Upload 10 files
Browse files- config.json +32 -0
- generation_config.json +6 -0
- model.safetensors +3 -0
- rng_state.pth +3 -0
- scheduler.pt +3 -0
- special_tokens_map.json +7 -0
- tokenizer.json +219 -0
- tokenizer_config.json +52 -0
- trainer_state.json +1101 -0
- training_args.bin +3 -0
config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "../MistralModels/models/Mixtral-8x7B-v0.1-verytiny-4096",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"MixtralForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 1,
|
| 8 |
+
"eos_token_id": 2,
|
| 9 |
+
"head_dim": 8,
|
| 10 |
+
"hidden_act": "silu",
|
| 11 |
+
"hidden_size": 64,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"intermediate_size": 64,
|
| 14 |
+
"max_position_embeddings": 256,
|
| 15 |
+
"model_type": "mixtral",
|
| 16 |
+
"num_attention_heads": 8,
|
| 17 |
+
"num_experts_per_tok": 1,
|
| 18 |
+
"num_hidden_layers": 4,
|
| 19 |
+
"num_key_value_heads": 8,
|
| 20 |
+
"num_local_experts": 8,
|
| 21 |
+
"output_router_logits": false,
|
| 22 |
+
"rms_norm_eps": 1e-05,
|
| 23 |
+
"rope_theta": 1000000.0,
|
| 24 |
+
"router_aux_loss_coef": 0.02,
|
| 25 |
+
"router_jitter_noise": 0.0,
|
| 26 |
+
"sliding_window": null,
|
| 27 |
+
"tie_word_embeddings": false,
|
| 28 |
+
"torch_dtype": "bfloat16",
|
| 29 |
+
"transformers_version": "4.47.1",
|
| 30 |
+
"use_cache": true,
|
| 31 |
+
"vocab_size": 69
|
| 32 |
+
}
|
generation_config.json
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 1,
|
| 4 |
+
"eos_token_id": 2,
|
| 5 |
+
"transformers_version": "4.47.1"
|
| 6 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:88b567c2b8ff9806e2321790eb08ef6bc9aa2add457dc0bbecb8e27f8ac595a2
|
| 3 |
+
size 954968
|
rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b3bc1ac846859ad13fd356d3ed41fd20834205a7e8764e4198636863fca3f64c
|
| 3 |
+
size 14244
|
scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:108adabe1cba638b6f7a5ccf74574e5d771eb5e6397a76b5c6fbbee6ba77030a
|
| 3 |
+
size 1064
|
special_tokens_map.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"cls_token": "[CLS]",
|
| 3 |
+
"mask_token": "[MASK]",
|
| 4 |
+
"pad_token": "[EOS]",
|
| 5 |
+
"sep_token": "[SEP]",
|
| 6 |
+
"unk_token": "[UNK]"
|
| 7 |
+
}
|
tokenizer.json
ADDED
|
@@ -0,0 +1,219 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"version": "1.0",
|
| 3 |
+
"truncation": {
|
| 4 |
+
"direction": "Right",
|
| 5 |
+
"max_length": 100,
|
| 6 |
+
"strategy": "LongestFirst",
|
| 7 |
+
"stride": 0
|
| 8 |
+
},
|
| 9 |
+
"padding": {
|
| 10 |
+
"strategy": "BatchLongest",
|
| 11 |
+
"direction": "Right",
|
| 12 |
+
"pad_to_multiple_of": null,
|
| 13 |
+
"pad_id": 64,
|
| 14 |
+
"pad_type_id": 0,
|
| 15 |
+
"pad_token": "[EOS]"
|
| 16 |
+
},
|
| 17 |
+
"added_tokens": [
|
| 18 |
+
{
|
| 19 |
+
"id": 64,
|
| 20 |
+
"content": "[UNK]",
|
| 21 |
+
"single_word": false,
|
| 22 |
+
"lstrip": false,
|
| 23 |
+
"rstrip": false,
|
| 24 |
+
"normalized": false,
|
| 25 |
+
"special": true
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"id": 65,
|
| 29 |
+
"content": "[CLS]",
|
| 30 |
+
"single_word": false,
|
| 31 |
+
"lstrip": false,
|
| 32 |
+
"rstrip": false,
|
| 33 |
+
"normalized": false,
|
| 34 |
+
"special": true
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"id": 66,
|
| 38 |
+
"content": "[SEP]",
|
| 39 |
+
"single_word": false,
|
| 40 |
+
"lstrip": false,
|
| 41 |
+
"rstrip": false,
|
| 42 |
+
"normalized": false,
|
| 43 |
+
"special": true
|
| 44 |
+
},
|
| 45 |
+
{
|
| 46 |
+
"id": 67,
|
| 47 |
+
"content": "[PAD]",
|
| 48 |
+
"single_word": false,
|
| 49 |
+
"lstrip": false,
|
| 50 |
+
"rstrip": false,
|
| 51 |
+
"normalized": false,
|
| 52 |
+
"special": true
|
| 53 |
+
},
|
| 54 |
+
{
|
| 55 |
+
"id": 68,
|
| 56 |
+
"content": "[MASK]",
|
| 57 |
+
"single_word": false,
|
| 58 |
+
"lstrip": false,
|
| 59 |
+
"rstrip": false,
|
| 60 |
+
"normalized": false,
|
| 61 |
+
"special": true
|
| 62 |
+
}
|
| 63 |
+
],
|
| 64 |
+
"normalizer": null,
|
| 65 |
+
"pre_tokenizer": {
|
| 66 |
+
"type": "Whitespace"
|
| 67 |
+
},
|
| 68 |
+
"post_processor": {
|
| 69 |
+
"type": "TemplateProcessing",
|
| 70 |
+
"single": [
|
| 71 |
+
{
|
| 72 |
+
"SpecialToken": {
|
| 73 |
+
"id": "[CLS]",
|
| 74 |
+
"type_id": 0
|
| 75 |
+
}
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"Sequence": {
|
| 79 |
+
"id": "A",
|
| 80 |
+
"type_id": 0
|
| 81 |
+
}
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"SpecialToken": {
|
| 85 |
+
"id": "[SEP]",
|
| 86 |
+
"type_id": 0
|
| 87 |
+
}
|
| 88 |
+
}
|
| 89 |
+
],
|
| 90 |
+
"pair": [
|
| 91 |
+
{
|
| 92 |
+
"SpecialToken": {
|
| 93 |
+
"id": "[CLS]",
|
| 94 |
+
"type_id": 0
|
| 95 |
+
}
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"Sequence": {
|
| 99 |
+
"id": "A",
|
| 100 |
+
"type_id": 0
|
| 101 |
+
}
|
| 102 |
+
},
|
| 103 |
+
{
|
| 104 |
+
"SpecialToken": {
|
| 105 |
+
"id": "[SEP]",
|
| 106 |
+
"type_id": 0
|
| 107 |
+
}
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"Sequence": {
|
| 111 |
+
"id": "B",
|
| 112 |
+
"type_id": 1
|
| 113 |
+
}
|
| 114 |
+
},
|
| 115 |
+
{
|
| 116 |
+
"SpecialToken": {
|
| 117 |
+
"id": "[SEP]",
|
| 118 |
+
"type_id": 1
|
| 119 |
+
}
|
| 120 |
+
}
|
| 121 |
+
],
|
| 122 |
+
"special_tokens": {
|
| 123 |
+
"[CLS]": {
|
| 124 |
+
"id": "[CLS]",
|
| 125 |
+
"ids": [
|
| 126 |
+
65
|
| 127 |
+
],
|
| 128 |
+
"tokens": [
|
| 129 |
+
"[CLS]"
|
| 130 |
+
]
|
| 131 |
+
},
|
| 132 |
+
"[SEP]": {
|
| 133 |
+
"id": "[SEP]",
|
| 134 |
+
"ids": [
|
| 135 |
+
66
|
| 136 |
+
],
|
| 137 |
+
"tokens": [
|
| 138 |
+
"[SEP]"
|
| 139 |
+
]
|
| 140 |
+
}
|
| 141 |
+
}
|
| 142 |
+
},
|
| 143 |
+
"decoder": null,
|
| 144 |
+
"model": {
|
| 145 |
+
"type": "WordLevel",
|
| 146 |
+
"vocab": {
|
| 147 |
+
"AAA": 0,
|
| 148 |
+
"AAT": 1,
|
| 149 |
+
"AAG": 2,
|
| 150 |
+
"AAC": 3,
|
| 151 |
+
"ATA": 4,
|
| 152 |
+
"ATT": 5,
|
| 153 |
+
"ATG": 6,
|
| 154 |
+
"ATC": 7,
|
| 155 |
+
"AGA": 8,
|
| 156 |
+
"AGT": 9,
|
| 157 |
+
"AGG": 10,
|
| 158 |
+
"AGC": 11,
|
| 159 |
+
"ACA": 12,
|
| 160 |
+
"ACT": 13,
|
| 161 |
+
"ACG": 14,
|
| 162 |
+
"ACC": 15,
|
| 163 |
+
"TAA": 16,
|
| 164 |
+
"TAT": 17,
|
| 165 |
+
"TAG": 18,
|
| 166 |
+
"TAC": 19,
|
| 167 |
+
"TTA": 20,
|
| 168 |
+
"TTT": 21,
|
| 169 |
+
"TTG": 22,
|
| 170 |
+
"TTC": 23,
|
| 171 |
+
"TGA": 24,
|
| 172 |
+
"TGT": 25,
|
| 173 |
+
"TGG": 26,
|
| 174 |
+
"TGC": 27,
|
| 175 |
+
"TCA": 28,
|
| 176 |
+
"TCT": 29,
|
| 177 |
+
"TCG": 30,
|
| 178 |
+
"TCC": 31,
|
| 179 |
+
"GAA": 32,
|
| 180 |
+
"GAT": 33,
|
| 181 |
+
"GAG": 34,
|
| 182 |
+
"GAC": 35,
|
| 183 |
+
"GTA": 36,
|
| 184 |
+
"GTT": 37,
|
| 185 |
+
"GTG": 38,
|
| 186 |
+
"GTC": 39,
|
| 187 |
+
"GGA": 40,
|
| 188 |
+
"GGT": 41,
|
| 189 |
+
"GGG": 42,
|
| 190 |
+
"GGC": 43,
|
| 191 |
+
"GCA": 44,
|
| 192 |
+
"GCT": 45,
|
| 193 |
+
"GCG": 46,
|
| 194 |
+
"GCC": 47,
|
| 195 |
+
"CAA": 48,
|
| 196 |
+
"CAT": 49,
|
| 197 |
+
"CAG": 50,
|
| 198 |
+
"CAC": 51,
|
| 199 |
+
"CTA": 52,
|
| 200 |
+
"CTT": 53,
|
| 201 |
+
"CTG": 54,
|
| 202 |
+
"CTC": 55,
|
| 203 |
+
"CGA": 56,
|
| 204 |
+
"CGT": 57,
|
| 205 |
+
"CGG": 58,
|
| 206 |
+
"CGC": 59,
|
| 207 |
+
"CCA": 60,
|
| 208 |
+
"CCT": 61,
|
| 209 |
+
"CCG": 62,
|
| 210 |
+
"CCC": 63,
|
| 211 |
+
"[UNK]": 64,
|
| 212 |
+
"[CLS]": 65,
|
| 213 |
+
"[SEP]": 66,
|
| 214 |
+
"[PAD]": 67,
|
| 215 |
+
"[MASK]": 68
|
| 216 |
+
},
|
| 217 |
+
"unk_token": "[UNK]"
|
| 218 |
+
}
|
| 219 |
+
}
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"added_tokens_decoder": {
|
| 3 |
+
"64": {
|
| 4 |
+
"content": "[UNK]",
|
| 5 |
+
"lstrip": false,
|
| 6 |
+
"normalized": false,
|
| 7 |
+
"rstrip": false,
|
| 8 |
+
"single_word": false,
|
| 9 |
+
"special": true
|
| 10 |
+
},
|
| 11 |
+
"65": {
|
| 12 |
+
"content": "[CLS]",
|
| 13 |
+
"lstrip": false,
|
| 14 |
+
"normalized": false,
|
| 15 |
+
"rstrip": false,
|
| 16 |
+
"single_word": false,
|
| 17 |
+
"special": true
|
| 18 |
+
},
|
| 19 |
+
"66": {
|
| 20 |
+
"content": "[SEP]",
|
| 21 |
+
"lstrip": false,
|
| 22 |
+
"normalized": false,
|
| 23 |
+
"rstrip": false,
|
| 24 |
+
"single_word": false,
|
| 25 |
+
"special": true
|
| 26 |
+
},
|
| 27 |
+
"67": {
|
| 28 |
+
"content": "[PAD]",
|
| 29 |
+
"lstrip": false,
|
| 30 |
+
"normalized": false,
|
| 31 |
+
"rstrip": false,
|
| 32 |
+
"single_word": false,
|
| 33 |
+
"special": true
|
| 34 |
+
},
|
| 35 |
+
"68": {
|
| 36 |
+
"content": "[MASK]",
|
| 37 |
+
"lstrip": false,
|
| 38 |
+
"normalized": false,
|
| 39 |
+
"rstrip": false,
|
| 40 |
+
"single_word": false,
|
| 41 |
+
"special": true
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"clean_up_tokenization_spaces": true,
|
| 45 |
+
"cls_token": "[CLS]",
|
| 46 |
+
"mask_token": "[MASK]",
|
| 47 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 48 |
+
"pad_token": "[EOS]",
|
| 49 |
+
"sep_token": "[SEP]",
|
| 50 |
+
"tokenizer_class": "PreTrainedTokenizerFast",
|
| 51 |
+
"unk_token": "[UNK]"
|
| 52 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,1101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": 3.7249155044555664,
|
| 3 |
+
"best_model_checkpoint": "./results/models/checkpoint-58650",
|
| 4 |
+
"epoch": 30.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 58650,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.2557544757033248,
|
| 13 |
+
"grad_norm": 0.047607421875,
|
| 14 |
+
"learning_rate": 0.003979539641943734,
|
| 15 |
+
"loss": 1.9283,
|
| 16 |
+
"step": 500
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.5115089514066496,
|
| 20 |
+
"grad_norm": 0.05810546875,
|
| 21 |
+
"learning_rate": 0.003959079283887468,
|
| 22 |
+
"loss": 1.89,
|
| 23 |
+
"step": 1000
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.7672634271099744,
|
| 27 |
+
"grad_norm": 0.048828125,
|
| 28 |
+
"learning_rate": 0.003938618925831202,
|
| 29 |
+
"loss": 1.8823,
|
| 30 |
+
"step": 1500
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 1.0,
|
| 34 |
+
"eval_loss": 3.7650158405303955,
|
| 35 |
+
"eval_runtime": 1.1856,
|
| 36 |
+
"eval_samples_per_second": 421.713,
|
| 37 |
+
"eval_steps_per_second": 0.843,
|
| 38 |
+
"step": 1955
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 1.0230179028132993,
|
| 42 |
+
"grad_norm": 0.048583984375,
|
| 43 |
+
"learning_rate": 0.003918158567774936,
|
| 44 |
+
"loss": 1.8791,
|
| 45 |
+
"step": 2000
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 1.278772378516624,
|
| 49 |
+
"grad_norm": 0.0458984375,
|
| 50 |
+
"learning_rate": 0.00389769820971867,
|
| 51 |
+
"loss": 1.8765,
|
| 52 |
+
"step": 2500
|
| 53 |
+
},
|
| 54 |
+
{
|
| 55 |
+
"epoch": 1.5345268542199488,
|
| 56 |
+
"grad_norm": 0.04443359375,
|
| 57 |
+
"learning_rate": 0.003877237851662404,
|
| 58 |
+
"loss": 1.8758,
|
| 59 |
+
"step": 3000
|
| 60 |
+
},
|
| 61 |
+
{
|
| 62 |
+
"epoch": 1.7902813299232738,
|
| 63 |
+
"grad_norm": 0.04638671875,
|
| 64 |
+
"learning_rate": 0.003856777493606138,
|
| 65 |
+
"loss": 1.874,
|
| 66 |
+
"step": 3500
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 2.0,
|
| 70 |
+
"eval_loss": 3.755614757537842,
|
| 71 |
+
"eval_runtime": 1.2689,
|
| 72 |
+
"eval_samples_per_second": 394.03,
|
| 73 |
+
"eval_steps_per_second": 0.788,
|
| 74 |
+
"step": 3910
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 2.0460358056265986,
|
| 78 |
+
"grad_norm": 0.0322265625,
|
| 79 |
+
"learning_rate": 0.0038363171355498722,
|
| 80 |
+
"loss": 1.8728,
|
| 81 |
+
"step": 4000
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"epoch": 2.3017902813299234,
|
| 85 |
+
"grad_norm": 0.03857421875,
|
| 86 |
+
"learning_rate": 0.0038158567774936062,
|
| 87 |
+
"loss": 1.8721,
|
| 88 |
+
"step": 4500
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"epoch": 2.557544757033248,
|
| 92 |
+
"grad_norm": 0.051025390625,
|
| 93 |
+
"learning_rate": 0.0037953964194373403,
|
| 94 |
+
"loss": 1.8714,
|
| 95 |
+
"step": 5000
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"epoch": 2.813299232736573,
|
| 99 |
+
"grad_norm": 0.043701171875,
|
| 100 |
+
"learning_rate": 0.0037749360613810743,
|
| 101 |
+
"loss": 1.8704,
|
| 102 |
+
"step": 5500
|
| 103 |
+
},
|
| 104 |
+
{
|
| 105 |
+
"epoch": 3.0,
|
| 106 |
+
"eval_loss": 3.749542474746704,
|
| 107 |
+
"eval_runtime": 1.1996,
|
| 108 |
+
"eval_samples_per_second": 416.822,
|
| 109 |
+
"eval_steps_per_second": 0.834,
|
| 110 |
+
"step": 5865
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"epoch": 3.0690537084398977,
|
| 114 |
+
"grad_norm": 0.045654296875,
|
| 115 |
+
"learning_rate": 0.0037544757033248083,
|
| 116 |
+
"loss": 1.8699,
|
| 117 |
+
"step": 6000
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"epoch": 3.3248081841432224,
|
| 121 |
+
"grad_norm": 0.038818359375,
|
| 122 |
+
"learning_rate": 0.0037340153452685423,
|
| 123 |
+
"loss": 1.8695,
|
| 124 |
+
"step": 6500
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"epoch": 3.580562659846547,
|
| 128 |
+
"grad_norm": 0.04541015625,
|
| 129 |
+
"learning_rate": 0.0037135549872122763,
|
| 130 |
+
"loss": 1.8688,
|
| 131 |
+
"step": 7000
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"epoch": 3.836317135549872,
|
| 135 |
+
"grad_norm": 0.037109375,
|
| 136 |
+
"learning_rate": 0.0036930946291560103,
|
| 137 |
+
"loss": 1.8686,
|
| 138 |
+
"step": 7500
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"epoch": 4.0,
|
| 142 |
+
"eval_loss": 3.7466166019439697,
|
| 143 |
+
"eval_runtime": 1.1726,
|
| 144 |
+
"eval_samples_per_second": 426.391,
|
| 145 |
+
"eval_steps_per_second": 0.853,
|
| 146 |
+
"step": 7820
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 4.092071611253197,
|
| 150 |
+
"grad_norm": 0.04833984375,
|
| 151 |
+
"learning_rate": 0.0036726342710997444,
|
| 152 |
+
"loss": 1.8681,
|
| 153 |
+
"step": 8000
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"epoch": 4.3478260869565215,
|
| 157 |
+
"grad_norm": 0.047607421875,
|
| 158 |
+
"learning_rate": 0.003652173913043478,
|
| 159 |
+
"loss": 1.8678,
|
| 160 |
+
"step": 8500
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 4.603580562659847,
|
| 164 |
+
"grad_norm": 0.056640625,
|
| 165 |
+
"learning_rate": 0.0036317135549872124,
|
| 166 |
+
"loss": 1.8674,
|
| 167 |
+
"step": 9000
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"epoch": 4.859335038363171,
|
| 171 |
+
"grad_norm": 0.0400390625,
|
| 172 |
+
"learning_rate": 0.0036112531969309464,
|
| 173 |
+
"loss": 1.867,
|
| 174 |
+
"step": 9500
|
| 175 |
+
},
|
| 176 |
+
{
|
| 177 |
+
"epoch": 5.0,
|
| 178 |
+
"eval_loss": 3.7444469928741455,
|
| 179 |
+
"eval_runtime": 1.1971,
|
| 180 |
+
"eval_samples_per_second": 417.668,
|
| 181 |
+
"eval_steps_per_second": 0.835,
|
| 182 |
+
"step": 9775
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"epoch": 5.115089514066496,
|
| 186 |
+
"grad_norm": 0.04736328125,
|
| 187 |
+
"learning_rate": 0.0035907928388746804,
|
| 188 |
+
"loss": 1.8668,
|
| 189 |
+
"step": 10000
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 5.370843989769821,
|
| 193 |
+
"grad_norm": 0.03564453125,
|
| 194 |
+
"learning_rate": 0.0035703324808184144,
|
| 195 |
+
"loss": 1.8668,
|
| 196 |
+
"step": 10500
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"epoch": 5.626598465473146,
|
| 200 |
+
"grad_norm": 0.047119140625,
|
| 201 |
+
"learning_rate": 0.003549872122762148,
|
| 202 |
+
"loss": 1.866,
|
| 203 |
+
"step": 11000
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"epoch": 5.882352941176471,
|
| 207 |
+
"grad_norm": 0.04052734375,
|
| 208 |
+
"learning_rate": 0.0035294117647058825,
|
| 209 |
+
"loss": 1.8662,
|
| 210 |
+
"step": 11500
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 6.0,
|
| 214 |
+
"eval_loss": 3.742023229598999,
|
| 215 |
+
"eval_runtime": 1.1689,
|
| 216 |
+
"eval_samples_per_second": 427.744,
|
| 217 |
+
"eval_steps_per_second": 0.855,
|
| 218 |
+
"step": 11730
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"epoch": 6.138107416879795,
|
| 222 |
+
"grad_norm": 0.047119140625,
|
| 223 |
+
"learning_rate": 0.0035089514066496165,
|
| 224 |
+
"loss": 1.8652,
|
| 225 |
+
"step": 12000
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"epoch": 6.3938618925831205,
|
| 229 |
+
"grad_norm": 0.041259765625,
|
| 230 |
+
"learning_rate": 0.0034884910485933505,
|
| 231 |
+
"loss": 1.8656,
|
| 232 |
+
"step": 12500
|
| 233 |
+
},
|
| 234 |
+
{
|
| 235 |
+
"epoch": 6.649616368286445,
|
| 236 |
+
"grad_norm": 0.040283203125,
|
| 237 |
+
"learning_rate": 0.0034680306905370845,
|
| 238 |
+
"loss": 1.8655,
|
| 239 |
+
"step": 13000
|
| 240 |
+
},
|
| 241 |
+
{
|
| 242 |
+
"epoch": 6.90537084398977,
|
| 243 |
+
"grad_norm": 0.04736328125,
|
| 244 |
+
"learning_rate": 0.003447570332480818,
|
| 245 |
+
"loss": 1.8653,
|
| 246 |
+
"step": 13500
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"epoch": 7.0,
|
| 250 |
+
"eval_loss": 3.7410786151885986,
|
| 251 |
+
"eval_runtime": 1.2028,
|
| 252 |
+
"eval_samples_per_second": 415.68,
|
| 253 |
+
"eval_steps_per_second": 0.831,
|
| 254 |
+
"step": 13685
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 7.161125319693094,
|
| 258 |
+
"grad_norm": 0.0439453125,
|
| 259 |
+
"learning_rate": 0.0034271099744245526,
|
| 260 |
+
"loss": 1.865,
|
| 261 |
+
"step": 14000
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 7.41687979539642,
|
| 265 |
+
"grad_norm": 0.04052734375,
|
| 266 |
+
"learning_rate": 0.0034066496163682866,
|
| 267 |
+
"loss": 1.8648,
|
| 268 |
+
"step": 14500
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 7.672634271099744,
|
| 272 |
+
"grad_norm": 0.060302734375,
|
| 273 |
+
"learning_rate": 0.0033861892583120206,
|
| 274 |
+
"loss": 1.865,
|
| 275 |
+
"step": 15000
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 7.928388746803069,
|
| 279 |
+
"grad_norm": 0.0400390625,
|
| 280 |
+
"learning_rate": 0.0033657289002557546,
|
| 281 |
+
"loss": 1.8641,
|
| 282 |
+
"step": 15500
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 8.0,
|
| 286 |
+
"eval_loss": 3.7378389835357666,
|
| 287 |
+
"eval_runtime": 1.1927,
|
| 288 |
+
"eval_samples_per_second": 419.217,
|
| 289 |
+
"eval_steps_per_second": 0.838,
|
| 290 |
+
"step": 15640
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"epoch": 8.184143222506394,
|
| 294 |
+
"grad_norm": 0.04345703125,
|
| 295 |
+
"learning_rate": 0.003345268542199488,
|
| 296 |
+
"loss": 1.8644,
|
| 297 |
+
"step": 16000
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"epoch": 8.43989769820972,
|
| 301 |
+
"grad_norm": 0.04736328125,
|
| 302 |
+
"learning_rate": 0.0033248081841432226,
|
| 303 |
+
"loss": 1.8638,
|
| 304 |
+
"step": 16500
|
| 305 |
+
},
|
| 306 |
+
{
|
| 307 |
+
"epoch": 8.695652173913043,
|
| 308 |
+
"grad_norm": 0.05126953125,
|
| 309 |
+
"learning_rate": 0.0033043478260869567,
|
| 310 |
+
"loss": 1.8639,
|
| 311 |
+
"step": 17000
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"epoch": 8.951406649616368,
|
| 315 |
+
"grad_norm": 0.05224609375,
|
| 316 |
+
"learning_rate": 0.0032838874680306907,
|
| 317 |
+
"loss": 1.8641,
|
| 318 |
+
"step": 17500
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 9.0,
|
| 322 |
+
"eval_loss": 3.7366719245910645,
|
| 323 |
+
"eval_runtime": 1.2549,
|
| 324 |
+
"eval_samples_per_second": 398.446,
|
| 325 |
+
"eval_steps_per_second": 0.797,
|
| 326 |
+
"step": 17595
|
| 327 |
+
},
|
| 328 |
+
{
|
| 329 |
+
"epoch": 9.207161125319693,
|
| 330 |
+
"grad_norm": 0.06396484375,
|
| 331 |
+
"learning_rate": 0.0032634271099744247,
|
| 332 |
+
"loss": 1.8636,
|
| 333 |
+
"step": 18000
|
| 334 |
+
},
|
| 335 |
+
{
|
| 336 |
+
"epoch": 9.462915601023019,
|
| 337 |
+
"grad_norm": 0.03857421875,
|
| 338 |
+
"learning_rate": 0.0032429667519181583,
|
| 339 |
+
"loss": 1.8655,
|
| 340 |
+
"step": 18500
|
| 341 |
+
},
|
| 342 |
+
{
|
| 343 |
+
"epoch": 9.718670076726342,
|
| 344 |
+
"grad_norm": 0.038818359375,
|
| 345 |
+
"learning_rate": 0.0032225063938618927,
|
| 346 |
+
"loss": 1.8655,
|
| 347 |
+
"step": 19000
|
| 348 |
+
},
|
| 349 |
+
{
|
| 350 |
+
"epoch": 9.974424552429667,
|
| 351 |
+
"grad_norm": 0.04150390625,
|
| 352 |
+
"learning_rate": 0.0032020460358056268,
|
| 353 |
+
"loss": 1.8646,
|
| 354 |
+
"step": 19500
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 10.0,
|
| 358 |
+
"eval_loss": 3.7368271350860596,
|
| 359 |
+
"eval_runtime": 1.1654,
|
| 360 |
+
"eval_samples_per_second": 429.02,
|
| 361 |
+
"eval_steps_per_second": 0.858,
|
| 362 |
+
"step": 19550
|
| 363 |
+
},
|
| 364 |
+
{
|
| 365 |
+
"epoch": 10.230179028132993,
|
| 366 |
+
"grad_norm": 0.051513671875,
|
| 367 |
+
"learning_rate": 0.0031815856777493608,
|
| 368 |
+
"loss": 1.8637,
|
| 369 |
+
"step": 20000
|
| 370 |
+
},
|
| 371 |
+
{
|
| 372 |
+
"epoch": 10.485933503836318,
|
| 373 |
+
"grad_norm": 0.048095703125,
|
| 374 |
+
"learning_rate": 0.003161125319693095,
|
| 375 |
+
"loss": 1.8641,
|
| 376 |
+
"step": 20500
|
| 377 |
+
},
|
| 378 |
+
{
|
| 379 |
+
"epoch": 10.741687979539641,
|
| 380 |
+
"grad_norm": 0.037841796875,
|
| 381 |
+
"learning_rate": 0.0031406649616368284,
|
| 382 |
+
"loss": 1.8633,
|
| 383 |
+
"step": 21000
|
| 384 |
+
},
|
| 385 |
+
{
|
| 386 |
+
"epoch": 10.997442455242966,
|
| 387 |
+
"grad_norm": 0.04931640625,
|
| 388 |
+
"learning_rate": 0.003120204603580563,
|
| 389 |
+
"loss": 1.8633,
|
| 390 |
+
"step": 21500
|
| 391 |
+
},
|
| 392 |
+
{
|
| 393 |
+
"epoch": 11.0,
|
| 394 |
+
"eval_loss": 3.736281156539917,
|
| 395 |
+
"eval_runtime": 1.3206,
|
| 396 |
+
"eval_samples_per_second": 378.604,
|
| 397 |
+
"eval_steps_per_second": 0.757,
|
| 398 |
+
"step": 21505
|
| 399 |
+
},
|
| 400 |
+
{
|
| 401 |
+
"epoch": 11.253196930946292,
|
| 402 |
+
"grad_norm": 0.04833984375,
|
| 403 |
+
"learning_rate": 0.003099744245524297,
|
| 404 |
+
"loss": 1.8636,
|
| 405 |
+
"step": 22000
|
| 406 |
+
},
|
| 407 |
+
{
|
| 408 |
+
"epoch": 11.508951406649617,
|
| 409 |
+
"grad_norm": 0.0478515625,
|
| 410 |
+
"learning_rate": 0.003079283887468031,
|
| 411 |
+
"loss": 1.863,
|
| 412 |
+
"step": 22500
|
| 413 |
+
},
|
| 414 |
+
{
|
| 415 |
+
"epoch": 11.764705882352942,
|
| 416 |
+
"grad_norm": 0.048583984375,
|
| 417 |
+
"learning_rate": 0.0030588235294117644,
|
| 418 |
+
"loss": 1.8629,
|
| 419 |
+
"step": 23000
|
| 420 |
+
},
|
| 421 |
+
{
|
| 422 |
+
"epoch": 12.0,
|
| 423 |
+
"eval_loss": 3.734565496444702,
|
| 424 |
+
"eval_runtime": 1.1835,
|
| 425 |
+
"eval_samples_per_second": 422.479,
|
| 426 |
+
"eval_steps_per_second": 0.845,
|
| 427 |
+
"step": 23460
|
| 428 |
+
},
|
| 429 |
+
{
|
| 430 |
+
"epoch": 12.020460358056265,
|
| 431 |
+
"grad_norm": 0.044189453125,
|
| 432 |
+
"learning_rate": 0.0030383631713554985,
|
| 433 |
+
"loss": 1.8625,
|
| 434 |
+
"step": 23500
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"epoch": 12.27621483375959,
|
| 438 |
+
"grad_norm": 0.048828125,
|
| 439 |
+
"learning_rate": 0.003017902813299233,
|
| 440 |
+
"loss": 1.8629,
|
| 441 |
+
"step": 24000
|
| 442 |
+
},
|
| 443 |
+
{
|
| 444 |
+
"epoch": 12.531969309462916,
|
| 445 |
+
"grad_norm": 0.04931640625,
|
| 446 |
+
"learning_rate": 0.002997442455242967,
|
| 447 |
+
"loss": 1.8624,
|
| 448 |
+
"step": 24500
|
| 449 |
+
},
|
| 450 |
+
{
|
| 451 |
+
"epoch": 12.787723785166241,
|
| 452 |
+
"grad_norm": 0.04345703125,
|
| 453 |
+
"learning_rate": 0.002976982097186701,
|
| 454 |
+
"loss": 1.8621,
|
| 455 |
+
"step": 25000
|
| 456 |
+
},
|
| 457 |
+
{
|
| 458 |
+
"epoch": 13.0,
|
| 459 |
+
"eval_loss": 3.7329983711242676,
|
| 460 |
+
"eval_runtime": 1.3627,
|
| 461 |
+
"eval_samples_per_second": 366.916,
|
| 462 |
+
"eval_steps_per_second": 0.734,
|
| 463 |
+
"step": 25415
|
| 464 |
+
},
|
| 465 |
+
{
|
| 466 |
+
"epoch": 13.043478260869565,
|
| 467 |
+
"grad_norm": 0.0546875,
|
| 468 |
+
"learning_rate": 0.0029565217391304345,
|
| 469 |
+
"loss": 1.8624,
|
| 470 |
+
"step": 25500
|
| 471 |
+
},
|
| 472 |
+
{
|
| 473 |
+
"epoch": 13.29923273657289,
|
| 474 |
+
"grad_norm": 0.048095703125,
|
| 475 |
+
"learning_rate": 0.002936061381074169,
|
| 476 |
+
"loss": 1.8624,
|
| 477 |
+
"step": 26000
|
| 478 |
+
},
|
| 479 |
+
{
|
| 480 |
+
"epoch": 13.554987212276215,
|
| 481 |
+
"grad_norm": 0.05517578125,
|
| 482 |
+
"learning_rate": 0.002915601023017903,
|
| 483 |
+
"loss": 1.8618,
|
| 484 |
+
"step": 26500
|
| 485 |
+
},
|
| 486 |
+
{
|
| 487 |
+
"epoch": 13.81074168797954,
|
| 488 |
+
"grad_norm": 0.04443359375,
|
| 489 |
+
"learning_rate": 0.002895140664961637,
|
| 490 |
+
"loss": 1.8619,
|
| 491 |
+
"step": 27000
|
| 492 |
+
},
|
| 493 |
+
{
|
| 494 |
+
"epoch": 14.0,
|
| 495 |
+
"eval_loss": 3.7325527667999268,
|
| 496 |
+
"eval_runtime": 1.1565,
|
| 497 |
+
"eval_samples_per_second": 432.347,
|
| 498 |
+
"eval_steps_per_second": 0.865,
|
| 499 |
+
"step": 27370
|
| 500 |
+
},
|
| 501 |
+
{
|
| 502 |
+
"epoch": 14.066496163682864,
|
| 503 |
+
"grad_norm": 0.043212890625,
|
| 504 |
+
"learning_rate": 0.002874680306905371,
|
| 505 |
+
"loss": 1.862,
|
| 506 |
+
"step": 27500
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 14.322250639386189,
|
| 510 |
+
"grad_norm": 0.06640625,
|
| 511 |
+
"learning_rate": 0.0028542199488491046,
|
| 512 |
+
"loss": 1.8616,
|
| 513 |
+
"step": 28000
|
| 514 |
+
},
|
| 515 |
+
{
|
| 516 |
+
"epoch": 14.578005115089514,
|
| 517 |
+
"grad_norm": 0.050048828125,
|
| 518 |
+
"learning_rate": 0.002833759590792839,
|
| 519 |
+
"loss": 1.8615,
|
| 520 |
+
"step": 28500
|
| 521 |
+
},
|
| 522 |
+
{
|
| 523 |
+
"epoch": 14.83375959079284,
|
| 524 |
+
"grad_norm": 0.046875,
|
| 525 |
+
"learning_rate": 0.002813299232736573,
|
| 526 |
+
"loss": 1.8623,
|
| 527 |
+
"step": 29000
|
| 528 |
+
},
|
| 529 |
+
{
|
| 530 |
+
"epoch": 15.0,
|
| 531 |
+
"eval_loss": 3.729966402053833,
|
| 532 |
+
"eval_runtime": 1.1923,
|
| 533 |
+
"eval_samples_per_second": 419.349,
|
| 534 |
+
"eval_steps_per_second": 0.839,
|
| 535 |
+
"step": 29325
|
| 536 |
+
},
|
| 537 |
+
{
|
| 538 |
+
"epoch": 15.089514066496164,
|
| 539 |
+
"grad_norm": 0.0458984375,
|
| 540 |
+
"learning_rate": 0.002792838874680307,
|
| 541 |
+
"loss": 1.8615,
|
| 542 |
+
"step": 29500
|
| 543 |
+
},
|
| 544 |
+
{
|
| 545 |
+
"epoch": 15.345268542199488,
|
| 546 |
+
"grad_norm": 0.042236328125,
|
| 547 |
+
"learning_rate": 0.002772378516624041,
|
| 548 |
+
"loss": 1.8616,
|
| 549 |
+
"step": 30000
|
| 550 |
+
},
|
| 551 |
+
{
|
| 552 |
+
"epoch": 15.601023017902813,
|
| 553 |
+
"grad_norm": 0.037353515625,
|
| 554 |
+
"learning_rate": 0.0027519181585677747,
|
| 555 |
+
"loss": 1.8618,
|
| 556 |
+
"step": 30500
|
| 557 |
+
},
|
| 558 |
+
{
|
| 559 |
+
"epoch": 15.856777493606138,
|
| 560 |
+
"grad_norm": 0.06591796875,
|
| 561 |
+
"learning_rate": 0.002731457800511509,
|
| 562 |
+
"loss": 1.8613,
|
| 563 |
+
"step": 31000
|
| 564 |
+
},
|
| 565 |
+
{
|
| 566 |
+
"epoch": 16.0,
|
| 567 |
+
"eval_loss": 3.7326087951660156,
|
| 568 |
+
"eval_runtime": 1.1839,
|
| 569 |
+
"eval_samples_per_second": 422.34,
|
| 570 |
+
"eval_steps_per_second": 0.845,
|
| 571 |
+
"step": 31280
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 16.11253196930946,
|
| 575 |
+
"grad_norm": 0.038818359375,
|
| 576 |
+
"learning_rate": 0.002710997442455243,
|
| 577 |
+
"loss": 1.8615,
|
| 578 |
+
"step": 31500
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 16.36828644501279,
|
| 582 |
+
"grad_norm": 0.0458984375,
|
| 583 |
+
"learning_rate": 0.002690537084398977,
|
| 584 |
+
"loss": 1.8619,
|
| 585 |
+
"step": 32000
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 16.624040920716112,
|
| 589 |
+
"grad_norm": 0.044677734375,
|
| 590 |
+
"learning_rate": 0.002670076726342711,
|
| 591 |
+
"loss": 1.8616,
|
| 592 |
+
"step": 32500
|
| 593 |
+
},
|
| 594 |
+
{
|
| 595 |
+
"epoch": 16.87979539641944,
|
| 596 |
+
"grad_norm": 0.04541015625,
|
| 597 |
+
"learning_rate": 0.0026496163682864448,
|
| 598 |
+
"loss": 1.8616,
|
| 599 |
+
"step": 33000
|
| 600 |
+
},
|
| 601 |
+
{
|
| 602 |
+
"epoch": 17.0,
|
| 603 |
+
"eval_loss": 3.729959011077881,
|
| 604 |
+
"eval_runtime": 1.2063,
|
| 605 |
+
"eval_samples_per_second": 414.478,
|
| 606 |
+
"eval_steps_per_second": 0.829,
|
| 607 |
+
"step": 33235
|
| 608 |
+
},
|
| 609 |
+
{
|
| 610 |
+
"epoch": 17.135549872122763,
|
| 611 |
+
"grad_norm": 0.04541015625,
|
| 612 |
+
"learning_rate": 0.0026291560102301792,
|
| 613 |
+
"loss": 1.8616,
|
| 614 |
+
"step": 33500
|
| 615 |
+
},
|
| 616 |
+
{
|
| 617 |
+
"epoch": 17.391304347826086,
|
| 618 |
+
"grad_norm": 0.048583984375,
|
| 619 |
+
"learning_rate": 0.0026086956521739132,
|
| 620 |
+
"loss": 1.8608,
|
| 621 |
+
"step": 34000
|
| 622 |
+
},
|
| 623 |
+
{
|
| 624 |
+
"epoch": 17.647058823529413,
|
| 625 |
+
"grad_norm": 0.047119140625,
|
| 626 |
+
"learning_rate": 0.0025882352941176473,
|
| 627 |
+
"loss": 1.8609,
|
| 628 |
+
"step": 34500
|
| 629 |
+
},
|
| 630 |
+
{
|
| 631 |
+
"epoch": 17.902813299232736,
|
| 632 |
+
"grad_norm": 0.041259765625,
|
| 633 |
+
"learning_rate": 0.002567774936061381,
|
| 634 |
+
"loss": 1.8606,
|
| 635 |
+
"step": 35000
|
| 636 |
+
},
|
| 637 |
+
{
|
| 638 |
+
"epoch": 18.0,
|
| 639 |
+
"eval_loss": 3.7307207584381104,
|
| 640 |
+
"eval_runtime": 1.1724,
|
| 641 |
+
"eval_samples_per_second": 426.468,
|
| 642 |
+
"eval_steps_per_second": 0.853,
|
| 643 |
+
"step": 35190
|
| 644 |
+
},
|
| 645 |
+
{
|
| 646 |
+
"epoch": 18.15856777493606,
|
| 647 |
+
"grad_norm": 0.04052734375,
|
| 648 |
+
"learning_rate": 0.002547314578005115,
|
| 649 |
+
"loss": 1.8613,
|
| 650 |
+
"step": 35500
|
| 651 |
+
},
|
| 652 |
+
{
|
| 653 |
+
"epoch": 18.414322250639387,
|
| 654 |
+
"grad_norm": 0.042236328125,
|
| 655 |
+
"learning_rate": 0.0025268542199488493,
|
| 656 |
+
"loss": 1.8608,
|
| 657 |
+
"step": 36000
|
| 658 |
+
},
|
| 659 |
+
{
|
| 660 |
+
"epoch": 18.67007672634271,
|
| 661 |
+
"grad_norm": 0.046142578125,
|
| 662 |
+
"learning_rate": 0.0025063938618925833,
|
| 663 |
+
"loss": 1.8607,
|
| 664 |
+
"step": 36500
|
| 665 |
+
},
|
| 666 |
+
{
|
| 667 |
+
"epoch": 18.925831202046037,
|
| 668 |
+
"grad_norm": 0.03955078125,
|
| 669 |
+
"learning_rate": 0.0024859335038363174,
|
| 670 |
+
"loss": 1.8606,
|
| 671 |
+
"step": 37000
|
| 672 |
+
},
|
| 673 |
+
{
|
| 674 |
+
"epoch": 19.0,
|
| 675 |
+
"eval_loss": 3.729092836380005,
|
| 676 |
+
"eval_runtime": 1.1564,
|
| 677 |
+
"eval_samples_per_second": 432.369,
|
| 678 |
+
"eval_steps_per_second": 0.865,
|
| 679 |
+
"step": 37145
|
| 680 |
+
},
|
| 681 |
+
{
|
| 682 |
+
"epoch": 19.18158567774936,
|
| 683 |
+
"grad_norm": 0.041259765625,
|
| 684 |
+
"learning_rate": 0.002465473145780051,
|
| 685 |
+
"loss": 1.8604,
|
| 686 |
+
"step": 37500
|
| 687 |
+
},
|
| 688 |
+
{
|
| 689 |
+
"epoch": 19.437340153452684,
|
| 690 |
+
"grad_norm": 0.047607421875,
|
| 691 |
+
"learning_rate": 0.002445012787723785,
|
| 692 |
+
"loss": 1.8607,
|
| 693 |
+
"step": 38000
|
| 694 |
+
},
|
| 695 |
+
{
|
| 696 |
+
"epoch": 19.69309462915601,
|
| 697 |
+
"grad_norm": 0.056396484375,
|
| 698 |
+
"learning_rate": 0.0024245524296675194,
|
| 699 |
+
"loss": 1.8605,
|
| 700 |
+
"step": 38500
|
| 701 |
+
},
|
| 702 |
+
{
|
| 703 |
+
"epoch": 19.948849104859335,
|
| 704 |
+
"grad_norm": 0.05517578125,
|
| 705 |
+
"learning_rate": 0.0024040920716112534,
|
| 706 |
+
"loss": 1.8606,
|
| 707 |
+
"step": 39000
|
| 708 |
+
},
|
| 709 |
+
{
|
| 710 |
+
"epoch": 20.0,
|
| 711 |
+
"eval_loss": 3.727680206298828,
|
| 712 |
+
"eval_runtime": 1.1607,
|
| 713 |
+
"eval_samples_per_second": 430.785,
|
| 714 |
+
"eval_steps_per_second": 0.862,
|
| 715 |
+
"step": 39100
|
| 716 |
+
},
|
| 717 |
+
{
|
| 718 |
+
"epoch": 20.20460358056266,
|
| 719 |
+
"grad_norm": 0.043212890625,
|
| 720 |
+
"learning_rate": 0.0023836317135549874,
|
| 721 |
+
"loss": 1.8605,
|
| 722 |
+
"step": 39500
|
| 723 |
+
},
|
| 724 |
+
{
|
| 725 |
+
"epoch": 20.460358056265985,
|
| 726 |
+
"grad_norm": 0.04345703125,
|
| 727 |
+
"learning_rate": 0.002363171355498721,
|
| 728 |
+
"loss": 1.8605,
|
| 729 |
+
"step": 40000
|
| 730 |
+
},
|
| 731 |
+
{
|
| 732 |
+
"epoch": 20.71611253196931,
|
| 733 |
+
"grad_norm": 0.037841796875,
|
| 734 |
+
"learning_rate": 0.002342710997442455,
|
| 735 |
+
"loss": 1.8606,
|
| 736 |
+
"step": 40500
|
| 737 |
+
},
|
| 738 |
+
{
|
| 739 |
+
"epoch": 20.971867007672635,
|
| 740 |
+
"grad_norm": 0.051513671875,
|
| 741 |
+
"learning_rate": 0.0023222506393861895,
|
| 742 |
+
"loss": 1.8601,
|
| 743 |
+
"step": 41000
|
| 744 |
+
},
|
| 745 |
+
{
|
| 746 |
+
"epoch": 21.0,
|
| 747 |
+
"eval_loss": 3.7270610332489014,
|
| 748 |
+
"eval_runtime": 1.1618,
|
| 749 |
+
"eval_samples_per_second": 430.348,
|
| 750 |
+
"eval_steps_per_second": 0.861,
|
| 751 |
+
"step": 41055
|
| 752 |
+
},
|
| 753 |
+
{
|
| 754 |
+
"epoch": 21.22762148337596,
|
| 755 |
+
"grad_norm": 0.05126953125,
|
| 756 |
+
"learning_rate": 0.0023017902813299235,
|
| 757 |
+
"loss": 1.8601,
|
| 758 |
+
"step": 41500
|
| 759 |
+
},
|
| 760 |
+
{
|
| 761 |
+
"epoch": 21.483375959079282,
|
| 762 |
+
"grad_norm": 0.056884765625,
|
| 763 |
+
"learning_rate": 0.0022813299232736575,
|
| 764 |
+
"loss": 1.8601,
|
| 765 |
+
"step": 42000
|
| 766 |
+
},
|
| 767 |
+
{
|
| 768 |
+
"epoch": 21.73913043478261,
|
| 769 |
+
"grad_norm": 0.045166015625,
|
| 770 |
+
"learning_rate": 0.002260869565217391,
|
| 771 |
+
"loss": 1.8602,
|
| 772 |
+
"step": 42500
|
| 773 |
+
},
|
| 774 |
+
{
|
| 775 |
+
"epoch": 21.994884910485933,
|
| 776 |
+
"grad_norm": 0.036376953125,
|
| 777 |
+
"learning_rate": 0.002240409207161125,
|
| 778 |
+
"loss": 1.86,
|
| 779 |
+
"step": 43000
|
| 780 |
+
},
|
| 781 |
+
{
|
| 782 |
+
"epoch": 22.0,
|
| 783 |
+
"eval_loss": 3.72686505317688,
|
| 784 |
+
"eval_runtime": 1.2281,
|
| 785 |
+
"eval_samples_per_second": 407.144,
|
| 786 |
+
"eval_steps_per_second": 0.814,
|
| 787 |
+
"step": 43010
|
| 788 |
+
},
|
| 789 |
+
{
|
| 790 |
+
"epoch": 22.25063938618926,
|
| 791 |
+
"grad_norm": 0.04931640625,
|
| 792 |
+
"learning_rate": 0.0022199488491048596,
|
| 793 |
+
"loss": 1.86,
|
| 794 |
+
"step": 43500
|
| 795 |
+
},
|
| 796 |
+
{
|
| 797 |
+
"epoch": 22.506393861892583,
|
| 798 |
+
"grad_norm": 0.0380859375,
|
| 799 |
+
"learning_rate": 0.0021994884910485936,
|
| 800 |
+
"loss": 1.8596,
|
| 801 |
+
"step": 44000
|
| 802 |
+
},
|
| 803 |
+
{
|
| 804 |
+
"epoch": 22.762148337595907,
|
| 805 |
+
"grad_norm": 0.039306640625,
|
| 806 |
+
"learning_rate": 0.0021790281329923276,
|
| 807 |
+
"loss": 1.8602,
|
| 808 |
+
"step": 44500
|
| 809 |
+
},
|
| 810 |
+
{
|
| 811 |
+
"epoch": 23.0,
|
| 812 |
+
"eval_loss": 3.7282092571258545,
|
| 813 |
+
"eval_runtime": 1.1696,
|
| 814 |
+
"eval_samples_per_second": 427.495,
|
| 815 |
+
"eval_steps_per_second": 0.855,
|
| 816 |
+
"step": 44965
|
| 817 |
+
},
|
| 818 |
+
{
|
| 819 |
+
"epoch": 23.017902813299234,
|
| 820 |
+
"grad_norm": 0.046142578125,
|
| 821 |
+
"learning_rate": 0.002158567774936061,
|
| 822 |
+
"loss": 1.86,
|
| 823 |
+
"step": 45000
|
| 824 |
+
},
|
| 825 |
+
{
|
| 826 |
+
"epoch": 23.273657289002557,
|
| 827 |
+
"grad_norm": 0.039306640625,
|
| 828 |
+
"learning_rate": 0.002138107416879795,
|
| 829 |
+
"loss": 1.8598,
|
| 830 |
+
"step": 45500
|
| 831 |
+
},
|
| 832 |
+
{
|
| 833 |
+
"epoch": 23.529411764705884,
|
| 834 |
+
"grad_norm": 0.04296875,
|
| 835 |
+
"learning_rate": 0.0021176470588235297,
|
| 836 |
+
"loss": 1.86,
|
| 837 |
+
"step": 46000
|
| 838 |
+
},
|
| 839 |
+
{
|
| 840 |
+
"epoch": 23.785166240409207,
|
| 841 |
+
"grad_norm": 0.054931640625,
|
| 842 |
+
"learning_rate": 0.0020971867007672637,
|
| 843 |
+
"loss": 1.8596,
|
| 844 |
+
"step": 46500
|
| 845 |
+
},
|
| 846 |
+
{
|
| 847 |
+
"epoch": 24.0,
|
| 848 |
+
"eval_loss": 3.7264657020568848,
|
| 849 |
+
"eval_runtime": 1.1709,
|
| 850 |
+
"eval_samples_per_second": 427.022,
|
| 851 |
+
"eval_steps_per_second": 0.854,
|
| 852 |
+
"step": 46920
|
| 853 |
+
},
|
| 854 |
+
{
|
| 855 |
+
"epoch": 24.04092071611253,
|
| 856 |
+
"grad_norm": 0.0419921875,
|
| 857 |
+
"learning_rate": 0.0020767263427109977,
|
| 858 |
+
"loss": 1.8599,
|
| 859 |
+
"step": 47000
|
| 860 |
+
},
|
| 861 |
+
{
|
| 862 |
+
"epoch": 24.296675191815858,
|
| 863 |
+
"grad_norm": 0.043701171875,
|
| 864 |
+
"learning_rate": 0.0020562659846547313,
|
| 865 |
+
"loss": 1.8603,
|
| 866 |
+
"step": 47500
|
| 867 |
+
},
|
| 868 |
+
{
|
| 869 |
+
"epoch": 24.55242966751918,
|
| 870 |
+
"grad_norm": 0.06884765625,
|
| 871 |
+
"learning_rate": 0.0020358056265984653,
|
| 872 |
+
"loss": 1.8592,
|
| 873 |
+
"step": 48000
|
| 874 |
+
},
|
| 875 |
+
{
|
| 876 |
+
"epoch": 24.808184143222505,
|
| 877 |
+
"grad_norm": 0.037353515625,
|
| 878 |
+
"learning_rate": 0.0020153452685421997,
|
| 879 |
+
"loss": 1.8596,
|
| 880 |
+
"step": 48500
|
| 881 |
+
},
|
| 882 |
+
{
|
| 883 |
+
"epoch": 25.0,
|
| 884 |
+
"eval_loss": 3.7268757820129395,
|
| 885 |
+
"eval_runtime": 1.2077,
|
| 886 |
+
"eval_samples_per_second": 414.025,
|
| 887 |
+
"eval_steps_per_second": 0.828,
|
| 888 |
+
"step": 48875
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 25.06393861892583,
|
| 892 |
+
"grad_norm": 0.0498046875,
|
| 893 |
+
"learning_rate": 0.0019948849104859333,
|
| 894 |
+
"loss": 1.8595,
|
| 895 |
+
"step": 49000
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 25.319693094629155,
|
| 899 |
+
"grad_norm": 0.0419921875,
|
| 900 |
+
"learning_rate": 0.0019744245524296678,
|
| 901 |
+
"loss": 1.8592,
|
| 902 |
+
"step": 49500
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 25.575447570332482,
|
| 906 |
+
"grad_norm": 0.0478515625,
|
| 907 |
+
"learning_rate": 0.001953964194373402,
|
| 908 |
+
"loss": 1.8595,
|
| 909 |
+
"step": 50000
|
| 910 |
+
},
|
| 911 |
+
{
|
| 912 |
+
"epoch": 25.831202046035806,
|
| 913 |
+
"grad_norm": 0.04052734375,
|
| 914 |
+
"learning_rate": 0.0019335038363171356,
|
| 915 |
+
"loss": 1.8598,
|
| 916 |
+
"step": 50500
|
| 917 |
+
},
|
| 918 |
+
{
|
| 919 |
+
"epoch": 26.0,
|
| 920 |
+
"eval_loss": 3.7266719341278076,
|
| 921 |
+
"eval_runtime": 1.1718,
|
| 922 |
+
"eval_samples_per_second": 426.711,
|
| 923 |
+
"eval_steps_per_second": 0.853,
|
| 924 |
+
"step": 50830
|
| 925 |
+
},
|
| 926 |
+
{
|
| 927 |
+
"epoch": 26.08695652173913,
|
| 928 |
+
"grad_norm": 0.046142578125,
|
| 929 |
+
"learning_rate": 0.0019130434782608696,
|
| 930 |
+
"loss": 1.8595,
|
| 931 |
+
"step": 51000
|
| 932 |
+
},
|
| 933 |
+
{
|
| 934 |
+
"epoch": 26.342710997442456,
|
| 935 |
+
"grad_norm": 0.043701171875,
|
| 936 |
+
"learning_rate": 0.0018925831202046036,
|
| 937 |
+
"loss": 1.8595,
|
| 938 |
+
"step": 51500
|
| 939 |
+
},
|
| 940 |
+
{
|
| 941 |
+
"epoch": 26.59846547314578,
|
| 942 |
+
"grad_norm": 0.03369140625,
|
| 943 |
+
"learning_rate": 0.0018721227621483376,
|
| 944 |
+
"loss": 1.8593,
|
| 945 |
+
"step": 52000
|
| 946 |
+
},
|
| 947 |
+
{
|
| 948 |
+
"epoch": 26.854219948849106,
|
| 949 |
+
"grad_norm": 0.03662109375,
|
| 950 |
+
"learning_rate": 0.0018516624040920714,
|
| 951 |
+
"loss": 1.8595,
|
| 952 |
+
"step": 52500
|
| 953 |
+
},
|
| 954 |
+
{
|
| 955 |
+
"epoch": 27.0,
|
| 956 |
+
"eval_loss": 3.72537899017334,
|
| 957 |
+
"eval_runtime": 1.2057,
|
| 958 |
+
"eval_samples_per_second": 414.687,
|
| 959 |
+
"eval_steps_per_second": 0.829,
|
| 960 |
+
"step": 52785
|
| 961 |
+
},
|
| 962 |
+
{
|
| 963 |
+
"epoch": 27.10997442455243,
|
| 964 |
+
"grad_norm": 0.044921875,
|
| 965 |
+
"learning_rate": 0.0018312020460358057,
|
| 966 |
+
"loss": 1.8596,
|
| 967 |
+
"step": 53000
|
| 968 |
+
},
|
| 969 |
+
{
|
| 970 |
+
"epoch": 27.365728900255753,
|
| 971 |
+
"grad_norm": 0.04248046875,
|
| 972 |
+
"learning_rate": 0.0018107416879795397,
|
| 973 |
+
"loss": 1.8592,
|
| 974 |
+
"step": 53500
|
| 975 |
+
},
|
| 976 |
+
{
|
| 977 |
+
"epoch": 27.62148337595908,
|
| 978 |
+
"grad_norm": 0.04833984375,
|
| 979 |
+
"learning_rate": 0.0017902813299232737,
|
| 980 |
+
"loss": 1.8598,
|
| 981 |
+
"step": 54000
|
| 982 |
+
},
|
| 983 |
+
{
|
| 984 |
+
"epoch": 27.877237851662404,
|
| 985 |
+
"grad_norm": 0.0400390625,
|
| 986 |
+
"learning_rate": 0.0017698209718670077,
|
| 987 |
+
"loss": 1.8592,
|
| 988 |
+
"step": 54500
|
| 989 |
+
},
|
| 990 |
+
{
|
| 991 |
+
"epoch": 28.0,
|
| 992 |
+
"eval_loss": 3.7256040573120117,
|
| 993 |
+
"eval_runtime": 1.1731,
|
| 994 |
+
"eval_samples_per_second": 426.237,
|
| 995 |
+
"eval_steps_per_second": 0.852,
|
| 996 |
+
"step": 54740
|
| 997 |
+
},
|
| 998 |
+
{
|
| 999 |
+
"epoch": 28.132992327365727,
|
| 1000 |
+
"grad_norm": 0.045654296875,
|
| 1001 |
+
"learning_rate": 0.0017493606138107415,
|
| 1002 |
+
"loss": 1.8593,
|
| 1003 |
+
"step": 55000
|
| 1004 |
+
},
|
| 1005 |
+
{
|
| 1006 |
+
"epoch": 28.388746803069054,
|
| 1007 |
+
"grad_norm": 0.0439453125,
|
| 1008 |
+
"learning_rate": 0.0017289002557544758,
|
| 1009 |
+
"loss": 1.8589,
|
| 1010 |
+
"step": 55500
|
| 1011 |
+
},
|
| 1012 |
+
{
|
| 1013 |
+
"epoch": 28.644501278772378,
|
| 1014 |
+
"grad_norm": 0.047607421875,
|
| 1015 |
+
"learning_rate": 0.0017084398976982098,
|
| 1016 |
+
"loss": 1.8591,
|
| 1017 |
+
"step": 56000
|
| 1018 |
+
},
|
| 1019 |
+
{
|
| 1020 |
+
"epoch": 28.900255754475705,
|
| 1021 |
+
"grad_norm": 0.05029296875,
|
| 1022 |
+
"learning_rate": 0.0016879795396419438,
|
| 1023 |
+
"loss": 1.8592,
|
| 1024 |
+
"step": 56500
|
| 1025 |
+
},
|
| 1026 |
+
{
|
| 1027 |
+
"epoch": 29.0,
|
| 1028 |
+
"eval_loss": 3.7257955074310303,
|
| 1029 |
+
"eval_runtime": 1.1794,
|
| 1030 |
+
"eval_samples_per_second": 423.928,
|
| 1031 |
+
"eval_steps_per_second": 0.848,
|
| 1032 |
+
"step": 56695
|
| 1033 |
+
},
|
| 1034 |
+
{
|
| 1035 |
+
"epoch": 29.156010230179028,
|
| 1036 |
+
"grad_norm": 0.045166015625,
|
| 1037 |
+
"learning_rate": 0.0016675191815856778,
|
| 1038 |
+
"loss": 1.8589,
|
| 1039 |
+
"step": 57000
|
| 1040 |
+
},
|
| 1041 |
+
{
|
| 1042 |
+
"epoch": 29.41176470588235,
|
| 1043 |
+
"grad_norm": 0.037109375,
|
| 1044 |
+
"learning_rate": 0.0016470588235294116,
|
| 1045 |
+
"loss": 1.8591,
|
| 1046 |
+
"step": 57500
|
| 1047 |
+
},
|
| 1048 |
+
{
|
| 1049 |
+
"epoch": 29.66751918158568,
|
| 1050 |
+
"grad_norm": 0.0390625,
|
| 1051 |
+
"learning_rate": 0.0016265984654731459,
|
| 1052 |
+
"loss": 1.8591,
|
| 1053 |
+
"step": 58000
|
| 1054 |
+
},
|
| 1055 |
+
{
|
| 1056 |
+
"epoch": 29.923273657289002,
|
| 1057 |
+
"grad_norm": 0.052734375,
|
| 1058 |
+
"learning_rate": 0.0016061381074168799,
|
| 1059 |
+
"loss": 1.8596,
|
| 1060 |
+
"step": 58500
|
| 1061 |
+
},
|
| 1062 |
+
{
|
| 1063 |
+
"epoch": 30.0,
|
| 1064 |
+
"eval_loss": 3.7249155044555664,
|
| 1065 |
+
"eval_runtime": 1.1467,
|
| 1066 |
+
"eval_samples_per_second": 436.021,
|
| 1067 |
+
"eval_steps_per_second": 0.872,
|
| 1068 |
+
"step": 58650
|
| 1069 |
+
}
|
| 1070 |
+
],
|
| 1071 |
+
"logging_steps": 500,
|
| 1072 |
+
"max_steps": 97750,
|
| 1073 |
+
"num_input_tokens_seen": 0,
|
| 1074 |
+
"num_train_epochs": 50,
|
| 1075 |
+
"save_steps": 500,
|
| 1076 |
+
"stateful_callbacks": {
|
| 1077 |
+
"EarlyStoppingCallback": {
|
| 1078 |
+
"args": {
|
| 1079 |
+
"early_stopping_patience": 3,
|
| 1080 |
+
"early_stopping_threshold": 0.0
|
| 1081 |
+
},
|
| 1082 |
+
"attributes": {
|
| 1083 |
+
"early_stopping_patience_counter": 0
|
| 1084 |
+
}
|
| 1085 |
+
},
|
| 1086 |
+
"TrainerControl": {
|
| 1087 |
+
"args": {
|
| 1088 |
+
"should_epoch_stop": false,
|
| 1089 |
+
"should_evaluate": false,
|
| 1090 |
+
"should_log": false,
|
| 1091 |
+
"should_save": true,
|
| 1092 |
+
"should_training_stop": false
|
| 1093 |
+
},
|
| 1094 |
+
"attributes": {}
|
| 1095 |
+
}
|
| 1096 |
+
},
|
| 1097 |
+
"total_flos": 8.390099826432e+16,
|
| 1098 |
+
"train_batch_size": 512,
|
| 1099 |
+
"trial_name": null,
|
| 1100 |
+
"trial_params": null
|
| 1101 |
+
}
|
training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d22b942b82fe652347305900a1d7e37b1a36bcf238d62c3b61838f38b667e1cf
|
| 3 |
+
size 5304
|