrip": false,
 
34
  "normalized": false,
35
  "rstrip": false,
36
  "single_word": false
37
+ },
38
+ {
39
+ "content": "<recommend>",
40
+ "lstrip": false,
41
+ "normalized": false,
42
+ "rstrip": false,
43
+ "single_word": false
44
+ },
45
+ {
46
+ "content": "<feedback>",
47
+ "lstrip": false,
48
+ "normalized": false,
49
+ "rstrip": false,
50
+ "single_word": false
51
+ },
52
+ {
53
+ "content": "<accept>",
54
+ "lstrip": false,
55
+ "normalized": false,
56
+ "rstrip": false,
57
+ "single_word": false
58
+ },
59
+ {
60
+ "content": "<movie_title>",
61
+ "lstrip": false,
62
+ "normalized": false,
63
+ "rstrip": false,
64
+ "single_word": false
65
+ },
66
+ {
67
+ "content": "</movie_title>",
68
+ "lstrip": false,
69
+ "normalized": false,
70
+ "rstrip": false,
71
+ "single_word": false
72
  }
73
  ],
74
  "bos_token": {
tokenizer.json CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b5ea5afcc70a5f73f9b545a5940b211fd23e2acd4d895a3ebc3144ca348a4633
3
- size 17211228
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e1119d769c8ea690273a0f56402ed5b2a88e4babc23d026a23bbdc11ad4a8231
3
+ size 17212268
tokenizer_config.json CHANGED
@@ -2081,7 +2081,7 @@
2081
  "special": true
2082
  },
2083
  "128260": {
2084
- "content": "<answer>",
2085
  "lstrip": false,
2086
  "normalized": false,
2087
  "rstrip": false,
@@ -2089,7 +2089,7 @@
2089
  "special": true
2090
  },
2091
  "128261": {
2092
- "content": "</answer>",
2093
  "lstrip": false,
2094
  "normalized": false,
2095
  "rstrip": false,
@@ -2103,14 +2103,59 @@
2103
  "rstrip": false,
2104
  "single_word": false,
2105
  "special": true
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2106
  }
2107
  },
2108
  "additional_special_tokens": [
2109
  "<response>",
2110
  "</response>",
2111
- "<answer>",
2112
- "</answer>",
2113
- "<inquire>"
 
 
 
 
 
2114
  ],
2115
  "bos_token": "<|im_start|>",
2116
  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
@@ -2123,5 +2168,5 @@
2123
  ],
2124
  "model_max_length": 131072,
2125
  "pad_token": "<|finetune_right_pad_id|>",
2126
- "tokenizer_class": "PreTrainedTokenizerFast"
2127
  }
 
2081
  "special": true
2082
  },
2083
  "128260": {
2084
+ "content": "<action>",
2085
  "lstrip": false,
2086
  "normalized": false,
2087
  "rstrip": false,
 
2089
  "special": true
2090
  },
2091
  "128261": {
2092
+ "content": "</action>",
2093
  "lstrip": false,
2094
  "normalized": false,
2095
  "rstrip": false,
 
2103
  "rstrip": false,
2104
  "single_word": false,
2105
  "special": true
2106
+ },
2107
+ "128263": {
2108
+ "content": "<recommend>",
2109
+ "lstrip": false,
2110
+ "normalized": false,
2111
+ "rstrip": false,
2112
+ "single_word": false,
2113
+ "special": true
2114
+ },
2115
+ "128264": {
2116
+ "content": "<feedback>",
2117
+ "lstrip": false,
2118
+ "normalized": false,
2119
+ "rstrip": false,
2120
+ "single_word": false,
2121
+ "special": true
2122
+ },
2123
+ "128265": {
2124
+ "content": "<accept>",
2125
+ "lstrip": false,
2126
+ "normalized": false,
2127
+ "rstrip": false,
2128
+ "single_word": false,
2129
+ "special": true
2130
+ },
2131
+ "128266": {
2132
+ "content": "<movie_title>",
2133
+ "lstrip": false,
2134
+ "normalized": false,
2135
+ "rstrip": false,
2136
+ "single_word": false,
2137
+ "special": true
2138
+ },
2139
+ "128267": {
2140
+ "content": "</movie_title>",
2141
+ "lstrip": false,
2142
+ "normalized": false,
2143
+ "rstrip": false,
2144
+ "single_word": false,
2145
+ "special": true
2146
  }
2147
  },
2148
  "additional_special_tokens": [
2149
  "<response>",
2150
  "</response>",
2151
+ "<action>",
2152
+ "</action>",
2153
+ "<inquire>",
2154
+ "<recommend>",
2155
+ "<feedback>",
2156
+ "<accept>",
2157
+ "<movie_title>",
2158
+ "</movie_title>"
2159
  ],
2160
  "bos_token": "<|im_start|>",
2161
  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
 
2168
  ],
2169
  "model_max_length": 131072,
2170
  "pad_token": "<|finetune_right_pad_id|>",
2171
+ "tokenizer_class": "PreTrainedTokenizer"
2172
  }
train_results.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
- "epoch": 1.9986168741355463,
3
- "total_flos": 3.364677087628624e+18,
4
- "train_loss": 0.7622144496341822,
5
- "train_runtime": 4614.5072,
6
- "train_samples": 46269,
7
- "train_samples_per_second": 20.054,
8
- "train_steps_per_second": 0.156
9
  }
 
1
  {
2
+ "epoch": 1.9982692973347178,
3
+ "total_flos": 4.75579504107769e+18,
4
+ "train_loss": 0.8681613599824773,
5
+ "train_runtime": 10264.6857,
6
+ "train_samples": 46218,
7
+ "train_samples_per_second": 9.005,
8
+ "train_steps_per_second": 0.141
9
  }
trainer_state.json CHANGED
@@ -1,1056 +1,2065 @@
1
  {
 
2
  "best_metric": null,
3
  "best_model_checkpoint": null,
4
- "epoch": 1.9986168741355463,
5
  "eval_steps": 500,
6
- "global_step": 722,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
10
  "log_history": [
11
  {
12
- "epoch": 0.0027662517289073307,
13
- "grad_norm": 20.252717971801758,
14
- "learning_rate": 2.7397260273972604e-06,
15
- "loss": 2.8393,
16
  "step": 1
17
  },
18
  {
19
- "epoch": 0.013831258644536652,
20
- "grad_norm": 3.2012741565704346,
21
- "learning_rate": 1.3698630136986302e-05,
22
- "loss": 2.5142,
23
  "step": 5
24
  },
25
  {
26
- "epoch": 0.027662517289073305,
27
- "grad_norm": 4.701728820800781,
28
- "learning_rate": 2.7397260273972603e-05,
29
- "loss": 2.0682,
30
  "step": 10
31
  },
32
  {
33
- "epoch": 0.04149377593360996,
34
- "grad_norm": 1.333525538444519,
35
- "learning_rate": 4.1095890410958905e-05,
36
- "loss": 1.8592,
37
  "step": 15
38
  },
39
  {
40
- "epoch": 0.05532503457814661,
41
- "grad_norm": 1.3727704286575317,
42
- "learning_rate": 5.479452054794521e-05,
43
- "loss": 1.611,
44
  "step": 20
45
  },
46
  {
47
- "epoch": 0.06915629322268327,
48
- "grad_norm": 1.287765622138977,
49
- "learning_rate": 6.84931506849315e-05,
50
- "loss": 1.2818,
51
  "step": 25
52
  },
53
  {
54
- "epoch": 0.08298755186721991,
55
- "grad_norm": 0.443893164396286,
56
- "learning_rate": 8.219178082191781e-05,
57
- "loss": 1.0442,
58
  "step": 30
59
  },
60
  {
61
- "epoch": 0.09681881051175657,
62
- "grad_norm": 1.4994786977767944,
63
- "learning_rate": 9.58904109589041e-05,
64
- "loss": 1.0328,
65
  "step": 35
66
  },
67
  {
68
- "epoch": 0.11065006915629322,
69
- "grad_norm": 3.60703444480896,
70
- "learning_rate": 0.00010958904109589041,
71
- "loss": 1.264,
72
  "step": 40
73
  },
74
  {
75
- "epoch": 0.12448132780082988,
76
- "grad_norm": 2.5533716678619385,
77
- "learning_rate": 0.0001232876712328767,
78
- "loss": 1.5079,
79
  "step": 45
80
  },
81
  {
82
- "epoch": 0.13831258644536654,
83
- "grad_norm": 25.258333206176758,
84
- "learning_rate": 0.000136986301369863,
85
- "loss": 1.3297,
86
  "step": 50
87
  },
88
  {
89
- "epoch": 0.15214384508990317,
90
- "grad_norm": 2.7347092628479004,
91
- "learning_rate": 0.00015068493150684933,
92
- "loss": 1.4471,
93
  "step": 55
94
  },
95
  {
96
- "epoch": 0.16597510373443983,
97
- "grad_norm": 3.742867946624756,
98
- "learning_rate": 0.00016438356164383562,
99
- "loss": 1.2809,
100
  "step": 60
101
  },
102
  {
103
- "epoch": 0.1798063623789765,
104
- "grad_norm": 3.686124563217163,
105
- "learning_rate": 0.00017808219178082192,
106
- "loss": 1.6302,
107
  "step": 65
108
  },
109
  {
110
- "epoch": 0.19363762102351315,
111
- "grad_norm": 8.028692245483398,
112
- "learning_rate": 0.0001917808219178082,
113
- "loss": 2.139,
114
  "step": 70
115
  },
116
  {
117
- "epoch": 0.2074688796680498,
118
- "grad_norm": 4.435426712036133,
119
- "learning_rate": 0.00019999531362588743,
120
- "loss": 1.4489,
121
  "step": 75
122
  },
123
  {
124
- "epoch": 0.22130013831258644,
125
- "grad_norm": 2.330904483795166,
126
- "learning_rate": 0.00019994259696141126,
127
- "loss": 1.4819,
128
  "step": 80
129
  },
130
  {
131
- "epoch": 0.2351313969571231,
132
- "grad_norm": 1.6414345502853394,
133
- "learning_rate": 0.0001998313366477513,
134
- "loss": 1.3125,
135
  "step": 85
136
  },
137
  {
138
- "epoch": 0.24896265560165975,
139
- "grad_norm": 15.830405235290527,
140
- "learning_rate": 0.00019966159785816663,
141
- "loss": 1.1538,
142
  "step": 90
143
  },
144
  {
145
- "epoch": 0.2627939142461964,
146
- "grad_norm": 0.8891340494155884,
147
- "learning_rate": 0.00019943348002101371,
148
- "loss": 1.1039,
149
  "step": 95
150
  },
151
  {
152
- "epoch": 0.2766251728907331,
153
- "grad_norm": 1.8807998895645142,
154
- "learning_rate": 0.00019914711676150378,
155
- "loss": 1.1167,
156
  "step": 100
157
  },
158
  {
159
- "epoch": 0.29045643153526973,
160
- "grad_norm": 0.5741276741027832,
161
- "learning_rate": 0.0001988026758234289,
162
- "loss": 1.1358,
163
  "step": 105
164
  },
165
  {
166
- "epoch": 0.30428769017980634,
167
- "grad_norm": 0.26256272196769714,
168
- "learning_rate": 0.00019840035897090215,
169
- "loss": 1.0063,
170
  "step": 110
171
  },
172
  {
173
- "epoch": 0.318118948824343,
174
- "grad_norm": 0.3350989818572998,
175
- "learning_rate": 0.00019794040187017005,
176
- "loss": 0.9463,
177
  "step": 115
178
  },
179
  {
180
- "epoch": 0.33195020746887965,
181
- "grad_norm": 0.25455090403556824,
182
- "learning_rate": 0.00019742307395156507,
183
- "loss": 0.9406,
184
  "step": 120
185
  },
186
  {
187
- "epoch": 0.3457814661134163,
188
- "grad_norm": 0.24111931025981903,
189
- "learning_rate": 0.0001968486782516813,
190
- "loss": 0.896,
191
  "step": 125
192
  },
193
  {
194
- "epoch": 0.359612724757953,
195
- "grad_norm": 0.2141195684671402,
196
- "learning_rate": 0.00019621755123586354,
197
- "loss": 0.9039,
198
  "step": 130
199
  },
200
  {
201
- "epoch": 0.37344398340248963,
202
- "grad_norm": 0.19233381748199463,
203
- "learning_rate": 0.00019553006260111515,
204
- "loss": 0.9018,
205
  "step": 135
206
  },
207
  {
208
- "epoch": 0.3872752420470263,
209
- "grad_norm": 0.18128527700901031,
210
- "learning_rate": 0.0001947866150595396,
211
- "loss": 0.8879,
212
  "step": 140
213
  },
214
  {
215
- "epoch": 0.40110650069156295,
216
- "grad_norm": 0.18775707483291626,
217
- "learning_rate": 0.00019398764410244275,
218
- "loss": 0.8892,
219
  "step": 145
220
  },
221
  {
222
- "epoch": 0.4149377593360996,
223
- "grad_norm": 0.18092310428619385,
224
- "learning_rate": 0.00019313361774523385,
225
- "loss": 0.8646,
226
  "step": 150
227
  },
228
  {
229
- "epoch": 0.4287690179806362,
230
- "grad_norm": 0.19820688664913177,
231
- "learning_rate": 0.00019222503625327496,
232
- "loss": 0.865,
233
  "step": 155
234
  },
235
  {
236
- "epoch": 0.4426002766251729,
237
- "grad_norm": 0.19297532737255096,
238
- "learning_rate": 0.00019126243184883898,
239
- "loss": 0.8473,
240
  "step": 160
241
  },
242
  {
243
- "epoch": 0.45643153526970953,
244
- "grad_norm": 0.17955026030540466,
245
- "learning_rate": 0.00019024636839934855,
246
- "loss": 0.8512,
247
  "step": 165
248
  },
249
  {
250
- "epoch": 0.4702627939142462,
251
- "grad_norm": 0.18673136830329895,
252
- "learning_rate": 0.00018917744108707776,
253
- "loss": 0.8421,
254
  "step": 170
255
  },
256
  {
257
- "epoch": 0.48409405255878285,
258
- "grad_norm": 0.1972828507423401,
259
- "learning_rate": 0.0001880562760605105,
260
- "loss": 0.8432,
261
  "step": 175
262
  },
263
  {
264
- "epoch": 0.4979253112033195,
265
- "grad_norm": 0.1867324709892273,
266
- "learning_rate": 0.00018688353006756004,
267
- "loss": 0.8299,
268
  "step": 180
269
  },
270
  {
271
- "epoch": 0.5117565698478561,
272
- "grad_norm": 0.16586913168430328,
273
- "learning_rate": 0.0001856598900708637,
274
- "loss": 0.837,
275
  "step": 185
276
  },
277
  {
278
- "epoch": 0.5255878284923928,
279
- "grad_norm": 0.1780562847852707,
280
- "learning_rate": 0.00018438607284537907,
281
- "loss": 0.8328,
282
  "step": 190
283
  },
284
  {
285
- "epoch": 0.5394190871369294,
286
- "grad_norm": 0.17686933279037476,
287
- "learning_rate": 0.00018306282455851655,
288
- "loss": 0.8238,
289
  "step": 195
290
  },
291
  {
292
- "epoch": 0.5532503457814661,
293
- "grad_norm": 0.1651686578989029,
294
- "learning_rate": 0.00018169092033305516,
295
- "loss": 0.81,
296
  "step": 200
297
  },
298
  {
299
- "epoch": 0.5670816044260027,
300
- "grad_norm": 0.15609298646450043,
301
- "learning_rate": 0.00018027116379309638,
302
- "loss": 0.8119,
303
  "step": 205
304
  },
305
  {
306
- "epoch": 0.5809128630705395,
307
- "grad_norm": 0.15646931529045105,
308
- "learning_rate": 0.00017880438659332332,
309
- "loss": 0.811,
310
  "step": 210
311
  },
312
  {
313
- "epoch": 0.5947441217150761,
314
- "grad_norm": 0.15522946417331696,
315
- "learning_rate": 0.00017729144793183992,
316
- "loss": 0.8105,
317
  "step": 215
318
  },
319
  {
320
- "epoch": 0.6085753803596127,
321
- "grad_norm": 0.14471209049224854,
322
- "learning_rate": 0.0001757332340468762,
323
- "loss": 0.8026,
324
  "step": 220
325
  },
326
  {
327
- "epoch": 0.6224066390041494,
328
- "grad_norm": 0.16337046027183533,
329
- "learning_rate": 0.00017413065769765406,
330
- "loss": 0.8051,
331
  "step": 225
332
  },
333
  {
334
- "epoch": 0.636237897648686,
335
- "grad_norm": 0.16175448894500732,
336
- "learning_rate": 0.00017248465762971776,
337
- "loss": 0.7999,
338
  "step": 230
339
  },
340
  {
341
- "epoch": 0.6500691562932227,
342
- "grad_norm": 0.17871476709842682,
343
- "learning_rate": 0.00017079619802504238,
344
- "loss": 0.8067,
345
  "step": 235
346
  },
347
  {
348
- "epoch": 0.6639004149377593,
349
- "grad_norm": 0.1535872220993042,
350
- "learning_rate": 0.00016906626793724224,
351
- "loss": 0.7893,
352
  "step": 240
353
  },
354
  {
355
- "epoch": 0.677731673582296,
356
- "grad_norm": 0.14963175356388092,
357
- "learning_rate": 0.00016729588071221055,
358
- "loss": 0.7867,
359
  "step": 245
360
  },
361
  {
362
- "epoch": 0.6915629322268326,
363
- "grad_norm": 0.14980217814445496,
364
- "learning_rate": 0.00016548607339452853,
365
- "loss": 0.7809,
366
  "step": 250
367
  },
368
  {
369
- "epoch": 0.7053941908713693,
370
- "grad_norm": 0.1530940979719162,
371
- "learning_rate": 0.0001636379061199933,
372
- "loss": 0.7745,
373
  "step": 255
374
  },
375
  {
376
- "epoch": 0.719225449515906,
377
- "grad_norm": 0.13913874328136444,
378
- "learning_rate": 0.0001617524614946192,
379
- "loss": 0.7844,
380
  "step": 260
381
  },
382
  {
383
- "epoch": 0.7330567081604425,
384
- "grad_norm": 0.1462012678384781,
385
- "learning_rate": 0.00015983084396047653,
386
- "loss": 0.781,
387
  "step": 265
388
  },
389
  {
390
- "epoch": 0.7468879668049793,
391
- "grad_norm": 0.14574629068374634,
392
- "learning_rate": 0.00015787417914873967,
393
- "loss": 0.7801,
394
  "step": 270
395
  },
396
  {
397
- "epoch": 0.7607192254495159,
398
- "grad_norm": 0.13526415824890137,
399
- "learning_rate": 0.00015588361322032283,
400
- "loss": 0.7629,
401
  "step": 275
402
  },
403
  {
404
- "epoch": 0.7745504840940526,
405
- "grad_norm": 0.14606699347496033,
406
- "learning_rate": 0.00015386031219449047,
407
- "loss": 0.77,
408
  "step": 280
409
  },
410
  {
411
- "epoch": 0.7883817427385892,
412
- "grad_norm": 0.14022809267044067,
413
- "learning_rate": 0.0001518054612658348,
414
- "loss": 0.7627,
415
  "step": 285
416
  },
417
  {
418
- "epoch": 0.8022130013831259,
419
- "grad_norm": 0.1374536156654358,
420
- "learning_rate": 0.00014972026411002107,
421
- "loss": 0.7599,
422
  "step": 290
423
  },
424
  {
425
- "epoch": 0.8160442600276625,
426
- "grad_norm": 0.15299181640148163,
427
- "learning_rate": 0.00014760594217870737,
428
- "loss": 0.754,
429
  "step": 295
430
  },
431
  {
432
- "epoch": 0.8298755186721992,
433
- "grad_norm": 0.13170954585075378,
434
- "learning_rate": 0.00014546373398405143,
435
- "loss": 0.7542,
436
  "step": 300
437
  },
438
  {
439
- "epoch": 0.8437067773167358,
440
- "grad_norm": 0.13624456524848938,
441
- "learning_rate": 0.00014329489437322397,
442
- "loss": 0.7527,
443
  "step": 305
444
  },
445
  {
446
- "epoch": 0.8575380359612724,
447
- "grad_norm": 0.13965509831905365,
448
- "learning_rate": 0.0001411006937933532,
449
- "loss": 0.7544,
450
  "step": 310
451
  },
452
  {
453
- "epoch": 0.8713692946058091,
454
- "grad_norm": 0.140866219997406,
455
- "learning_rate": 0.00013888241754733208,
456
- "loss": 0.7439,
457
  "step": 315
458
  },
459
  {
460
- "epoch": 0.8852005532503457,
461
- "grad_norm": 0.1314583420753479,
462
- "learning_rate": 0.0001366413650409223,
463
- "loss": 0.7431,
464
  "step": 320
465
  },
466
  {
467
- "epoch": 0.8990318118948825,
468
- "grad_norm": 0.123719722032547,
469
- "learning_rate": 0.00013437884902159822,
470
- "loss": 0.7377,
471
  "step": 325
472
  },
473
  {
474
- "epoch": 0.9128630705394191,
475
- "grad_norm": 0.15139903128147125,
476
- "learning_rate": 0.00013209619480957497,
477
- "loss": 0.7316,
478
  "step": 330
479
  },
480
  {
481
- "epoch": 0.9266943291839558,
482
- "grad_norm": 0.13264508545398712,
483
- "learning_rate": 0.00012979473952147205,
484
- "loss": 0.7251,
485
  "step": 335
486
  },
487
  {
488
- "epoch": 0.9405255878284924,
489
- "grad_norm": 0.13327768445014954,
490
- "learning_rate": 0.00012747583128706698,
491
- "loss": 0.7523,
492
  "step": 340
493
  },
494
  {
495
- "epoch": 0.9543568464730291,
496
- "grad_norm": 0.12970297038555145,
497
- "learning_rate": 0.0001251408284595974,
498
- "loss": 0.7293,
499
  "step": 345
500
  },
501
  {
502
- "epoch": 0.9681881051175657,
503
- "grad_norm": 0.14814430475234985,
504
- "learning_rate": 0.00012279109882007492,
505
- "loss": 0.7262,
506
  "step": 350
507
  },
508
  {
509
- "epoch": 0.9820193637621023,
510
- "grad_norm": 0.14566916227340698,
511
- "learning_rate": 0.00012042801877607625,
512
- "loss": 0.7223,
513
  "step": 355
514
  },
515
  {
516
- "epoch": 0.995850622406639,
517
- "grad_norm": 0.12835277616977692,
518
- "learning_rate": 0.00011805297255548118,
519
- "loss": 0.7246,
520
  "step": 360
521
  },
522
  {
523
- "epoch": 0.9986168741355463,
524
- "eval_loss": 0.7382652759552002,
525
- "eval_runtime": 51.8183,
526
- "eval_samples_per_second": 88.965,
527
- "eval_steps_per_second": 2.798,
528
- "step": 361
529
- },
530
- {
531
- "epoch": 1.0110650069156293,
532
- "grad_norm": 0.14208853244781494,
533
- "learning_rate": 0.00011566735139562947,
534
- "loss": 0.7608,
535
  "step": 365
536
  },
537
  {
538
- "epoch": 1.0248962655601659,
539
- "grad_norm": 0.12897470593452454,
540
- "learning_rate": 0.00011327255272837221,
541
- "loss": 0.581,
542
  "step": 370
543
  },
544
  {
545
- "epoch": 1.0387275242047027,
546
- "grad_norm": 0.13211384415626526,
547
- "learning_rate": 0.00011086997936149408,
548
- "loss": 0.5594,
549
  "step": 375
550
  },
551
  {
552
- "epoch": 1.0525587828492393,
553
- "grad_norm": 0.12891365587711334,
554
- "learning_rate": 0.00010846103865698696,
555
- "loss": 0.5629,
556
  "step": 380
557
  },
558
  {
559
- "epoch": 1.066390041493776,
560
- "grad_norm": 0.13562671840190887,
561
- "learning_rate": 0.00010604714170665544,
562
- "loss": 0.5539,
563
  "step": 385
564
  },
565
  {
566
- "epoch": 1.0802213001383125,
567
- "grad_norm": 0.12082191556692123,
568
- "learning_rate": 0.00010362970250553796,
569
- "loss": 0.5562,
570
  "step": 390
571
  },
572
  {
573
- "epoch": 1.0940525587828493,
574
- "grad_norm": 0.12038177996873856,
575
- "learning_rate": 0.00010121013712362684,
576
- "loss": 0.5506,
577
  "step": 395
578
  },
579
  {
580
- "epoch": 1.107883817427386,
581
- "grad_norm": 0.11669060587882996,
582
- "learning_rate": 9.878986287637318e-05,
583
- "loss": 0.552,
584
  "step": 400
585
  },
586
  {
587
- "epoch": 1.1217150760719226,
588
- "grad_norm": 0.13290895521640778,
589
- "learning_rate": 9.637029749446205e-05,
590
- "loss": 0.5591,
591
  "step": 405
592
  },
593
  {
594
- "epoch": 1.1355463347164592,
595
- "grad_norm": 0.14374125003814697,
596
- "learning_rate": 9.395285829334458e-05,
597
- "loss": 0.5549,
598
  "step": 410
599
  },
600
  {
601
- "epoch": 1.1493775933609958,
602
- "grad_norm": 0.12170559167861938,
603
- "learning_rate": 9.153896134301309e-05,
604
- "loss": 0.5551,
605
  "step": 415
606
  },
607
  {
608
- "epoch": 1.1632088520055326,
609
- "grad_norm": 0.12605440616607666,
610
- "learning_rate": 8.913002063850593e-05,
611
- "loss": 0.5548,
612
  "step": 420
613
  },
614
  {
615
- "epoch": 1.1770401106500692,
616
- "grad_norm": 0.12290360033512115,
617
- "learning_rate": 8.672744727162781e-05,
618
- "loss": 0.5582,
619
  "step": 425
620
  },
621
  {
622
- "epoch": 1.1908713692946058,
623
- "grad_norm": 0.11714824289083481,
624
- "learning_rate": 8.433264860437056e-05,
625
- "loss": 0.5484,
626
  "step": 430
627
  },
628
  {
629
- "epoch": 1.2047026279391424,
630
- "grad_norm": 0.12463142722845078,
631
- "learning_rate": 8.194702744451886e-05,
632
- "loss": 0.5443,
633
  "step": 435
634
  },
635
  {
636
- "epoch": 1.2185338865836792,
637
- "grad_norm": 0.11625911295413971,
638
- "learning_rate": 7.957198122392377e-05,
639
- "loss": 0.5373,
640
  "step": 440
641
  },
642
  {
643
- "epoch": 1.2323651452282158,
644
- "grad_norm": 0.13417629897594452,
645
- "learning_rate": 7.72089011799251e-05,
646
- "loss": 0.548,
647
  "step": 445
648
  },
649
  {
650
- "epoch": 1.2461964038727524,
651
- "grad_norm": 0.13118138909339905,
652
- "learning_rate": 7.485917154040263e-05,
653
- "loss": 0.5549,
654
  "step": 450
655
  },
656
  {
657
- "epoch": 1.260027662517289,
658
- "grad_norm": 0.12697063386440277,
659
- "learning_rate": 7.252416871293304e-05,
660
- "loss": 0.5428,
661
  "step": 455
662
  },
663
  {
664
- "epoch": 1.2738589211618256,
665
- "grad_norm": 0.12316182255744934,
666
- "learning_rate": 7.020526047852797e-05,
667
- "loss": 0.5465,
668
  "step": 460
669
  },
670
  {
671
- "epoch": 1.2876901798063622,
672
- "grad_norm": 0.12391973286867142,
673
- "learning_rate": 6.790380519042507e-05,
674
- "loss": 0.5314,
675
  "step": 465
676
  },
677
  {
678
- "epoch": 1.301521438450899,
679
- "grad_norm": 0.1304333359003067,
680
- "learning_rate": 6.562115097840182e-05,
681
- "loss": 0.5395,
682
  "step": 470
683
  },
684
  {
685
- "epoch": 1.3153526970954357,
686
- "grad_norm": 0.13805896043777466,
687
- "learning_rate": 6.335863495907772e-05,
688
- "loss": 0.5357,
689
  "step": 475
690
  },
691
  {
692
- "epoch": 1.3291839557399723,
693
- "grad_norm": 0.11262813210487366,
694
- "learning_rate": 6.111758245266794e-05,
695
- "loss": 0.5315,
696
  "step": 480
697
  },
698
  {
699
- "epoch": 1.343015214384509,
700
- "grad_norm": 0.1169748455286026,
701
- "learning_rate": 5.889930620664681e-05,
702
- "loss": 0.5377,
703
  "step": 485
704
  },
705
  {
706
- "epoch": 1.3568464730290457,
707
- "grad_norm": 0.1196884959936142,
708
- "learning_rate": 5.670510562677607e-05,
709
- "loss": 0.5328,
710
  "step": 490
711
  },
712
  {
713
- "epoch": 1.3706777316735823,
714
- "grad_norm": 0.1325497031211853,
715
- "learning_rate": 5.453626601594857e-05,
716
- "loss": 0.5361,
717
  "step": 495
718
  },
719
  {
720
- "epoch": 1.384508990318119,
721
- "grad_norm": 0.12347881495952606,
722
- "learning_rate": 5.239405782129261e-05,
723
- "loss": 0.5308,
724
  "step": 500
725
  },
726
  {
727
- "epoch": 1.3983402489626555,
728
- "grad_norm": 0.12328892946243286,
729
- "learning_rate": 5.027973588997896e-05,
730
- "loss": 0.5324,
731
  "step": 505
732
  },
733
  {
734
- "epoch": 1.4121715076071921,
735
- "grad_norm": 0.12255866080522537,
736
- "learning_rate": 4.819453873416526e-05,
737
- "loss": 0.5314,
738
  "step": 510
739
  },
740
  {
741
- "epoch": 1.426002766251729,
742
- "grad_norm": 0.11596041172742844,
743
- "learning_rate": 4.6139687805509535e-05,
744
- "loss": 0.5247,
745
  "step": 515
746
  },
747
  {
748
- "epoch": 1.4398340248962656,
749
- "grad_norm": 0.1198066920042038,
750
- "learning_rate": 4.411638677967718e-05,
751
- "loss": 0.5176,
752
  "step": 520
753
  },
754
  {
755
- "epoch": 1.4536652835408022,
756
- "grad_norm": 0.11880145221948624,
757
- "learning_rate": 4.212582085126038e-05,
758
- "loss": 0.5209,
759
  "step": 525
760
  },
761
  {
762
- "epoch": 1.467496542185339,
763
- "grad_norm": 0.11095953732728958,
764
- "learning_rate": 4.016915603952347e-05,
765
- "loss": 0.5276,
766
  "step": 530
767
  },
768
  {
769
- "epoch": 1.4813278008298756,
770
- "grad_norm": 0.11372353136539459,
771
- "learning_rate": 3.824753850538082e-05,
772
- "loss": 0.5196,
773
  "step": 535
774
  },
775
  {
776
- "epoch": 1.4951590594744122,
777
- "grad_norm": 0.1171097531914711,
778
- "learning_rate": 3.636209388000673e-05,
779
- "loss": 0.5273,
780
  "step": 540
781
  },
782
  {
783
- "epoch": 1.5089903181189488,
784
- "grad_norm": 0.11904237419366837,
785
- "learning_rate": 3.45139266054715e-05,
786
- "loss": 0.5202,
787
  "step": 545
788
  },
789
  {
790
- "epoch": 1.5228215767634854,
791
- "grad_norm": 0.11875548213720322,
792
- "learning_rate": 3.270411928778948e-05,
793
- "loss": 0.5161,
794
  "step": 550
795
  },
796
  {
797
- "epoch": 1.536652835408022,
798
- "grad_norm": 0.10985347628593445,
799
- "learning_rate": 3.093373206275775e-05,
800
- "loss": 0.5154,
801
  "step": 555
802
  },
803
  {
804
- "epoch": 1.5504840940525588,
805
- "grad_norm": 0.11928029358386993,
806
- "learning_rate": 2.9203801974957666e-05,
807
- "loss": 0.5075,
808
  "step": 560
809
  },
810
  {
811
- "epoch": 1.5643153526970954,
812
- "grad_norm": 0.11182394623756409,
813
- "learning_rate": 2.751534237028227e-05,
814
- "loss": 0.518,
815
  "step": 565
816
  },
817
  {
818
- "epoch": 1.5781466113416323,
819
- "grad_norm": 0.11065185815095901,
820
- "learning_rate": 2.5869342302345945e-05,
821
- "loss": 0.5112,
822
  "step": 570
823
  },
824
  {
825
- "epoch": 1.5919778699861689,
826
- "grad_norm": 0.10788684338331223,
827
- "learning_rate": 2.4266765953123814e-05,
828
- "loss": 0.5201,
829
  "step": 575
830
  },
831
  {
832
- "epoch": 1.6058091286307055,
833
- "grad_norm": 0.12274058163166046,
834
- "learning_rate": 2.2708552068160115e-05,
835
- "loss": 0.5122,
836
  "step": 580
837
  },
838
  {
839
- "epoch": 1.619640387275242,
840
- "grad_norm": 0.11127958446741104,
841
- "learning_rate": 2.1195613406676706e-05,
842
- "loss": 0.5183,
843
  "step": 585
844
  },
845
  {
846
- "epoch": 1.6334716459197787,
847
- "grad_norm": 0.11190652847290039,
848
- "learning_rate": 1.9728836206903656e-05,
849
- "loss": 0.5052,
850
  "step": 590
851
  },
852
  {
853
- "epoch": 1.6473029045643153,
854
- "grad_norm": 0.10597892105579376,
855
- "learning_rate": 1.8309079666944883e-05,
856
- "loss": 0.5038,
857
  "step": 595
858
  },
859
  {
860
- "epoch": 1.6611341632088519,
861
- "grad_norm": 0.11393982917070389,
862
- "learning_rate": 1.6937175441483455e-05,
863
- "loss": 0.4965,
864
  "step": 600
865
  },
866
  {
867
- "epoch": 1.6749654218533887,
868
- "grad_norm": 0.11752679198980331,
869
- "learning_rate": 1.561392715462098e-05,
870
- "loss": 0.5137,
871
  "step": 605
872
  },
873
  {
874
- "epoch": 1.6887966804979253,
875
- "grad_norm": 0.10845957696437836,
876
- "learning_rate": 1.4340109929136291e-05,
877
- "loss": 0.5051,
878
  "step": 610
879
  },
880
  {
881
- "epoch": 1.702627939142462,
882
- "grad_norm": 0.11502846330404282,
883
- "learning_rate": 1.3116469932439968e-05,
884
- "loss": 0.5065,
885
  "step": 615
886
  },
887
  {
888
- "epoch": 1.7164591977869987,
889
- "grad_norm": 0.10717900097370148,
890
- "learning_rate": 1.1943723939489516e-05,
891
- "loss": 0.4963,
892
  "step": 620
893
  },
894
  {
895
- "epoch": 1.7302904564315353,
896
- "grad_norm": 0.11207237094640732,
897
- "learning_rate": 1.0822558912922265e-05,
898
- "loss": 0.4953,
899
  "step": 625
900
  },
901
  {
902
- "epoch": 1.744121715076072,
903
- "grad_norm": 0.10951482504606247,
904
- "learning_rate": 9.753631600651458e-06,
905
- "loss": 0.4875,
906
  "step": 630
907
  },
908
  {
909
- "epoch": 1.7579529737206085,
910
- "grad_norm": 0.10522555559873581,
911
- "learning_rate": 8.737568151161024e-06,
912
- "loss": 0.5041,
913
  "step": 635
914
  },
915
  {
916
- "epoch": 1.7717842323651452,
917
- "grad_norm": 0.10585460811853409,
918
- "learning_rate": 7.774963746725073e-06,
919
- "loss": 0.5084,
920
  "step": 640
921
  },
922
  {
923
- "epoch": 1.7856154910096818,
924
- "grad_norm": 0.10949090868234634,
925
- "learning_rate": 6.866382254766157e-06,
926
- "loss": 0.5026,
927
  "step": 645
928
  },
929
  {
930
- "epoch": 1.7994467496542186,
931
- "grad_norm": 0.10402841866016388,
932
- "learning_rate": 6.0123558975572645e-06,
933
- "loss": 0.4941,
934
  "step": 650
935
  },
936
  {
937
- "epoch": 1.8132780082987552,
938
- "grad_norm": 0.1075565442442894,
939
- "learning_rate": 5.213384940460408e-06,
940
- "loss": 0.5069,
941
  "step": 655
942
  },
943
  {
944
- "epoch": 1.8271092669432918,
945
- "grad_norm": 0.11545684933662415,
946
- "learning_rate": 4.46993739888486e-06,
947
- "loss": 0.5089,
948
  "step": 660
949
  },
950
  {
951
- "epoch": 1.8409405255878286,
952
- "grad_norm": 0.11043102294206619,
953
- "learning_rate": 3.7824487641364594e-06,
954
- "loss": 0.4983,
955
  "step": 665
956
  },
957
  {
958
- "epoch": 1.8547717842323652,
959
- "grad_norm": 0.10490711033344269,
960
- "learning_rate": 3.151321748318692e-06,
961
- "loss": 0.5023,
962
  "step": 670
963
  },
964
  {
965
- "epoch": 1.8686030428769018,
966
- "grad_norm": 0.10829063504934311,
967
- "learning_rate": 2.5769260484349466e-06,
968
- "loss": 0.4941,
969
  "step": 675
970
  },
971
  {
972
- "epoch": 1.8824343015214384,
973
- "grad_norm": 0.10498882085084915,
974
- "learning_rate": 2.059598129829976e-06,
975
- "loss": 0.5005,
976
  "step": 680
977
  },
978
  {
979
- "epoch": 1.896265560165975,
980
- "grad_norm": 0.10581523925065994,
981
- "learning_rate": 1.5996410290978314e-06,
982
- "loss": 0.4905,
983
  "step": 685
984
  },
985
  {
986
- "epoch": 1.9100968188105116,
987
- "grad_norm": 0.11104903370141983,
988
- "learning_rate": 1.1973241765711352e-06,
989
- "loss": 0.4912,
990
  "step": 690
991
  },
992
  {
993
- "epoch": 1.9239280774550485,
994
- "grad_norm": 0.10330861806869507,
995
- "learning_rate": 8.52883238496227e-07,
996
- "loss": 0.4814,
997
  "step": 695
998
  },
999
  {
1000
- "epoch": 1.937759336099585,
1001
- "grad_norm": 0.10697363317012787,
1002
- "learning_rate": 5.665199789862907e-07,
1003
- "loss": 0.4984,
1004
  "step": 700
1005
  },
1006
  {
1007
- "epoch": 1.9515905947441217,
1008
- "grad_norm": 0.10598309338092804,
1009
- "learning_rate": 3.3840214183337157e-07,
1010
- "loss": 0.4924,
1011
  "step": 705
1012
  },
1013
  {
1014
- "epoch": 1.9654218533886585,
1015
- "grad_norm": 0.10365109890699387,
1016
- "learning_rate": 1.686633522487213e-07,
1017
- "loss": 0.4958,
1018
  "step": 710
1019
  },
1020
  {
1021
- "epoch": 1.979253112033195,
1022
- "grad_norm": 0.10521161556243896,
1023
- "learning_rate": 5.740303858874363e-08,
1024
- "loss": 0.5027,
1025
  "step": 715
1026
  },
1027
  {
1028
- "epoch": 1.9930843706777317,
1029
- "grad_norm": 0.10533251613378525,
1030
- "learning_rate": 4.686374112583547e-09,
1031
- "loss": 0.4935,
1032
  "step": 720
1033
  },
1034
  {
1035
- "epoch": 1.9986168741355463,
1036
- "eval_loss": 0.6898206472396851,
1037
- "eval_runtime": 51.7983,
1038
- "eval_samples_per_second": 88.999,
1039
- "eval_steps_per_second": 2.799,
1040
- "step": 722
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1041
  },
1042
  {
1043
- "epoch": 1.9986168741355463,
1044
- "step": 722,
1045
- "total_flos": 3.364677087628624e+18,
1046
- "train_loss": 0.7622144496341822,
1047
- "train_runtime": 4614.5072,
1048
- "train_samples_per_second": 20.054,
1049
- "train_steps_per_second": 0.156
1050
  }
1051
  ],
1052
  "logging_steps": 5,
1053
- "max_steps": 722,
1054
  "num_input_tokens_seen": 0,
1055
  "num_train_epochs": 2,
1056
  "save_steps": 500,
@@ -1066,7 +2075,7 @@
1066
  "attributes": {}
1067
  }
1068
  },
1069
- "total_flos": 3.364677087628624e+18,
1070
  "train_batch_size": 4,
1071
  "trial_name": null,
1072
  "trial_params": null
 
1
  {
2
+ "best_global_step": null,
3
  "best_metric": null,
4
  "best_model_checkpoint": null,
5
+ "epoch": 1.9982692973347178,
6
  "eval_steps": 500,
7
+ "global_step": 1444,
8
  "is_hyper_param_search": false,
9
  "is_local_process_zero": true,
10
  "is_world_process_zero": true,
11
  "log_history": [
12
  {
13
+ "epoch": 0.0013845621322256837,
14
+ "grad_norm": 78.96892547607422,
15
+ "learning_rate": 0.0,
16
+ "loss": 3.6138,
17
  "step": 1
18
  },
19
  {
20
+ "epoch": 0.006922810661128418,
21
+ "grad_norm": 26.50972557067871,
22
+ "learning_rate": 5.517241379310345e-06,
23
+ "loss": 3.3919,
24
  "step": 5
25
  },
26
  {
27
+ "epoch": 0.013845621322256836,
28
+ "grad_norm": 10.996688842773438,
29
+ "learning_rate": 1.2413793103448277e-05,
30
+ "loss": 2.8025,
31
  "step": 10
32
  },
33
  {
34
+ "epoch": 0.020768431983385256,
35
+ "grad_norm": 7.297196388244629,
36
+ "learning_rate": 1.9310344827586207e-05,
37
+ "loss": 2.5925,
38
  "step": 15
39
  },
40
  {
41
+ "epoch": 0.027691242644513673,
42
+ "grad_norm": 6.842594146728516,
43
+ "learning_rate": 2.620689655172414e-05,
44
+ "loss": 2.3888,
45
  "step": 20
46
  },
47
  {
48
+ "epoch": 0.034614053305642094,
49
+ "grad_norm": 6.920436382293701,
50
+ "learning_rate": 3.310344827586207e-05,
51
+ "loss": 2.1351,
52
  "step": 25
53
  },
54
  {
55
+ "epoch": 0.04153686396677051,
56
+ "grad_norm": 6.916154384613037,
57
+ "learning_rate": 4e-05,
58
+ "loss": 1.7918,
59
  "step": 30
60
  },
61
  {
62
+ "epoch": 0.04845967462789893,
63
+ "grad_norm": 3.957890033721924,
64
+ "learning_rate": 4.689655172413793e-05,
65
+ "loss": 1.4435,
66
  "step": 35
67
  },
68
  {
69
+ "epoch": 0.055382485289027346,
70
+ "grad_norm": 2.863860607147217,
71
+ "learning_rate": 5.379310344827586e-05,
72
+ "loss": 1.3002,
73
  "step": 40
74
  },
75
  {
76
+ "epoch": 0.06230529595015576,
77
+ "grad_norm": 15.180214881896973,
78
+ "learning_rate": 6.068965517241379e-05,
79
+ "loss": 1.3723,
80
  "step": 45
81
  },
82
  {
83
+ "epoch": 0.06922810661128419,
84
+ "grad_norm": 3.9463565349578857,
85
+ "learning_rate": 6.758620689655173e-05,
86
+ "loss": 1.3394,
87
  "step": 50
88
  },
89
  {
90
+ "epoch": 0.0761509172724126,
91
+ "grad_norm": 35.227447509765625,
92
+ "learning_rate": 7.448275862068966e-05,
93
+ "loss": 1.3752,
94
  "step": 55
95
  },
96
  {
97
+ "epoch": 0.08307372793354102,
98
+ "grad_norm": 15.025148391723633,
99
+ "learning_rate": 8.137931034482759e-05,
100
+ "loss": 1.4322,
101
  "step": 60
102
  },
103
  {
104
+ "epoch": 0.08999653859466944,
105
+ "grad_norm": 45.38770294189453,
106
+ "learning_rate": 8.827586206896552e-05,
107
+ "loss": 3.0681,
108
  "step": 65
109
  },
110
  {
111
+ "epoch": 0.09691934925579786,
112
+ "grad_norm": 8.21139907836914,
113
+ "learning_rate": 9.517241379310345e-05,
114
+ "loss": 1.3441,
115
  "step": 70
116
  },
117
  {
118
+ "epoch": 0.10384215991692627,
119
+ "grad_norm": 3.560570240020752,
120
+ "learning_rate": 0.0001020689655172414,
121
+ "loss": 1.2695,
122
  "step": 75
123
  },
124
  {
125
+ "epoch": 0.11076497057805469,
126
+ "grad_norm": 11.044499397277832,
127
+ "learning_rate": 0.00010896551724137931,
128
+ "loss": 1.2145,
129
  "step": 80
130
  },
131
  {
132
+ "epoch": 0.11768778123918311,
133
+ "grad_norm": 3.8201582431793213,
134
+ "learning_rate": 0.00011586206896551725,
135
+ "loss": 1.1547,
136
  "step": 85
137
  },
138
  {
139
+ "epoch": 0.12461059190031153,
140
+ "grad_norm": 3.9128692150115967,
141
+ "learning_rate": 0.00012275862068965518,
142
+ "loss": 1.1285,
143
  "step": 90
144
  },
145
  {
146
+ "epoch": 0.13153340256143994,
147
+ "grad_norm": 6.907658576965332,
148
+ "learning_rate": 0.0001296551724137931,
149
+ "loss": 1.1376,
150
  "step": 95
151
  },
152
  {
153
+ "epoch": 0.13845621322256838,
154
+ "grad_norm": 2.5651073455810547,
155
+ "learning_rate": 0.00013655172413793104,
156
+ "loss": 1.1429,
157
  "step": 100
158
  },
159
  {
160
+ "epoch": 0.14537902388369678,
161
+ "grad_norm": 2.3567330837249756,
162
+ "learning_rate": 0.00014344827586206896,
163
+ "loss": 1.1356,
164
  "step": 105
165
  },
166
  {
167
+ "epoch": 0.1523018345448252,
168
+ "grad_norm": 47.36102294921875,
169
+ "learning_rate": 0.0001503448275862069,
170
+ "loss": 1.2263,
171
  "step": 110
172
  },
173
  {
174
+ "epoch": 0.1592246452059536,
175
+ "grad_norm": 6.215350151062012,
176
+ "learning_rate": 0.00015724137931034485,
177
+ "loss": 1.4307,
178
  "step": 115
179
  },
180
  {
181
+ "epoch": 0.16614745586708204,
182
+ "grad_norm": 2.4405245780944824,
183
+ "learning_rate": 0.00016413793103448276,
184
+ "loss": 1.2431,
185
  "step": 120
186
  },
187
  {
188
+ "epoch": 0.17307026652821045,
189
+ "grad_norm": 6.445745944976807,
190
+ "learning_rate": 0.0001710344827586207,
191
+ "loss": 1.4538,
192
  "step": 125
193
  },
194
  {
195
+ "epoch": 0.17999307718933888,
196
+ "grad_norm": 7.712639808654785,
197
+ "learning_rate": 0.00017793103448275862,
198
+ "loss": 1.3982,
199
  "step": 130
200
  },
201
  {
202
+ "epoch": 0.18691588785046728,
203
+ "grad_norm": 8.239908218383789,
204
+ "learning_rate": 0.00018482758620689654,
205
+ "loss": 1.4976,
206
  "step": 135
207
  },
208
  {
209
+ "epoch": 0.19383869851159571,
210
+ "grad_norm": 2.3590519428253174,
211
+ "learning_rate": 0.0001917241379310345,
212
+ "loss": 1.2595,
213
  "step": 140
214
  },
215
  {
216
+ "epoch": 0.20076150917272412,
217
+ "grad_norm": 2.654240608215332,
218
+ "learning_rate": 0.00019862068965517243,
219
+ "loss": 1.2232,
220
  "step": 145
221
  },
222
  {
223
+ "epoch": 0.20768431983385255,
224
+ "grad_norm": 2.5592308044433594,
225
+ "learning_rate": 0.0001999953208384105,
226
+ "loss": 1.2181,
227
  "step": 150
228
  },
229
  {
230
+ "epoch": 0.21460713049498095,
231
+ "grad_norm": 1.9601577520370483,
232
+ "learning_rate": 0.0001999763124949378,
233
+ "loss": 1.1714,
234
  "step": 155
235
  },
236
  {
237
+ "epoch": 0.22152994115610938,
238
+ "grad_norm": 1.5926353931427002,
239
+ "learning_rate": 0.0001999426852993044,
240
+ "loss": 1.173,
241
  "step": 160
242
  },
243
  {
244
+ "epoch": 0.2284527518172378,
245
+ "grad_norm": 1.8037822246551514,
246
+ "learning_rate": 0.00019989444416858502,
247
+ "loss": 1.1432,
248
  "step": 165
249
  },
250
  {
251
+ "epoch": 0.23537556247836622,
252
+ "grad_norm": 2.564453601837158,
253
+ "learning_rate": 0.0001998315961567502,
254
+ "loss": 1.184,
255
  "step": 170
256
  },
257
  {
258
+ "epoch": 0.24229837313949462,
259
+ "grad_norm": 1.347174882888794,
260
+ "learning_rate": 0.00019975415045363467,
261
+ "loss": 1.1879,
262
  "step": 175
263
  },
264
  {
265
+ "epoch": 0.24922118380062305,
266
+ "grad_norm": 1.8346368074417114,
267
+ "learning_rate": 0.00019966211838359378,
268
+ "loss": 1.1895,
269
  "step": 180
270
  },
271
  {
272
+ "epoch": 0.25614399446175146,
273
+ "grad_norm": 1.5952632427215576,
274
+ "learning_rate": 0.00019955551340384743,
275
+ "loss": 1.1717,
276
  "step": 185
277
  },
278
  {
279
+ "epoch": 0.2630668051228799,
280
+ "grad_norm": 10.082040786743164,
281
+ "learning_rate": 0.00019943435110251248,
282
+ "loss": 1.2318,
283
  "step": 190
284
  },
285
  {
286
+ "epoch": 0.2699896157840083,
287
+ "grad_norm": 9.962459564208984,
288
+ "learning_rate": 0.0001992986491963232,
289
+ "loss": 1.2411,
290
  "step": 195
291
  },
292
  {
293
+ "epoch": 0.27691242644513675,
294
+ "grad_norm": 1.3912514448165894,
295
+ "learning_rate": 0.00019914842752804103,
296
+ "loss": 1.2747,
297
  "step": 200
298
  },
299
  {
300
+ "epoch": 0.2838352371062651,
301
+ "grad_norm": 2.4310293197631836,
302
+ "learning_rate": 0.00019898370806355263,
303
+ "loss": 1.1451,
304
  "step": 205
305
  },
306
  {
307
+ "epoch": 0.29075804776739356,
308
+ "grad_norm": 2.632899761199951,
309
+ "learning_rate": 0.00019880451488865836,
310
+ "loss": 1.1757,
311
  "step": 210
312
  },
313
  {
314
+ "epoch": 0.297680858428522,
315
+ "grad_norm": 2.9560701847076416,
316
+ "learning_rate": 0.00019861087420555018,
317
+ "loss": 1.2112,
318
  "step": 215
319
  },
320
  {
321
+ "epoch": 0.3046036690896504,
322
+ "grad_norm": 1.2769361734390259,
323
+ "learning_rate": 0.0001984028143289803,
324
+ "loss": 1.1494,
325
  "step": 220
326
  },
327
  {
328
+ "epoch": 0.3115264797507788,
329
+ "grad_norm": 0.9540227055549622,
330
+ "learning_rate": 0.00019818036568212106,
331
+ "loss": 1.1274,
332
  "step": 225
333
  },
334
  {
335
+ "epoch": 0.3184492904119072,
336
+ "grad_norm": 1.050113320350647,
337
+ "learning_rate": 0.00019794356079211604,
338
+ "loss": 1.1254,
339
  "step": 230
340
  },
341
  {
342
+ "epoch": 0.32537210107303566,
343
+ "grad_norm": 1.335493564605713,
344
+ "learning_rate": 0.00019769243428532422,
345
+ "loss": 1.1688,
346
  "step": 235
347
  },
348
  {
349
+ "epoch": 0.3322949117341641,
350
+ "grad_norm": 1.0514490604400635,
351
+ "learning_rate": 0.00019742702288225652,
352
+ "loss": 1.153,
353
  "step": 240
354
  },
355
  {
356
+ "epoch": 0.33921772239529246,
357
+ "grad_norm": 1.2041540145874023,
358
+ "learning_rate": 0.00019714736539220648,
359
+ "loss": 1.1369,
360
  "step": 245
361
  },
362
  {
363
+ "epoch": 0.3461405330564209,
364
+ "grad_norm": 1.1564425230026245,
365
+ "learning_rate": 0.00019685350270757555,
366
+ "loss": 1.1567,
367
  "step": 250
368
  },
369
  {
370
+ "epoch": 0.3530633437175493,
371
+ "grad_norm": 1.2910791635513306,
372
+ "learning_rate": 0.0001965454777978936,
373
+ "loss": 1.1633,
374
  "step": 255
375
  },
376
  {
377
+ "epoch": 0.35998615437867776,
378
+ "grad_norm": 1.6193257570266724,
379
+ "learning_rate": 0.00019622333570353567,
380
+ "loss": 1.1348,
381
  "step": 260
382
  },
383
  {
384
+ "epoch": 0.36690896503980613,
385
+ "grad_norm": 0.8881533741950989,
386
+ "learning_rate": 0.00019588712352913625,
387
+ "loss": 1.1118,
388
  "step": 265
389
  },
390
  {
391
+ "epoch": 0.37383177570093457,
392
+ "grad_norm": 1.275267243385315,
393
+ "learning_rate": 0.00019553689043670127,
394
+ "loss": 1.1278,
395
  "step": 270
396
  },
397
  {
398
+ "epoch": 0.380754586362063,
399
+ "grad_norm": 1.1025657653808594,
400
+ "learning_rate": 0.00019517268763841962,
401
+ "loss": 1.0943,
402
  "step": 275
403
  },
404
  {
405
+ "epoch": 0.38767739702319143,
406
+ "grad_norm": 0.9336686730384827,
407
+ "learning_rate": 0.00019479456838917473,
408
+ "loss": 1.1002,
409
  "step": 280
410
  },
411
  {
412
+ "epoch": 0.39460020768431986,
413
+ "grad_norm": 0.8029949069023132,
414
+ "learning_rate": 0.0001944025879787574,
415
+ "loss": 1.089,
416
  "step": 285
417
  },
418
  {
419
+ "epoch": 0.40152301834544823,
420
+ "grad_norm": 1.0376158952713013,
421
+ "learning_rate": 0.0001939968037237812,
422
+ "loss": 1.1215,
423
  "step": 290
424
  },
425
  {
426
+ "epoch": 0.40844582900657667,
427
+ "grad_norm": 0.8140847682952881,
428
+ "learning_rate": 0.00019357727495930146,
429
+ "loss": 1.0881,
430
  "step": 295
431
  },
432
  {
433
+ "epoch": 0.4153686396677051,
434
+ "grad_norm": 0.8856680393218994,
435
+ "learning_rate": 0.00019314406303013904,
436
+ "loss": 1.0907,
437
  "step": 300
438
  },
439
  {
440
+ "epoch": 0.42229145032883353,
441
+ "grad_norm": 0.880487859249115,
442
+ "learning_rate": 0.00019269723128191048,
443
+ "loss": 1.0793,
444
  "step": 305
445
  },
446
  {
447
+ "epoch": 0.4292142609899619,
448
+ "grad_norm": 0.8125194907188416,
449
+ "learning_rate": 0.00019223684505176517,
450
+ "loss": 1.0859,
451
  "step": 310
452
  },
453
  {
454
+ "epoch": 0.43613707165109034,
455
+ "grad_norm": 1.1872682571411133,
456
+ "learning_rate": 0.00019176297165883165,
457
+ "loss": 1.0933,
458
  "step": 315
459
  },
460
  {
461
+ "epoch": 0.44305988231221877,
462
+ "grad_norm": 0.8477077484130859,
463
+ "learning_rate": 0.00019127568039437406,
464
+ "loss": 1.0726,
465
  "step": 320
466
  },
467
  {
468
+ "epoch": 0.4499826929733472,
469
+ "grad_norm": 0.8214722275733948,
470
+ "learning_rate": 0.00019077504251166006,
471
+ "loss": 1.0685,
472
  "step": 325
473
  },
474
  {
475
+ "epoch": 0.4569055036344756,
476
+ "grad_norm": 0.8920738101005554,
477
+ "learning_rate": 0.00019026113121554204,
478
+ "loss": 1.0697,
479
  "step": 330
480
  },
481
  {
482
+ "epoch": 0.463828314295604,
483
+ "grad_norm": 0.748255729675293,
484
+ "learning_rate": 0.00018973402165175268,
485
+ "loss": 1.0703,
486
  "step": 335
487
  },
488
  {
489
+ "epoch": 0.47075112495673244,
490
+ "grad_norm": 0.7890346050262451,
491
+ "learning_rate": 0.0001891937908959172,
492
+ "loss": 1.0471,
493
  "step": 340
494
  },
495
  {
496
+ "epoch": 0.47767393561786087,
497
+ "grad_norm": 0.7698908448219299,
498
+ "learning_rate": 0.00018864051794228282,
499
+ "loss": 1.0529,
500
  "step": 345
501
  },
502
  {
503
+ "epoch": 0.48459674627898924,
504
+ "grad_norm": 0.7833818197250366,
505
+ "learning_rate": 0.00018807428369216822,
506
+ "loss": 1.0187,
507
  "step": 350
508
  },
509
  {
510
+ "epoch": 0.4915195569401177,
511
+ "grad_norm": 0.7760437726974487,
512
+ "learning_rate": 0.00018749517094213375,
513
+ "loss": 1.0155,
514
  "step": 355
515
  },
516
  {
517
+ "epoch": 0.4984423676012461,
518
+ "grad_norm": 0.7187851071357727,
519
+ "learning_rate": 0.00018690326437187472,
520
+ "loss": 1.0361,
521
  "step": 360
522
  },
523
  {
524
+ "epoch": 0.5053651782623745,
525
+ "grad_norm": 0.7711403965950012,
526
+ "learning_rate": 0.00018629865053183911,
527
+ "loss": 1.0248,
 
 
 
 
 
 
 
 
528
  "step": 365
529
  },
530
  {
531
+ "epoch": 0.5122879889235029,
532
+ "grad_norm": 0.8515886068344116,
533
+ "learning_rate": 0.0001856814178305722,
534
+ "loss": 1.0077,
535
  "step": 370
536
  },
537
  {
538
+ "epoch": 0.5192107995846313,
539
+ "grad_norm": 0.8935133218765259,
540
+ "learning_rate": 0.00018505165652178893,
541
+ "loss": 0.9975,
542
  "step": 375
543
  },
544
  {
545
+ "epoch": 0.5261336102457598,
546
+ "grad_norm": 0.7678277492523193,
547
+ "learning_rate": 0.00018440945869117675,
548
+ "loss": 1.0241,
549
  "step": 380
550
  },
551
  {
552
+ "epoch": 0.5330564209068882,
553
+ "grad_norm": 0.8560886383056641,
554
+ "learning_rate": 0.00018375491824293067,
555
+ "loss": 1.0204,
556
  "step": 385
557
  },
558
  {
559
+ "epoch": 0.5399792315680166,
560
+ "grad_norm": 0.973434329032898,
561
+ "learning_rate": 0.00018308813088602227,
562
+ "loss": 1.0232,
563
  "step": 390
564
  },
565
  {
566
+ "epoch": 0.5469020422291451,
567
+ "grad_norm": 0.7952532172203064,
568
+ "learning_rate": 0.00018240919412020466,
569
+ "loss": 0.9945,
570
  "step": 395
571
  },
572
  {
573
+ "epoch": 0.5538248528902735,
574
+ "grad_norm": 0.7281926870346069,
575
+ "learning_rate": 0.00018171820722175604,
576
+ "loss": 1.0214,
577
  "step": 400
578
  },
579
  {
580
+ "epoch": 0.5607476635514018,
581
+ "grad_norm": 0.6621217131614685,
582
+ "learning_rate": 0.00018101527122896293,
583
+ "loss": 1.0157,
584
  "step": 405
585
  },
586
  {
587
+ "epoch": 0.5676704742125303,
588
+ "grad_norm": 0.6945295929908752,
589
+ "learning_rate": 0.0001803004889273463,
590
+ "loss": 0.9936,
591
  "step": 410
592
  },
593
  {
594
+ "epoch": 0.5745932848736587,
595
+ "grad_norm": 0.6977776885032654,
596
+ "learning_rate": 0.0001795739648346317,
597
+ "loss": 0.9647,
598
  "step": 415
599
  },
600
  {
601
+ "epoch": 0.5815160955347871,
602
+ "grad_norm": 0.7883012890815735,
603
+ "learning_rate": 0.00017883580518546647,
604
+ "loss": 0.9637,
605
  "step": 420
606
  },
607
  {
608
+ "epoch": 0.5884389061959155,
609
+ "grad_norm": 0.6024463176727295,
610
+ "learning_rate": 0.00017808611791588584,
611
+ "loss": 0.9791,
612
  "step": 425
613
  },
614
  {
615
+ "epoch": 0.595361716857044,
616
+ "grad_norm": 0.8087362051010132,
617
+ "learning_rate": 0.00017732501264752996,
618
+ "loss": 0.9714,
619
  "step": 430
620
  },
621
  {
622
+ "epoch": 0.6022845275181724,
623
+ "grad_norm": 0.6511737108230591,
624
+ "learning_rate": 0.00017655260067161503,
625
+ "loss": 0.9914,
626
  "step": 435
627
  },
628
  {
629
+ "epoch": 0.6092073381793008,
630
+ "grad_norm": 0.6464141607284546,
631
+ "learning_rate": 0.00017576899493265954,
632
+ "loss": 0.9766,
633
  "step": 440
634
  },
635
  {
636
+ "epoch": 0.6161301488404292,
637
+ "grad_norm": 0.6870825290679932,
638
+ "learning_rate": 0.00017497431001196943,
639
+ "loss": 0.9775,
640
  "step": 445
641
  },
642
  {
643
+ "epoch": 0.6230529595015576,
644
+ "grad_norm": 0.7621681094169617,
645
+ "learning_rate": 0.00017416866211088366,
646
+ "loss": 0.9823,
647
  "step": 450
648
  },
649
  {
650
+ "epoch": 0.629975770162686,
651
+ "grad_norm": 0.7556800842285156,
652
+ "learning_rate": 0.00017335216903378267,
653
+ "loss": 0.9605,
654
  "step": 455
655
  },
656
  {
657
+ "epoch": 0.6368985808238145,
658
+ "grad_norm": 0.715800404548645,
659
+ "learning_rate": 0.00017252495017086296,
660
+ "loss": 0.9465,
661
  "step": 460
662
  },
663
  {
664
+ "epoch": 0.6438213914849429,
665
+ "grad_norm": 0.6417484283447266,
666
+ "learning_rate": 0.0001716871264806791,
667
+ "loss": 0.9631,
668
  "step": 465
669
  },
670
  {
671
+ "epoch": 0.6507442021460713,
672
+ "grad_norm": 0.7063789367675781,
673
+ "learning_rate": 0.0001708388204724572,
674
+ "loss": 0.9628,
675
  "step": 470
676
  },
677
  {
678
+ "epoch": 0.6576670128071997,
679
+ "grad_norm": 0.741763710975647,
680
+ "learning_rate": 0.00016998015618818087,
681
+ "loss": 0.9786,
682
  "step": 475
683
  },
684
  {
685
+ "epoch": 0.6645898234683282,
686
+ "grad_norm": 0.5884097814559937,
687
+ "learning_rate": 0.00016911125918445363,
688
+ "loss": 0.9563,
689
  "step": 480
690
  },
691
  {
692
+ "epoch": 0.6715126341294566,
693
+ "grad_norm": 0.5698336362838745,
694
+ "learning_rate": 0.00016823225651413953,
695
+ "loss": 0.9456,
696
  "step": 485
697
  },
698
  {
699
+ "epoch": 0.6784354447905849,
700
+ "grad_norm": 0.5873739719390869,
701
+ "learning_rate": 0.00016734327670778508,
702
+ "loss": 0.9356,
703
  "step": 490
704
  },
705
  {
706
+ "epoch": 0.6853582554517134,
707
+ "grad_norm": 0.6806433200836182,
708
+ "learning_rate": 0.00016644444975482523,
709
+ "loss": 0.9311,
710
  "step": 495
711
  },
712
  {
713
+ "epoch": 0.6922810661128418,
714
+ "grad_norm": 0.595560610294342,
715
+ "learning_rate": 0.0001655359070845757,
716
+ "loss": 0.9354,
717
  "step": 500
718
  },
719
  {
720
+ "epoch": 0.6992038767739702,
721
+ "grad_norm": 0.593410849571228,
722
+ "learning_rate": 0.0001646177815470151,
723
+ "loss": 0.9353,
724
  "step": 505
725
  },
726
  {
727
+ "epoch": 0.7061266874350987,
728
+ "grad_norm": 0.5478509068489075,
729
+ "learning_rate": 0.00016369020739335924,
730
+ "loss": 0.9195,
731
  "step": 510
732
  },
733
  {
734
+ "epoch": 0.7130494980962271,
735
+ "grad_norm": 0.5979152321815491,
736
+ "learning_rate": 0.00016275332025643028,
737
+ "loss": 0.9331,
738
  "step": 515
739
  },
740
  {
741
+ "epoch": 0.7199723087573555,
742
+ "grad_norm": 0.6606078147888184,
743
+ "learning_rate": 0.00016180725713082434,
744
+ "loss": 0.9369,
745
  "step": 520
746
  },
747
  {
748
+ "epoch": 0.726895119418484,
749
+ "grad_norm": 0.5818477869033813,
750
+ "learning_rate": 0.0001608521563528795,
751
+ "loss": 0.9204,
752
  "step": 525
753
  },
754
  {
755
+ "epoch": 0.7338179300796123,
756
+ "grad_norm": 0.6692835688591003,
757
+ "learning_rate": 0.00015988815758044792,
758
+ "loss": 0.9115,
759
  "step": 530
760
  },
761
  {
762
+ "epoch": 0.7407407407407407,
763
+ "grad_norm": 0.731097936630249,
764
+ "learning_rate": 0.00015891540177247467,
765
+ "loss": 0.9457,
766
  "step": 535
767
  },
768
  {
769
+ "epoch": 0.7476635514018691,
770
+ "grad_norm": 0.7018098831176758,
771
+ "learning_rate": 0.00015793403116838613,
772
+ "loss": 0.9256,
773
  "step": 540
774
  },
775
  {
776
+ "epoch": 0.7545863620629976,
777
+ "grad_norm": 0.6631230711936951,
778
+ "learning_rate": 0.00015694418926729146,
779
+ "loss": 0.912,
780
  "step": 545
781
  },
782
  {
783
+ "epoch": 0.761509172724126,
784
+ "grad_norm": 0.6354436874389648,
785
+ "learning_rate": 0.00015594602080699956,
786
+ "loss": 0.9141,
787
  "step": 550
788
  },
789
  {
790
+ "epoch": 0.7684319833852544,
791
+ "grad_norm": 0.7462336421012878,
792
+ "learning_rate": 0.00015493967174285512,
793
+ "loss": 0.9094,
794
  "step": 555
795
  },
796
  {
797
+ "epoch": 0.7753547940463829,
798
+ "grad_norm": 0.5586750507354736,
799
+ "learning_rate": 0.00015392528922639662,
800
+ "loss": 0.9188,
801
  "step": 560
802
  },
803
  {
804
+ "epoch": 0.7822776047075113,
805
+ "grad_norm": 0.5627267956733704,
806
+ "learning_rate": 0.00015290302158383923,
807
+ "loss": 0.9077,
808
  "step": 565
809
  },
810
  {
811
+ "epoch": 0.7892004153686397,
812
+ "grad_norm": 0.603247880935669,
813
+ "learning_rate": 0.0001518730182943862,
814
+ "loss": 0.9043,
815
  "step": 570
816
  },
817
  {
818
+ "epoch": 0.796123226029768,
819
+ "grad_norm": 0.5933719277381897,
820
+ "learning_rate": 0.0001508354299683715,
821
+ "loss": 0.9011,
822
  "step": 575
823
  },
824
  {
825
+ "epoch": 0.8030460366908965,
826
+ "grad_norm": 0.5786928534507751,
827
+ "learning_rate": 0.0001497904083252371,
828
+ "loss": 0.9074,
829
  "step": 580
830
  },
831
  {
832
+ "epoch": 0.8099688473520249,
833
+ "grad_norm": 0.591249942779541,
834
+ "learning_rate": 0.00014873810617134806,
835
+ "loss": 0.9001,
836
  "step": 585
837
  },
838
  {
839
+ "epoch": 0.8168916580131533,
840
+ "grad_norm": 0.5570294857025146,
841
+ "learning_rate": 0.00014767867737764876,
842
+ "loss": 0.8942,
843
  "step": 590
844
  },
845
  {
846
+ "epoch": 0.8238144686742818,
847
+ "grad_norm": 0.6055375337600708,
848
+ "learning_rate": 0.0001466122768571633,
849
+ "loss": 0.8992,
850
  "step": 595
851
  },
852
  {
853
+ "epoch": 0.8307372793354102,
854
+ "grad_norm": 0.5531699657440186,
855
+ "learning_rate": 0.0001455390605423437,
856
+ "loss": 0.9092,
857
  "step": 600
858
  },
859
  {
860
+ "epoch": 0.8376600899965386,
861
+ "grad_norm": 5.621354103088379,
862
+ "learning_rate": 0.00014445918536226904,
863
+ "loss": 0.8969,
864
  "step": 605
865
  },
866
  {
867
+ "epoch": 0.8445829006576671,
868
+ "grad_norm": 0.6026459336280823,
869
+ "learning_rate": 0.00014337280921969858,
870
+ "loss": 0.8898,
871
  "step": 610
872
  },
873
  {
874
+ "epoch": 0.8515057113187954,
875
+ "grad_norm": 0.5794016718864441,
876
+ "learning_rate": 0.0001422800909679829,
877
+ "loss": 0.8964,
878
  "step": 615
879
  },
880
  {
881
+ "epoch": 0.8584285219799238,
882
+ "grad_norm": 0.5771220922470093,
883
+ "learning_rate": 0.00014118119038783588,
884
+ "loss": 0.8905,
885
  "step": 620
886
  },
887
  {
888
+ "epoch": 0.8653513326410522,
889
+ "grad_norm": 0.6543847918510437,
890
+ "learning_rate": 0.00014007626816397087,
891
+ "loss": 0.9124,
892
  "step": 625
893
  },
894
  {
895
+ "epoch": 0.8722741433021807,
896
+ "grad_norm": 0.5902742147445679,
897
+ "learning_rate": 0.00013896548586160503,
898
+ "loss": 0.8901,
899
  "step": 630
900
  },
901
  {
902
+ "epoch": 0.8791969539633091,
903
+ "grad_norm": 0.6009337902069092,
904
+ "learning_rate": 0.00013784900590283473,
905
+ "loss": 0.8904,
906
  "step": 635
907
  },
908
  {
909
+ "epoch": 0.8861197646244375,
910
+ "grad_norm": 0.5576380491256714,
911
+ "learning_rate": 0.00013672699154288569,
912
+ "loss": 0.8768,
913
  "step": 640
914
  },
915
  {
916
+ "epoch": 0.893042575285566,
917
+ "grad_norm": 0.6333903074264526,
918
+ "learning_rate": 0.00013559960684624112,
919
+ "loss": 0.8993,
920
  "step": 645
921
  },
922
  {
923
+ "epoch": 0.8999653859466944,
924
+ "grad_norm": 0.6231611371040344,
925
+ "learning_rate": 0.0001344670166626519,
926
+ "loss": 0.8701,
927
  "step": 650
928
  },
929
  {
930
+ "epoch": 0.9068881966078228,
931
+ "grad_norm": 0.5712286829948425,
932
+ "learning_rate": 0.0001333293866030317,
933
+ "loss": 0.8729,
934
  "step": 655
935
  },
936
  {
937
+ "epoch": 0.9138110072689511,
938
+ "grad_norm": 0.5796388983726501,
939
+ "learning_rate": 0.0001321868830152407,
940
+ "loss": 0.8521,
941
  "step": 660
942
  },
943
  {
944
+ "epoch": 0.9207338179300796,
945
+ "grad_norm": 0.5513543486595154,
946
+ "learning_rate": 0.00013103967295976179,
947
+ "loss": 0.8777,
948
  "step": 665
949
  },
950
  {
951
+ "epoch": 0.927656628591208,
952
+ "grad_norm": 0.5434843897819519,
953
+ "learning_rate": 0.0001298879241852724,
954
+ "loss": 0.8689,
955
  "step": 670
956
  },
957
  {
958
+ "epoch": 0.9345794392523364,
959
+ "grad_norm": 0.5376566648483276,
960
+ "learning_rate": 0.0001287318051041159,
961
+ "loss": 0.8776,
962
  "step": 675
963
  },
964
  {
965
+ "epoch": 0.9415022499134649,
966
+ "grad_norm": 0.5433745384216309,
967
+ "learning_rate": 0.00012757148476767553,
968
+ "loss": 0.8643,
969
  "step": 680
970
  },
971
  {
972
+ "epoch": 0.9484250605745933,
973
+ "grad_norm": 0.6016582250595093,
974
+ "learning_rate": 0.0001264071328416555,
975
+ "loss": 0.8659,
976
  "step": 685
977
  },
978
  {
979
+ "epoch": 0.9553478712357217,
980
+ "grad_norm": 0.5252034664154053,
981
+ "learning_rate": 0.00012523891958127168,
982
+ "loss": 0.8675,
983
  "step": 690
984
  },
985
  {
986
+ "epoch": 0.9622706818968502,
987
+ "grad_norm": 0.5640143752098083,
988
+ "learning_rate": 0.0001240670158063565,
989
+ "loss": 0.8582,
990
  "step": 695
991
  },
992
  {
993
+ "epoch": 0.9691934925579785,
994
+ "grad_norm": 0.5769084095954895,
995
+ "learning_rate": 0.00012289159287638105,
996
+ "loss": 0.8686,
997
  "step": 700
998
  },
999
  {
1000
+ "epoch": 0.9761163032191069,
1001
+ "grad_norm": 0.5791013240814209,
1002
+ "learning_rate": 0.00012171282266539845,
1003
+ "loss": 0.8651,
1004
  "step": 705
1005
  },
1006
  {
1007
+ "epoch": 0.9830391138802353,
1008
+ "grad_norm": 0.5205267071723938,
1009
+ "learning_rate": 0.00012053087753691172,
1010
+ "loss": 0.8487,
1011
  "step": 710
1012
  },
1013
  {
1014
+ "epoch": 0.9899619245413638,
1015
+ "grad_norm": 0.6154927611351013,
1016
+ "learning_rate": 0.00011934593031867038,
1017
+ "loss": 0.8658,
1018
  "step": 715
1019
  },
1020
  {
1021
+ "epoch": 0.9968847352024922,
1022
+ "grad_norm": 0.5034484267234802,
1023
+ "learning_rate": 0.00011815815427739911,
1024
+ "loss": 0.8655,
1025
  "step": 720
1026
  },
1027
  {
1028
+ "epoch": 1.0,
1029
+ "eval_loss": 0.8625231385231018,
1030
+ "eval_runtime": 54.6603,
1031
+ "eval_samples_per_second": 37.852,
1032
+ "eval_steps_per_second": 2.378,
1033
+ "step": 723
1034
+ },
1035
+ {
1036
+ "epoch": 1.0027691242644514,
1037
+ "grad_norm": 0.5045917630195618,
1038
+ "learning_rate": 0.00011696772309346182,
1039
+ "loss": 0.7816,
1040
+ "step": 725
1041
+ },
1042
+ {
1043
+ "epoch": 1.0096919349255797,
1044
+ "grad_norm": 0.6295811533927917,
1045
+ "learning_rate": 0.0001157748108354659,
1046
+ "loss": 0.6759,
1047
+ "step": 730
1048
+ },
1049
+ {
1050
+ "epoch": 1.0166147455867083,
1051
+ "grad_norm": 0.6809110045433044,
1052
+ "learning_rate": 0.00011457959193480925,
1053
+ "loss": 0.6842,
1054
+ "step": 735
1055
+ },
1056
+ {
1057
+ "epoch": 1.0235375562478366,
1058
+ "grad_norm": 0.5183382630348206,
1059
+ "learning_rate": 0.00011338224116017423,
1060
+ "loss": 0.6853,
1061
+ "step": 740
1062
+ },
1063
+ {
1064
+ "epoch": 1.0304603669089651,
1065
+ "grad_norm": 0.4994449317455292,
1066
+ "learning_rate": 0.00011218293359197267,
1067
+ "loss": 0.6822,
1068
+ "step": 745
1069
+ },
1070
+ {
1071
+ "epoch": 1.0373831775700935,
1072
+ "grad_norm": 0.5263209342956543,
1073
+ "learning_rate": 0.00011098184459674484,
1074
+ "loss": 0.6475,
1075
+ "step": 750
1076
+ },
1077
+ {
1078
+ "epoch": 1.0443059882312218,
1079
+ "grad_norm": 0.5542846322059631,
1080
+ "learning_rate": 0.00010977914980151691,
1081
+ "loss": 0.6848,
1082
+ "step": 755
1083
+ },
1084
+ {
1085
+ "epoch": 1.0512287988923503,
1086
+ "grad_norm": 0.539220929145813,
1087
+ "learning_rate": 0.00010857502506812029,
1088
+ "loss": 0.6583,
1089
+ "step": 760
1090
+ },
1091
+ {
1092
+ "epoch": 1.0581516095534786,
1093
+ "grad_norm": 0.5188484191894531,
1094
+ "learning_rate": 0.00010736964646747629,
1095
+ "loss": 0.6609,
1096
+ "step": 765
1097
+ },
1098
+ {
1099
+ "epoch": 1.0650744202146072,
1100
+ "grad_norm": 0.5194864869117737,
1101
+ "learning_rate": 0.00010616319025385089,
1102
+ "loss": 0.6779,
1103
+ "step": 770
1104
+ },
1105
+ {
1106
+ "epoch": 1.0719972308757355,
1107
+ "grad_norm": 0.532119870185852,
1108
+ "learning_rate": 0.00010495583283908177,
1109
+ "loss": 0.6696,
1110
+ "step": 775
1111
+ },
1112
+ {
1113
+ "epoch": 1.078920041536864,
1114
+ "grad_norm": 0.5120254755020142,
1115
+ "learning_rate": 0.00010374775076678332,
1116
+ "loss": 0.6471,
1117
+ "step": 780
1118
+ },
1119
+ {
1120
+ "epoch": 1.0858428521979924,
1121
+ "grad_norm": 0.5310303568840027,
1122
+ "learning_rate": 0.00010253912068653146,
1123
+ "loss": 0.6754,
1124
+ "step": 785
1125
+ },
1126
+ {
1127
+ "epoch": 1.092765662859121,
1128
+ "grad_norm": 0.4805733859539032,
1129
+ "learning_rate": 0.00010133011932803378,
1130
+ "loss": 0.6555,
1131
+ "step": 790
1132
+ },
1133
+ {
1134
+ "epoch": 1.0996884735202492,
1135
+ "grad_norm": 0.5308163166046143,
1136
+ "learning_rate": 0.00010012092347528705,
1137
+ "loss": 0.6753,
1138
+ "step": 795
1139
+ },
1140
+ {
1141
+ "epoch": 1.1066112841813776,
1142
+ "grad_norm": 0.5158370137214661,
1143
+ "learning_rate": 9.891170994072793e-05,
1144
+ "loss": 0.6559,
1145
+ "step": 800
1146
+ },
1147
+ {
1148
+ "epoch": 1.113534094842506,
1149
+ "grad_norm": 0.5924445390701294,
1150
+ "learning_rate": 9.770265553937831e-05,
1151
+ "loss": 0.6539,
1152
+ "step": 805
1153
+ },
1154
+ {
1155
+ "epoch": 1.1204569055036344,
1156
+ "grad_norm": 0.5329247713088989,
1157
+ "learning_rate": 9.649393706299127e-05,
1158
+ "loss": 0.6444,
1159
+ "step": 810
1160
+ },
1161
+ {
1162
+ "epoch": 1.127379716164763,
1163
+ "grad_norm": 0.544358491897583,
1164
+ "learning_rate": 9.52857312541998e-05,
1165
+ "loss": 0.6558,
1166
+ "step": 815
1167
+ },
1168
+ {
1169
+ "epoch": 1.1343025268258913,
1170
+ "grad_norm": 0.4904942214488983,
1171
+ "learning_rate": 9.407821478067304e-05,
1172
+ "loss": 0.6471,
1173
+ "step": 820
1174
+ },
1175
+ {
1176
+ "epoch": 1.1412253374870198,
1177
+ "grad_norm": 0.5689169764518738,
1178
+ "learning_rate": 9.287156420928342e-05,
1179
+ "loss": 0.6513,
1180
+ "step": 825
1181
+ },
1182
+ {
1183
+ "epoch": 1.1481481481481481,
1184
+ "grad_norm": 0.5149237513542175,
1185
+ "learning_rate": 9.166595598028832e-05,
1186
+ "loss": 0.6663,
1187
+ "step": 830
1188
+ },
1189
+ {
1190
+ "epoch": 1.1550709588092767,
1191
+ "grad_norm": 0.5315957069396973,
1192
+ "learning_rate": 9.046156638153056e-05,
1193
+ "loss": 0.6504,
1194
+ "step": 835
1195
+ },
1196
+ {
1197
+ "epoch": 1.161993769470405,
1198
+ "grad_norm": 0.4693305492401123,
1199
+ "learning_rate": 8.925857152266103e-05,
1200
+ "loss": 0.6503,
1201
+ "step": 840
1202
+ },
1203
+ {
1204
+ "epoch": 1.1689165801315333,
1205
+ "grad_norm": 0.5185849666595459,
1206
+ "learning_rate": 8.805714730938728e-05,
1207
+ "loss": 0.6598,
1208
+ "step": 845
1209
+ },
1210
+ {
1211
+ "epoch": 1.1758393907926619,
1212
+ "grad_norm": 0.5028591156005859,
1213
+ "learning_rate": 8.685746941775219e-05,
1214
+ "loss": 0.6461,
1215
+ "step": 850
1216
+ },
1217
+ {
1218
+ "epoch": 1.1827622014537902,
1219
+ "grad_norm": 0.5199161767959595,
1220
+ "learning_rate": 8.565971326844584e-05,
1221
+ "loss": 0.6468,
1222
+ "step": 855
1223
+ },
1224
+ {
1225
+ "epoch": 1.1896850121149187,
1226
+ "grad_norm": 0.5406040549278259,
1227
+ "learning_rate": 8.44640540011553e-05,
1228
+ "loss": 0.6497,
1229
+ "step": 860
1230
+ },
1231
+ {
1232
+ "epoch": 1.196607822776047,
1233
+ "grad_norm": 0.4970882833003998,
1234
+ "learning_rate": 8.327066644895469e-05,
1235
+ "loss": 0.6542,
1236
+ "step": 865
1237
+ },
1238
+ {
1239
+ "epoch": 1.2035306334371756,
1240
+ "grad_norm": 0.5521699786186218,
1241
+ "learning_rate": 8.207972511274095e-05,
1242
+ "loss": 0.6413,
1243
+ "step": 870
1244
+ },
1245
+ {
1246
+ "epoch": 1.210453444098304,
1247
+ "grad_norm": 0.473796010017395,
1248
+ "learning_rate": 8.089140413571747e-05,
1249
+ "loss": 0.6411,
1250
+ "step": 875
1251
+ },
1252
+ {
1253
+ "epoch": 1.2173762547594325,
1254
+ "grad_norm": 0.4899975061416626,
1255
+ "learning_rate": 7.970587727793051e-05,
1256
+ "loss": 0.637,
1257
+ "step": 880
1258
+ },
1259
+ {
1260
+ "epoch": 1.2242990654205608,
1261
+ "grad_norm": 0.4936094284057617,
1262
+ "learning_rate": 7.852331789086135e-05,
1263
+ "loss": 0.6401,
1264
+ "step": 885
1265
+ },
1266
+ {
1267
+ "epoch": 1.231221876081689,
1268
+ "grad_norm": 0.5000462532043457,
1269
+ "learning_rate": 7.73438988920784e-05,
1270
+ "loss": 0.6482,
1271
+ "step": 890
1272
+ },
1273
+ {
1274
+ "epoch": 1.2381446867428176,
1275
+ "grad_norm": 0.5123869180679321,
1276
+ "learning_rate": 7.616779273995252e-05,
1277
+ "loss": 0.6492,
1278
+ "step": 895
1279
+ },
1280
+ {
1281
+ "epoch": 1.245067497403946,
1282
+ "grad_norm": 0.48594337701797485,
1283
+ "learning_rate": 7.499517140843968e-05,
1284
+ "loss": 0.6448,
1285
+ "step": 900
1286
+ },
1287
+ {
1288
+ "epoch": 1.2519903080650745,
1289
+ "grad_norm": 0.4969227910041809,
1290
+ "learning_rate": 7.382620636193438e-05,
1291
+ "loss": 0.636,
1292
+ "step": 905
1293
+ },
1294
+ {
1295
+ "epoch": 1.2589131187262028,
1296
+ "grad_norm": 0.4929391145706177,
1297
+ "learning_rate": 7.266106853019759e-05,
1298
+ "loss": 0.636,
1299
+ "step": 910
1300
+ },
1301
+ {
1302
+ "epoch": 1.2658359293873311,
1303
+ "grad_norm": 0.48114389181137085,
1304
+ "learning_rate": 7.149992828336274e-05,
1305
+ "loss": 0.6399,
1306
+ "step": 915
1307
+ },
1308
+ {
1309
+ "epoch": 1.2727587400484597,
1310
+ "grad_norm": 0.5209516286849976,
1311
+ "learning_rate": 7.034295540702397e-05,
1312
+ "loss": 0.6329,
1313
+ "step": 920
1314
+ },
1315
+ {
1316
+ "epoch": 1.2796815507095882,
1317
+ "grad_norm": 0.4901733994483948,
1318
+ "learning_rate": 6.919031907740922e-05,
1319
+ "loss": 0.6543,
1320
+ "step": 925
1321
+ },
1322
+ {
1323
+ "epoch": 1.2866043613707165,
1324
+ "grad_norm": 0.5102095603942871,
1325
+ "learning_rate": 6.804218783664288e-05,
1326
+ "loss": 0.6462,
1327
+ "step": 930
1328
+ },
1329
+ {
1330
+ "epoch": 1.2935271720318449,
1331
+ "grad_norm": 0.46668604016304016,
1332
+ "learning_rate": 6.689872956810102e-05,
1333
+ "loss": 0.6346,
1334
+ "step": 935
1335
+ },
1336
+ {
1337
+ "epoch": 1.3004499826929734,
1338
+ "grad_norm": 0.48058241605758667,
1339
+ "learning_rate": 6.5760111471863e-05,
1340
+ "loss": 0.6302,
1341
+ "step": 940
1342
+ },
1343
+ {
1344
+ "epoch": 1.3073727933541017,
1345
+ "grad_norm": 0.5134970545768738,
1346
+ "learning_rate": 6.46265000402628e-05,
1347
+ "loss": 0.6389,
1348
+ "step": 945
1349
+ },
1350
+ {
1351
+ "epoch": 1.3142956040152303,
1352
+ "grad_norm": 0.49927303194999695,
1353
+ "learning_rate": 6.349806103354417e-05,
1354
+ "loss": 0.6265,
1355
+ "step": 950
1356
+ },
1357
+ {
1358
+ "epoch": 1.3212184146763586,
1359
+ "grad_norm": 0.5016152858734131,
1360
+ "learning_rate": 6.23749594556226e-05,
1361
+ "loss": 0.6255,
1362
+ "step": 955
1363
+ },
1364
+ {
1365
+ "epoch": 1.328141225337487,
1366
+ "grad_norm": 0.4818898141384125,
1367
+ "learning_rate": 6.125735952995775e-05,
1368
+ "loss": 0.6299,
1369
+ "step": 960
1370
+ },
1371
+ {
1372
+ "epoch": 1.3350640359986155,
1373
+ "grad_norm": 0.5072540044784546,
1374
+ "learning_rate": 6.0145424675540394e-05,
1375
+ "loss": 0.6256,
1376
+ "step": 965
1377
+ },
1378
+ {
1379
+ "epoch": 1.341986846659744,
1380
+ "grad_norm": 0.47364550828933716,
1381
+ "learning_rate": 5.9039317482996735e-05,
1382
+ "loss": 0.6369,
1383
+ "step": 970
1384
+ },
1385
+ {
1386
+ "epoch": 1.3489096573208723,
1387
+ "grad_norm": 0.4561605453491211,
1388
+ "learning_rate": 5.793919969081374e-05,
1389
+ "loss": 0.6136,
1390
+ "step": 975
1391
+ },
1392
+ {
1393
+ "epoch": 1.3558324679820006,
1394
+ "grad_norm": 0.5107749700546265,
1395
+ "learning_rate": 5.68452321616894e-05,
1396
+ "loss": 0.6104,
1397
+ "step": 980
1398
+ },
1399
+ {
1400
+ "epoch": 1.3627552786431292,
1401
+ "grad_norm": 0.4962555170059204,
1402
+ "learning_rate": 5.5757574859010886e-05,
1403
+ "loss": 0.6367,
1404
+ "step": 985
1405
+ },
1406
+ {
1407
+ "epoch": 1.3696780893042575,
1408
+ "grad_norm": 0.5515092015266418,
1409
+ "learning_rate": 5.467638682346403e-05,
1410
+ "loss": 0.6087,
1411
+ "step": 990
1412
+ },
1413
+ {
1414
+ "epoch": 1.376600899965386,
1415
+ "grad_norm": 0.5041881799697876,
1416
+ "learning_rate": 5.3601826149777966e-05,
1417
+ "loss": 0.6424,
1418
+ "step": 995
1419
+ },
1420
+ {
1421
+ "epoch": 1.3835237106265144,
1422
+ "grad_norm": 0.4918469488620758,
1423
+ "learning_rate": 5.25340499636082e-05,
1424
+ "loss": 0.6283,
1425
+ "step": 1000
1426
+ },
1427
+ {
1428
+ "epoch": 1.3904465212876427,
1429
+ "grad_norm": 0.5199057459831238,
1430
+ "learning_rate": 5.147321439856091e-05,
1431
+ "loss": 0.6248,
1432
+ "step": 1005
1433
+ },
1434
+ {
1435
+ "epoch": 1.3973693319487712,
1436
+ "grad_norm": 0.46590614318847656,
1437
+ "learning_rate": 5.041947457336274e-05,
1438
+ "loss": 0.6309,
1439
+ "step": 1010
1440
+ },
1441
+ {
1442
+ "epoch": 1.4042921426098995,
1443
+ "grad_norm": 0.45809245109558105,
1444
+ "learning_rate": 4.937298456917889e-05,
1445
+ "loss": 0.6095,
1446
+ "step": 1015
1447
+ },
1448
+ {
1449
+ "epoch": 1.411214953271028,
1450
+ "grad_norm": 0.5258966088294983,
1451
+ "learning_rate": 4.8333897407082896e-05,
1452
+ "loss": 0.6133,
1453
+ "step": 1020
1454
+ },
1455
+ {
1456
+ "epoch": 1.4181377639321564,
1457
+ "grad_norm": 0.4614286720752716,
1458
+ "learning_rate": 4.7302365025681206e-05,
1459
+ "loss": 0.623,
1460
+ "step": 1025
1461
+ },
1462
+ {
1463
+ "epoch": 1.425060574593285,
1464
+ "grad_norm": 0.493114173412323,
1465
+ "learning_rate": 4.627853825889642e-05,
1466
+ "loss": 0.6017,
1467
+ "step": 1030
1468
+ },
1469
+ {
1470
+ "epoch": 1.4319833852544133,
1471
+ "grad_norm": 0.481393039226532,
1472
+ "learning_rate": 4.526256681391192e-05,
1473
+ "loss": 0.616,
1474
+ "step": 1035
1475
+ },
1476
+ {
1477
+ "epoch": 1.4389061959155418,
1478
+ "grad_norm": 0.5155964493751526,
1479
+ "learning_rate": 4.4254599249281016e-05,
1480
+ "loss": 0.5972,
1481
+ "step": 1040
1482
+ },
1483
+ {
1484
+ "epoch": 1.4458290065766701,
1485
+ "grad_norm": 0.5064235925674438,
1486
+ "learning_rate": 4.3254782953204375e-05,
1487
+ "loss": 0.6093,
1488
+ "step": 1045
1489
+ },
1490
+ {
1491
+ "epoch": 1.4527518172377984,
1492
+ "grad_norm": 0.4666582942008972,
1493
+ "learning_rate": 4.2263264121978505e-05,
1494
+ "loss": 0.6126,
1495
+ "step": 1050
1496
+ },
1497
+ {
1498
+ "epoch": 1.459674627898927,
1499
+ "grad_norm": 0.5063011646270752,
1500
+ "learning_rate": 4.12801877386183e-05,
1501
+ "loss": 0.6026,
1502
+ "step": 1055
1503
+ },
1504
+ {
1505
+ "epoch": 1.4665974385600553,
1506
+ "grad_norm": 0.47389891743659973,
1507
+ "learning_rate": 4.030569755165736e-05,
1508
+ "loss": 0.6116,
1509
+ "step": 1060
1510
+ },
1511
+ {
1512
+ "epoch": 1.4735202492211839,
1513
+ "grad_norm": 0.46991026401519775,
1514
+ "learning_rate": 3.9339936054128466e-05,
1515
+ "loss": 0.6106,
1516
+ "step": 1065
1517
+ },
1518
+ {
1519
+ "epoch": 1.4804430598823122,
1520
+ "grad_norm": 0.49021977186203003,
1521
+ "learning_rate": 3.8383044462727826e-05,
1522
+ "loss": 0.5915,
1523
+ "step": 1070
1524
+ },
1525
+ {
1526
+ "epoch": 1.4873658705434407,
1527
+ "grad_norm": 0.5076444149017334,
1528
+ "learning_rate": 3.7435162697166107e-05,
1529
+ "loss": 0.6187,
1530
+ "step": 1075
1531
+ },
1532
+ {
1533
+ "epoch": 1.494288681204569,
1534
+ "grad_norm": 0.4995920956134796,
1535
+ "learning_rate": 3.649642935970859e-05,
1536
+ "loss": 0.6204,
1537
+ "step": 1080
1538
+ },
1539
+ {
1540
+ "epoch": 1.5012114918656976,
1541
+ "grad_norm": 0.4842764735221863,
1542
+ "learning_rate": 3.556698171490871e-05,
1543
+ "loss": 0.6013,
1544
+ "step": 1085
1545
+ },
1546
+ {
1547
+ "epoch": 1.508134302526826,
1548
+ "grad_norm": 0.48687830567359924,
1549
+ "learning_rate": 3.464695566953644e-05,
1550
+ "loss": 0.5962,
1551
+ "step": 1090
1552
+ },
1553
+ {
1554
+ "epoch": 1.5150571131879542,
1555
+ "grad_norm": 0.5043684840202332,
1556
+ "learning_rate": 3.373648575270576e-05,
1557
+ "loss": 0.6021,
1558
+ "step": 1095
1559
+ },
1560
+ {
1561
+ "epoch": 1.5219799238490828,
1562
+ "grad_norm": 0.4804685711860657,
1563
+ "learning_rate": 3.283570509620344e-05,
1564
+ "loss": 0.606,
1565
+ "step": 1100
1566
+ },
1567
+ {
1568
+ "epoch": 1.5289027345102113,
1569
+ "grad_norm": 0.4706571400165558,
1570
+ "learning_rate": 3.1944745415021915e-05,
1571
+ "loss": 0.607,
1572
+ "step": 1105
1573
+ },
1574
+ {
1575
+ "epoch": 1.5358255451713396,
1576
+ "grad_norm": 0.43756231665611267,
1577
+ "learning_rate": 3.106373698809958e-05,
1578
+ "loss": 0.6149,
1579
+ "step": 1110
1580
+ },
1581
+ {
1582
+ "epoch": 1.542748355832468,
1583
+ "grad_norm": 0.47284871339797974,
1584
+ "learning_rate": 3.0192808639271065e-05,
1585
+ "loss": 0.6173,
1586
+ "step": 1115
1587
+ },
1588
+ {
1589
+ "epoch": 1.5496711664935963,
1590
+ "grad_norm": 0.46921491622924805,
1591
+ "learning_rate": 2.933208771843018e-05,
1592
+ "loss": 0.6036,
1593
+ "step": 1120
1594
+ },
1595
+ {
1596
+ "epoch": 1.5565939771547248,
1597
+ "grad_norm": 0.4850091338157654,
1598
+ "learning_rate": 2.8481700082908268e-05,
1599
+ "loss": 0.599,
1600
+ "step": 1125
1601
+ },
1602
+ {
1603
+ "epoch": 1.5635167878158533,
1604
+ "grad_norm": 0.48241791129112244,
1605
+ "learning_rate": 2.7641770079071127e-05,
1606
+ "loss": 0.6,
1607
+ "step": 1130
1608
+ },
1609
+ {
1610
+ "epoch": 1.5704395984769817,
1611
+ "grad_norm": 0.4461960792541504,
1612
+ "learning_rate": 2.681242052413656e-05,
1613
+ "loss": 0.5818,
1614
+ "step": 1135
1615
+ },
1616
+ {
1617
+ "epoch": 1.57736240913811,
1618
+ "grad_norm": 0.49775242805480957,
1619
+ "learning_rate": 2.59937726882158e-05,
1620
+ "loss": 0.5829,
1621
+ "step": 1140
1622
+ },
1623
+ {
1624
+ "epoch": 1.5842852197992385,
1625
+ "grad_norm": 0.5127538442611694,
1626
+ "learning_rate": 2.5185946276580918e-05,
1627
+ "loss": 0.6164,
1628
+ "step": 1145
1629
+ },
1630
+ {
1631
+ "epoch": 1.591208030460367,
1632
+ "grad_norm": 0.47250741720199585,
1633
+ "learning_rate": 2.4389059412161087e-05,
1634
+ "loss": 0.5978,
1635
+ "step": 1150
1636
+ },
1637
+ {
1638
+ "epoch": 1.5981308411214954,
1639
+ "grad_norm": 0.4654916524887085,
1640
+ "learning_rate": 2.3603228618270356e-05,
1641
+ "loss": 0.5952,
1642
+ "step": 1155
1643
+ },
1644
+ {
1645
+ "epoch": 1.6050536517826237,
1646
+ "grad_norm": 0.45718511939048767,
1647
+ "learning_rate": 2.2828568801569283e-05,
1648
+ "loss": 0.5719,
1649
+ "step": 1160
1650
+ },
1651
+ {
1652
+ "epoch": 1.611976462443752,
1653
+ "grad_norm": 0.443266361951828,
1654
+ "learning_rate": 2.2065193235262782e-05,
1655
+ "loss": 0.5734,
1656
+ "step": 1165
1657
+ },
1658
+ {
1659
+ "epoch": 1.6188992731048806,
1660
+ "grad_norm": 0.46032777428627014,
1661
+ "learning_rate": 2.1313213542537058e-05,
1662
+ "loss": 0.5949,
1663
+ "step": 1170
1664
+ },
1665
+ {
1666
+ "epoch": 1.6258220837660091,
1667
+ "grad_norm": 0.4900462329387665,
1668
+ "learning_rate": 2.0572739680237717e-05,
1669
+ "loss": 0.6047,
1670
+ "step": 1175
1671
+ },
1672
+ {
1673
+ "epoch": 1.6327448944271374,
1674
+ "grad_norm": 0.46509256958961487,
1675
+ "learning_rate": 1.984387992279153e-05,
1676
+ "loss": 0.6182,
1677
+ "step": 1180
1678
+ },
1679
+ {
1680
+ "epoch": 1.6396677050882658,
1681
+ "grad_norm": 0.4681301712989807,
1682
+ "learning_rate": 1.912674084637409e-05,
1683
+ "loss": 0.5983,
1684
+ "step": 1185
1685
+ },
1686
+ {
1687
+ "epoch": 1.6465905157493943,
1688
+ "grad_norm": 0.4606594145298004,
1689
+ "learning_rate": 1.8421427313326046e-05,
1690
+ "loss": 0.5796,
1691
+ "step": 1190
1692
+ },
1693
+ {
1694
+ "epoch": 1.6535133264105226,
1695
+ "grad_norm": 0.47154656052589417,
1696
+ "learning_rate": 1.7728042456819704e-05,
1697
+ "loss": 0.5992,
1698
+ "step": 1195
1699
+ },
1700
+ {
1701
+ "epoch": 1.6604361370716512,
1702
+ "grad_norm": 0.4498773515224457,
1703
+ "learning_rate": 1.7046687665778717e-05,
1704
+ "loss": 0.5811,
1705
+ "step": 1200
1706
+ },
1707
+ {
1708
+ "epoch": 1.6673589477327795,
1709
+ "grad_norm": 0.45348694920539856,
1710
+ "learning_rate": 1.6377462570052438e-05,
1711
+ "loss": 0.5784,
1712
+ "step": 1205
1713
+ },
1714
+ {
1715
+ "epoch": 1.6742817583939078,
1716
+ "grad_norm": 0.536761999130249,
1717
+ "learning_rate": 1.5720465025847987e-05,
1718
+ "loss": 0.5861,
1719
+ "step": 1210
1720
+ },
1721
+ {
1722
+ "epoch": 1.6812045690550363,
1723
+ "grad_norm": 0.4612574577331543,
1724
+ "learning_rate": 1.5075791101421166e-05,
1725
+ "loss": 0.5904,
1726
+ "step": 1215
1727
+ },
1728
+ {
1729
+ "epoch": 1.6881273797161649,
1730
+ "grad_norm": 0.47577938437461853,
1731
+ "learning_rate": 1.4443535063029279e-05,
1732
+ "loss": 0.593,
1733
+ "step": 1220
1734
+ },
1735
+ {
1736
+ "epoch": 1.6950501903772932,
1737
+ "grad_norm": 0.4678874611854553,
1738
+ "learning_rate": 1.3823789361147044e-05,
1739
+ "loss": 0.5761,
1740
+ "step": 1225
1741
+ },
1742
+ {
1743
+ "epoch": 1.7019730010384215,
1744
+ "grad_norm": 0.4748591482639313,
1745
+ "learning_rate": 1.3216644616948282e-05,
1746
+ "loss": 0.5878,
1747
+ "step": 1230
1748
+ },
1749
+ {
1750
+ "epoch": 1.70889581169955,
1751
+ "grad_norm": 0.46718382835388184,
1752
+ "learning_rate": 1.2622189609054979e-05,
1753
+ "loss": 0.5748,
1754
+ "step": 1235
1755
+ },
1756
+ {
1757
+ "epoch": 1.7158186223606784,
1758
+ "grad_norm": 0.49095532298088074,
1759
+ "learning_rate": 1.2040511260555831e-05,
1760
+ "loss": 0.5797,
1761
+ "step": 1240
1762
+ },
1763
+ {
1764
+ "epoch": 1.722741433021807,
1765
+ "grad_norm": 0.4614841341972351,
1766
+ "learning_rate": 1.1471694626296049e-05,
1767
+ "loss": 0.5787,
1768
+ "step": 1245
1769
+ },
1770
+ {
1771
+ "epoch": 1.7296642436829353,
1772
+ "grad_norm": 0.45303580164909363,
1773
+ "learning_rate": 1.0915822880440308e-05,
1774
+ "loss": 0.5796,
1775
+ "step": 1250
1776
+ },
1777
+ {
1778
+ "epoch": 1.7365870543440636,
1779
+ "grad_norm": 0.46378132700920105,
1780
+ "learning_rate": 1.037297730431085e-05,
1781
+ "loss": 0.573,
1782
+ "step": 1255
1783
+ },
1784
+ {
1785
+ "epoch": 1.7435098650051921,
1786
+ "grad_norm": 0.45119667053222656,
1787
+ "learning_rate": 9.843237274502282e-06,
1788
+ "loss": 0.5825,
1789
+ "step": 1260
1790
+ },
1791
+ {
1792
+ "epoch": 1.7504326756663207,
1793
+ "grad_norm": 0.47857195138931274,
1794
+ "learning_rate": 9.326680251274777e-06,
1795
+ "loss": 0.5904,
1796
+ "step": 1265
1797
+ },
1798
+ {
1799
+ "epoch": 1.757355486327449,
1800
+ "grad_norm": 0.4850415885448456,
1801
+ "learning_rate": 8.823381767227667e-06,
1802
+ "loss": 0.5877,
1803
+ "step": 1270
1804
+ },
1805
+ {
1806
+ "epoch": 1.7642782969885773,
1807
+ "grad_norm": 0.4598202407360077,
1808
+ "learning_rate": 8.333415416254831e-06,
1809
+ "loss": 0.5749,
1810
+ "step": 1275
1811
+ },
1812
+ {
1813
+ "epoch": 1.7712011076497056,
1814
+ "grad_norm": 0.4723171293735504,
1815
+ "learning_rate": 7.856852842783547e-06,
1816
+ "loss": 0.5721,
1817
+ "step": 1280
1818
+ },
1819
+ {
1820
+ "epoch": 1.7781239183108342,
1821
+ "grad_norm": 0.4513912498950958,
1822
+ "learning_rate": 7.393763731298387e-06,
1823
+ "loss": 0.5689,
1824
+ "step": 1285
1825
+ },
1826
+ {
1827
+ "epoch": 1.7850467289719627,
1828
+ "grad_norm": 0.4459558427333832,
1829
+ "learning_rate": 6.944215796151765e-06,
1830
+ "loss": 0.5726,
1831
+ "step": 1290
1832
+ },
1833
+ {
1834
+ "epoch": 1.791969539633091,
1835
+ "grad_norm": 0.4863142669200897,
1836
+ "learning_rate": 6.5082747716625255e-06,
1837
+ "loss": 0.5853,
1838
+ "step": 1295
1839
+ },
1840
+ {
1841
+ "epoch": 1.7988923502942193,
1842
+ "grad_norm": 0.44126665592193604,
1843
+ "learning_rate": 6.086004402504098e-06,
1844
+ "loss": 0.5599,
1845
+ "step": 1300
1846
+ },
1847
+ {
1848
+ "epoch": 1.8058151609553479,
1849
+ "grad_norm": 0.46068254113197327,
1850
+ "learning_rate": 5.6774664343834495e-06,
1851
+ "loss": 0.5939,
1852
+ "step": 1305
1853
+ },
1854
+ {
1855
+ "epoch": 1.8127379716164764,
1856
+ "grad_norm": 0.46407076716423035,
1857
+ "learning_rate": 5.2827206050125585e-06,
1858
+ "loss": 0.5782,
1859
+ "step": 1310
1860
+ },
1861
+ {
1862
+ "epoch": 1.8196607822776047,
1863
+ "grad_norm": 0.4844149649143219,
1864
+ "learning_rate": 4.9018246353732755e-06,
1865
+ "loss": 0.5958,
1866
+ "step": 1315
1867
+ },
1868
+ {
1869
+ "epoch": 1.826583592938733,
1870
+ "grad_norm": 0.44412821531295776,
1871
+ "learning_rate": 4.534834221277262e-06,
1872
+ "loss": 0.568,
1873
+ "step": 1320
1874
+ },
1875
+ {
1876
+ "epoch": 1.8335064035998614,
1877
+ "grad_norm": 0.4620952308177948,
1878
+ "learning_rate": 4.181803025221898e-06,
1879
+ "loss": 0.5797,
1880
+ "step": 1325
1881
+ },
1882
+ {
1883
+ "epoch": 1.84042921426099,
1884
+ "grad_norm": 0.46902209520339966,
1885
+ "learning_rate": 3.842782668543599e-06,
1886
+ "loss": 0.5784,
1887
+ "step": 1330
1888
+ },
1889
+ {
1890
+ "epoch": 1.8473520249221185,
1891
+ "grad_norm": 0.4106810390949249,
1892
+ "learning_rate": 3.5178227238696347e-06,
1893
+ "loss": 0.5638,
1894
+ "step": 1335
1895
+ },
1896
+ {
1897
+ "epoch": 1.8542748355832468,
1898
+ "grad_norm": 0.47497299313545227,
1899
+ "learning_rate": 3.2069707078694057e-06,
1900
+ "loss": 0.5582,
1901
+ "step": 1340
1902
+ },
1903
+ {
1904
+ "epoch": 1.8611976462443751,
1905
+ "grad_norm": 0.44151052832603455,
1906
+ "learning_rate": 2.9102720743064527e-06,
1907
+ "loss": 0.5876,
1908
+ "step": 1345
1909
+ },
1910
+ {
1911
+ "epoch": 1.8681204569055037,
1912
+ "grad_norm": 0.4456900656223297,
1913
+ "learning_rate": 2.6277702073919997e-06,
1914
+ "loss": 0.5659,
1915
+ "step": 1350
1916
+ },
1917
+ {
1918
+ "epoch": 1.8750432675666322,
1919
+ "grad_norm": 0.44609665870666504,
1920
+ "learning_rate": 2.3595064154412374e-06,
1921
+ "loss": 0.5683,
1922
+ "step": 1355
1923
+ },
1924
+ {
1925
+ "epoch": 1.8819660782277605,
1926
+ "grad_norm": 0.4577263593673706,
1927
+ "learning_rate": 2.1055199248330546e-06,
1928
+ "loss": 0.5716,
1929
+ "step": 1360
1930
+ },
1931
+ {
1932
+ "epoch": 1.8888888888888888,
1933
+ "grad_norm": 0.4626077711582184,
1934
+ "learning_rate": 1.8658478742742625e-06,
1935
+ "loss": 0.5688,
1936
+ "step": 1365
1937
+ },
1938
+ {
1939
+ "epoch": 1.8958116995500172,
1940
+ "grad_norm": 0.4604962170124054,
1941
+ "learning_rate": 1.6405253093690343e-06,
1942
+ "loss": 0.5702,
1943
+ "step": 1370
1944
+ },
1945
+ {
1946
+ "epoch": 1.9027345102111457,
1947
+ "grad_norm": 0.46187230944633484,
1948
+ "learning_rate": 1.429585177494419e-06,
1949
+ "loss": 0.5671,
1950
+ "step": 1375
1951
+ },
1952
+ {
1953
+ "epoch": 1.9096573208722742,
1954
+ "grad_norm": 0.44326135516166687,
1955
+ "learning_rate": 1.2330583229827053e-06,
1956
+ "loss": 0.5812,
1957
+ "step": 1380
1958
+ },
1959
+ {
1960
+ "epoch": 1.9165801315334026,
1961
+ "grad_norm": 0.46138623356819153,
1962
+ "learning_rate": 1.0509734826112394e-06,
1963
+ "loss": 0.5651,
1964
+ "step": 1385
1965
+ },
1966
+ {
1967
+ "epoch": 1.9235029421945309,
1968
+ "grad_norm": 0.46161016821861267,
1969
+ "learning_rate": 8.833572814004321e-07,
1970
+ "loss": 0.5634,
1971
+ "step": 1390
1972
+ },
1973
+ {
1974
+ "epoch": 1.9304257528556594,
1975
+ "grad_norm": 0.4505715072154999,
1976
+ "learning_rate": 7.302342287205943e-07,
1977
+ "loss": 0.5688,
1978
+ "step": 1395
1979
+ },
1980
+ {
1981
+ "epoch": 1.937348563516788,
1982
+ "grad_norm": 0.4546772837638855,
1983
+ "learning_rate": 5.916267147080934e-07,
1984
+ "loss": 0.5727,
1985
+ "step": 1400
1986
+ },
1987
+ {
1988
+ "epoch": 1.9442713741779163,
1989
+ "grad_norm": 0.4859260022640228,
1990
+ "learning_rate": 4.6755500699138253e-07,
1991
+ "loss": 0.5614,
1992
+ "step": 1405
1993
+ },
1994
+ {
1995
+ "epoch": 1.9511941848390446,
1996
+ "grad_norm": 0.483917236328125,
1997
+ "learning_rate": 3.5803724772740523e-07,
1998
+ "loss": 0.5635,
1999
+ "step": 1410
2000
+ },
2001
+ {
2002
+ "epoch": 1.958116995500173,
2003
+ "grad_norm": 0.45442262291908264,
2004
+ "learning_rate": 2.630894509488058e-07,
2005
+ "loss": 0.5754,
2006
+ "step": 1415
2007
+ },
2008
+ {
2009
+ "epoch": 1.9650398061613015,
2010
+ "grad_norm": 0.45076045393943787,
2011
+ "learning_rate": 1.8272550022230272e-07,
2012
+ "loss": 0.5665,
2013
+ "step": 1420
2014
+ },
2015
+ {
2016
+ "epoch": 1.97196261682243,
2017
+ "grad_norm": 0.4759106934070587,
2018
+ "learning_rate": 1.1695714661856816e-07,
2019
+ "loss": 0.5583,
2020
+ "step": 1425
2021
+ },
2022
+ {
2023
+ "epoch": 1.9788854274835583,
2024
+ "grad_norm": 0.48755067586898804,
2025
+ "learning_rate": 6.579400699397998e-08,
2026
+ "loss": 0.57,
2027
+ "step": 1430
2028
+ },
2029
+ {
2030
+ "epoch": 1.9858082381446867,
2031
+ "grad_norm": 0.4441988170146942,
2032
+ "learning_rate": 2.9243562584402305e-08,
2033
+ "loss": 0.5784,
2034
+ "step": 1435
2035
+ },
2036
+ {
2037
+ "epoch": 1.9927310488058152,
2038
+ "grad_norm": 0.44851863384246826,
2039
+ "learning_rate": 7.311157911249478e-09,
2040
+ "loss": 0.5849,
2041
+ "step": 1440
2042
+ },
2043
+ {
2044
+ "epoch": 1.9982692973347178,
2045
+ "eval_loss": 0.7809522747993469,
2046
+ "eval_runtime": 54.6607,
2047
+ "eval_samples_per_second": 37.852,
2048
+ "eval_steps_per_second": 2.378,
2049
+ "step": 1444
2050
  },
2051
  {
2052
+ "epoch": 1.9982692973347178,
2053
+ "step": 1444,
2054
+ "total_flos": 4.75579504107769e+18,
2055
+ "train_loss": 0.8681613599824773,
2056
+ "train_runtime": 10264.6857,
2057
+ "train_samples_per_second": 9.005,
2058
+ "train_steps_per_second": 0.141
2059
  }
2060
  ],
2061
  "logging_steps": 5,
2062
+ "max_steps": 1444,
2063
  "num_input_tokens_seen": 0,
2064
  "num_train_epochs": 2,
2065
  "save_steps": 500,
 
2075
  "attributes": {}
2076
  }
2077
  },
2078
+ "total_flos": 4.75579504107769e+18,
2079
  "train_batch_size": 4,
2080
  "trial_name": null,
2081
  "trial_params": null
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3e9a343aa9f12c033062e11705d146dffaacd0bf53572b9e68d2ca60f23368e7
3
  size 7096
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8e280e4b69974082ee15290acf5bdc448e88447100b57fab17908e143c7db85a
3
  size 7096