Hugging Face's logo

PurCL
/
src_prober_codellama-7b-last1unfreeze

Text2Text Generation
Transformers
Safetensors
rad-prober
Generated from Trainer
Inference Endpoints
Model card Files Community
sheepy928 commited on
Commit
6ef8bde
·
verified ·
1 Parent(s): f999923

Training in progress, step 5000, checkpoint

Browse files
Files changed (11) hide show
  1. checkpoint-5000/config.json +195 -0
  2. checkpoint-5000/generation_config.json +4 -0
  3. checkpoint-5000/model.safetensors +3 -0
  4. checkpoint-5000/optimizer.pt +3 -0
  5. checkpoint-5000/rng_state_0.pth +3 -0
  6. checkpoint-5000/rng_state_1.pth +3 -0
  7. checkpoint-5000/rng_state_2.pth +3 -0
  8. checkpoint-5000/rng_state_3.pth +3 -0
  9. checkpoint-5000/scheduler.pt +3 -0
  10. checkpoint-5000/trainer_state.json +3601 -0
  11. checkpoint-5000/training_args.bin +3 -0
checkpoint-5000/config.json ADDED
@@ -0,0 +1,195 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "SrcProberForConditionalGeneration"
4
+ ],
5
+ "asm_encoder_config": {
6
+ "_name_or_path": "",
7
+ "add_cross_attention": false,
8
+ "architectures": [
9
+ "LongelmForMaskedLMWithEdgePrediction"
10
+ ],
11
+ "attention_probs_dropout_prob": 0.1,
12
+ "bad_words_ids": null,
13
+ "begin_suppress_tokens": null,
14
+ "block_size": 8,
15
+ "bos_token_id": 1,
16
+ "chunk_size_feed_forward": 0,
17
+ "classifier_dropout": null,
18
+ "cross_attention_hidden_size": null,
19
+ "decoder_start_token_id": null,
20
+ "diversity_penalty": 0.0,
21
+ "do_sample": false,
22
+ "early_stopping": false,
23
+ "encoder_no_repeat_ngram_size": 0,
24
+ "eos_token_id": 2,
25
+ "ep_add_linear_projection": true,
26
+ "exponential_decay_length_penalty": null,
27
+ "finetuning_task": null,
28
+ "forced_bos_token_id": null,
29
+ "forced_eos_token_id": null,
30
+ "global_memory_size": 1,
31
+ "hidden_act": "gelu",
32
+ "hidden_dropout_prob": 0.1,
33
+ "hidden_size": 768,
34
+ "id2label": {
35
+ "0": "LABEL_0",
36
+ "1": "LABEL_1"
37
+ },
38
+ "initializer_range": 0.02,
39
+ "intermediate_size": 3072,
40
+ "is_decoder": false,
41
+ "is_encoder_decoder": false,
42
+ "label2id": {
43
+ "LABEL_0": 0,
44
+ "LABEL_1": 1
45
+ },
46
+ "layer_norm_eps": 1e-12,
47
+ "length_penalty": 1.0,
48
+ "max_blocks": 200,
49
+ "max_length": 20,
50
+ "max_position_embeddings": 1805,
51
+ "max_relative_position_embeddings": 8,
52
+ "min_length": 0,
53
+ "model_type": "longelm",
54
+ "no_repeat_ngram_size": 0,
55
+ "node_size": 1,
56
+ "num_attention_heads": 12,
57
+ "num_beam_groups": 1,
58
+ "num_beams": 1,
59
+ "num_hidden_layers": 12,
60
+ "num_return_sequences": 1,
61
+ "output_attentions": false,
62
+ "output_hidden_states": false,
63
+ "output_scores": false,
64
+ "pad_token_id": 3,
65
+ "position_embedding_type": "mixed",
66
+ "prefix": null,
67
+ "problem_type": null,
68
+ "pruned_heads": {},
69
+ "remove_invalid_values": false,
70
+ "repetition_penalty": 1.0,
71
+ "return_dict": true,
72
+ "return_dict_in_generate": false,
73
+ "sep_token_id": null,
74
+ "suppress_tokens": null,
75
+ "task_specific_params": null,
76
+ "temperature": 1.0,
77
+ "tf_legacy_loss": false,
78
+ "tie_encoder_decoder": false,
79
+ "tie_word_embeddings": true,
80
+ "tokenizer_class": null,
81
+ "top_k": 50,
82
+ "top_p": 1.0,
83
+ "torch_dtype": "float32",
84
+ "torchscript": false,
85
+ "type_vocab_size": 2,
86
+ "typical_p": 1.0,
87
+ "use_bfloat16": false,
88
+ "use_cache": true,
89
+ "vocab_size": 50265
90
+ },
91
+ "asm_feature_select_strategy": "nodes",
92
+ "asm_token_index": 32016,
93
+ "ignore_index": -100,
94
+ "model_type": "rad-prober",
95
+ "pad_token_id": 2,
96
+ "projector_hidden_act": "gelu",
97
+ "src_lm_config": {
98
+ "_name_or_path": "codellama/CodeLlama-7b-hf",
99
+ "add_cross_attention": false,
100
+ "architectures": [
101
+ "LlamaForCausalLM"
102
+ ],
103
+ "attention_bias": false,
104
+ "attention_dropout": 0.0,
105
+ "bad_words_ids": null,
106
+ "begin_suppress_tokens": null,
107
+ "bos_token_id": 1,
108
+ "chunk_size_feed_forward": 0,
109
+ "cross_attention_hidden_size": null,
110
+ "decoder_start_token_id": null,
111
+ "diversity_penalty": 0.0,
112
+ "do_sample": false,
113
+ "early_stopping": false,
114
+ "encoder_no_repeat_ngram_size": 0,
115
+ "eos_token_id": 2,
116
+ "exponential_decay_length_penalty": null,
117
+ "finetuning_task": null,
118
+ "forced_bos_token_id": null,
119
+ "forced_eos_token_id": null,
120
+ "hidden_act": "silu",
121
+ "hidden_size": 4096,
122
+ "id2label": {
123
+ "0": "LABEL_0",
124
+ "1": "LABEL_1"
125
+ },
126
+ "initializer_range": 0.02,
127
+ "intermediate_size": 11008,
128
+ "is_decoder": false,
129
+ "is_encoder_decoder": false,
130
+ "label2id": {
131
+ "LABEL_0": 0,
132
+ "LABEL_1": 1
133
+ },
134
+ "length_penalty": 1.0,
135
+ "max_length": 20,
136
+ "max_position_embeddings": 16384,
137
+ "min_length": 0,
138
+ "model_type": "llama",
139
+ "no_repeat_ngram_size": 0,
140
+ "num_attention_heads": 32,
141
+ "num_beam_groups": 1,
142
+ "num_beams": 1,
143
+ "num_hidden_layers": 32,
144
+ "num_key_value_heads": 32,
145
+ "num_return_sequences": 1,
146
+ "output_attentions": false,
147
+ "output_hidden_states": false,
148
+ "output_scores": false,
149
+ "pad_token_id": null,
150
+ "prefix": null,
151
+ "pretraining_tp": 1,
152
+ "problem_type": null,
153
+ "pruned_heads": {},
154
+ "quantization_config": {
155
+ "_load_in_4bit": false,
156
+ "_load_in_8bit": true,
157
+ "bnb_4bit_compute_dtype": "float32",
158
+ "bnb_4bit_quant_storage": "uint8",
159
+ "bnb_4bit_quant_type": "fp4",
160
+ "bnb_4bit_use_double_quant": false,
161
+ "llm_int8_enable_fp32_cpu_offload": false,
162
+ "llm_int8_has_fp16_weight": false,
163
+ "llm_int8_skip_modules": null,
164
+ "llm_int8_threshold": 6.0,
165
+ "load_in_4bit": false,
166
+ "load_in_8bit": true,
167
+ "quant_method": "bitsandbytes"
168
+ },
169
+ "remove_invalid_values": false,
170
+ "repetition_penalty": 1.0,
171
+ "return_dict": true,
172
+ "return_dict_in_generate": false,
173
+ "rms_norm_eps": 1e-05,
174
+ "rope_scaling": null,
175
+ "rope_theta": 1000000,
176
+ "sep_token_id": null,
177
+ "suppress_tokens": null,
178
+ "task_specific_params": null,
179
+ "temperature": 1.0,
180
+ "tf_legacy_loss": false,
181
+ "tie_encoder_decoder": false,
182
+ "tie_word_embeddings": false,
183
+ "tokenizer_class": null,
184
+ "top_k": 50,
185
+ "top_p": 1.0,
186
+ "torch_dtype": "bfloat16",
187
+ "torchscript": false,
188
+ "typical_p": 1.0,
189
+ "use_bfloat16": false,
190
+ "use_cache": true,
191
+ "vocab_size": 32017
192
+ },
193
+ "torch_dtype": "float32",
194
+ "transformers_version": "4.39.3"
195
+ }
checkpoint-5000/generation_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "transformers_version": "4.39.3"
4
+ }
checkpoint-5000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e4cbb896df6b65ee49dfe8f0a46c4c7cd6d5bd1a31c4d7c113f620a6311ed101
3
+ size 579945600
checkpoint-5000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:061338ea1f4edcc844ccf7afec6c6c51504650e4a85eed0a9fcfa006f0dcfa51
3
+ size 216170911
checkpoint-5000/rng_state_0.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:997f05c38a6359893eee4a5b07b6e6c21d0af2e54ff32cd8ec61aeba7eae7ddb
3
+ size 14960
checkpoint-5000/rng_state_1.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6aeeac2fa4c99ec6106b308e017a147d9d39f5045968bd97249031c945f825eb
3
+ size 14960
checkpoint-5000/rng_state_2.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bfd4cc6a0c0886257afee402c28b398044f7be658a452ed4a490ebca25cb7fef
3
+ size 14960
checkpoint-5000/rng_state_3.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e2401713a1083d805760ffe227f0a8379bc9ef08e13f0952d4401f14f67aca8
3
+ size 14960
checkpoint-5000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8df647b59e085f1f5d43cccc9e067052484b24a4bcca9410f478960320a180db
3
+ size 1064
checkpoint-5000/trainer_state.json ADDED
@@ -0,0 +1,3601 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 2.416626389560174,
5
+ "eval_steps": 500,
6
+ "global_step": 5000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.0,
13
+ "grad_norm": 75.8236083984375,
14
+ "learning_rate": 4.0000000000000003e-07,
15
+ "loss": 3.7148,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.01,
20
+ "grad_norm": 52.23726272583008,
21
+ "learning_rate": 9e-07,
22
+ "loss": 3.2721,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.01,
27
+ "grad_norm": 35.391998291015625,
28
+ "learning_rate": 1.4000000000000001e-06,
29
+ "loss": 2.4262,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.02,
34
+ "grad_norm": 15.225934028625488,
35
+ "learning_rate": 1.9e-06,
36
+ "loss": 1.8828,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.02,
41
+ "grad_norm": 13.488654136657715,
42
+ "learning_rate": 2.4000000000000003e-06,
43
+ "loss": 1.5398,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.03,
48
+ "grad_norm": 8.773133277893066,
49
+ "learning_rate": 2.8500000000000002e-06,
50
+ "loss": 1.2772,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.03,
55
+ "grad_norm": 6.736795902252197,
56
+ "learning_rate": 3.3500000000000005e-06,
57
+ "loss": 1.1159,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.04,
62
+ "grad_norm": 2.9463019371032715,
63
+ "learning_rate": 3.85e-06,
64
+ "loss": 0.9883,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.04,
69
+ "grad_norm": 1.3851969242095947,
70
+ "learning_rate": 4.35e-06,
71
+ "loss": 0.902,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.05,
76
+ "grad_norm": 1.4642181396484375,
77
+ "learning_rate": 4.85e-06,
78
+ "loss": 0.8701,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.05,
83
+ "grad_norm": 0.6514687538146973,
84
+ "learning_rate": 5.3500000000000004e-06,
85
+ "loss": 0.8548,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.06,
90
+ "grad_norm": 0.4448617994785309,
91
+ "learning_rate": 5.850000000000001e-06,
92
+ "loss": 0.8038,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.06,
97
+ "grad_norm": 0.5425733327865601,
98
+ "learning_rate": 6.35e-06,
99
+ "loss": 0.825,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.07,
104
+ "grad_norm": 0.47352057695388794,
105
+ "learning_rate": 6.8500000000000005e-06,
106
+ "loss": 0.812,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.07,
111
+ "grad_norm": 0.40401744842529297,
112
+ "learning_rate": 7.35e-06,
113
+ "loss": 0.7912,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.08,
118
+ "grad_norm": 0.3112718164920807,
119
+ "learning_rate": 7.850000000000001e-06,
120
+ "loss": 0.8057,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.08,
125
+ "grad_norm": 0.45351892709732056,
126
+ "learning_rate": 8.350000000000001e-06,
127
+ "loss": 0.7912,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.09,
132
+ "grad_norm": 0.38080641627311707,
133
+ "learning_rate": 8.85e-06,
134
+ "loss": 0.7956,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.09,
139
+ "grad_norm": 0.22949431836605072,
140
+ "learning_rate": 9.35e-06,
141
+ "loss": 0.7785,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.1,
146
+ "grad_norm": 0.8331744074821472,
147
+ "learning_rate": 9.85e-06,
148
+ "loss": 0.7769,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.1,
153
+ "grad_norm": 0.30213871598243713,
154
+ "learning_rate": 1.035e-05,
155
+ "loss": 0.7797,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.11,
160
+ "grad_norm": 0.32015591859817505,
161
+ "learning_rate": 1.0850000000000001e-05,
162
+ "loss": 0.7941,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.11,
167
+ "grad_norm": 0.22447015345096588,
168
+ "learning_rate": 1.1350000000000001e-05,
169
+ "loss": 0.8014,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.12,
174
+ "grad_norm": 0.2546006143093109,
175
+ "learning_rate": 1.185e-05,
176
+ "loss": 0.771,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.12,
181
+ "grad_norm": 0.23833389580249786,
182
+ "learning_rate": 1.235e-05,
183
+ "loss": 0.7956,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.13,
188
+ "grad_norm": 0.1831669956445694,
189
+ "learning_rate": 1.285e-05,
190
+ "loss": 0.7833,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.13,
195
+ "grad_norm": 0.19860319793224335,
196
+ "learning_rate": 1.3350000000000001e-05,
197
+ "loss": 0.7956,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.14,
202
+ "grad_norm": 0.21220380067825317,
203
+ "learning_rate": 1.3850000000000001e-05,
204
+ "loss": 0.7942,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.14,
209
+ "grad_norm": 0.26025038957595825,
210
+ "learning_rate": 1.435e-05,
211
+ "loss": 0.7877,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.14,
216
+ "grad_norm": 0.2029470056295395,
217
+ "learning_rate": 1.485e-05,
218
+ "loss": 0.7637,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.15,
223
+ "grad_norm": 0.21639183163642883,
224
+ "learning_rate": 1.535e-05,
225
+ "loss": 0.7885,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.15,
230
+ "grad_norm": 0.23670387268066406,
231
+ "learning_rate": 1.5850000000000002e-05,
232
+ "loss": 0.7558,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.16,
237
+ "grad_norm": 0.2855335772037506,
238
+ "learning_rate": 1.635e-05,
239
+ "loss": 0.7854,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.16,
244
+ "grad_norm": 0.21515759825706482,
245
+ "learning_rate": 1.6850000000000003e-05,
246
+ "loss": 0.7577,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.17,
251
+ "grad_norm": 0.18981383740901947,
252
+ "learning_rate": 1.7349999999999998e-05,
253
+ "loss": 0.7802,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.17,
258
+ "grad_norm": 0.2674611508846283,
259
+ "learning_rate": 1.785e-05,
260
+ "loss": 0.7809,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.18,
265
+ "grad_norm": 0.24821774661540985,
266
+ "learning_rate": 1.8350000000000002e-05,
267
+ "loss": 0.7965,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.18,
272
+ "grad_norm": 0.3644929528236389,
273
+ "learning_rate": 1.885e-05,
274
+ "loss": 0.7957,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.19,
279
+ "grad_norm": 0.3568972647190094,
280
+ "learning_rate": 1.9350000000000003e-05,
281
+ "loss": 0.7882,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.19,
286
+ "grad_norm": 0.3236319422721863,
287
+ "learning_rate": 1.985e-05,
288
+ "loss": 0.7841,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.2,
293
+ "grad_norm": 0.24553723633289337,
294
+ "learning_rate": 2.035e-05,
295
+ "loss": 0.7824,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.2,
300
+ "grad_norm": 0.1855400949716568,
301
+ "learning_rate": 2.085e-05,
302
+ "loss": 0.7645,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.21,
307
+ "grad_norm": 0.26212435960769653,
308
+ "learning_rate": 2.135e-05,
309
+ "loss": 0.7824,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.21,
314
+ "grad_norm": 0.3167509138584137,
315
+ "learning_rate": 2.1850000000000003e-05,
316
+ "loss": 0.7908,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.22,
321
+ "grad_norm": 0.21845023334026337,
322
+ "learning_rate": 2.235e-05,
323
+ "loss": 0.7888,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.22,
328
+ "grad_norm": 0.30318892002105713,
329
+ "learning_rate": 2.2850000000000003e-05,
330
+ "loss": 0.7682,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.23,
335
+ "grad_norm": 0.28828638792037964,
336
+ "learning_rate": 2.3350000000000002e-05,
337
+ "loss": 0.7588,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.23,
342
+ "grad_norm": 0.2422240674495697,
343
+ "learning_rate": 2.385e-05,
344
+ "loss": 0.7787,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.24,
349
+ "grad_norm": 0.3579326272010803,
350
+ "learning_rate": 2.435e-05,
351
+ "loss": 0.7932,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.24,
356
+ "grad_norm": 0.3549540042877197,
357
+ "learning_rate": 2.485e-05,
358
+ "loss": 0.7634,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.24,
363
+ "eval_loss": 0.7744565606117249,
364
+ "eval_runtime": 43.9626,
365
+ "eval_samples_per_second": 45.493,
366
+ "eval_steps_per_second": 0.364,
367
+ "step": 500
368
+ },
369
+ {
370
+ "epoch": 0.25,
371
+ "grad_norm": 0.2931281328201294,
372
+ "learning_rate": 2.5350000000000003e-05,
373
+ "loss": 0.7763,
374
+ "step": 510
375
+ },
376
+ {
377
+ "epoch": 0.25,
378
+ "grad_norm": 0.21228083968162537,
379
+ "learning_rate": 2.585e-05,
380
+ "loss": 0.7748,
381
+ "step": 520
382
+ },
383
+ {
384
+ "epoch": 0.26,
385
+ "grad_norm": 0.26342302560806274,
386
+ "learning_rate": 2.6350000000000004e-05,
387
+ "loss": 0.789,
388
+ "step": 530
389
+ },
390
+ {
391
+ "epoch": 0.26,
392
+ "grad_norm": 0.2840302288532257,
393
+ "learning_rate": 2.6850000000000002e-05,
394
+ "loss": 0.768,
395
+ "step": 540
396
+ },
397
+ {
398
+ "epoch": 0.27,
399
+ "grad_norm": 0.273771196603775,
400
+ "learning_rate": 2.7350000000000004e-05,
401
+ "loss": 0.7707,
402
+ "step": 550
403
+ },
404
+ {
405
+ "epoch": 0.27,
406
+ "grad_norm": 1.6022837162017822,
407
+ "learning_rate": 2.7850000000000003e-05,
408
+ "loss": 0.7705,
409
+ "step": 560
410
+ },
411
+ {
412
+ "epoch": 0.28,
413
+ "grad_norm": 0.3082919716835022,
414
+ "learning_rate": 2.8349999999999998e-05,
415
+ "loss": 0.7849,
416
+ "step": 570
417
+ },
418
+ {
419
+ "epoch": 0.28,
420
+ "grad_norm": 0.25341132283210754,
421
+ "learning_rate": 2.885e-05,
422
+ "loss": 0.7658,
423
+ "step": 580
424
+ },
425
+ {
426
+ "epoch": 0.29,
427
+ "grad_norm": 0.2250708043575287,
428
+ "learning_rate": 2.935e-05,
429
+ "loss": 0.756,
430
+ "step": 590
431
+ },
432
+ {
433
+ "epoch": 0.29,
434
+ "grad_norm": 0.23522531986236572,
435
+ "learning_rate": 2.985e-05,
436
+ "loss": 0.7692,
437
+ "step": 600
438
+ },
439
+ {
440
+ "epoch": 0.29,
441
+ "grad_norm": 0.2448827624320984,
442
+ "learning_rate": 3.035e-05,
443
+ "loss": 0.7635,
444
+ "step": 610
445
+ },
446
+ {
447
+ "epoch": 0.3,
448
+ "grad_norm": 0.19717586040496826,
449
+ "learning_rate": 3.0850000000000004e-05,
450
+ "loss": 0.7548,
451
+ "step": 620
452
+ },
453
+ {
454
+ "epoch": 0.3,
455
+ "grad_norm": 0.24299098551273346,
456
+ "learning_rate": 3.135e-05,
457
+ "loss": 0.7673,
458
+ "step": 630
459
+ },
460
+ {
461
+ "epoch": 0.31,
462
+ "grad_norm": 0.2555953562259674,
463
+ "learning_rate": 3.185e-05,
464
+ "loss": 0.7528,
465
+ "step": 640
466
+ },
467
+ {
468
+ "epoch": 0.31,
469
+ "grad_norm": 0.2102127969264984,
470
+ "learning_rate": 3.235e-05,
471
+ "loss": 0.7678,
472
+ "step": 650
473
+ },
474
+ {
475
+ "epoch": 0.32,
476
+ "grad_norm": 0.21767646074295044,
477
+ "learning_rate": 3.2850000000000006e-05,
478
+ "loss": 0.7564,
479
+ "step": 660
480
+ },
481
+ {
482
+ "epoch": 0.32,
483
+ "grad_norm": 0.3582068085670471,
484
+ "learning_rate": 3.3350000000000004e-05,
485
+ "loss": 0.7645,
486
+ "step": 670
487
+ },
488
+ {
489
+ "epoch": 0.33,
490
+ "grad_norm": 0.27804431319236755,
491
+ "learning_rate": 3.385e-05,
492
+ "loss": 0.7645,
493
+ "step": 680
494
+ },
495
+ {
496
+ "epoch": 0.33,
497
+ "grad_norm": 0.2464047521352768,
498
+ "learning_rate": 3.435e-05,
499
+ "loss": 0.7325,
500
+ "step": 690
501
+ },
502
+ {
503
+ "epoch": 0.34,
504
+ "grad_norm": 0.22323842346668243,
505
+ "learning_rate": 3.485e-05,
506
+ "loss": 0.7579,
507
+ "step": 700
508
+ },
509
+ {
510
+ "epoch": 0.34,
511
+ "grad_norm": 0.28329703211784363,
512
+ "learning_rate": 3.535e-05,
513
+ "loss": 0.7399,
514
+ "step": 710
515
+ },
516
+ {
517
+ "epoch": 0.35,
518
+ "grad_norm": 0.2471865713596344,
519
+ "learning_rate": 3.585e-05,
520
+ "loss": 0.765,
521
+ "step": 720
522
+ },
523
+ {
524
+ "epoch": 0.35,
525
+ "grad_norm": 0.314908504486084,
526
+ "learning_rate": 3.635e-05,
527
+ "loss": 0.7529,
528
+ "step": 730
529
+ },
530
+ {
531
+ "epoch": 0.36,
532
+ "grad_norm": 0.21466995775699615,
533
+ "learning_rate": 3.685e-05,
534
+ "loss": 0.7365,
535
+ "step": 740
536
+ },
537
+ {
538
+ "epoch": 0.36,
539
+ "grad_norm": 0.2545580565929413,
540
+ "learning_rate": 3.735e-05,
541
+ "loss": 0.7458,
542
+ "step": 750
543
+ },
544
+ {
545
+ "epoch": 0.37,
546
+ "grad_norm": 0.2651059329509735,
547
+ "learning_rate": 3.7850000000000005e-05,
548
+ "loss": 0.7578,
549
+ "step": 760
550
+ },
551
+ {
552
+ "epoch": 0.37,
553
+ "grad_norm": 0.2408379763364792,
554
+ "learning_rate": 3.8350000000000004e-05,
555
+ "loss": 0.7283,
556
+ "step": 770
557
+ },
558
+ {
559
+ "epoch": 0.38,
560
+ "grad_norm": 0.2575875520706177,
561
+ "learning_rate": 3.885e-05,
562
+ "loss": 0.7367,
563
+ "step": 780
564
+ },
565
+ {
566
+ "epoch": 0.38,
567
+ "grad_norm": 0.28182050585746765,
568
+ "learning_rate": 3.935e-05,
569
+ "loss": 0.7784,
570
+ "step": 790
571
+ },
572
+ {
573
+ "epoch": 0.39,
574
+ "grad_norm": 0.20579397678375244,
575
+ "learning_rate": 3.9850000000000006e-05,
576
+ "loss": 0.7533,
577
+ "step": 800
578
+ },
579
+ {
580
+ "epoch": 0.39,
581
+ "grad_norm": 0.1906532198190689,
582
+ "learning_rate": 4.0350000000000005e-05,
583
+ "loss": 0.7544,
584
+ "step": 810
585
+ },
586
+ {
587
+ "epoch": 0.4,
588
+ "grad_norm": 0.18311142921447754,
589
+ "learning_rate": 4.085e-05,
590
+ "loss": 0.75,
591
+ "step": 820
592
+ },
593
+ {
594
+ "epoch": 0.4,
595
+ "grad_norm": 0.1910839080810547,
596
+ "learning_rate": 4.135e-05,
597
+ "loss": 0.7538,
598
+ "step": 830
599
+ },
600
+ {
601
+ "epoch": 0.41,
602
+ "grad_norm": 0.2178005427122116,
603
+ "learning_rate": 4.185e-05,
604
+ "loss": 0.7456,
605
+ "step": 840
606
+ },
607
+ {
608
+ "epoch": 0.41,
609
+ "grad_norm": 0.27642378211021423,
610
+ "learning_rate": 4.235e-05,
611
+ "loss": 0.744,
612
+ "step": 850
613
+ },
614
+ {
615
+ "epoch": 0.42,
616
+ "grad_norm": 0.26685312390327454,
617
+ "learning_rate": 4.285e-05,
618
+ "loss": 0.7611,
619
+ "step": 860
620
+ },
621
+ {
622
+ "epoch": 0.42,
623
+ "grad_norm": 0.2021845579147339,
624
+ "learning_rate": 4.335e-05,
625
+ "loss": 0.758,
626
+ "step": 870
627
+ },
628
+ {
629
+ "epoch": 0.43,
630
+ "grad_norm": 0.20516808331012726,
631
+ "learning_rate": 4.385e-05,
632
+ "loss": 0.7404,
633
+ "step": 880
634
+ },
635
+ {
636
+ "epoch": 0.43,
637
+ "grad_norm": 0.20619668066501617,
638
+ "learning_rate": 4.435e-05,
639
+ "loss": 0.7379,
640
+ "step": 890
641
+ },
642
+ {
643
+ "epoch": 0.43,
644
+ "grad_norm": 0.22439967095851898,
645
+ "learning_rate": 4.4850000000000006e-05,
646
+ "loss": 0.7495,
647
+ "step": 900
648
+ },
649
+ {
650
+ "epoch": 0.44,
651
+ "grad_norm": 0.2192295491695404,
652
+ "learning_rate": 4.5350000000000005e-05,
653
+ "loss": 0.7377,
654
+ "step": 910
655
+ },
656
+ {
657
+ "epoch": 0.44,
658
+ "grad_norm": 0.26503369212150574,
659
+ "learning_rate": 4.585e-05,
660
+ "loss": 0.7481,
661
+ "step": 920
662
+ },
663
+ {
664
+ "epoch": 0.45,
665
+ "grad_norm": 0.2249353975057602,
666
+ "learning_rate": 4.635e-05,
667
+ "loss": 0.742,
668
+ "step": 930
669
+ },
670
+ {
671
+ "epoch": 0.45,
672
+ "grad_norm": 0.21413160860538483,
673
+ "learning_rate": 4.685000000000001e-05,
674
+ "loss": 0.7485,
675
+ "step": 940
676
+ },
677
+ {
678
+ "epoch": 0.46,
679
+ "grad_norm": 0.20971478521823883,
680
+ "learning_rate": 4.735e-05,
681
+ "loss": 0.7339,
682
+ "step": 950
683
+ },
684
+ {
685
+ "epoch": 0.46,
686
+ "grad_norm": 0.20295798778533936,
687
+ "learning_rate": 4.785e-05,
688
+ "loss": 0.722,
689
+ "step": 960
690
+ },
691
+ {
692
+ "epoch": 0.47,
693
+ "grad_norm": 0.1957596242427826,
694
+ "learning_rate": 4.835e-05,
695
+ "loss": 0.7538,
696
+ "step": 970
697
+ },
698
+ {
699
+ "epoch": 0.47,
700
+ "grad_norm": 0.18014560639858246,
701
+ "learning_rate": 4.885e-05,
702
+ "loss": 0.7251,
703
+ "step": 980
704
+ },
705
+ {
706
+ "epoch": 0.48,
707
+ "grad_norm": 0.19334861636161804,
708
+ "learning_rate": 4.935e-05,
709
+ "loss": 0.7444,
710
+ "step": 990
711
+ },
712
+ {
713
+ "epoch": 0.48,
714
+ "grad_norm": 0.1753440499305725,
715
+ "learning_rate": 4.9850000000000006e-05,
716
+ "loss": 0.7376,
717
+ "step": 1000
718
+ },
719
+ {
720
+ "epoch": 0.48,
721
+ "eval_loss": 0.73988938331604,
722
+ "eval_runtime": 42.9535,
723
+ "eval_samples_per_second": 46.562,
724
+ "eval_steps_per_second": 0.372,
725
+ "step": 1000
726
+ },
727
+ {
728
+ "epoch": 0.49,
729
+ "grad_norm": 0.24455028772354126,
730
+ "learning_rate": 4.99999307775404e-05,
731
+ "loss": 0.7404,
732
+ "step": 1010
733
+ },
734
+ {
735
+ "epoch": 0.49,
736
+ "grad_norm": 0.2149570733308792,
737
+ "learning_rate": 4.999959172968145e-05,
738
+ "loss": 0.7342,
739
+ "step": 1020
740
+ },
741
+ {
742
+ "epoch": 0.5,
743
+ "grad_norm": 0.1931283175945282,
744
+ "learning_rate": 4.999897014592085e-05,
745
+ "loss": 0.735,
746
+ "step": 1030
747
+ },
748
+ {
749
+ "epoch": 0.5,
750
+ "grad_norm": 0.2331906110048294,
751
+ "learning_rate": 4.999806603328352e-05,
752
+ "loss": 0.7145,
753
+ "step": 1040
754
+ },
755
+ {
756
+ "epoch": 0.51,
757
+ "grad_norm": 0.21177324652671814,
758
+ "learning_rate": 4.999687940198738e-05,
759
+ "loss": 0.7308,
760
+ "step": 1050
761
+ },
762
+ {
763
+ "epoch": 0.51,
764
+ "grad_norm": 0.2026621401309967,
765
+ "learning_rate": 4.9995410265443286e-05,
766
+ "loss": 0.7445,
767
+ "step": 1060
768
+ },
769
+ {
770
+ "epoch": 0.52,
771
+ "grad_norm": 0.19811251759529114,
772
+ "learning_rate": 4.999365864025487e-05,
773
+ "loss": 0.738,
774
+ "step": 1070
775
+ },
776
+ {
777
+ "epoch": 0.52,
778
+ "grad_norm": 0.1981566846370697,
779
+ "learning_rate": 4.9991624546218334e-05,
780
+ "loss": 0.7286,
781
+ "step": 1080
782
+ },
783
+ {
784
+ "epoch": 0.53,
785
+ "grad_norm": 0.17122291028499603,
786
+ "learning_rate": 4.9989308006322235e-05,
787
+ "loss": 0.737,
788
+ "step": 1090
789
+ },
790
+ {
791
+ "epoch": 0.53,
792
+ "grad_norm": 0.2182038575410843,
793
+ "learning_rate": 4.9986709046747225e-05,
794
+ "loss": 0.7354,
795
+ "step": 1100
796
+ },
797
+ {
798
+ "epoch": 0.54,
799
+ "grad_norm": 0.1823730617761612,
800
+ "learning_rate": 4.9983827696865764e-05,
801
+ "loss": 0.725,
802
+ "step": 1110
803
+ },
804
+ {
805
+ "epoch": 0.54,
806
+ "grad_norm": 0.2195768803358078,
807
+ "learning_rate": 4.998066398924177e-05,
808
+ "loss": 0.7173,
809
+ "step": 1120
810
+ },
811
+ {
812
+ "epoch": 0.55,
813
+ "grad_norm": 0.18349726498126984,
814
+ "learning_rate": 4.997721795963028e-05,
815
+ "loss": 0.7159,
816
+ "step": 1130
817
+ },
818
+ {
819
+ "epoch": 0.55,
820
+ "grad_norm": 0.2440386265516281,
821
+ "learning_rate": 4.997348964697699e-05,
822
+ "loss": 0.7171,
823
+ "step": 1140
824
+ },
825
+ {
826
+ "epoch": 0.56,
827
+ "grad_norm": 0.20371629297733307,
828
+ "learning_rate": 4.9969479093417894e-05,
829
+ "loss": 0.7179,
830
+ "step": 1150
831
+ },
832
+ {
833
+ "epoch": 0.56,
834
+ "grad_norm": 0.2982828617095947,
835
+ "learning_rate": 4.9965186344278746e-05,
836
+ "loss": 0.7222,
837
+ "step": 1160
838
+ },
839
+ {
840
+ "epoch": 0.57,
841
+ "grad_norm": 0.2097097784280777,
842
+ "learning_rate": 4.9960611448074555e-05,
843
+ "loss": 0.7174,
844
+ "step": 1170
845
+ },
846
+ {
847
+ "epoch": 0.57,
848
+ "grad_norm": 0.2264368236064911,
849
+ "learning_rate": 4.995575445650907e-05,
850
+ "loss": 0.7079,
851
+ "step": 1180
852
+ },
853
+ {
854
+ "epoch": 0.58,
855
+ "grad_norm": 0.24737605452537537,
856
+ "learning_rate": 4.995061542447418e-05,
857
+ "loss": 0.7272,
858
+ "step": 1190
859
+ },
860
+ {
861
+ "epoch": 0.58,
862
+ "grad_norm": 0.19530241191387177,
863
+ "learning_rate": 4.9945194410049254e-05,
864
+ "loss": 0.7216,
865
+ "step": 1200
866
+ },
867
+ {
868
+ "epoch": 0.58,
869
+ "grad_norm": 0.2791181802749634,
870
+ "learning_rate": 4.993949147450055e-05,
871
+ "loss": 0.712,
872
+ "step": 1210
873
+ },
874
+ {
875
+ "epoch": 0.59,
876
+ "grad_norm": 0.2095661610364914,
877
+ "learning_rate": 4.993350668228049e-05,
878
+ "loss": 0.7067,
879
+ "step": 1220
880
+ },
881
+ {
882
+ "epoch": 0.59,
883
+ "grad_norm": 0.17129307985305786,
884
+ "learning_rate": 4.9927240101026915e-05,
885
+ "loss": 0.7087,
886
+ "step": 1230
887
+ },
888
+ {
889
+ "epoch": 0.6,
890
+ "grad_norm": 0.2154540866613388,
891
+ "learning_rate": 4.992069180156235e-05,
892
+ "loss": 0.7411,
893
+ "step": 1240
894
+ },
895
+ {
896
+ "epoch": 0.6,
897
+ "grad_norm": 0.22642803192138672,
898
+ "learning_rate": 4.991386185789319e-05,
899
+ "loss": 0.7266,
900
+ "step": 1250
901
+ },
902
+ {
903
+ "epoch": 0.61,
904
+ "grad_norm": 0.19676081836223602,
905
+ "learning_rate": 4.9906750347208866e-05,
906
+ "loss": 0.7012,
907
+ "step": 1260
908
+ },
909
+ {
910
+ "epoch": 0.61,
911
+ "grad_norm": 0.2095811814069748,
912
+ "learning_rate": 4.989935734988098e-05,
913
+ "loss": 0.7162,
914
+ "step": 1270
915
+ },
916
+ {
917
+ "epoch": 0.62,
918
+ "grad_norm": 0.21455398201942444,
919
+ "learning_rate": 4.9891682949462385e-05,
920
+ "loss": 0.7177,
921
+ "step": 1280
922
+ },
923
+ {
924
+ "epoch": 0.62,
925
+ "grad_norm": 0.20205985009670258,
926
+ "learning_rate": 4.988372723268623e-05,
927
+ "loss": 0.718,
928
+ "step": 1290
929
+ },
930
+ {
931
+ "epoch": 0.63,
932
+ "grad_norm": 0.18780440092086792,
933
+ "learning_rate": 4.987549028946502e-05,
934
+ "loss": 0.7103,
935
+ "step": 1300
936
+ },
937
+ {
938
+ "epoch": 0.63,
939
+ "grad_norm": 0.20362792909145355,
940
+ "learning_rate": 4.986697221288956e-05,
941
+ "loss": 0.7304,
942
+ "step": 1310
943
+ },
944
+ {
945
+ "epoch": 0.64,
946
+ "grad_norm": 0.2820493280887604,
947
+ "learning_rate": 4.985817309922792e-05,
948
+ "loss": 0.7182,
949
+ "step": 1320
950
+ },
951
+ {
952
+ "epoch": 0.64,
953
+ "grad_norm": 0.20787598192691803,
954
+ "learning_rate": 4.984909304792435e-05,
955
+ "loss": 0.7142,
956
+ "step": 1330
957
+ },
958
+ {
959
+ "epoch": 0.65,
960
+ "grad_norm": 0.21624509990215302,
961
+ "learning_rate": 4.983973216159813e-05,
962
+ "loss": 0.7208,
963
+ "step": 1340
964
+ },
965
+ {
966
+ "epoch": 0.65,
967
+ "grad_norm": 0.19343602657318115,
968
+ "learning_rate": 4.983009054604246e-05,
969
+ "loss": 0.7097,
970
+ "step": 1350
971
+ },
972
+ {
973
+ "epoch": 0.66,
974
+ "grad_norm": 0.254402756690979,
975
+ "learning_rate": 4.9820168310223215e-05,
976
+ "loss": 0.7193,
977
+ "step": 1360
978
+ },
979
+ {
980
+ "epoch": 0.66,
981
+ "grad_norm": 0.1940561830997467,
982
+ "learning_rate": 4.980996556627774e-05,
983
+ "loss": 0.6988,
984
+ "step": 1370
985
+ },
986
+ {
987
+ "epoch": 0.67,
988
+ "grad_norm": 0.22593462467193604,
989
+ "learning_rate": 4.9799482429513576e-05,
990
+ "loss": 0.7054,
991
+ "step": 1380
992
+ },
993
+ {
994
+ "epoch": 0.67,
995
+ "grad_norm": 0.22020135819911957,
996
+ "learning_rate": 4.978871901840716e-05,
997
+ "loss": 0.7221,
998
+ "step": 1390
999
+ },
1000
+ {
1001
+ "epoch": 0.68,
1002
+ "grad_norm": 0.22807644307613373,
1003
+ "learning_rate": 4.977767545460248e-05,
1004
+ "loss": 0.7213,
1005
+ "step": 1400
1006
+ },
1007
+ {
1008
+ "epoch": 0.68,
1009
+ "grad_norm": 0.1925562173128128,
1010
+ "learning_rate": 4.9766351862909703e-05,
1011
+ "loss": 0.7127,
1012
+ "step": 1410
1013
+ },
1014
+ {
1015
+ "epoch": 0.69,
1016
+ "grad_norm": 0.20845136046409607,
1017
+ "learning_rate": 4.9754748371303775e-05,
1018
+ "loss": 0.7207,
1019
+ "step": 1420
1020
+ },
1021
+ {
1022
+ "epoch": 0.69,
1023
+ "grad_norm": 0.18583528697490692,
1024
+ "learning_rate": 4.974286511092294e-05,
1025
+ "loss": 0.7098,
1026
+ "step": 1430
1027
+ },
1028
+ {
1029
+ "epoch": 0.7,
1030
+ "grad_norm": 0.2019319236278534,
1031
+ "learning_rate": 4.973070221606732e-05,
1032
+ "loss": 0.7029,
1033
+ "step": 1440
1034
+ },
1035
+ {
1036
+ "epoch": 0.7,
1037
+ "grad_norm": 0.1966458261013031,
1038
+ "learning_rate": 4.971825982419731e-05,
1039
+ "loss": 0.7241,
1040
+ "step": 1450
1041
+ },
1042
+ {
1043
+ "epoch": 0.71,
1044
+ "grad_norm": 0.20924623310565948,
1045
+ "learning_rate": 4.970553807593214e-05,
1046
+ "loss": 0.7126,
1047
+ "step": 1460
1048
+ },
1049
+ {
1050
+ "epoch": 0.71,
1051
+ "grad_norm": 0.20655803382396698,
1052
+ "learning_rate": 4.969253711504814e-05,
1053
+ "loss": 0.6903,
1054
+ "step": 1470
1055
+ },
1056
+ {
1057
+ "epoch": 0.72,
1058
+ "grad_norm": 0.2179209142923355,
1059
+ "learning_rate": 4.967925708847727e-05,
1060
+ "loss": 0.683,
1061
+ "step": 1480
1062
+ },
1063
+ {
1064
+ "epoch": 0.72,
1065
+ "grad_norm": 0.2198410928249359,
1066
+ "learning_rate": 4.966569814630534e-05,
1067
+ "loss": 0.7087,
1068
+ "step": 1490
1069
+ },
1070
+ {
1071
+ "epoch": 0.72,
1072
+ "grad_norm": 0.17970815300941467,
1073
+ "learning_rate": 4.9651860441770374e-05,
1074
+ "loss": 0.7006,
1075
+ "step": 1500
1076
+ },
1077
+ {
1078
+ "epoch": 0.72,
1079
+ "eval_loss": 0.7137542366981506,
1080
+ "eval_runtime": 43.0052,
1081
+ "eval_samples_per_second": 46.506,
1082
+ "eval_steps_per_second": 0.372,
1083
+ "step": 1500
1084
+ },
1085
+ {
1086
+ "epoch": 0.73,
1087
+ "grad_norm": 0.1931023746728897,
1088
+ "learning_rate": 4.963774413126086e-05,
1089
+ "loss": 0.7245,
1090
+ "step": 1510
1091
+ },
1092
+ {
1093
+ "epoch": 0.73,
1094
+ "grad_norm": 0.21812495589256287,
1095
+ "learning_rate": 4.9623349374313973e-05,
1096
+ "loss": 0.7119,
1097
+ "step": 1520
1098
+ },
1099
+ {
1100
+ "epoch": 0.74,
1101
+ "grad_norm": 0.20784322917461395,
1102
+ "learning_rate": 4.960867633361381e-05,
1103
+ "loss": 0.704,
1104
+ "step": 1530
1105
+ },
1106
+ {
1107
+ "epoch": 0.74,
1108
+ "grad_norm": 0.18697650730609894,
1109
+ "learning_rate": 4.9593725174989507e-05,
1110
+ "loss": 0.7129,
1111
+ "step": 1540
1112
+ },
1113
+ {
1114
+ "epoch": 0.75,
1115
+ "grad_norm": 0.25061744451522827,
1116
+ "learning_rate": 4.957849606741338e-05,
1117
+ "loss": 0.6816,
1118
+ "step": 1550
1119
+ },
1120
+ {
1121
+ "epoch": 0.75,
1122
+ "grad_norm": 0.1888163685798645,
1123
+ "learning_rate": 4.956298918299903e-05,
1124
+ "loss": 0.6905,
1125
+ "step": 1560
1126
+ },
1127
+ {
1128
+ "epoch": 0.76,
1129
+ "grad_norm": 0.1991235315799713,
1130
+ "learning_rate": 4.954720469699938e-05,
1131
+ "loss": 0.7258,
1132
+ "step": 1570
1133
+ },
1134
+ {
1135
+ "epoch": 0.76,
1136
+ "grad_norm": 0.18779252469539642,
1137
+ "learning_rate": 4.953114278780471e-05,
1138
+ "loss": 0.6918,
1139
+ "step": 1580
1140
+ },
1141
+ {
1142
+ "epoch": 0.77,
1143
+ "grad_norm": 0.21268677711486816,
1144
+ "learning_rate": 4.951480363694061e-05,
1145
+ "loss": 0.6995,
1146
+ "step": 1590
1147
+ },
1148
+ {
1149
+ "epoch": 0.77,
1150
+ "grad_norm": 0.20242677628993988,
1151
+ "learning_rate": 4.949818742906597e-05,
1152
+ "loss": 0.7132,
1153
+ "step": 1600
1154
+ },
1155
+ {
1156
+ "epoch": 0.78,
1157
+ "grad_norm": 0.2314455509185791,
1158
+ "learning_rate": 4.9481294351970874e-05,
1159
+ "loss": 0.6866,
1160
+ "step": 1610
1161
+ },
1162
+ {
1163
+ "epoch": 0.78,
1164
+ "grad_norm": 0.19638539850711823,
1165
+ "learning_rate": 4.9464124596574465e-05,
1166
+ "loss": 0.6993,
1167
+ "step": 1620
1168
+ },
1169
+ {
1170
+ "epoch": 0.79,
1171
+ "grad_norm": 0.185603529214859,
1172
+ "learning_rate": 4.944667835692281e-05,
1173
+ "loss": 0.6971,
1174
+ "step": 1630
1175
+ },
1176
+ {
1177
+ "epoch": 0.79,
1178
+ "grad_norm": 0.1594298928976059,
1179
+ "learning_rate": 4.942895583018669e-05,
1180
+ "loss": 0.7258,
1181
+ "step": 1640
1182
+ },
1183
+ {
1184
+ "epoch": 0.8,
1185
+ "grad_norm": 0.21844029426574707,
1186
+ "learning_rate": 4.9410957216659374e-05,
1187
+ "loss": 0.7022,
1188
+ "step": 1650
1189
+ },
1190
+ {
1191
+ "epoch": 0.8,
1192
+ "grad_norm": 0.19146642088890076,
1193
+ "learning_rate": 4.9392682719754364e-05,
1194
+ "loss": 0.6954,
1195
+ "step": 1660
1196
+ },
1197
+ {
1198
+ "epoch": 0.81,
1199
+ "grad_norm": 0.20112347602844238,
1200
+ "learning_rate": 4.9374132546003096e-05,
1201
+ "loss": 0.7017,
1202
+ "step": 1670
1203
+ },
1204
+ {
1205
+ "epoch": 0.81,
1206
+ "grad_norm": 0.22593769431114197,
1207
+ "learning_rate": 4.935530690505259e-05,
1208
+ "loss": 0.7397,
1209
+ "step": 1680
1210
+ },
1211
+ {
1212
+ "epoch": 0.82,
1213
+ "grad_norm": 0.2611504793167114,
1214
+ "learning_rate": 4.933620600966311e-05,
1215
+ "loss": 0.7048,
1216
+ "step": 1690
1217
+ },
1218
+ {
1219
+ "epoch": 0.82,
1220
+ "grad_norm": 0.23995471000671387,
1221
+ "learning_rate": 4.931683007570571e-05,
1222
+ "loss": 0.709,
1223
+ "step": 1700
1224
+ },
1225
+ {
1226
+ "epoch": 0.83,
1227
+ "grad_norm": 0.22278735041618347,
1228
+ "learning_rate": 4.9297179322159856e-05,
1229
+ "loss": 0.7116,
1230
+ "step": 1710
1231
+ },
1232
+ {
1233
+ "epoch": 0.83,
1234
+ "grad_norm": 0.2096502035856247,
1235
+ "learning_rate": 4.927725397111093e-05,
1236
+ "loss": 0.7061,
1237
+ "step": 1720
1238
+ },
1239
+ {
1240
+ "epoch": 0.84,
1241
+ "grad_norm": 0.22335517406463623,
1242
+ "learning_rate": 4.9257054247747644e-05,
1243
+ "loss": 0.6982,
1244
+ "step": 1730
1245
+ },
1246
+ {
1247
+ "epoch": 0.84,
1248
+ "grad_norm": 0.18597687780857086,
1249
+ "learning_rate": 4.923658038035965e-05,
1250
+ "loss": 0.6927,
1251
+ "step": 1740
1252
+ },
1253
+ {
1254
+ "epoch": 0.85,
1255
+ "grad_norm": 0.1971946507692337,
1256
+ "learning_rate": 4.9215832600334816e-05,
1257
+ "loss": 0.6768,
1258
+ "step": 1750
1259
+ },
1260
+ {
1261
+ "epoch": 0.85,
1262
+ "grad_norm": 0.2101408839225769,
1263
+ "learning_rate": 4.9194811142156674e-05,
1264
+ "loss": 0.6933,
1265
+ "step": 1760
1266
+ },
1267
+ {
1268
+ "epoch": 0.86,
1269
+ "grad_norm": 0.17352430522441864,
1270
+ "learning_rate": 4.917351624340178e-05,
1271
+ "loss": 0.6918,
1272
+ "step": 1770
1273
+ },
1274
+ {
1275
+ "epoch": 0.86,
1276
+ "grad_norm": 0.18568973243236542,
1277
+ "learning_rate": 4.915194814473699e-05,
1278
+ "loss": 0.6711,
1279
+ "step": 1780
1280
+ },
1281
+ {
1282
+ "epoch": 0.87,
1283
+ "grad_norm": 0.22991472482681274,
1284
+ "learning_rate": 4.913010708991679e-05,
1285
+ "loss": 0.6938,
1286
+ "step": 1790
1287
+ },
1288
+ {
1289
+ "epoch": 0.87,
1290
+ "grad_norm": 0.201872780919075,
1291
+ "learning_rate": 4.910799332578048e-05,
1292
+ "loss": 0.7051,
1293
+ "step": 1800
1294
+ },
1295
+ {
1296
+ "epoch": 0.87,
1297
+ "grad_norm": 0.2069864720106125,
1298
+ "learning_rate": 4.908560710224943e-05,
1299
+ "loss": 0.6983,
1300
+ "step": 1810
1301
+ },
1302
+ {
1303
+ "epoch": 0.88,
1304
+ "grad_norm": 0.2075955718755722,
1305
+ "learning_rate": 4.906294867232426e-05,
1306
+ "loss": 0.6996,
1307
+ "step": 1820
1308
+ },
1309
+ {
1310
+ "epoch": 0.88,
1311
+ "grad_norm": 0.19524317979812622,
1312
+ "learning_rate": 4.904001829208194e-05,
1313
+ "loss": 0.691,
1314
+ "step": 1830
1315
+ },
1316
+ {
1317
+ "epoch": 0.89,
1318
+ "grad_norm": 0.1967206597328186,
1319
+ "learning_rate": 4.901681622067293e-05,
1320
+ "loss": 0.6964,
1321
+ "step": 1840
1322
+ },
1323
+ {
1324
+ "epoch": 0.89,
1325
+ "grad_norm": 0.2195650041103363,
1326
+ "learning_rate": 4.899334272031825e-05,
1327
+ "loss": 0.6982,
1328
+ "step": 1850
1329
+ },
1330
+ {
1331
+ "epoch": 0.9,
1332
+ "grad_norm": 0.23072399199008942,
1333
+ "learning_rate": 4.8969598056306475e-05,
1334
+ "loss": 0.6838,
1335
+ "step": 1860
1336
+ },
1337
+ {
1338
+ "epoch": 0.9,
1339
+ "grad_norm": 0.1950792670249939,
1340
+ "learning_rate": 4.89455824969908e-05,
1341
+ "loss": 0.6984,
1342
+ "step": 1870
1343
+ },
1344
+ {
1345
+ "epoch": 0.91,
1346
+ "grad_norm": 0.2289438396692276,
1347
+ "learning_rate": 4.892129631378597e-05,
1348
+ "loss": 0.7012,
1349
+ "step": 1880
1350
+ },
1351
+ {
1352
+ "epoch": 0.91,
1353
+ "grad_norm": 0.22316494584083557,
1354
+ "learning_rate": 4.8896739781165215e-05,
1355
+ "loss": 0.6904,
1356
+ "step": 1890
1357
+ },
1358
+ {
1359
+ "epoch": 0.92,
1360
+ "grad_norm": 0.24066634476184845,
1361
+ "learning_rate": 4.8871913176657135e-05,
1362
+ "loss": 0.6974,
1363
+ "step": 1900
1364
+ },
1365
+ {
1366
+ "epoch": 0.92,
1367
+ "grad_norm": 0.18922199308872223,
1368
+ "learning_rate": 4.8846816780842606e-05,
1369
+ "loss": 0.686,
1370
+ "step": 1910
1371
+ },
1372
+ {
1373
+ "epoch": 0.93,
1374
+ "grad_norm": 0.2425553798675537,
1375
+ "learning_rate": 4.882145087735158e-05,
1376
+ "loss": 0.715,
1377
+ "step": 1920
1378
+ },
1379
+ {
1380
+ "epoch": 0.93,
1381
+ "grad_norm": 0.2267230898141861,
1382
+ "learning_rate": 4.879581575285988e-05,
1383
+ "loss": 0.698,
1384
+ "step": 1930
1385
+ },
1386
+ {
1387
+ "epoch": 0.94,
1388
+ "grad_norm": 0.22119645774364471,
1389
+ "learning_rate": 4.876991169708592e-05,
1390
+ "loss": 0.6942,
1391
+ "step": 1940
1392
+ },
1393
+ {
1394
+ "epoch": 0.94,
1395
+ "grad_norm": 0.22854161262512207,
1396
+ "learning_rate": 4.874373900278756e-05,
1397
+ "loss": 0.6912,
1398
+ "step": 1950
1399
+ },
1400
+ {
1401
+ "epoch": 0.95,
1402
+ "grad_norm": 0.1927136778831482,
1403
+ "learning_rate": 4.871729796575863e-05,
1404
+ "loss": 0.6829,
1405
+ "step": 1960
1406
+ },
1407
+ {
1408
+ "epoch": 0.95,
1409
+ "grad_norm": 0.2238394170999527,
1410
+ "learning_rate": 4.869058888482571e-05,
1411
+ "loss": 0.6831,
1412
+ "step": 1970
1413
+ },
1414
+ {
1415
+ "epoch": 0.96,
1416
+ "grad_norm": 0.17643775045871735,
1417
+ "learning_rate": 4.86636120618447e-05,
1418
+ "loss": 0.6875,
1419
+ "step": 1980
1420
+ },
1421
+ {
1422
+ "epoch": 0.96,
1423
+ "grad_norm": 0.2161882072687149,
1424
+ "learning_rate": 4.863636780169742e-05,
1425
+ "loss": 0.6893,
1426
+ "step": 1990
1427
+ },
1428
+ {
1429
+ "epoch": 0.97,
1430
+ "grad_norm": 0.20689886808395386,
1431
+ "learning_rate": 4.860885641228815e-05,
1432
+ "loss": 0.6721,
1433
+ "step": 2000
1434
+ },
1435
+ {
1436
+ "epoch": 0.97,
1437
+ "eval_loss": 0.701475203037262,
1438
+ "eval_runtime": 43.016,
1439
+ "eval_samples_per_second": 46.494,
1440
+ "eval_steps_per_second": 0.372,
1441
+ "step": 2000
1442
+ },
1443
+ {
1444
+ "epoch": 0.97,
1445
+ "grad_norm": 0.18866655230522156,
1446
+ "learning_rate": 4.8581078204540196e-05,
1447
+ "loss": 0.6941,
1448
+ "step": 2010
1449
+ },
1450
+ {
1451
+ "epoch": 0.98,
1452
+ "grad_norm": 0.20633092522621155,
1453
+ "learning_rate": 4.855303349239231e-05,
1454
+ "loss": 0.6943,
1455
+ "step": 2020
1456
+ },
1457
+ {
1458
+ "epoch": 0.98,
1459
+ "grad_norm": 0.2335944026708603,
1460
+ "learning_rate": 4.85247225927952e-05,
1461
+ "loss": 0.6986,
1462
+ "step": 2030
1463
+ },
1464
+ {
1465
+ "epoch": 0.99,
1466
+ "grad_norm": 0.18950659036636353,
1467
+ "learning_rate": 4.8496145825707905e-05,
1468
+ "loss": 0.6894,
1469
+ "step": 2040
1470
+ },
1471
+ {
1472
+ "epoch": 0.99,
1473
+ "grad_norm": 0.22117875516414642,
1474
+ "learning_rate": 4.8467303514094204e-05,
1475
+ "loss": 0.6956,
1476
+ "step": 2050
1477
+ },
1478
+ {
1479
+ "epoch": 1.0,
1480
+ "grad_norm": 0.23478396236896515,
1481
+ "learning_rate": 4.8438195983918967e-05,
1482
+ "loss": 0.6919,
1483
+ "step": 2060
1484
+ },
1485
+ {
1486
+ "epoch": 1.0,
1487
+ "grad_norm": 0.24754443764686584,
1488
+ "learning_rate": 4.8408823564144454e-05,
1489
+ "loss": 0.6904,
1490
+ "step": 2070
1491
+ },
1492
+ {
1493
+ "epoch": 1.01,
1494
+ "grad_norm": 0.19141903519630432,
1495
+ "learning_rate": 4.837918658672661e-05,
1496
+ "loss": 0.6914,
1497
+ "step": 2080
1498
+ },
1499
+ {
1500
+ "epoch": 1.01,
1501
+ "grad_norm": 0.21387813985347748,
1502
+ "learning_rate": 4.834928538661131e-05,
1503
+ "loss": 0.6965,
1504
+ "step": 2090
1505
+ },
1506
+ {
1507
+ "epoch": 1.01,
1508
+ "grad_norm": 0.20495550334453583,
1509
+ "learning_rate": 4.831912030173058e-05,
1510
+ "loss": 0.6949,
1511
+ "step": 2100
1512
+ },
1513
+ {
1514
+ "epoch": 1.02,
1515
+ "grad_norm": 0.35174649953842163,
1516
+ "learning_rate": 4.8288691672998765e-05,
1517
+ "loss": 0.6813,
1518
+ "step": 2110
1519
+ },
1520
+ {
1521
+ "epoch": 1.02,
1522
+ "grad_norm": 0.2461618334054947,
1523
+ "learning_rate": 4.825799984430867e-05,
1524
+ "loss": 0.6687,
1525
+ "step": 2120
1526
+ },
1527
+ {
1528
+ "epoch": 1.03,
1529
+ "grad_norm": 0.20039792358875275,
1530
+ "learning_rate": 4.8227045162527714e-05,
1531
+ "loss": 0.6737,
1532
+ "step": 2130
1533
+ },
1534
+ {
1535
+ "epoch": 1.03,
1536
+ "grad_norm": 0.19774436950683594,
1537
+ "learning_rate": 4.8195827977493955e-05,
1538
+ "loss": 0.6815,
1539
+ "step": 2140
1540
+ },
1541
+ {
1542
+ "epoch": 1.04,
1543
+ "grad_norm": 0.21472251415252686,
1544
+ "learning_rate": 4.8164348642012194e-05,
1545
+ "loss": 0.6957,
1546
+ "step": 2150
1547
+ },
1548
+ {
1549
+ "epoch": 1.04,
1550
+ "grad_norm": 0.1990274041891098,
1551
+ "learning_rate": 4.813260751184992e-05,
1552
+ "loss": 0.6916,
1553
+ "step": 2160
1554
+ },
1555
+ {
1556
+ "epoch": 1.05,
1557
+ "grad_norm": 0.20387563109397888,
1558
+ "learning_rate": 4.810060494573335e-05,
1559
+ "loss": 0.6814,
1560
+ "step": 2170
1561
+ },
1562
+ {
1563
+ "epoch": 1.05,
1564
+ "grad_norm": 0.1792985051870346,
1565
+ "learning_rate": 4.806834130534336e-05,
1566
+ "loss": 0.698,
1567
+ "step": 2180
1568
+ },
1569
+ {
1570
+ "epoch": 1.06,
1571
+ "grad_norm": 0.22255714237689972,
1572
+ "learning_rate": 4.803581695531134e-05,
1573
+ "loss": 0.6765,
1574
+ "step": 2190
1575
+ },
1576
+ {
1577
+ "epoch": 1.06,
1578
+ "grad_norm": 0.19549421966075897,
1579
+ "learning_rate": 4.8003032263215185e-05,
1580
+ "loss": 0.6867,
1581
+ "step": 2200
1582
+ },
1583
+ {
1584
+ "epoch": 1.07,
1585
+ "grad_norm": 0.1929677277803421,
1586
+ "learning_rate": 4.796998759957504e-05,
1587
+ "loss": 0.6935,
1588
+ "step": 2210
1589
+ },
1590
+ {
1591
+ "epoch": 1.07,
1592
+ "grad_norm": 0.22705228626728058,
1593
+ "learning_rate": 4.793668333784915e-05,
1594
+ "loss": 0.6897,
1595
+ "step": 2220
1596
+ },
1597
+ {
1598
+ "epoch": 1.08,
1599
+ "grad_norm": 0.21798522770404816,
1600
+ "learning_rate": 4.790311985442966e-05,
1601
+ "loss": 0.6694,
1602
+ "step": 2230
1603
+ },
1604
+ {
1605
+ "epoch": 1.08,
1606
+ "grad_norm": 0.19095788896083832,
1607
+ "learning_rate": 4.7869297528638315e-05,
1608
+ "loss": 0.6699,
1609
+ "step": 2240
1610
+ },
1611
+ {
1612
+ "epoch": 1.09,
1613
+ "grad_norm": 0.23002374172210693,
1614
+ "learning_rate": 4.7835216742722225e-05,
1615
+ "loss": 0.6807,
1616
+ "step": 2250
1617
+ },
1618
+ {
1619
+ "epoch": 1.09,
1620
+ "grad_norm": 0.2107260823249817,
1621
+ "learning_rate": 4.780087788184947e-05,
1622
+ "loss": 0.6855,
1623
+ "step": 2260
1624
+ },
1625
+ {
1626
+ "epoch": 1.1,
1627
+ "grad_norm": 0.23853424191474915,
1628
+ "learning_rate": 4.776628133410487e-05,
1629
+ "loss": 0.6946,
1630
+ "step": 2270
1631
+ },
1632
+ {
1633
+ "epoch": 1.1,
1634
+ "grad_norm": 0.25372380018234253,
1635
+ "learning_rate": 4.7731427490485455e-05,
1636
+ "loss": 0.6724,
1637
+ "step": 2280
1638
+ },
1639
+ {
1640
+ "epoch": 1.11,
1641
+ "grad_norm": 0.182255819439888,
1642
+ "learning_rate": 4.7696316744896146e-05,
1643
+ "loss": 0.6976,
1644
+ "step": 2290
1645
+ },
1646
+ {
1647
+ "epoch": 1.11,
1648
+ "grad_norm": 0.20137862861156464,
1649
+ "learning_rate": 4.7660949494145276e-05,
1650
+ "loss": 0.6853,
1651
+ "step": 2300
1652
+ },
1653
+ {
1654
+ "epoch": 1.12,
1655
+ "grad_norm": 0.19899927079677582,
1656
+ "learning_rate": 4.7625326137940106e-05,
1657
+ "loss": 0.6696,
1658
+ "step": 2310
1659
+ },
1660
+ {
1661
+ "epoch": 1.12,
1662
+ "grad_norm": 0.24147948622703552,
1663
+ "learning_rate": 4.758944707888228e-05,
1664
+ "loss": 0.6615,
1665
+ "step": 2320
1666
+ },
1667
+ {
1668
+ "epoch": 1.13,
1669
+ "grad_norm": 0.19519510865211487,
1670
+ "learning_rate": 4.7553312722463305e-05,
1671
+ "loss": 0.685,
1672
+ "step": 2330
1673
+ },
1674
+ {
1675
+ "epoch": 1.13,
1676
+ "grad_norm": 0.20128916203975677,
1677
+ "learning_rate": 4.751692347705998e-05,
1678
+ "loss": 0.6829,
1679
+ "step": 2340
1680
+ },
1681
+ {
1682
+ "epoch": 1.14,
1683
+ "grad_norm": 0.16112156212329865,
1684
+ "learning_rate": 4.748027975392976e-05,
1685
+ "loss": 0.6714,
1686
+ "step": 2350
1687
+ },
1688
+ {
1689
+ "epoch": 1.14,
1690
+ "grad_norm": 0.2659178674221039,
1691
+ "learning_rate": 4.744338196720608e-05,
1692
+ "loss": 0.6834,
1693
+ "step": 2360
1694
+ },
1695
+ {
1696
+ "epoch": 1.15,
1697
+ "grad_norm": 0.20477250218391418,
1698
+ "learning_rate": 4.740623053389374e-05,
1699
+ "loss": 0.6664,
1700
+ "step": 2370
1701
+ },
1702
+ {
1703
+ "epoch": 1.15,
1704
+ "grad_norm": 0.18069259822368622,
1705
+ "learning_rate": 4.7368825873864154e-05,
1706
+ "loss": 0.669,
1707
+ "step": 2380
1708
+ },
1709
+ {
1710
+ "epoch": 1.16,
1711
+ "grad_norm": 0.2146090418100357,
1712
+ "learning_rate": 4.733116840985058e-05,
1713
+ "loss": 0.6836,
1714
+ "step": 2390
1715
+ },
1716
+ {
1717
+ "epoch": 1.16,
1718
+ "grad_norm": 0.20656266808509827,
1719
+ "learning_rate": 4.729325856744341e-05,
1720
+ "loss": 0.6818,
1721
+ "step": 2400
1722
+ },
1723
+ {
1724
+ "epoch": 1.16,
1725
+ "grad_norm": 0.1764717549085617,
1726
+ "learning_rate": 4.725509677508528e-05,
1727
+ "loss": 0.7012,
1728
+ "step": 2410
1729
+ },
1730
+ {
1731
+ "epoch": 1.17,
1732
+ "grad_norm": 0.19059453904628754,
1733
+ "learning_rate": 4.721668346406631e-05,
1734
+ "loss": 0.6885,
1735
+ "step": 2420
1736
+ },
1737
+ {
1738
+ "epoch": 1.17,
1739
+ "grad_norm": 0.1867195963859558,
1740
+ "learning_rate": 4.7178019068519165e-05,
1741
+ "loss": 0.6704,
1742
+ "step": 2430
1743
+ },
1744
+ {
1745
+ "epoch": 1.18,
1746
+ "grad_norm": 0.19844096899032593,
1747
+ "learning_rate": 4.713910402541416e-05,
1748
+ "loss": 0.6746,
1749
+ "step": 2440
1750
+ },
1751
+ {
1752
+ "epoch": 1.18,
1753
+ "grad_norm": 0.21426193416118622,
1754
+ "learning_rate": 4.709993877455436e-05,
1755
+ "loss": 0.6627,
1756
+ "step": 2450
1757
+ },
1758
+ {
1759
+ "epoch": 1.19,
1760
+ "grad_norm": 0.19326262176036835,
1761
+ "learning_rate": 4.706052375857058e-05,
1762
+ "loss": 0.6652,
1763
+ "step": 2460
1764
+ },
1765
+ {
1766
+ "epoch": 1.19,
1767
+ "grad_norm": 0.229305237531662,
1768
+ "learning_rate": 4.7020859422916365e-05,
1769
+ "loss": 0.6746,
1770
+ "step": 2470
1771
+ },
1772
+ {
1773
+ "epoch": 1.2,
1774
+ "grad_norm": 0.20060321688652039,
1775
+ "learning_rate": 4.698094621586299e-05,
1776
+ "loss": 0.6658,
1777
+ "step": 2480
1778
+ },
1779
+ {
1780
+ "epoch": 1.2,
1781
+ "grad_norm": 0.21455462276935577,
1782
+ "learning_rate": 4.694078458849438e-05,
1783
+ "loss": 0.6966,
1784
+ "step": 2490
1785
+ },
1786
+ {
1787
+ "epoch": 1.21,
1788
+ "grad_norm": 0.18379837274551392,
1789
+ "learning_rate": 4.690037499470202e-05,
1790
+ "loss": 0.6753,
1791
+ "step": 2500
1792
+ },
1793
+ {
1794
+ "epoch": 1.21,
1795
+ "eval_loss": 0.6940956711769104,
1796
+ "eval_runtime": 42.7889,
1797
+ "eval_samples_per_second": 46.741,
1798
+ "eval_steps_per_second": 0.374,
1799
+ "step": 2500
1800
+ },
1801
+ {
1802
+ "epoch": 1.21,
1803
+ "grad_norm": 0.19967205822467804,
1804
+ "learning_rate": 4.68597178911798e-05,
1805
+ "loss": 0.697,
1806
+ "step": 2510
1807
+ },
1808
+ {
1809
+ "epoch": 1.22,
1810
+ "grad_norm": 0.1688542366027832,
1811
+ "learning_rate": 4.681881373741888e-05,
1812
+ "loss": 0.681,
1813
+ "step": 2520
1814
+ },
1815
+ {
1816
+ "epoch": 1.22,
1817
+ "grad_norm": 0.19124309718608856,
1818
+ "learning_rate": 4.67776629957025e-05,
1819
+ "loss": 0.6855,
1820
+ "step": 2530
1821
+ },
1822
+ {
1823
+ "epoch": 1.23,
1824
+ "grad_norm": 0.22720417380332947,
1825
+ "learning_rate": 4.6736266131100706e-05,
1826
+ "loss": 0.6747,
1827
+ "step": 2540
1828
+ },
1829
+ {
1830
+ "epoch": 1.23,
1831
+ "grad_norm": 0.21285779774188995,
1832
+ "learning_rate": 4.6694623611465185e-05,
1833
+ "loss": 0.687,
1834
+ "step": 2550
1835
+ },
1836
+ {
1837
+ "epoch": 1.24,
1838
+ "grad_norm": 0.1932476907968521,
1839
+ "learning_rate": 4.6652735907423886e-05,
1840
+ "loss": 0.6745,
1841
+ "step": 2560
1842
+ },
1843
+ {
1844
+ "epoch": 1.24,
1845
+ "grad_norm": 0.2050684541463852,
1846
+ "learning_rate": 4.661060349237574e-05,
1847
+ "loss": 0.6865,
1848
+ "step": 2570
1849
+ },
1850
+ {
1851
+ "epoch": 1.25,
1852
+ "grad_norm": 0.20403894782066345,
1853
+ "learning_rate": 4.656822684248533e-05,
1854
+ "loss": 0.6897,
1855
+ "step": 2580
1856
+ },
1857
+ {
1858
+ "epoch": 1.25,
1859
+ "grad_norm": 0.20080254971981049,
1860
+ "learning_rate": 4.652560643667747e-05,
1861
+ "loss": 0.6768,
1862
+ "step": 2590
1863
+ },
1864
+ {
1865
+ "epoch": 1.26,
1866
+ "grad_norm": 0.20147019624710083,
1867
+ "learning_rate": 4.648274275663183e-05,
1868
+ "loss": 0.6911,
1869
+ "step": 2600
1870
+ },
1871
+ {
1872
+ "epoch": 1.26,
1873
+ "grad_norm": 0.17825543880462646,
1874
+ "learning_rate": 4.643963628677743e-05,
1875
+ "loss": 0.6635,
1876
+ "step": 2610
1877
+ },
1878
+ {
1879
+ "epoch": 1.27,
1880
+ "grad_norm": 0.20095126330852509,
1881
+ "learning_rate": 4.6396287514287275e-05,
1882
+ "loss": 0.664,
1883
+ "step": 2620
1884
+ },
1885
+ {
1886
+ "epoch": 1.27,
1887
+ "grad_norm": 0.20084308087825775,
1888
+ "learning_rate": 4.6352696929072727e-05,
1889
+ "loss": 0.6653,
1890
+ "step": 2630
1891
+ },
1892
+ {
1893
+ "epoch": 1.28,
1894
+ "grad_norm": 0.21178288757801056,
1895
+ "learning_rate": 4.630886502377805e-05,
1896
+ "loss": 0.6981,
1897
+ "step": 2640
1898
+ },
1899
+ {
1900
+ "epoch": 1.28,
1901
+ "grad_norm": 0.1963459551334381,
1902
+ "learning_rate": 4.62647922937748e-05,
1903
+ "loss": 0.6666,
1904
+ "step": 2650
1905
+ },
1906
+ {
1907
+ "epoch": 1.29,
1908
+ "grad_norm": 0.21702295541763306,
1909
+ "learning_rate": 4.6220479237156254e-05,
1910
+ "loss": 0.6868,
1911
+ "step": 2660
1912
+ },
1913
+ {
1914
+ "epoch": 1.29,
1915
+ "grad_norm": 0.20558403432369232,
1916
+ "learning_rate": 4.6175926354731785e-05,
1917
+ "loss": 0.6794,
1918
+ "step": 2670
1919
+ },
1920
+ {
1921
+ "epoch": 1.3,
1922
+ "grad_norm": 0.20398521423339844,
1923
+ "learning_rate": 4.613113415002115e-05,
1924
+ "loss": 0.69,
1925
+ "step": 2680
1926
+ },
1927
+ {
1928
+ "epoch": 1.3,
1929
+ "grad_norm": 0.2166917473077774,
1930
+ "learning_rate": 4.6086103129248846e-05,
1931
+ "loss": 0.6604,
1932
+ "step": 2690
1933
+ },
1934
+ {
1935
+ "epoch": 1.3,
1936
+ "grad_norm": 0.2121194452047348,
1937
+ "learning_rate": 4.604083380133841e-05,
1938
+ "loss": 0.6653,
1939
+ "step": 2700
1940
+ },
1941
+ {
1942
+ "epoch": 1.31,
1943
+ "grad_norm": 0.252469927072525,
1944
+ "learning_rate": 4.5995326677906605e-05,
1945
+ "loss": 0.6678,
1946
+ "step": 2710
1947
+ },
1948
+ {
1949
+ "epoch": 1.31,
1950
+ "grad_norm": 0.1864558607339859,
1951
+ "learning_rate": 4.5949582273257656e-05,
1952
+ "loss": 0.6806,
1953
+ "step": 2720
1954
+ },
1955
+ {
1956
+ "epoch": 1.32,
1957
+ "grad_norm": 0.19438904523849487,
1958
+ "learning_rate": 4.59036011043775e-05,
1959
+ "loss": 0.6832,
1960
+ "step": 2730
1961
+ },
1962
+ {
1963
+ "epoch": 1.32,
1964
+ "grad_norm": 0.1954311728477478,
1965
+ "learning_rate": 4.5857383690927844e-05,
1966
+ "loss": 0.6772,
1967
+ "step": 2740
1968
+ },
1969
+ {
1970
+ "epoch": 1.33,
1971
+ "grad_norm": 0.38669708371162415,
1972
+ "learning_rate": 4.5810930555240364e-05,
1973
+ "loss": 0.6663,
1974
+ "step": 2750
1975
+ },
1976
+ {
1977
+ "epoch": 1.33,
1978
+ "grad_norm": 0.5850951075553894,
1979
+ "learning_rate": 4.576424222231078e-05,
1980
+ "loss": 0.6578,
1981
+ "step": 2760
1982
+ },
1983
+ {
1984
+ "epoch": 1.34,
1985
+ "grad_norm": 0.22347845137119293,
1986
+ "learning_rate": 4.571731921979292e-05,
1987
+ "loss": 0.6735,
1988
+ "step": 2770
1989
+ },
1990
+ {
1991
+ "epoch": 1.34,
1992
+ "grad_norm": 0.22810417413711548,
1993
+ "learning_rate": 4.567016207799276e-05,
1994
+ "loss": 0.6628,
1995
+ "step": 2780
1996
+ },
1997
+ {
1998
+ "epoch": 1.35,
1999
+ "grad_norm": 0.18823856115341187,
2000
+ "learning_rate": 4.562277132986241e-05,
2001
+ "loss": 0.6803,
2002
+ "step": 2790
2003
+ },
2004
+ {
2005
+ "epoch": 1.35,
2006
+ "grad_norm": 0.219554141163826,
2007
+ "learning_rate": 4.557514751099415e-05,
2008
+ "loss": 0.6701,
2009
+ "step": 2800
2010
+ },
2011
+ {
2012
+ "epoch": 1.36,
2013
+ "grad_norm": 0.1830976903438568,
2014
+ "learning_rate": 4.55272911596143e-05,
2015
+ "loss": 0.6864,
2016
+ "step": 2810
2017
+ },
2018
+ {
2019
+ "epoch": 1.36,
2020
+ "grad_norm": 0.18995565176010132,
2021
+ "learning_rate": 4.5479202816577195e-05,
2022
+ "loss": 0.6683,
2023
+ "step": 2820
2024
+ },
2025
+ {
2026
+ "epoch": 1.37,
2027
+ "grad_norm": 0.19047056138515472,
2028
+ "learning_rate": 4.543088302535903e-05,
2029
+ "loss": 0.6855,
2030
+ "step": 2830
2031
+ },
2032
+ {
2033
+ "epoch": 1.37,
2034
+ "grad_norm": 0.18707235157489777,
2035
+ "learning_rate": 4.538233233205177e-05,
2036
+ "loss": 0.6786,
2037
+ "step": 2840
2038
+ },
2039
+ {
2040
+ "epoch": 1.38,
2041
+ "grad_norm": 0.19992856681346893,
2042
+ "learning_rate": 4.533355128535693e-05,
2043
+ "loss": 0.6676,
2044
+ "step": 2850
2045
+ },
2046
+ {
2047
+ "epoch": 1.38,
2048
+ "grad_norm": 0.20344853401184082,
2049
+ "learning_rate": 4.5284540436579395e-05,
2050
+ "loss": 0.6938,
2051
+ "step": 2860
2052
+ },
2053
+ {
2054
+ "epoch": 1.39,
2055
+ "grad_norm": 0.1867518275976181,
2056
+ "learning_rate": 4.5235300339621164e-05,
2057
+ "loss": 0.6736,
2058
+ "step": 2870
2059
+ },
2060
+ {
2061
+ "epoch": 1.39,
2062
+ "grad_norm": 0.23014211654663086,
2063
+ "learning_rate": 4.518583155097517e-05,
2064
+ "loss": 0.674,
2065
+ "step": 2880
2066
+ },
2067
+ {
2068
+ "epoch": 1.4,
2069
+ "grad_norm": 0.1905904859304428,
2070
+ "learning_rate": 4.5136134629718855e-05,
2071
+ "loss": 0.6704,
2072
+ "step": 2890
2073
+ },
2074
+ {
2075
+ "epoch": 1.4,
2076
+ "grad_norm": 0.2236865758895874,
2077
+ "learning_rate": 4.5086210137508e-05,
2078
+ "loss": 0.6633,
2079
+ "step": 2900
2080
+ },
2081
+ {
2082
+ "epoch": 1.41,
2083
+ "grad_norm": 0.19701044261455536,
2084
+ "learning_rate": 4.5036058638570264e-05,
2085
+ "loss": 0.6626,
2086
+ "step": 2910
2087
+ },
2088
+ {
2089
+ "epoch": 1.41,
2090
+ "grad_norm": 0.19464708864688873,
2091
+ "learning_rate": 4.4985680699698855e-05,
2092
+ "loss": 0.6595,
2093
+ "step": 2920
2094
+ },
2095
+ {
2096
+ "epoch": 1.42,
2097
+ "grad_norm": 0.19034789502620697,
2098
+ "learning_rate": 4.493507689024614e-05,
2099
+ "loss": 0.6688,
2100
+ "step": 2930
2101
+ },
2102
+ {
2103
+ "epoch": 1.42,
2104
+ "grad_norm": 0.19750766456127167,
2105
+ "learning_rate": 4.488424778211717e-05,
2106
+ "loss": 0.6521,
2107
+ "step": 2940
2108
+ },
2109
+ {
2110
+ "epoch": 1.43,
2111
+ "grad_norm": 0.20544354617595673,
2112
+ "learning_rate": 4.483319394976325e-05,
2113
+ "loss": 0.6947,
2114
+ "step": 2950
2115
+ },
2116
+ {
2117
+ "epoch": 1.43,
2118
+ "grad_norm": 0.22409361600875854,
2119
+ "learning_rate": 4.478191597017541e-05,
2120
+ "loss": 0.674,
2121
+ "step": 2960
2122
+ },
2123
+ {
2124
+ "epoch": 1.44,
2125
+ "grad_norm": 0.19409336149692535,
2126
+ "learning_rate": 4.473041442287793e-05,
2127
+ "loss": 0.6978,
2128
+ "step": 2970
2129
+ },
2130
+ {
2131
+ "epoch": 1.44,
2132
+ "grad_norm": 0.20303764939308167,
2133
+ "learning_rate": 4.4678689889921755e-05,
2134
+ "loss": 0.6771,
2135
+ "step": 2980
2136
+ },
2137
+ {
2138
+ "epoch": 1.45,
2139
+ "grad_norm": 0.193495512008667,
2140
+ "learning_rate": 4.462674295587794e-05,
2141
+ "loss": 0.6693,
2142
+ "step": 2990
2143
+ },
2144
+ {
2145
+ "epoch": 1.45,
2146
+ "grad_norm": 0.18171487748622894,
2147
+ "learning_rate": 4.457457420783103e-05,
2148
+ "loss": 0.6716,
2149
+ "step": 3000
2150
+ },
2151
+ {
2152
+ "epoch": 1.45,
2153
+ "eval_loss": 0.6894034147262573,
2154
+ "eval_runtime": 43.0285,
2155
+ "eval_samples_per_second": 46.481,
2156
+ "eval_steps_per_second": 0.372,
2157
+ "step": 3000
2158
+ },
2159
+ {
2160
+ "epoch": 1.45,
2161
+ "grad_norm": 0.22247907519340515,
2162
+ "learning_rate": 4.452218423537241e-05,
2163
+ "loss": 0.6753,
2164
+ "step": 3010
2165
+ },
2166
+ {
2167
+ "epoch": 1.46,
2168
+ "grad_norm": 0.2469698041677475,
2169
+ "learning_rate": 4.4469573630593686e-05,
2170
+ "loss": 0.6715,
2171
+ "step": 3020
2172
+ },
2173
+ {
2174
+ "epoch": 1.46,
2175
+ "grad_norm": 0.18804265558719635,
2176
+ "learning_rate": 4.4416742988079945e-05,
2177
+ "loss": 0.6546,
2178
+ "step": 3030
2179
+ },
2180
+ {
2181
+ "epoch": 1.47,
2182
+ "grad_norm": 0.19240662455558777,
2183
+ "learning_rate": 4.436369290490307e-05,
2184
+ "loss": 0.6857,
2185
+ "step": 3040
2186
+ },
2187
+ {
2188
+ "epoch": 1.47,
2189
+ "grad_norm": 0.22342869639396667,
2190
+ "learning_rate": 4.431042398061499e-05,
2191
+ "loss": 0.6938,
2192
+ "step": 3050
2193
+ },
2194
+ {
2195
+ "epoch": 1.48,
2196
+ "grad_norm": 0.20553608238697052,
2197
+ "learning_rate": 4.425693681724086e-05,
2198
+ "loss": 0.6687,
2199
+ "step": 3060
2200
+ },
2201
+ {
2202
+ "epoch": 1.48,
2203
+ "grad_norm": 0.19881366193294525,
2204
+ "learning_rate": 4.420323201927231e-05,
2205
+ "loss": 0.6644,
2206
+ "step": 3070
2207
+ },
2208
+ {
2209
+ "epoch": 1.49,
2210
+ "grad_norm": 0.18151846528053284,
2211
+ "learning_rate": 4.41493101936606e-05,
2212
+ "loss": 0.6667,
2213
+ "step": 3080
2214
+ },
2215
+ {
2216
+ "epoch": 1.49,
2217
+ "grad_norm": 0.1906978040933609,
2218
+ "learning_rate": 4.409517194980974e-05,
2219
+ "loss": 0.6578,
2220
+ "step": 3090
2221
+ },
2222
+ {
2223
+ "epoch": 1.5,
2224
+ "grad_norm": 0.18123480677604675,
2225
+ "learning_rate": 4.4040817899569644e-05,
2226
+ "loss": 0.681,
2227
+ "step": 3100
2228
+ },
2229
+ {
2230
+ "epoch": 1.5,
2231
+ "grad_norm": 0.18974031507968903,
2232
+ "learning_rate": 4.3986248657229134e-05,
2233
+ "loss": 0.6645,
2234
+ "step": 3110
2235
+ },
2236
+ {
2237
+ "epoch": 1.51,
2238
+ "grad_norm": 0.2028564214706421,
2239
+ "learning_rate": 4.3931464839509105e-05,
2240
+ "loss": 0.6913,
2241
+ "step": 3120
2242
+ },
2243
+ {
2244
+ "epoch": 1.51,
2245
+ "grad_norm": 0.17814776301383972,
2246
+ "learning_rate": 4.387646706555548e-05,
2247
+ "loss": 0.6949,
2248
+ "step": 3130
2249
+ },
2250
+ {
2251
+ "epoch": 1.52,
2252
+ "grad_norm": 0.21407385170459747,
2253
+ "learning_rate": 4.382125595693224e-05,
2254
+ "loss": 0.6656,
2255
+ "step": 3140
2256
+ },
2257
+ {
2258
+ "epoch": 1.52,
2259
+ "grad_norm": 0.21102961897850037,
2260
+ "learning_rate": 4.376583213761438e-05,
2261
+ "loss": 0.6612,
2262
+ "step": 3150
2263
+ },
2264
+ {
2265
+ "epoch": 1.53,
2266
+ "grad_norm": 0.19947603344917297,
2267
+ "learning_rate": 4.371019623398088e-05,
2268
+ "loss": 0.6894,
2269
+ "step": 3160
2270
+ },
2271
+ {
2272
+ "epoch": 1.53,
2273
+ "grad_norm": 0.20861291885375977,
2274
+ "learning_rate": 4.365434887480763e-05,
2275
+ "loss": 0.6739,
2276
+ "step": 3170
2277
+ },
2278
+ {
2279
+ "epoch": 1.54,
2280
+ "grad_norm": 0.2098263055086136,
2281
+ "learning_rate": 4.359829069126028e-05,
2282
+ "loss": 0.6739,
2283
+ "step": 3180
2284
+ },
2285
+ {
2286
+ "epoch": 1.54,
2287
+ "grad_norm": 0.19079270958900452,
2288
+ "learning_rate": 4.3542022316887166e-05,
2289
+ "loss": 0.6739,
2290
+ "step": 3190
2291
+ },
2292
+ {
2293
+ "epoch": 1.55,
2294
+ "grad_norm": 0.21104255318641663,
2295
+ "learning_rate": 4.34855443876121e-05,
2296
+ "loss": 0.652,
2297
+ "step": 3200
2298
+ },
2299
+ {
2300
+ "epoch": 1.55,
2301
+ "grad_norm": 0.19791875779628754,
2302
+ "learning_rate": 4.342885754172721e-05,
2303
+ "loss": 0.6551,
2304
+ "step": 3210
2305
+ },
2306
+ {
2307
+ "epoch": 1.56,
2308
+ "grad_norm": 0.22481437027454376,
2309
+ "learning_rate": 4.337196241988573e-05,
2310
+ "loss": 0.6649,
2311
+ "step": 3220
2312
+ },
2313
+ {
2314
+ "epoch": 1.56,
2315
+ "grad_norm": 0.19483722746372223,
2316
+ "learning_rate": 4.3314859665094745e-05,
2317
+ "loss": 0.6762,
2318
+ "step": 3230
2319
+ },
2320
+ {
2321
+ "epoch": 1.57,
2322
+ "grad_norm": 0.18499700725078583,
2323
+ "learning_rate": 4.3257549922707926e-05,
2324
+ "loss": 0.6754,
2325
+ "step": 3240
2326
+ },
2327
+ {
2328
+ "epoch": 1.57,
2329
+ "grad_norm": 0.21884921193122864,
2330
+ "learning_rate": 4.320003384041823e-05,
2331
+ "loss": 0.6757,
2332
+ "step": 3250
2333
+ },
2334
+ {
2335
+ "epoch": 1.58,
2336
+ "grad_norm": 0.20222313702106476,
2337
+ "learning_rate": 4.314231206825061e-05,
2338
+ "loss": 0.6527,
2339
+ "step": 3260
2340
+ },
2341
+ {
2342
+ "epoch": 1.58,
2343
+ "grad_norm": 0.18140584230422974,
2344
+ "learning_rate": 4.3084385258554635e-05,
2345
+ "loss": 0.6731,
2346
+ "step": 3270
2347
+ },
2348
+ {
2349
+ "epoch": 1.59,
2350
+ "grad_norm": 0.21039867401123047,
2351
+ "learning_rate": 4.302625406599713e-05,
2352
+ "loss": 0.6722,
2353
+ "step": 3280
2354
+ },
2355
+ {
2356
+ "epoch": 1.59,
2357
+ "grad_norm": 0.18162797391414642,
2358
+ "learning_rate": 4.296791914755478e-05,
2359
+ "loss": 0.6563,
2360
+ "step": 3290
2361
+ },
2362
+ {
2363
+ "epoch": 1.59,
2364
+ "grad_norm": 0.178864985704422,
2365
+ "learning_rate": 4.29093811625067e-05,
2366
+ "loss": 0.6666,
2367
+ "step": 3300
2368
+ },
2369
+ {
2370
+ "epoch": 1.6,
2371
+ "grad_norm": 0.21808697283267975,
2372
+ "learning_rate": 4.285064077242699e-05,
2373
+ "loss": 0.6689,
2374
+ "step": 3310
2375
+ },
2376
+ {
2377
+ "epoch": 1.6,
2378
+ "grad_norm": 0.20632699131965637,
2379
+ "learning_rate": 4.279169864117727e-05,
2380
+ "loss": 0.6664,
2381
+ "step": 3320
2382
+ },
2383
+ {
2384
+ "epoch": 1.61,
2385
+ "grad_norm": 0.20482878386974335,
2386
+ "learning_rate": 4.273255543489912e-05,
2387
+ "loss": 0.6407,
2388
+ "step": 3330
2389
+ },
2390
+ {
2391
+ "epoch": 1.61,
2392
+ "grad_norm": 0.18613804876804352,
2393
+ "learning_rate": 4.267321182200664e-05,
2394
+ "loss": 0.666,
2395
+ "step": 3340
2396
+ },
2397
+ {
2398
+ "epoch": 1.62,
2399
+ "grad_norm": 0.23219124972820282,
2400
+ "learning_rate": 4.2613668473178836e-05,
2401
+ "loss": 0.6802,
2402
+ "step": 3350
2403
+ },
2404
+ {
2405
+ "epoch": 1.62,
2406
+ "grad_norm": 0.17973752319812775,
2407
+ "learning_rate": 4.255392606135202e-05,
2408
+ "loss": 0.6571,
2409
+ "step": 3360
2410
+ },
2411
+ {
2412
+ "epoch": 1.63,
2413
+ "grad_norm": 0.21650458872318268,
2414
+ "learning_rate": 4.2493985261712285e-05,
2415
+ "loss": 0.6727,
2416
+ "step": 3370
2417
+ },
2418
+ {
2419
+ "epoch": 1.63,
2420
+ "grad_norm": 0.19476266205310822,
2421
+ "learning_rate": 4.2433846751687815e-05,
2422
+ "loss": 0.6836,
2423
+ "step": 3380
2424
+ },
2425
+ {
2426
+ "epoch": 1.64,
2427
+ "grad_norm": 0.20223549008369446,
2428
+ "learning_rate": 4.237351121094121e-05,
2429
+ "loss": 0.6666,
2430
+ "step": 3390
2431
+ },
2432
+ {
2433
+ "epoch": 1.64,
2434
+ "grad_norm": 0.2078579217195511,
2435
+ "learning_rate": 4.231297932136189e-05,
2436
+ "loss": 0.673,
2437
+ "step": 3400
2438
+ },
2439
+ {
2440
+ "epoch": 1.65,
2441
+ "grad_norm": 0.193430095911026,
2442
+ "learning_rate": 4.225225176705829e-05,
2443
+ "loss": 0.6718,
2444
+ "step": 3410
2445
+ },
2446
+ {
2447
+ "epoch": 1.65,
2448
+ "grad_norm": 0.18017272651195526,
2449
+ "learning_rate": 4.2191329234350194e-05,
2450
+ "loss": 0.6806,
2451
+ "step": 3420
2452
+ },
2453
+ {
2454
+ "epoch": 1.66,
2455
+ "grad_norm": 0.20289075374603271,
2456
+ "learning_rate": 4.213021241176096e-05,
2457
+ "loss": 0.667,
2458
+ "step": 3430
2459
+ },
2460
+ {
2461
+ "epoch": 1.66,
2462
+ "grad_norm": 0.2014080286026001,
2463
+ "learning_rate": 4.2068901990009726e-05,
2464
+ "loss": 0.6441,
2465
+ "step": 3440
2466
+ },
2467
+ {
2468
+ "epoch": 1.67,
2469
+ "grad_norm": 0.17385190725326538,
2470
+ "learning_rate": 4.200739866200363e-05,
2471
+ "loss": 0.6564,
2472
+ "step": 3450
2473
+ },
2474
+ {
2475
+ "epoch": 1.67,
2476
+ "grad_norm": 0.20427776873111725,
2477
+ "learning_rate": 4.1945703122829924e-05,
2478
+ "loss": 0.6509,
2479
+ "step": 3460
2480
+ },
2481
+ {
2482
+ "epoch": 1.68,
2483
+ "grad_norm": 0.2093549370765686,
2484
+ "learning_rate": 4.1883816069748214e-05,
2485
+ "loss": 0.6741,
2486
+ "step": 3470
2487
+ },
2488
+ {
2489
+ "epoch": 1.68,
2490
+ "grad_norm": 0.21946479380130768,
2491
+ "learning_rate": 4.182173820218249e-05,
2492
+ "loss": 0.6778,
2493
+ "step": 3480
2494
+ },
2495
+ {
2496
+ "epoch": 1.69,
2497
+ "grad_norm": 0.2244766503572464,
2498
+ "learning_rate": 4.175947022171326e-05,
2499
+ "loss": 0.681,
2500
+ "step": 3490
2501
+ },
2502
+ {
2503
+ "epoch": 1.69,
2504
+ "grad_norm": 0.22939437627792358,
2505
+ "learning_rate": 4.169701283206961e-05,
2506
+ "loss": 0.6595,
2507
+ "step": 3500
2508
+ },
2509
+ {
2510
+ "epoch": 1.69,
2511
+ "eval_loss": 0.6864724159240723,
2512
+ "eval_runtime": 43.0148,
2513
+ "eval_samples_per_second": 46.496,
2514
+ "eval_steps_per_second": 0.372,
2515
+ "step": 3500
2516
+ },
2517
+ {
2518
+ "epoch": 1.7,
2519
+ "grad_norm": 0.2838324010372162,
2520
+ "learning_rate": 4.163436673912127e-05,
2521
+ "loss": 0.666,
2522
+ "step": 3510
2523
+ },
2524
+ {
2525
+ "epoch": 1.7,
2526
+ "grad_norm": 0.1862134337425232,
2527
+ "learning_rate": 4.1571532650870614e-05,
2528
+ "loss": 0.6671,
2529
+ "step": 3520
2530
+ },
2531
+ {
2532
+ "epoch": 1.71,
2533
+ "grad_norm": 0.2170250117778778,
2534
+ "learning_rate": 4.1508511277444674e-05,
2535
+ "loss": 0.6566,
2536
+ "step": 3530
2537
+ },
2538
+ {
2539
+ "epoch": 1.71,
2540
+ "grad_norm": 0.17831502854824066,
2541
+ "learning_rate": 4.1445303331087096e-05,
2542
+ "loss": 0.6776,
2543
+ "step": 3540
2544
+ },
2545
+ {
2546
+ "epoch": 1.72,
2547
+ "grad_norm": 0.17921501398086548,
2548
+ "learning_rate": 4.138190952615012e-05,
2549
+ "loss": 0.6876,
2550
+ "step": 3550
2551
+ },
2552
+ {
2553
+ "epoch": 1.72,
2554
+ "grad_norm": 0.19673947989940643,
2555
+ "learning_rate": 4.131833057908648e-05,
2556
+ "loss": 0.6596,
2557
+ "step": 3560
2558
+ },
2559
+ {
2560
+ "epoch": 1.73,
2561
+ "grad_norm": 0.1829831749200821,
2562
+ "learning_rate": 4.1254567208441296e-05,
2563
+ "loss": 0.6712,
2564
+ "step": 3570
2565
+ },
2566
+ {
2567
+ "epoch": 1.73,
2568
+ "grad_norm": 0.3023044764995575,
2569
+ "learning_rate": 4.119062013484402e-05,
2570
+ "loss": 0.6747,
2571
+ "step": 3580
2572
+ },
2573
+ {
2574
+ "epoch": 1.74,
2575
+ "grad_norm": 0.21133002638816833,
2576
+ "learning_rate": 4.11264900810002e-05,
2577
+ "loss": 0.674,
2578
+ "step": 3590
2579
+ },
2580
+ {
2581
+ "epoch": 1.74,
2582
+ "grad_norm": 0.19868247210979462,
2583
+ "learning_rate": 4.106217777168339e-05,
2584
+ "loss": 0.68,
2585
+ "step": 3600
2586
+ },
2587
+ {
2588
+ "epoch": 1.74,
2589
+ "grad_norm": 0.21643079817295074,
2590
+ "learning_rate": 4.0997683933726894e-05,
2591
+ "loss": 0.6555,
2592
+ "step": 3610
2593
+ },
2594
+ {
2595
+ "epoch": 1.75,
2596
+ "grad_norm": 0.1953793615102768,
2597
+ "learning_rate": 4.093300929601562e-05,
2598
+ "loss": 0.6704,
2599
+ "step": 3620
2600
+ },
2601
+ {
2602
+ "epoch": 1.75,
2603
+ "grad_norm": 0.21795640885829926,
2604
+ "learning_rate": 4.086815458947777e-05,
2605
+ "loss": 0.6583,
2606
+ "step": 3630
2607
+ },
2608
+ {
2609
+ "epoch": 1.76,
2610
+ "grad_norm": 0.22441154718399048,
2611
+ "learning_rate": 4.080312054707665e-05,
2612
+ "loss": 0.6772,
2613
+ "step": 3640
2614
+ },
2615
+ {
2616
+ "epoch": 1.76,
2617
+ "grad_norm": 0.20651431381702423,
2618
+ "learning_rate": 4.073790790380232e-05,
2619
+ "loss": 0.6793,
2620
+ "step": 3650
2621
+ },
2622
+ {
2623
+ "epoch": 1.77,
2624
+ "grad_norm": 0.21724598109722137,
2625
+ "learning_rate": 4.0672517396663354e-05,
2626
+ "loss": 0.6742,
2627
+ "step": 3660
2628
+ },
2629
+ {
2630
+ "epoch": 1.77,
2631
+ "grad_norm": 0.27277424931526184,
2632
+ "learning_rate": 4.0606949764678434e-05,
2633
+ "loss": 0.6495,
2634
+ "step": 3670
2635
+ },
2636
+ {
2637
+ "epoch": 1.78,
2638
+ "grad_norm": 0.21489045023918152,
2639
+ "learning_rate": 4.054120574886808e-05,
2640
+ "loss": 0.6516,
2641
+ "step": 3680
2642
+ },
2643
+ {
2644
+ "epoch": 1.78,
2645
+ "grad_norm": 0.19597062468528748,
2646
+ "learning_rate": 4.0475286092246215e-05,
2647
+ "loss": 0.6498,
2648
+ "step": 3690
2649
+ },
2650
+ {
2651
+ "epoch": 1.79,
2652
+ "grad_norm": 0.20840667188167572,
2653
+ "learning_rate": 4.0409191539811774e-05,
2654
+ "loss": 0.6394,
2655
+ "step": 3700
2656
+ },
2657
+ {
2658
+ "epoch": 1.79,
2659
+ "grad_norm": 0.23823177814483643,
2660
+ "learning_rate": 4.0342922838540334e-05,
2661
+ "loss": 0.6798,
2662
+ "step": 3710
2663
+ },
2664
+ {
2665
+ "epoch": 1.8,
2666
+ "grad_norm": 0.20505908131599426,
2667
+ "learning_rate": 4.0276480737375606e-05,
2668
+ "loss": 0.6607,
2669
+ "step": 3720
2670
+ },
2671
+ {
2672
+ "epoch": 1.8,
2673
+ "grad_norm": 0.28554779291152954,
2674
+ "learning_rate": 4.0209865987221014e-05,
2675
+ "loss": 0.6467,
2676
+ "step": 3730
2677
+ },
2678
+ {
2679
+ "epoch": 1.81,
2680
+ "grad_norm": 0.24168336391448975,
2681
+ "learning_rate": 4.014307934093119e-05,
2682
+ "loss": 0.6625,
2683
+ "step": 3740
2684
+ },
2685
+ {
2686
+ "epoch": 1.81,
2687
+ "grad_norm": 0.21519850194454193,
2688
+ "learning_rate": 4.007612155330348e-05,
2689
+ "loss": 0.6431,
2690
+ "step": 3750
2691
+ },
2692
+ {
2693
+ "epoch": 1.82,
2694
+ "grad_norm": 0.1954280287027359,
2695
+ "learning_rate": 4.000899338106939e-05,
2696
+ "loss": 0.6669,
2697
+ "step": 3760
2698
+ },
2699
+ {
2700
+ "epoch": 1.82,
2701
+ "grad_norm": 0.22829805314540863,
2702
+ "learning_rate": 3.9941695582886065e-05,
2703
+ "loss": 0.6507,
2704
+ "step": 3770
2705
+ },
2706
+ {
2707
+ "epoch": 1.83,
2708
+ "grad_norm": 0.30064237117767334,
2709
+ "learning_rate": 3.9874228919327685e-05,
2710
+ "loss": 0.664,
2711
+ "step": 3780
2712
+ },
2713
+ {
2714
+ "epoch": 1.83,
2715
+ "grad_norm": 0.21391743421554565,
2716
+ "learning_rate": 3.980659415287689e-05,
2717
+ "loss": 0.6577,
2718
+ "step": 3790
2719
+ },
2720
+ {
2721
+ "epoch": 1.84,
2722
+ "grad_norm": 0.237900048494339,
2723
+ "learning_rate": 3.9738792047916143e-05,
2724
+ "loss": 0.6678,
2725
+ "step": 3800
2726
+ },
2727
+ {
2728
+ "epoch": 1.84,
2729
+ "grad_norm": 0.2035730481147766,
2730
+ "learning_rate": 3.967082337071911e-05,
2731
+ "loss": 0.6822,
2732
+ "step": 3810
2733
+ },
2734
+ {
2735
+ "epoch": 1.85,
2736
+ "grad_norm": 0.25883620977401733,
2737
+ "learning_rate": 3.960268888944202e-05,
2738
+ "loss": 0.6657,
2739
+ "step": 3820
2740
+ },
2741
+ {
2742
+ "epoch": 1.85,
2743
+ "grad_norm": 0.21861082315444946,
2744
+ "learning_rate": 3.9534389374114905e-05,
2745
+ "loss": 0.6608,
2746
+ "step": 3830
2747
+ },
2748
+ {
2749
+ "epoch": 1.86,
2750
+ "grad_norm": 0.2185363620519638,
2751
+ "learning_rate": 3.946592559663299e-05,
2752
+ "loss": 0.6763,
2753
+ "step": 3840
2754
+ },
2755
+ {
2756
+ "epoch": 1.86,
2757
+ "grad_norm": 0.2056983858346939,
2758
+ "learning_rate": 3.9397298330747905e-05,
2759
+ "loss": 0.6703,
2760
+ "step": 3850
2761
+ },
2762
+ {
2763
+ "epoch": 1.87,
2764
+ "grad_norm": 0.19903923571109772,
2765
+ "learning_rate": 3.932850835205899e-05,
2766
+ "loss": 0.6887,
2767
+ "step": 3860
2768
+ },
2769
+ {
2770
+ "epoch": 1.87,
2771
+ "grad_norm": 0.19114772975444794,
2772
+ "learning_rate": 3.9259556438004476e-05,
2773
+ "loss": 0.6563,
2774
+ "step": 3870
2775
+ },
2776
+ {
2777
+ "epoch": 1.88,
2778
+ "grad_norm": 0.20320531725883484,
2779
+ "learning_rate": 3.9190443367852736e-05,
2780
+ "loss": 0.6471,
2781
+ "step": 3880
2782
+ },
2783
+ {
2784
+ "epoch": 1.88,
2785
+ "grad_norm": 0.23587201535701752,
2786
+ "learning_rate": 3.912116992269348e-05,
2787
+ "loss": 0.6705,
2788
+ "step": 3890
2789
+ },
2790
+ {
2791
+ "epoch": 1.88,
2792
+ "grad_norm": 0.19107288122177124,
2793
+ "learning_rate": 3.9051736885428886e-05,
2794
+ "loss": 0.6696,
2795
+ "step": 3900
2796
+ },
2797
+ {
2798
+ "epoch": 1.89,
2799
+ "grad_norm": 0.20316600799560547,
2800
+ "learning_rate": 3.898214504076482e-05,
2801
+ "loss": 0.6724,
2802
+ "step": 3910
2803
+ },
2804
+ {
2805
+ "epoch": 1.89,
2806
+ "grad_norm": 0.2201535403728485,
2807
+ "learning_rate": 3.89123951752019e-05,
2808
+ "loss": 0.658,
2809
+ "step": 3920
2810
+ },
2811
+ {
2812
+ "epoch": 1.9,
2813
+ "grad_norm": 0.20604528486728668,
2814
+ "learning_rate": 3.884248807702665e-05,
2815
+ "loss": 0.6746,
2816
+ "step": 3930
2817
+ },
2818
+ {
2819
+ "epoch": 1.9,
2820
+ "grad_norm": 0.24204471707344055,
2821
+ "learning_rate": 3.8772424536302564e-05,
2822
+ "loss": 0.6737,
2823
+ "step": 3940
2824
+ },
2825
+ {
2826
+ "epoch": 1.91,
2827
+ "grad_norm": 0.20363375544548035,
2828
+ "learning_rate": 3.870220534486121e-05,
2829
+ "loss": 0.669,
2830
+ "step": 3950
2831
+ },
2832
+ {
2833
+ "epoch": 1.91,
2834
+ "grad_norm": 0.22277967631816864,
2835
+ "learning_rate": 3.8631831296293246e-05,
2836
+ "loss": 0.6676,
2837
+ "step": 3960
2838
+ },
2839
+ {
2840
+ "epoch": 1.92,
2841
+ "grad_norm": 0.18391238152980804,
2842
+ "learning_rate": 3.856130318593947e-05,
2843
+ "loss": 0.6651,
2844
+ "step": 3970
2845
+ },
2846
+ {
2847
+ "epoch": 1.92,
2848
+ "grad_norm": 0.185078427195549,
2849
+ "learning_rate": 3.849062181088183e-05,
2850
+ "loss": 0.6432,
2851
+ "step": 3980
2852
+ },
2853
+ {
2854
+ "epoch": 1.93,
2855
+ "grad_norm": 0.2277977466583252,
2856
+ "learning_rate": 3.841978796993442e-05,
2857
+ "loss": 0.6518,
2858
+ "step": 3990
2859
+ },
2860
+ {
2861
+ "epoch": 1.93,
2862
+ "grad_norm": 0.22232377529144287,
2863
+ "learning_rate": 3.834880246363443e-05,
2864
+ "loss": 0.6743,
2865
+ "step": 4000
2866
+ },
2867
+ {
2868
+ "epoch": 1.93,
2869
+ "eval_loss": 0.6847629547119141,
2870
+ "eval_runtime": 42.7658,
2871
+ "eval_samples_per_second": 46.766,
2872
+ "eval_steps_per_second": 0.374,
2873
+ "step": 4000
2874
+ },
2875
+ {
2876
+ "epoch": 1.94,
2877
+ "grad_norm": 0.20745177567005157,
2878
+ "learning_rate": 3.8277666094233115e-05,
2879
+ "loss": 0.6787,
2880
+ "step": 4010
2881
+ },
2882
+ {
2883
+ "epoch": 1.94,
2884
+ "grad_norm": 0.21468785405158997,
2885
+ "learning_rate": 3.820637966568675e-05,
2886
+ "loss": 0.6603,
2887
+ "step": 4020
2888
+ },
2889
+ {
2890
+ "epoch": 1.95,
2891
+ "grad_norm": 0.18202567100524902,
2892
+ "learning_rate": 3.81349439836475e-05,
2893
+ "loss": 0.6487,
2894
+ "step": 4030
2895
+ },
2896
+ {
2897
+ "epoch": 1.95,
2898
+ "grad_norm": 0.21424496173858643,
2899
+ "learning_rate": 3.806335985545434e-05,
2900
+ "loss": 0.6674,
2901
+ "step": 4040
2902
+ },
2903
+ {
2904
+ "epoch": 1.96,
2905
+ "grad_norm": 0.19061969220638275,
2906
+ "learning_rate": 3.7991628090123933e-05,
2907
+ "loss": 0.651,
2908
+ "step": 4050
2909
+ },
2910
+ {
2911
+ "epoch": 1.96,
2912
+ "grad_norm": 0.20192697644233704,
2913
+ "learning_rate": 3.7919749498341477e-05,
2914
+ "loss": 0.661,
2915
+ "step": 4060
2916
+ },
2917
+ {
2918
+ "epoch": 1.97,
2919
+ "grad_norm": 0.1942390650510788,
2920
+ "learning_rate": 3.784772489245155e-05,
2921
+ "loss": 0.6688,
2922
+ "step": 4070
2923
+ },
2924
+ {
2925
+ "epoch": 1.97,
2926
+ "grad_norm": 0.19686420261859894,
2927
+ "learning_rate": 3.777555508644893e-05,
2928
+ "loss": 0.6492,
2929
+ "step": 4080
2930
+ },
2931
+ {
2932
+ "epoch": 1.98,
2933
+ "grad_norm": 0.2590291202068329,
2934
+ "learning_rate": 3.770324089596937e-05,
2935
+ "loss": 0.6597,
2936
+ "step": 4090
2937
+ },
2938
+ {
2939
+ "epoch": 1.98,
2940
+ "grad_norm": 0.208131343126297,
2941
+ "learning_rate": 3.763078313828043e-05,
2942
+ "loss": 0.6695,
2943
+ "step": 4100
2944
+ },
2945
+ {
2946
+ "epoch": 1.99,
2947
+ "grad_norm": 0.215104341506958,
2948
+ "learning_rate": 3.755818263227219e-05,
2949
+ "loss": 0.6521,
2950
+ "step": 4110
2951
+ },
2952
+ {
2953
+ "epoch": 1.99,
2954
+ "grad_norm": 0.22334003448486328,
2955
+ "learning_rate": 3.748544019844803e-05,
2956
+ "loss": 0.6775,
2957
+ "step": 4120
2958
+ },
2959
+ {
2960
+ "epoch": 2.0,
2961
+ "grad_norm": 0.2614120841026306,
2962
+ "learning_rate": 3.741255665891534e-05,
2963
+ "loss": 0.6643,
2964
+ "step": 4130
2965
+ },
2966
+ {
2967
+ "epoch": 2.0,
2968
+ "grad_norm": 0.22348462045192719,
2969
+ "learning_rate": 3.733953283737625e-05,
2970
+ "loss": 0.6598,
2971
+ "step": 4140
2972
+ },
2973
+ {
2974
+ "epoch": 2.01,
2975
+ "grad_norm": 0.21082797646522522,
2976
+ "learning_rate": 3.726636955911825e-05,
2977
+ "loss": 0.6697,
2978
+ "step": 4150
2979
+ },
2980
+ {
2981
+ "epoch": 2.01,
2982
+ "grad_norm": 0.20834925770759583,
2983
+ "learning_rate": 3.719306765100499e-05,
2984
+ "loss": 0.6562,
2985
+ "step": 4160
2986
+ },
2987
+ {
2988
+ "epoch": 2.02,
2989
+ "grad_norm": 0.223398357629776,
2990
+ "learning_rate": 3.711962794146682e-05,
2991
+ "loss": 0.6403,
2992
+ "step": 4170
2993
+ },
2994
+ {
2995
+ "epoch": 2.02,
2996
+ "grad_norm": 0.1990179866552353,
2997
+ "learning_rate": 3.704605126049147e-05,
2998
+ "loss": 0.6662,
2999
+ "step": 4180
3000
+ },
3001
+ {
3002
+ "epoch": 2.03,
3003
+ "grad_norm": 0.20885860919952393,
3004
+ "learning_rate": 3.6972338439614676e-05,
3005
+ "loss": 0.6669,
3006
+ "step": 4190
3007
+ },
3008
+ {
3009
+ "epoch": 2.03,
3010
+ "grad_norm": 0.21816882491111755,
3011
+ "learning_rate": 3.6898490311910774e-05,
3012
+ "loss": 0.6378,
3013
+ "step": 4200
3014
+ },
3015
+ {
3016
+ "epoch": 2.03,
3017
+ "grad_norm": 0.21286895871162415,
3018
+ "learning_rate": 3.6824507711983294e-05,
3019
+ "loss": 0.633,
3020
+ "step": 4210
3021
+ },
3022
+ {
3023
+ "epoch": 2.04,
3024
+ "grad_norm": 0.20113909244537354,
3025
+ "learning_rate": 3.6750391475955506e-05,
3026
+ "loss": 0.6561,
3027
+ "step": 4220
3028
+ },
3029
+ {
3030
+ "epoch": 2.04,
3031
+ "grad_norm": 0.1805492788553238,
3032
+ "learning_rate": 3.6676142441461e-05,
3033
+ "loss": 0.6478,
3034
+ "step": 4230
3035
+ },
3036
+ {
3037
+ "epoch": 2.05,
3038
+ "grad_norm": 0.198676198720932,
3039
+ "learning_rate": 3.66017614476342e-05,
3040
+ "loss": 0.6477,
3041
+ "step": 4240
3042
+ },
3043
+ {
3044
+ "epoch": 2.05,
3045
+ "grad_norm": 0.18248610198497772,
3046
+ "learning_rate": 3.6527249335100856e-05,
3047
+ "loss": 0.6476,
3048
+ "step": 4250
3049
+ },
3050
+ {
3051
+ "epoch": 2.06,
3052
+ "grad_norm": 0.20229819416999817,
3053
+ "learning_rate": 3.645260694596862e-05,
3054
+ "loss": 0.6675,
3055
+ "step": 4260
3056
+ },
3057
+ {
3058
+ "epoch": 2.06,
3059
+ "grad_norm": 0.24611905217170715,
3060
+ "learning_rate": 3.637783512381745e-05,
3061
+ "loss": 0.6546,
3062
+ "step": 4270
3063
+ },
3064
+ {
3065
+ "epoch": 2.07,
3066
+ "grad_norm": 0.1821327954530716,
3067
+ "learning_rate": 3.6302934713690114e-05,
3068
+ "loss": 0.6482,
3069
+ "step": 4280
3070
+ },
3071
+ {
3072
+ "epoch": 2.07,
3073
+ "grad_norm": 0.19002540409564972,
3074
+ "learning_rate": 3.622790656208263e-05,
3075
+ "loss": 0.6468,
3076
+ "step": 4290
3077
+ },
3078
+ {
3079
+ "epoch": 2.08,
3080
+ "grad_norm": 0.20899777114391327,
3081
+ "learning_rate": 3.615275151693471e-05,
3082
+ "loss": 0.6649,
3083
+ "step": 4300
3084
+ },
3085
+ {
3086
+ "epoch": 2.08,
3087
+ "grad_norm": 0.1918490082025528,
3088
+ "learning_rate": 3.607747042762016e-05,
3089
+ "loss": 0.6554,
3090
+ "step": 4310
3091
+ },
3092
+ {
3093
+ "epoch": 2.09,
3094
+ "grad_norm": 0.2790839970111847,
3095
+ "learning_rate": 3.600206414493728e-05,
3096
+ "loss": 0.6607,
3097
+ "step": 4320
3098
+ },
3099
+ {
3100
+ "epoch": 2.09,
3101
+ "grad_norm": 0.21416689455509186,
3102
+ "learning_rate": 3.592653352109929e-05,
3103
+ "loss": 0.6537,
3104
+ "step": 4330
3105
+ },
3106
+ {
3107
+ "epoch": 2.1,
3108
+ "grad_norm": 0.19238290190696716,
3109
+ "learning_rate": 3.5850879409724624e-05,
3110
+ "loss": 0.6566,
3111
+ "step": 4340
3112
+ },
3113
+ {
3114
+ "epoch": 2.1,
3115
+ "grad_norm": 0.24673143029212952,
3116
+ "learning_rate": 3.577510266582737e-05,
3117
+ "loss": 0.6599,
3118
+ "step": 4350
3119
+ },
3120
+ {
3121
+ "epoch": 2.11,
3122
+ "grad_norm": 0.1863541454076767,
3123
+ "learning_rate": 3.569920414580754e-05,
3124
+ "loss": 0.6593,
3125
+ "step": 4360
3126
+ },
3127
+ {
3128
+ "epoch": 2.11,
3129
+ "grad_norm": 0.2196236401796341,
3130
+ "learning_rate": 3.56231847074414e-05,
3131
+ "loss": 0.6483,
3132
+ "step": 4370
3133
+ },
3134
+ {
3135
+ "epoch": 2.12,
3136
+ "grad_norm": 0.18300732970237732,
3137
+ "learning_rate": 3.5547045209871806e-05,
3138
+ "loss": 0.6359,
3139
+ "step": 4380
3140
+ },
3141
+ {
3142
+ "epoch": 2.12,
3143
+ "grad_norm": 0.18719661235809326,
3144
+ "learning_rate": 3.5470786513598476e-05,
3145
+ "loss": 0.6707,
3146
+ "step": 4390
3147
+ },
3148
+ {
3149
+ "epoch": 2.13,
3150
+ "grad_norm": 0.1748167872428894,
3151
+ "learning_rate": 3.539440948046827e-05,
3152
+ "loss": 0.6516,
3153
+ "step": 4400
3154
+ },
3155
+ {
3156
+ "epoch": 2.13,
3157
+ "grad_norm": 0.21993610262870789,
3158
+ "learning_rate": 3.531791497366543e-05,
3159
+ "loss": 0.6539,
3160
+ "step": 4410
3161
+ },
3162
+ {
3163
+ "epoch": 2.14,
3164
+ "grad_norm": 0.20704199373722076,
3165
+ "learning_rate": 3.524130385770186e-05,
3166
+ "loss": 0.6734,
3167
+ "step": 4420
3168
+ },
3169
+ {
3170
+ "epoch": 2.14,
3171
+ "grad_norm": 0.23367135226726532,
3172
+ "learning_rate": 3.516457699840733e-05,
3173
+ "loss": 0.6855,
3174
+ "step": 4430
3175
+ },
3176
+ {
3177
+ "epoch": 2.15,
3178
+ "grad_norm": 0.1824340671300888,
3179
+ "learning_rate": 3.50877352629197e-05,
3180
+ "loss": 0.6575,
3181
+ "step": 4440
3182
+ },
3183
+ {
3184
+ "epoch": 2.15,
3185
+ "grad_norm": 0.22375932335853577,
3186
+ "learning_rate": 3.50107795196751e-05,
3187
+ "loss": 0.6487,
3188
+ "step": 4450
3189
+ },
3190
+ {
3191
+ "epoch": 2.16,
3192
+ "grad_norm": 0.23603610694408417,
3193
+ "learning_rate": 3.4933710638398156e-05,
3194
+ "loss": 0.6505,
3195
+ "step": 4460
3196
+ },
3197
+ {
3198
+ "epoch": 2.16,
3199
+ "grad_norm": 0.22663302719593048,
3200
+ "learning_rate": 3.485652949009212e-05,
3201
+ "loss": 0.6569,
3202
+ "step": 4470
3203
+ },
3204
+ {
3205
+ "epoch": 2.17,
3206
+ "grad_norm": 0.21095499396324158,
3207
+ "learning_rate": 3.4779236947029055e-05,
3208
+ "loss": 0.6657,
3209
+ "step": 4480
3210
+ },
3211
+ {
3212
+ "epoch": 2.17,
3213
+ "grad_norm": 0.2422657608985901,
3214
+ "learning_rate": 3.470183388273995e-05,
3215
+ "loss": 0.645,
3216
+ "step": 4490
3217
+ },
3218
+ {
3219
+ "epoch": 2.17,
3220
+ "grad_norm": 0.19883093237876892,
3221
+ "learning_rate": 3.462432117200489e-05,
3222
+ "loss": 0.6647,
3223
+ "step": 4500
3224
+ },
3225
+ {
3226
+ "epoch": 2.17,
3227
+ "eval_loss": 0.6818509697914124,
3228
+ "eval_runtime": 42.9789,
3229
+ "eval_samples_per_second": 46.534,
3230
+ "eval_steps_per_second": 0.372,
3231
+ "step": 4500
3232
+ },
3233
+ {
3234
+ "epoch": 2.18,
3235
+ "grad_norm": 0.23546861112117767,
3236
+ "learning_rate": 3.454669969084312e-05,
3237
+ "loss": 0.6594,
3238
+ "step": 4510
3239
+ },
3240
+ {
3241
+ "epoch": 2.18,
3242
+ "grad_norm": 0.20284776389598846,
3243
+ "learning_rate": 3.446897031650316e-05,
3244
+ "loss": 0.65,
3245
+ "step": 4520
3246
+ },
3247
+ {
3248
+ "epoch": 2.19,
3249
+ "grad_norm": 0.19946908950805664,
3250
+ "learning_rate": 3.4391133927452925e-05,
3251
+ "loss": 0.6573,
3252
+ "step": 4530
3253
+ },
3254
+ {
3255
+ "epoch": 2.19,
3256
+ "grad_norm": 0.2393037974834442,
3257
+ "learning_rate": 3.431319140336975e-05,
3258
+ "loss": 0.6502,
3259
+ "step": 4540
3260
+ },
3261
+ {
3262
+ "epoch": 2.2,
3263
+ "grad_norm": 0.2206243872642517,
3264
+ "learning_rate": 3.423514362513048e-05,
3265
+ "loss": 0.6704,
3266
+ "step": 4550
3267
+ },
3268
+ {
3269
+ "epoch": 2.2,
3270
+ "grad_norm": 0.20536164939403534,
3271
+ "learning_rate": 3.415699147480149e-05,
3272
+ "loss": 0.6629,
3273
+ "step": 4560
3274
+ },
3275
+ {
3276
+ "epoch": 2.21,
3277
+ "grad_norm": 0.22283408045768738,
3278
+ "learning_rate": 3.407873583562873e-05,
3279
+ "loss": 0.6789,
3280
+ "step": 4570
3281
+ },
3282
+ {
3283
+ "epoch": 2.21,
3284
+ "grad_norm": 0.21526776254177094,
3285
+ "learning_rate": 3.4000377592027754e-05,
3286
+ "loss": 0.6608,
3287
+ "step": 4580
3288
+ },
3289
+ {
3290
+ "epoch": 2.22,
3291
+ "grad_norm": 0.2566728889942169,
3292
+ "learning_rate": 3.3921917629573695e-05,
3293
+ "loss": 0.6624,
3294
+ "step": 4590
3295
+ },
3296
+ {
3297
+ "epoch": 2.22,
3298
+ "grad_norm": 0.21391907334327698,
3299
+ "learning_rate": 3.384335683499129e-05,
3300
+ "loss": 0.6536,
3301
+ "step": 4600
3302
+ },
3303
+ {
3304
+ "epoch": 2.23,
3305
+ "grad_norm": 0.21451455354690552,
3306
+ "learning_rate": 3.376469609614484e-05,
3307
+ "loss": 0.648,
3308
+ "step": 4610
3309
+ },
3310
+ {
3311
+ "epoch": 2.23,
3312
+ "grad_norm": 0.19541354477405548,
3313
+ "learning_rate": 3.368593630202818e-05,
3314
+ "loss": 0.657,
3315
+ "step": 4620
3316
+ },
3317
+ {
3318
+ "epoch": 2.24,
3319
+ "grad_norm": 0.2338065356016159,
3320
+ "learning_rate": 3.360707834275459e-05,
3321
+ "loss": 0.6686,
3322
+ "step": 4630
3323
+ },
3324
+ {
3325
+ "epoch": 2.24,
3326
+ "grad_norm": 0.1928481012582779,
3327
+ "learning_rate": 3.3528123109546844e-05,
3328
+ "loss": 0.6496,
3329
+ "step": 4640
3330
+ },
3331
+ {
3332
+ "epoch": 2.25,
3333
+ "grad_norm": 0.23110058903694153,
3334
+ "learning_rate": 3.344907149472702e-05,
3335
+ "loss": 0.6556,
3336
+ "step": 4650
3337
+ },
3338
+ {
3339
+ "epoch": 2.25,
3340
+ "grad_norm": 0.22316649556159973,
3341
+ "learning_rate": 3.336992439170649e-05,
3342
+ "loss": 0.663,
3343
+ "step": 4660
3344
+ },
3345
+ {
3346
+ "epoch": 2.26,
3347
+ "grad_norm": 0.1999579668045044,
3348
+ "learning_rate": 3.3290682694975775e-05,
3349
+ "loss": 0.6436,
3350
+ "step": 4670
3351
+ },
3352
+ {
3353
+ "epoch": 2.26,
3354
+ "grad_norm": 0.276755154132843,
3355
+ "learning_rate": 3.321134730009446e-05,
3356
+ "loss": 0.6355,
3357
+ "step": 4680
3358
+ },
3359
+ {
3360
+ "epoch": 2.27,
3361
+ "grad_norm": 0.24972842633724213,
3362
+ "learning_rate": 3.313191910368111e-05,
3363
+ "loss": 0.6454,
3364
+ "step": 4690
3365
+ },
3366
+ {
3367
+ "epoch": 2.27,
3368
+ "grad_norm": 0.2169172465801239,
3369
+ "learning_rate": 3.3052399003403046e-05,
3370
+ "loss": 0.6419,
3371
+ "step": 4700
3372
+ },
3373
+ {
3374
+ "epoch": 2.28,
3375
+ "grad_norm": 0.20602068305015564,
3376
+ "learning_rate": 3.297278789796629e-05,
3377
+ "loss": 0.6459,
3378
+ "step": 4710
3379
+ },
3380
+ {
3381
+ "epoch": 2.28,
3382
+ "grad_norm": 0.25669389963150024,
3383
+ "learning_rate": 3.289308668710532e-05,
3384
+ "loss": 0.6675,
3385
+ "step": 4720
3386
+ },
3387
+ {
3388
+ "epoch": 2.29,
3389
+ "grad_norm": 0.21008487045764923,
3390
+ "learning_rate": 3.281329627157302e-05,
3391
+ "loss": 0.642,
3392
+ "step": 4730
3393
+ },
3394
+ {
3395
+ "epoch": 2.29,
3396
+ "grad_norm": 0.21287119388580322,
3397
+ "learning_rate": 3.2733417553130384e-05,
3398
+ "loss": 0.6348,
3399
+ "step": 4740
3400
+ },
3401
+ {
3402
+ "epoch": 2.3,
3403
+ "grad_norm": 0.21862877905368805,
3404
+ "learning_rate": 3.2653451434536394e-05,
3405
+ "loss": 0.6579,
3406
+ "step": 4750
3407
+ },
3408
+ {
3409
+ "epoch": 2.3,
3410
+ "grad_norm": 0.19740484654903412,
3411
+ "learning_rate": 3.2573398819537767e-05,
3412
+ "loss": 0.6428,
3413
+ "step": 4760
3414
+ },
3415
+ {
3416
+ "epoch": 2.31,
3417
+ "grad_norm": 0.2695826292037964,
3418
+ "learning_rate": 3.2493260612858803e-05,
3419
+ "loss": 0.658,
3420
+ "step": 4770
3421
+ },
3422
+ {
3423
+ "epoch": 2.31,
3424
+ "grad_norm": 0.19435593485832214,
3425
+ "learning_rate": 3.2413037720191096e-05,
3426
+ "loss": 0.6615,
3427
+ "step": 4780
3428
+ },
3429
+ {
3430
+ "epoch": 2.32,
3431
+ "grad_norm": 0.21996192634105682,
3432
+ "learning_rate": 3.233273104818337e-05,
3433
+ "loss": 0.6588,
3434
+ "step": 4790
3435
+ },
3436
+ {
3437
+ "epoch": 2.32,
3438
+ "grad_norm": 0.26496437191963196,
3439
+ "learning_rate": 3.225234150443114e-05,
3440
+ "loss": 0.6548,
3441
+ "step": 4800
3442
+ },
3443
+ {
3444
+ "epoch": 2.32,
3445
+ "grad_norm": 0.2110251486301422,
3446
+ "learning_rate": 3.217186999746654e-05,
3447
+ "loss": 0.6618,
3448
+ "step": 4810
3449
+ },
3450
+ {
3451
+ "epoch": 2.33,
3452
+ "grad_norm": 0.19939462840557098,
3453
+ "learning_rate": 3.209131743674803e-05,
3454
+ "loss": 0.6531,
3455
+ "step": 4820
3456
+ },
3457
+ {
3458
+ "epoch": 2.33,
3459
+ "grad_norm": 0.19813968241214752,
3460
+ "learning_rate": 3.201068473265007e-05,
3461
+ "loss": 0.6484,
3462
+ "step": 4830
3463
+ },
3464
+ {
3465
+ "epoch": 2.34,
3466
+ "grad_norm": 0.2121964991092682,
3467
+ "learning_rate": 3.192997279645291e-05,
3468
+ "loss": 0.6556,
3469
+ "step": 4840
3470
+ },
3471
+ {
3472
+ "epoch": 2.34,
3473
+ "grad_norm": 0.19998233020305634,
3474
+ "learning_rate": 3.1849182540332214e-05,
3475
+ "loss": 0.6554,
3476
+ "step": 4850
3477
+ },
3478
+ {
3479
+ "epoch": 2.35,
3480
+ "grad_norm": 0.23478324711322784,
3481
+ "learning_rate": 3.176831487734882e-05,
3482
+ "loss": 0.637,
3483
+ "step": 4860
3484
+ },
3485
+ {
3486
+ "epoch": 2.35,
3487
+ "grad_norm": 0.19482719898223877,
3488
+ "learning_rate": 3.168737072143838e-05,
3489
+ "loss": 0.6464,
3490
+ "step": 4870
3491
+ },
3492
+ {
3493
+ "epoch": 2.36,
3494
+ "grad_norm": 0.21184729039669037,
3495
+ "learning_rate": 3.160635098740103e-05,
3496
+ "loss": 0.6584,
3497
+ "step": 4880
3498
+ },
3499
+ {
3500
+ "epoch": 2.36,
3501
+ "grad_norm": 0.27235257625579834,
3502
+ "learning_rate": 3.152525659089106e-05,
3503
+ "loss": 0.6597,
3504
+ "step": 4890
3505
+ },
3506
+ {
3507
+ "epoch": 2.37,
3508
+ "grad_norm": 0.22343479096889496,
3509
+ "learning_rate": 3.1444088448406584e-05,
3510
+ "loss": 0.6673,
3511
+ "step": 4900
3512
+ },
3513
+ {
3514
+ "epoch": 2.37,
3515
+ "grad_norm": 0.19164462387561798,
3516
+ "learning_rate": 3.136284747727916e-05,
3517
+ "loss": 0.6617,
3518
+ "step": 4910
3519
+ },
3520
+ {
3521
+ "epoch": 2.38,
3522
+ "grad_norm": 0.21662816405296326,
3523
+ "learning_rate": 3.128153459566341e-05,
3524
+ "loss": 0.6505,
3525
+ "step": 4920
3526
+ },
3527
+ {
3528
+ "epoch": 2.38,
3529
+ "grad_norm": 0.23168237507343292,
3530
+ "learning_rate": 3.1200150722526697e-05,
3531
+ "loss": 0.6479,
3532
+ "step": 4930
3533
+ },
3534
+ {
3535
+ "epoch": 2.39,
3536
+ "grad_norm": 0.19140580296516418,
3537
+ "learning_rate": 3.111869677763865e-05,
3538
+ "loss": 0.6413,
3539
+ "step": 4940
3540
+ },
3541
+ {
3542
+ "epoch": 2.39,
3543
+ "grad_norm": 0.25583407282829285,
3544
+ "learning_rate": 3.1037173681560874e-05,
3545
+ "loss": 0.6443,
3546
+ "step": 4950
3547
+ },
3548
+ {
3549
+ "epoch": 2.4,
3550
+ "grad_norm": 0.1979406476020813,
3551
+ "learning_rate": 3.095558235563647e-05,
3552
+ "loss": 0.6668,
3553
+ "step": 4960
3554
+ },
3555
+ {
3556
+ "epoch": 2.4,
3557
+ "grad_norm": 0.21006809175014496,
3558
+ "learning_rate": 3.0873923721979645e-05,
3559
+ "loss": 0.6505,
3560
+ "step": 4970
3561
+ },
3562
+ {
3563
+ "epoch": 2.41,
3564
+ "grad_norm": 0.2352473884820938,
3565
+ "learning_rate": 3.07921987034653e-05,
3566
+ "loss": 0.6552,
3567
+ "step": 4980
3568
+ },
3569
+ {
3570
+ "epoch": 2.41,
3571
+ "grad_norm": 0.22581614553928375,
3572
+ "learning_rate": 3.07104082237186e-05,
3573
+ "loss": 0.6489,
3574
+ "step": 4990
3575
+ },
3576
+ {
3577
+ "epoch": 2.42,
3578
+ "grad_norm": 0.21772272884845734,
3579
+ "learning_rate": 3.06285532071045e-05,
3580
+ "loss": 0.6721,
3581
+ "step": 5000
3582
+ },
3583
+ {
3584
+ "epoch": 2.42,
3585
+ "eval_loss": 0.6797133684158325,
3586
+ "eval_runtime": 43.0653,
3587
+ "eval_samples_per_second": 46.441,
3588
+ "eval_steps_per_second": 0.372,
3589
+ "step": 5000
3590
+ }
3591
+ ],
3592
+ "logging_steps": 10,
3593
+ "max_steps": 10345,
3594
+ "num_input_tokens_seen": 0,
3595
+ "num_train_epochs": 5,
3596
+ "save_steps": 500,
3597
+ "total_flos": 1.2657613455646261e+19,
3598
+ "train_batch_size": 8,
3599
+ "trial_name": null,
3600
+ "trial_params": null
3601
+ }
checkpoint-5000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3814e1058daec6f70a62fab0452b08b4d1c663bae76d77a07b33001875b2afcd
3
+ size 5048