UltimoUno commited on
Commit
98b927b
1 Parent(s): 52a76a6

Uploaded checkpoint-5000

Browse files
config.json ADDED
@@ -0,0 +1,31 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "deepseek-ai/deepseek-coder-1.3b-instruct",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 32013,
9
+ "eos_token_id": 32021,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 5504,
14
+ "max_position_embeddings": 16384,
15
+ "model_type": "llama",
16
+ "num_attention_heads": 16,
17
+ "num_hidden_layers": 24,
18
+ "num_key_value_heads": 16,
19
+ "pretraining_tp": 1,
20
+ "rms_norm_eps": 1e-06,
21
+ "rope_scaling": {
22
+ "factor": 4.0,
23
+ "type": "linear"
24
+ },
25
+ "rope_theta": 100000,
26
+ "tie_word_embeddings": false,
27
+ "torch_dtype": "bfloat16",
28
+ "transformers_version": "4.38.2",
29
+ "use_cache": true,
30
+ "vocab_size": 32256
31
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 32013,
4
+ "eos_token_id": 32021,
5
+ "transformers_version": "4.38.2"
6
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ba416897d4288dac0b42f57a85dbce1773b7dd1e5cf62da3cd2b991761cd523d
3
+ size 2692969128
optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cd8264dbd84d04a7d2583064ae67bfb2780ff95c109b32c977c53fbfe62cc0fe
3
+ size 5386075202
rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f2e155718eaa0de9768e57816e6e6874ba931ec4489afb1c2874db91c97f7a96
3
+ size 14244
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:540fe86e7a7bc9e04f9fb5883b9895a8c37cfe49e162eed520e2d90e4046a9f0
3
+ size 1064
trainer_state.json ADDED
@@ -0,0 +1,3529 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 1.4290132522583008,
3
+ "best_model_checkpoint": "runs/deepseek_20240422-210351/checkpoint-5000",
4
+ "epoch": 0.125,
5
+ "eval_steps": 5000,
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": 21.625,
14
+ "learning_rate": 4.0000000000000003e-07,
15
+ "loss": 2.2157,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.0,
20
+ "grad_norm": 41.0,
21
+ "learning_rate": 8.000000000000001e-07,
22
+ "loss": 2.1199,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.0,
27
+ "grad_norm": 30.875,
28
+ "learning_rate": 1.2000000000000002e-06,
29
+ "loss": 2.2071,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.0,
34
+ "grad_norm": 51.0,
35
+ "learning_rate": 1.6000000000000001e-06,
36
+ "loss": 2.2955,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.0,
41
+ "grad_norm": 61.25,
42
+ "learning_rate": 2.0000000000000003e-06,
43
+ "loss": 2.1046,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.0,
48
+ "grad_norm": 32.25,
49
+ "learning_rate": 2.4000000000000003e-06,
50
+ "loss": 2.162,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.0,
55
+ "grad_norm": 28.5,
56
+ "learning_rate": 2.8000000000000003e-06,
57
+ "loss": 2.2084,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.0,
62
+ "grad_norm": 24.0,
63
+ "learning_rate": 3.2000000000000003e-06,
64
+ "loss": 2.1635,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.0,
69
+ "grad_norm": 31.5,
70
+ "learning_rate": 3.6000000000000003e-06,
71
+ "loss": 2.0963,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.0,
76
+ "grad_norm": 30.0,
77
+ "learning_rate": 4.000000000000001e-06,
78
+ "loss": 1.9403,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.0,
83
+ "grad_norm": 44.5,
84
+ "learning_rate": 4.4e-06,
85
+ "loss": 1.8556,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.0,
90
+ "grad_norm": 22.375,
91
+ "learning_rate": 4.800000000000001e-06,
92
+ "loss": 1.9425,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.0,
97
+ "grad_norm": 44.0,
98
+ "learning_rate": 5.2e-06,
99
+ "loss": 1.8286,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.0,
104
+ "grad_norm": 21.75,
105
+ "learning_rate": 5.600000000000001e-06,
106
+ "loss": 2.0117,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.0,
111
+ "grad_norm": 51.75,
112
+ "learning_rate": 6e-06,
113
+ "loss": 1.8744,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.0,
118
+ "grad_norm": 23.0,
119
+ "learning_rate": 6.4000000000000006e-06,
120
+ "loss": 1.8316,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.0,
125
+ "grad_norm": 25.25,
126
+ "learning_rate": 6.800000000000001e-06,
127
+ "loss": 1.63,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.0,
132
+ "grad_norm": 44.5,
133
+ "learning_rate": 7.2000000000000005e-06,
134
+ "loss": 1.7395,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.0,
139
+ "grad_norm": 20.375,
140
+ "learning_rate": 7.600000000000001e-06,
141
+ "loss": 1.8581,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.01,
146
+ "grad_norm": 44.75,
147
+ "learning_rate": 8.000000000000001e-06,
148
+ "loss": 1.7136,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.01,
153
+ "grad_norm": 59.75,
154
+ "learning_rate": 8.400000000000001e-06,
155
+ "loss": 1.5627,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.01,
160
+ "grad_norm": 33.25,
161
+ "learning_rate": 8.8e-06,
162
+ "loss": 1.6858,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.01,
167
+ "grad_norm": 22.0,
168
+ "learning_rate": 9.200000000000002e-06,
169
+ "loss": 1.9664,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.01,
174
+ "grad_norm": 18.625,
175
+ "learning_rate": 9.600000000000001e-06,
176
+ "loss": 1.5142,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.01,
181
+ "grad_norm": 15.5625,
182
+ "learning_rate": 1e-05,
183
+ "loss": 1.7143,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.01,
188
+ "grad_norm": 13.5625,
189
+ "learning_rate": 1.04e-05,
190
+ "loss": 1.8581,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.01,
195
+ "grad_norm": 21.125,
196
+ "learning_rate": 1.0800000000000002e-05,
197
+ "loss": 1.625,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.01,
202
+ "grad_norm": 15.75,
203
+ "learning_rate": 1.1200000000000001e-05,
204
+ "loss": 1.6071,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.01,
209
+ "grad_norm": 14.5,
210
+ "learning_rate": 1.16e-05,
211
+ "loss": 1.8191,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.01,
216
+ "grad_norm": 22.25,
217
+ "learning_rate": 1.2e-05,
218
+ "loss": 1.5947,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.01,
223
+ "grad_norm": 22.625,
224
+ "learning_rate": 1.2400000000000002e-05,
225
+ "loss": 1.6599,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.01,
230
+ "grad_norm": 88.5,
231
+ "learning_rate": 1.2800000000000001e-05,
232
+ "loss": 1.6836,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.01,
237
+ "grad_norm": 12.5625,
238
+ "learning_rate": 1.3200000000000002e-05,
239
+ "loss": 1.8363,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.01,
244
+ "grad_norm": 15.625,
245
+ "learning_rate": 1.3600000000000002e-05,
246
+ "loss": 1.723,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.01,
251
+ "grad_norm": 18.5,
252
+ "learning_rate": 1.4e-05,
253
+ "loss": 1.6806,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.01,
258
+ "grad_norm": 10.875,
259
+ "learning_rate": 1.4400000000000001e-05,
260
+ "loss": 1.618,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.01,
265
+ "grad_norm": 11.5,
266
+ "learning_rate": 1.48e-05,
267
+ "loss": 1.6253,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.01,
272
+ "grad_norm": 23.375,
273
+ "learning_rate": 1.5200000000000002e-05,
274
+ "loss": 1.7142,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.01,
279
+ "grad_norm": 9.9375,
280
+ "learning_rate": 1.5600000000000003e-05,
281
+ "loss": 1.6268,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.01,
286
+ "grad_norm": 15.375,
287
+ "learning_rate": 1.6000000000000003e-05,
288
+ "loss": 1.652,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.01,
293
+ "grad_norm": 29.375,
294
+ "learning_rate": 1.64e-05,
295
+ "loss": 1.6204,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.01,
300
+ "grad_norm": 33.75,
301
+ "learning_rate": 1.6800000000000002e-05,
302
+ "loss": 1.5077,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.01,
307
+ "grad_norm": 49.0,
308
+ "learning_rate": 1.72e-05,
309
+ "loss": 1.7802,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.01,
314
+ "grad_norm": 13.8125,
315
+ "learning_rate": 1.76e-05,
316
+ "loss": 1.4695,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.01,
321
+ "grad_norm": 25.75,
322
+ "learning_rate": 1.8e-05,
323
+ "loss": 1.6256,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.01,
328
+ "grad_norm": 16.25,
329
+ "learning_rate": 1.8400000000000003e-05,
330
+ "loss": 1.6964,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.01,
335
+ "grad_norm": 20.375,
336
+ "learning_rate": 1.88e-05,
337
+ "loss": 1.7083,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.01,
342
+ "grad_norm": 22.625,
343
+ "learning_rate": 1.9200000000000003e-05,
344
+ "loss": 1.6634,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.01,
349
+ "grad_norm": 14.4375,
350
+ "learning_rate": 1.9600000000000002e-05,
351
+ "loss": 1.5189,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.01,
356
+ "grad_norm": 49.25,
357
+ "learning_rate": 2e-05,
358
+ "loss": 1.6792,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.01,
363
+ "grad_norm": 28.125,
364
+ "learning_rate": 1.999322033898305e-05,
365
+ "loss": 1.6381,
366
+ "step": 510
367
+ },
368
+ {
369
+ "epoch": 0.01,
370
+ "grad_norm": 13.4375,
371
+ "learning_rate": 1.9986440677966104e-05,
372
+ "loss": 1.5793,
373
+ "step": 520
374
+ },
375
+ {
376
+ "epoch": 0.01,
377
+ "grad_norm": 20.25,
378
+ "learning_rate": 1.9979661016949154e-05,
379
+ "loss": 1.5979,
380
+ "step": 530
381
+ },
382
+ {
383
+ "epoch": 0.01,
384
+ "grad_norm": 10.875,
385
+ "learning_rate": 1.9972881355932204e-05,
386
+ "loss": 1.7109,
387
+ "step": 540
388
+ },
389
+ {
390
+ "epoch": 0.01,
391
+ "grad_norm": 19.75,
392
+ "learning_rate": 1.9966101694915257e-05,
393
+ "loss": 1.5867,
394
+ "step": 550
395
+ },
396
+ {
397
+ "epoch": 0.01,
398
+ "grad_norm": 25.375,
399
+ "learning_rate": 1.9959322033898307e-05,
400
+ "loss": 1.6398,
401
+ "step": 560
402
+ },
403
+ {
404
+ "epoch": 0.01,
405
+ "grad_norm": 19.5,
406
+ "learning_rate": 1.995254237288136e-05,
407
+ "loss": 1.4327,
408
+ "step": 570
409
+ },
410
+ {
411
+ "epoch": 0.01,
412
+ "grad_norm": 11.0625,
413
+ "learning_rate": 1.994576271186441e-05,
414
+ "loss": 1.5478,
415
+ "step": 580
416
+ },
417
+ {
418
+ "epoch": 0.01,
419
+ "grad_norm": 23.625,
420
+ "learning_rate": 1.993898305084746e-05,
421
+ "loss": 1.8115,
422
+ "step": 590
423
+ },
424
+ {
425
+ "epoch": 0.01,
426
+ "grad_norm": 11.625,
427
+ "learning_rate": 1.9932203389830512e-05,
428
+ "loss": 1.6309,
429
+ "step": 600
430
+ },
431
+ {
432
+ "epoch": 0.02,
433
+ "grad_norm": 10.4375,
434
+ "learning_rate": 1.992542372881356e-05,
435
+ "loss": 1.5757,
436
+ "step": 610
437
+ },
438
+ {
439
+ "epoch": 0.02,
440
+ "grad_norm": 11.375,
441
+ "learning_rate": 1.991864406779661e-05,
442
+ "loss": 1.6906,
443
+ "step": 620
444
+ },
445
+ {
446
+ "epoch": 0.02,
447
+ "grad_norm": 12.625,
448
+ "learning_rate": 1.991186440677966e-05,
449
+ "loss": 1.6413,
450
+ "step": 630
451
+ },
452
+ {
453
+ "epoch": 0.02,
454
+ "grad_norm": 17.125,
455
+ "learning_rate": 1.990508474576271e-05,
456
+ "loss": 1.6699,
457
+ "step": 640
458
+ },
459
+ {
460
+ "epoch": 0.02,
461
+ "grad_norm": 16.875,
462
+ "learning_rate": 1.9898305084745764e-05,
463
+ "loss": 1.5938,
464
+ "step": 650
465
+ },
466
+ {
467
+ "epoch": 0.02,
468
+ "grad_norm": 12.75,
469
+ "learning_rate": 1.9891525423728814e-05,
470
+ "loss": 1.8807,
471
+ "step": 660
472
+ },
473
+ {
474
+ "epoch": 0.02,
475
+ "grad_norm": 13.5,
476
+ "learning_rate": 1.9884745762711867e-05,
477
+ "loss": 1.5723,
478
+ "step": 670
479
+ },
480
+ {
481
+ "epoch": 0.02,
482
+ "grad_norm": 10.5,
483
+ "learning_rate": 1.9877966101694917e-05,
484
+ "loss": 1.4536,
485
+ "step": 680
486
+ },
487
+ {
488
+ "epoch": 0.02,
489
+ "grad_norm": 17.0,
490
+ "learning_rate": 1.9871186440677966e-05,
491
+ "loss": 1.6836,
492
+ "step": 690
493
+ },
494
+ {
495
+ "epoch": 0.02,
496
+ "grad_norm": 18.125,
497
+ "learning_rate": 1.986440677966102e-05,
498
+ "loss": 1.5296,
499
+ "step": 700
500
+ },
501
+ {
502
+ "epoch": 0.02,
503
+ "grad_norm": 15.375,
504
+ "learning_rate": 1.985762711864407e-05,
505
+ "loss": 1.6754,
506
+ "step": 710
507
+ },
508
+ {
509
+ "epoch": 0.02,
510
+ "grad_norm": 17.625,
511
+ "learning_rate": 1.985084745762712e-05,
512
+ "loss": 1.5548,
513
+ "step": 720
514
+ },
515
+ {
516
+ "epoch": 0.02,
517
+ "grad_norm": 8.875,
518
+ "learning_rate": 1.9844067796610172e-05,
519
+ "loss": 1.6737,
520
+ "step": 730
521
+ },
522
+ {
523
+ "epoch": 0.02,
524
+ "grad_norm": 13.0,
525
+ "learning_rate": 1.9837288135593222e-05,
526
+ "loss": 1.5516,
527
+ "step": 740
528
+ },
529
+ {
530
+ "epoch": 0.02,
531
+ "grad_norm": 13.875,
532
+ "learning_rate": 1.9830508474576275e-05,
533
+ "loss": 1.6862,
534
+ "step": 750
535
+ },
536
+ {
537
+ "epoch": 0.02,
538
+ "grad_norm": 8.8125,
539
+ "learning_rate": 1.9823728813559324e-05,
540
+ "loss": 1.6523,
541
+ "step": 760
542
+ },
543
+ {
544
+ "epoch": 0.02,
545
+ "grad_norm": 10.5625,
546
+ "learning_rate": 1.9816949152542374e-05,
547
+ "loss": 1.6752,
548
+ "step": 770
549
+ },
550
+ {
551
+ "epoch": 0.02,
552
+ "grad_norm": 11.0625,
553
+ "learning_rate": 1.9810169491525427e-05,
554
+ "loss": 1.565,
555
+ "step": 780
556
+ },
557
+ {
558
+ "epoch": 0.02,
559
+ "grad_norm": 11.375,
560
+ "learning_rate": 1.9803389830508477e-05,
561
+ "loss": 1.447,
562
+ "step": 790
563
+ },
564
+ {
565
+ "epoch": 0.02,
566
+ "grad_norm": 16.875,
567
+ "learning_rate": 1.9796610169491527e-05,
568
+ "loss": 1.618,
569
+ "step": 800
570
+ },
571
+ {
572
+ "epoch": 0.02,
573
+ "grad_norm": 12.375,
574
+ "learning_rate": 1.978983050847458e-05,
575
+ "loss": 1.5672,
576
+ "step": 810
577
+ },
578
+ {
579
+ "epoch": 0.02,
580
+ "grad_norm": 8.3125,
581
+ "learning_rate": 1.9783050847457626e-05,
582
+ "loss": 1.5638,
583
+ "step": 820
584
+ },
585
+ {
586
+ "epoch": 0.02,
587
+ "grad_norm": 10.0625,
588
+ "learning_rate": 1.977627118644068e-05,
589
+ "loss": 1.6477,
590
+ "step": 830
591
+ },
592
+ {
593
+ "epoch": 0.02,
594
+ "grad_norm": 27.5,
595
+ "learning_rate": 1.976949152542373e-05,
596
+ "loss": 1.5532,
597
+ "step": 840
598
+ },
599
+ {
600
+ "epoch": 0.02,
601
+ "grad_norm": 17.625,
602
+ "learning_rate": 1.9762711864406782e-05,
603
+ "loss": 1.6517,
604
+ "step": 850
605
+ },
606
+ {
607
+ "epoch": 0.02,
608
+ "grad_norm": 9.75,
609
+ "learning_rate": 1.9755932203389832e-05,
610
+ "loss": 1.663,
611
+ "step": 860
612
+ },
613
+ {
614
+ "epoch": 0.02,
615
+ "grad_norm": 8.3125,
616
+ "learning_rate": 1.974915254237288e-05,
617
+ "loss": 1.6119,
618
+ "step": 870
619
+ },
620
+ {
621
+ "epoch": 0.02,
622
+ "grad_norm": 19.625,
623
+ "learning_rate": 1.9742372881355935e-05,
624
+ "loss": 1.5863,
625
+ "step": 880
626
+ },
627
+ {
628
+ "epoch": 0.02,
629
+ "grad_norm": 9.875,
630
+ "learning_rate": 1.9735593220338984e-05,
631
+ "loss": 1.6251,
632
+ "step": 890
633
+ },
634
+ {
635
+ "epoch": 0.02,
636
+ "grad_norm": 13.875,
637
+ "learning_rate": 1.9728813559322034e-05,
638
+ "loss": 1.6313,
639
+ "step": 900
640
+ },
641
+ {
642
+ "epoch": 0.02,
643
+ "grad_norm": 9.125,
644
+ "learning_rate": 1.9722033898305087e-05,
645
+ "loss": 1.5873,
646
+ "step": 910
647
+ },
648
+ {
649
+ "epoch": 0.02,
650
+ "grad_norm": 25.375,
651
+ "learning_rate": 1.9715254237288137e-05,
652
+ "loss": 1.6079,
653
+ "step": 920
654
+ },
655
+ {
656
+ "epoch": 0.02,
657
+ "grad_norm": 24.25,
658
+ "learning_rate": 1.970847457627119e-05,
659
+ "loss": 1.4977,
660
+ "step": 930
661
+ },
662
+ {
663
+ "epoch": 0.02,
664
+ "grad_norm": 9.3125,
665
+ "learning_rate": 1.970169491525424e-05,
666
+ "loss": 1.6284,
667
+ "step": 940
668
+ },
669
+ {
670
+ "epoch": 0.02,
671
+ "grad_norm": 10.375,
672
+ "learning_rate": 1.969491525423729e-05,
673
+ "loss": 1.6224,
674
+ "step": 950
675
+ },
676
+ {
677
+ "epoch": 0.02,
678
+ "grad_norm": 12.6875,
679
+ "learning_rate": 1.9688135593220342e-05,
680
+ "loss": 1.5419,
681
+ "step": 960
682
+ },
683
+ {
684
+ "epoch": 0.02,
685
+ "grad_norm": 15.0,
686
+ "learning_rate": 1.9681355932203392e-05,
687
+ "loss": 1.5991,
688
+ "step": 970
689
+ },
690
+ {
691
+ "epoch": 0.02,
692
+ "grad_norm": 9.0625,
693
+ "learning_rate": 1.9674576271186442e-05,
694
+ "loss": 1.583,
695
+ "step": 980
696
+ },
697
+ {
698
+ "epoch": 0.02,
699
+ "grad_norm": 16.625,
700
+ "learning_rate": 1.9667796610169495e-05,
701
+ "loss": 1.5774,
702
+ "step": 990
703
+ },
704
+ {
705
+ "epoch": 0.03,
706
+ "grad_norm": 14.625,
707
+ "learning_rate": 1.9661016949152545e-05,
708
+ "loss": 1.7158,
709
+ "step": 1000
710
+ },
711
+ {
712
+ "epoch": 0.03,
713
+ "grad_norm": 15.0,
714
+ "learning_rate": 1.9654237288135594e-05,
715
+ "loss": 1.6369,
716
+ "step": 1010
717
+ },
718
+ {
719
+ "epoch": 0.03,
720
+ "grad_norm": 9.0625,
721
+ "learning_rate": 1.9647457627118644e-05,
722
+ "loss": 1.6978,
723
+ "step": 1020
724
+ },
725
+ {
726
+ "epoch": 0.03,
727
+ "grad_norm": 9.3125,
728
+ "learning_rate": 1.9640677966101697e-05,
729
+ "loss": 1.6175,
730
+ "step": 1030
731
+ },
732
+ {
733
+ "epoch": 0.03,
734
+ "grad_norm": 11.1875,
735
+ "learning_rate": 1.9633898305084747e-05,
736
+ "loss": 1.6845,
737
+ "step": 1040
738
+ },
739
+ {
740
+ "epoch": 0.03,
741
+ "grad_norm": 14.1875,
742
+ "learning_rate": 1.9627118644067796e-05,
743
+ "loss": 1.5605,
744
+ "step": 1050
745
+ },
746
+ {
747
+ "epoch": 0.03,
748
+ "grad_norm": 11.1875,
749
+ "learning_rate": 1.962033898305085e-05,
750
+ "loss": 1.734,
751
+ "step": 1060
752
+ },
753
+ {
754
+ "epoch": 0.03,
755
+ "grad_norm": 11.8125,
756
+ "learning_rate": 1.96135593220339e-05,
757
+ "loss": 1.4532,
758
+ "step": 1070
759
+ },
760
+ {
761
+ "epoch": 0.03,
762
+ "grad_norm": 42.0,
763
+ "learning_rate": 1.960677966101695e-05,
764
+ "loss": 1.7891,
765
+ "step": 1080
766
+ },
767
+ {
768
+ "epoch": 0.03,
769
+ "grad_norm": 18.75,
770
+ "learning_rate": 1.9600000000000002e-05,
771
+ "loss": 1.6711,
772
+ "step": 1090
773
+ },
774
+ {
775
+ "epoch": 0.03,
776
+ "grad_norm": 13.4375,
777
+ "learning_rate": 1.9593220338983052e-05,
778
+ "loss": 1.7482,
779
+ "step": 1100
780
+ },
781
+ {
782
+ "epoch": 0.03,
783
+ "grad_norm": 16.25,
784
+ "learning_rate": 1.95864406779661e-05,
785
+ "loss": 1.7121,
786
+ "step": 1110
787
+ },
788
+ {
789
+ "epoch": 0.03,
790
+ "grad_norm": 12.5,
791
+ "learning_rate": 1.9579661016949155e-05,
792
+ "loss": 1.5464,
793
+ "step": 1120
794
+ },
795
+ {
796
+ "epoch": 0.03,
797
+ "grad_norm": 18.375,
798
+ "learning_rate": 1.9572881355932204e-05,
799
+ "loss": 1.5346,
800
+ "step": 1130
801
+ },
802
+ {
803
+ "epoch": 0.03,
804
+ "grad_norm": 13.4375,
805
+ "learning_rate": 1.9566101694915257e-05,
806
+ "loss": 1.5374,
807
+ "step": 1140
808
+ },
809
+ {
810
+ "epoch": 0.03,
811
+ "grad_norm": 12.875,
812
+ "learning_rate": 1.9559322033898307e-05,
813
+ "loss": 1.6641,
814
+ "step": 1150
815
+ },
816
+ {
817
+ "epoch": 0.03,
818
+ "grad_norm": 10.25,
819
+ "learning_rate": 1.9552542372881357e-05,
820
+ "loss": 1.5511,
821
+ "step": 1160
822
+ },
823
+ {
824
+ "epoch": 0.03,
825
+ "grad_norm": 8.75,
826
+ "learning_rate": 1.954576271186441e-05,
827
+ "loss": 1.5672,
828
+ "step": 1170
829
+ },
830
+ {
831
+ "epoch": 0.03,
832
+ "grad_norm": 29.125,
833
+ "learning_rate": 1.953898305084746e-05,
834
+ "loss": 1.4583,
835
+ "step": 1180
836
+ },
837
+ {
838
+ "epoch": 0.03,
839
+ "grad_norm": 16.5,
840
+ "learning_rate": 1.953220338983051e-05,
841
+ "loss": 1.6112,
842
+ "step": 1190
843
+ },
844
+ {
845
+ "epoch": 0.03,
846
+ "grad_norm": 11.0,
847
+ "learning_rate": 1.9525423728813562e-05,
848
+ "loss": 1.6229,
849
+ "step": 1200
850
+ },
851
+ {
852
+ "epoch": 0.03,
853
+ "grad_norm": 6.6875,
854
+ "learning_rate": 1.9518644067796612e-05,
855
+ "loss": 1.5782,
856
+ "step": 1210
857
+ },
858
+ {
859
+ "epoch": 0.03,
860
+ "grad_norm": 12.9375,
861
+ "learning_rate": 1.9511864406779665e-05,
862
+ "loss": 1.5294,
863
+ "step": 1220
864
+ },
865
+ {
866
+ "epoch": 0.03,
867
+ "grad_norm": 6.96875,
868
+ "learning_rate": 1.950508474576271e-05,
869
+ "loss": 1.5371,
870
+ "step": 1230
871
+ },
872
+ {
873
+ "epoch": 0.03,
874
+ "grad_norm": 18.5,
875
+ "learning_rate": 1.9498305084745765e-05,
876
+ "loss": 1.7215,
877
+ "step": 1240
878
+ },
879
+ {
880
+ "epoch": 0.03,
881
+ "grad_norm": 17.75,
882
+ "learning_rate": 1.9491525423728814e-05,
883
+ "loss": 1.7009,
884
+ "step": 1250
885
+ },
886
+ {
887
+ "epoch": 0.03,
888
+ "grad_norm": 28.25,
889
+ "learning_rate": 1.9484745762711864e-05,
890
+ "loss": 1.4353,
891
+ "step": 1260
892
+ },
893
+ {
894
+ "epoch": 0.03,
895
+ "grad_norm": 11.0625,
896
+ "learning_rate": 1.9477966101694917e-05,
897
+ "loss": 1.5369,
898
+ "step": 1270
899
+ },
900
+ {
901
+ "epoch": 0.03,
902
+ "grad_norm": 13.25,
903
+ "learning_rate": 1.9471186440677967e-05,
904
+ "loss": 1.3927,
905
+ "step": 1280
906
+ },
907
+ {
908
+ "epoch": 0.03,
909
+ "grad_norm": 25.375,
910
+ "learning_rate": 1.9464406779661017e-05,
911
+ "loss": 1.7541,
912
+ "step": 1290
913
+ },
914
+ {
915
+ "epoch": 0.03,
916
+ "grad_norm": 15.3125,
917
+ "learning_rate": 1.945762711864407e-05,
918
+ "loss": 1.5196,
919
+ "step": 1300
920
+ },
921
+ {
922
+ "epoch": 0.03,
923
+ "grad_norm": 10.125,
924
+ "learning_rate": 1.945084745762712e-05,
925
+ "loss": 1.545,
926
+ "step": 1310
927
+ },
928
+ {
929
+ "epoch": 0.03,
930
+ "grad_norm": 10.3125,
931
+ "learning_rate": 1.9444067796610172e-05,
932
+ "loss": 1.547,
933
+ "step": 1320
934
+ },
935
+ {
936
+ "epoch": 0.03,
937
+ "grad_norm": 11.4375,
938
+ "learning_rate": 1.9437288135593222e-05,
939
+ "loss": 1.4786,
940
+ "step": 1330
941
+ },
942
+ {
943
+ "epoch": 0.03,
944
+ "grad_norm": 6.5,
945
+ "learning_rate": 1.9430508474576272e-05,
946
+ "loss": 1.6925,
947
+ "step": 1340
948
+ },
949
+ {
950
+ "epoch": 0.03,
951
+ "grad_norm": 10.4375,
952
+ "learning_rate": 1.9423728813559325e-05,
953
+ "loss": 1.3977,
954
+ "step": 1350
955
+ },
956
+ {
957
+ "epoch": 0.03,
958
+ "grad_norm": 11.0625,
959
+ "learning_rate": 1.9416949152542375e-05,
960
+ "loss": 1.5348,
961
+ "step": 1360
962
+ },
963
+ {
964
+ "epoch": 0.03,
965
+ "grad_norm": 10.875,
966
+ "learning_rate": 1.9410169491525424e-05,
967
+ "loss": 1.4933,
968
+ "step": 1370
969
+ },
970
+ {
971
+ "epoch": 0.03,
972
+ "grad_norm": 10.1875,
973
+ "learning_rate": 1.9403389830508477e-05,
974
+ "loss": 1.5145,
975
+ "step": 1380
976
+ },
977
+ {
978
+ "epoch": 0.03,
979
+ "grad_norm": 18.0,
980
+ "learning_rate": 1.9396610169491527e-05,
981
+ "loss": 1.3786,
982
+ "step": 1390
983
+ },
984
+ {
985
+ "epoch": 0.04,
986
+ "grad_norm": 28.25,
987
+ "learning_rate": 1.938983050847458e-05,
988
+ "loss": 1.7004,
989
+ "step": 1400
990
+ },
991
+ {
992
+ "epoch": 0.04,
993
+ "grad_norm": 24.5,
994
+ "learning_rate": 1.938305084745763e-05,
995
+ "loss": 1.5797,
996
+ "step": 1410
997
+ },
998
+ {
999
+ "epoch": 0.04,
1000
+ "grad_norm": 15.5,
1001
+ "learning_rate": 1.937627118644068e-05,
1002
+ "loss": 1.7938,
1003
+ "step": 1420
1004
+ },
1005
+ {
1006
+ "epoch": 0.04,
1007
+ "grad_norm": 10.875,
1008
+ "learning_rate": 1.9369491525423733e-05,
1009
+ "loss": 1.4801,
1010
+ "step": 1430
1011
+ },
1012
+ {
1013
+ "epoch": 0.04,
1014
+ "grad_norm": 20.75,
1015
+ "learning_rate": 1.936271186440678e-05,
1016
+ "loss": 1.5033,
1017
+ "step": 1440
1018
+ },
1019
+ {
1020
+ "epoch": 0.04,
1021
+ "grad_norm": 16.125,
1022
+ "learning_rate": 1.9355932203389832e-05,
1023
+ "loss": 1.4213,
1024
+ "step": 1450
1025
+ },
1026
+ {
1027
+ "epoch": 0.04,
1028
+ "grad_norm": 20.25,
1029
+ "learning_rate": 1.9349152542372882e-05,
1030
+ "loss": 1.5956,
1031
+ "step": 1460
1032
+ },
1033
+ {
1034
+ "epoch": 0.04,
1035
+ "grad_norm": 13.5625,
1036
+ "learning_rate": 1.934237288135593e-05,
1037
+ "loss": 1.7034,
1038
+ "step": 1470
1039
+ },
1040
+ {
1041
+ "epoch": 0.04,
1042
+ "grad_norm": 12.5,
1043
+ "learning_rate": 1.9335593220338985e-05,
1044
+ "loss": 1.5511,
1045
+ "step": 1480
1046
+ },
1047
+ {
1048
+ "epoch": 0.04,
1049
+ "grad_norm": 10.5,
1050
+ "learning_rate": 1.9328813559322034e-05,
1051
+ "loss": 1.4598,
1052
+ "step": 1490
1053
+ },
1054
+ {
1055
+ "epoch": 0.04,
1056
+ "grad_norm": 10.75,
1057
+ "learning_rate": 1.9322033898305087e-05,
1058
+ "loss": 1.5557,
1059
+ "step": 1500
1060
+ },
1061
+ {
1062
+ "epoch": 0.04,
1063
+ "grad_norm": 9.6875,
1064
+ "learning_rate": 1.9315254237288137e-05,
1065
+ "loss": 1.4942,
1066
+ "step": 1510
1067
+ },
1068
+ {
1069
+ "epoch": 0.04,
1070
+ "grad_norm": 35.5,
1071
+ "learning_rate": 1.9308474576271187e-05,
1072
+ "loss": 1.6278,
1073
+ "step": 1520
1074
+ },
1075
+ {
1076
+ "epoch": 0.04,
1077
+ "grad_norm": 7.96875,
1078
+ "learning_rate": 1.930169491525424e-05,
1079
+ "loss": 1.5048,
1080
+ "step": 1530
1081
+ },
1082
+ {
1083
+ "epoch": 0.04,
1084
+ "grad_norm": 11.25,
1085
+ "learning_rate": 1.929491525423729e-05,
1086
+ "loss": 1.515,
1087
+ "step": 1540
1088
+ },
1089
+ {
1090
+ "epoch": 0.04,
1091
+ "grad_norm": 7.875,
1092
+ "learning_rate": 1.928813559322034e-05,
1093
+ "loss": 1.6476,
1094
+ "step": 1550
1095
+ },
1096
+ {
1097
+ "epoch": 0.04,
1098
+ "grad_norm": 14.5625,
1099
+ "learning_rate": 1.9281355932203392e-05,
1100
+ "loss": 1.4581,
1101
+ "step": 1560
1102
+ },
1103
+ {
1104
+ "epoch": 0.04,
1105
+ "grad_norm": 8.0625,
1106
+ "learning_rate": 1.9274576271186442e-05,
1107
+ "loss": 1.5717,
1108
+ "step": 1570
1109
+ },
1110
+ {
1111
+ "epoch": 0.04,
1112
+ "grad_norm": 11.3125,
1113
+ "learning_rate": 1.9267796610169492e-05,
1114
+ "loss": 1.4752,
1115
+ "step": 1580
1116
+ },
1117
+ {
1118
+ "epoch": 0.04,
1119
+ "grad_norm": 12.875,
1120
+ "learning_rate": 1.9261016949152545e-05,
1121
+ "loss": 1.5378,
1122
+ "step": 1590
1123
+ },
1124
+ {
1125
+ "epoch": 0.04,
1126
+ "grad_norm": 23.125,
1127
+ "learning_rate": 1.9254237288135595e-05,
1128
+ "loss": 1.608,
1129
+ "step": 1600
1130
+ },
1131
+ {
1132
+ "epoch": 0.04,
1133
+ "grad_norm": 21.75,
1134
+ "learning_rate": 1.9247457627118648e-05,
1135
+ "loss": 1.4953,
1136
+ "step": 1610
1137
+ },
1138
+ {
1139
+ "epoch": 0.04,
1140
+ "grad_norm": 19.875,
1141
+ "learning_rate": 1.9240677966101698e-05,
1142
+ "loss": 1.4522,
1143
+ "step": 1620
1144
+ },
1145
+ {
1146
+ "epoch": 0.04,
1147
+ "grad_norm": 14.9375,
1148
+ "learning_rate": 1.9233898305084747e-05,
1149
+ "loss": 1.6297,
1150
+ "step": 1630
1151
+ },
1152
+ {
1153
+ "epoch": 0.04,
1154
+ "grad_norm": 11.6875,
1155
+ "learning_rate": 1.92271186440678e-05,
1156
+ "loss": 1.698,
1157
+ "step": 1640
1158
+ },
1159
+ {
1160
+ "epoch": 0.04,
1161
+ "grad_norm": 10.625,
1162
+ "learning_rate": 1.9220338983050847e-05,
1163
+ "loss": 1.5721,
1164
+ "step": 1650
1165
+ },
1166
+ {
1167
+ "epoch": 0.04,
1168
+ "grad_norm": 9.0,
1169
+ "learning_rate": 1.92135593220339e-05,
1170
+ "loss": 1.487,
1171
+ "step": 1660
1172
+ },
1173
+ {
1174
+ "epoch": 0.04,
1175
+ "grad_norm": 10.625,
1176
+ "learning_rate": 1.920677966101695e-05,
1177
+ "loss": 1.5897,
1178
+ "step": 1670
1179
+ },
1180
+ {
1181
+ "epoch": 0.04,
1182
+ "grad_norm": 15.25,
1183
+ "learning_rate": 1.9200000000000003e-05,
1184
+ "loss": 1.6502,
1185
+ "step": 1680
1186
+ },
1187
+ {
1188
+ "epoch": 0.04,
1189
+ "grad_norm": 24.625,
1190
+ "learning_rate": 1.9193220338983052e-05,
1191
+ "loss": 1.6367,
1192
+ "step": 1690
1193
+ },
1194
+ {
1195
+ "epoch": 0.04,
1196
+ "grad_norm": 8.3125,
1197
+ "learning_rate": 1.9186440677966102e-05,
1198
+ "loss": 1.4932,
1199
+ "step": 1700
1200
+ },
1201
+ {
1202
+ "epoch": 0.04,
1203
+ "grad_norm": 14.8125,
1204
+ "learning_rate": 1.9179661016949155e-05,
1205
+ "loss": 1.6045,
1206
+ "step": 1710
1207
+ },
1208
+ {
1209
+ "epoch": 0.04,
1210
+ "grad_norm": 12.0,
1211
+ "learning_rate": 1.9172881355932205e-05,
1212
+ "loss": 1.6201,
1213
+ "step": 1720
1214
+ },
1215
+ {
1216
+ "epoch": 0.04,
1217
+ "grad_norm": 9.3125,
1218
+ "learning_rate": 1.9166101694915254e-05,
1219
+ "loss": 1.395,
1220
+ "step": 1730
1221
+ },
1222
+ {
1223
+ "epoch": 0.04,
1224
+ "grad_norm": 14.0,
1225
+ "learning_rate": 1.9159322033898308e-05,
1226
+ "loss": 1.4311,
1227
+ "step": 1740
1228
+ },
1229
+ {
1230
+ "epoch": 0.04,
1231
+ "grad_norm": 10.875,
1232
+ "learning_rate": 1.9152542372881357e-05,
1233
+ "loss": 1.6827,
1234
+ "step": 1750
1235
+ },
1236
+ {
1237
+ "epoch": 0.04,
1238
+ "grad_norm": 10.25,
1239
+ "learning_rate": 1.9145762711864407e-05,
1240
+ "loss": 1.6859,
1241
+ "step": 1760
1242
+ },
1243
+ {
1244
+ "epoch": 0.04,
1245
+ "grad_norm": 16.125,
1246
+ "learning_rate": 1.913898305084746e-05,
1247
+ "loss": 1.5337,
1248
+ "step": 1770
1249
+ },
1250
+ {
1251
+ "epoch": 0.04,
1252
+ "grad_norm": 8.8125,
1253
+ "learning_rate": 1.913220338983051e-05,
1254
+ "loss": 1.7639,
1255
+ "step": 1780
1256
+ },
1257
+ {
1258
+ "epoch": 0.04,
1259
+ "grad_norm": 11.1875,
1260
+ "learning_rate": 1.9125423728813563e-05,
1261
+ "loss": 1.508,
1262
+ "step": 1790
1263
+ },
1264
+ {
1265
+ "epoch": 0.04,
1266
+ "grad_norm": 9.5625,
1267
+ "learning_rate": 1.9118644067796613e-05,
1268
+ "loss": 1.5,
1269
+ "step": 1800
1270
+ },
1271
+ {
1272
+ "epoch": 0.05,
1273
+ "grad_norm": 11.875,
1274
+ "learning_rate": 1.9111864406779662e-05,
1275
+ "loss": 1.6331,
1276
+ "step": 1810
1277
+ },
1278
+ {
1279
+ "epoch": 0.05,
1280
+ "grad_norm": 9.875,
1281
+ "learning_rate": 1.9105084745762715e-05,
1282
+ "loss": 1.3836,
1283
+ "step": 1820
1284
+ },
1285
+ {
1286
+ "epoch": 0.05,
1287
+ "grad_norm": 11.0625,
1288
+ "learning_rate": 1.9098305084745765e-05,
1289
+ "loss": 1.5407,
1290
+ "step": 1830
1291
+ },
1292
+ {
1293
+ "epoch": 0.05,
1294
+ "grad_norm": 12.25,
1295
+ "learning_rate": 1.9091525423728815e-05,
1296
+ "loss": 1.5822,
1297
+ "step": 1840
1298
+ },
1299
+ {
1300
+ "epoch": 0.05,
1301
+ "grad_norm": 10.125,
1302
+ "learning_rate": 1.9084745762711868e-05,
1303
+ "loss": 1.5684,
1304
+ "step": 1850
1305
+ },
1306
+ {
1307
+ "epoch": 0.05,
1308
+ "grad_norm": 13.875,
1309
+ "learning_rate": 1.9077966101694914e-05,
1310
+ "loss": 1.5036,
1311
+ "step": 1860
1312
+ },
1313
+ {
1314
+ "epoch": 0.05,
1315
+ "grad_norm": 15.0625,
1316
+ "learning_rate": 1.9071186440677967e-05,
1317
+ "loss": 1.5305,
1318
+ "step": 1870
1319
+ },
1320
+ {
1321
+ "epoch": 0.05,
1322
+ "grad_norm": 33.5,
1323
+ "learning_rate": 1.9064406779661017e-05,
1324
+ "loss": 1.4381,
1325
+ "step": 1880
1326
+ },
1327
+ {
1328
+ "epoch": 0.05,
1329
+ "grad_norm": 13.625,
1330
+ "learning_rate": 1.905762711864407e-05,
1331
+ "loss": 1.5901,
1332
+ "step": 1890
1333
+ },
1334
+ {
1335
+ "epoch": 0.05,
1336
+ "grad_norm": 16.625,
1337
+ "learning_rate": 1.905084745762712e-05,
1338
+ "loss": 1.4541,
1339
+ "step": 1900
1340
+ },
1341
+ {
1342
+ "epoch": 0.05,
1343
+ "grad_norm": 8.625,
1344
+ "learning_rate": 1.904406779661017e-05,
1345
+ "loss": 1.5044,
1346
+ "step": 1910
1347
+ },
1348
+ {
1349
+ "epoch": 0.05,
1350
+ "grad_norm": 15.75,
1351
+ "learning_rate": 1.9037288135593223e-05,
1352
+ "loss": 1.5519,
1353
+ "step": 1920
1354
+ },
1355
+ {
1356
+ "epoch": 0.05,
1357
+ "grad_norm": 8.0625,
1358
+ "learning_rate": 1.9030508474576272e-05,
1359
+ "loss": 1.5799,
1360
+ "step": 1930
1361
+ },
1362
+ {
1363
+ "epoch": 0.05,
1364
+ "grad_norm": 9.625,
1365
+ "learning_rate": 1.9023728813559322e-05,
1366
+ "loss": 1.5128,
1367
+ "step": 1940
1368
+ },
1369
+ {
1370
+ "epoch": 0.05,
1371
+ "grad_norm": 19.125,
1372
+ "learning_rate": 1.9016949152542375e-05,
1373
+ "loss": 1.5036,
1374
+ "step": 1950
1375
+ },
1376
+ {
1377
+ "epoch": 0.05,
1378
+ "grad_norm": 11.75,
1379
+ "learning_rate": 1.9010169491525425e-05,
1380
+ "loss": 1.4,
1381
+ "step": 1960
1382
+ },
1383
+ {
1384
+ "epoch": 0.05,
1385
+ "grad_norm": 8.625,
1386
+ "learning_rate": 1.9003389830508478e-05,
1387
+ "loss": 1.3903,
1388
+ "step": 1970
1389
+ },
1390
+ {
1391
+ "epoch": 0.05,
1392
+ "grad_norm": 10.6875,
1393
+ "learning_rate": 1.8996610169491528e-05,
1394
+ "loss": 1.5759,
1395
+ "step": 1980
1396
+ },
1397
+ {
1398
+ "epoch": 0.05,
1399
+ "grad_norm": 12.0625,
1400
+ "learning_rate": 1.8989830508474577e-05,
1401
+ "loss": 1.6143,
1402
+ "step": 1990
1403
+ },
1404
+ {
1405
+ "epoch": 0.05,
1406
+ "grad_norm": 43.5,
1407
+ "learning_rate": 1.898305084745763e-05,
1408
+ "loss": 1.5139,
1409
+ "step": 2000
1410
+ },
1411
+ {
1412
+ "epoch": 0.05,
1413
+ "grad_norm": 24.625,
1414
+ "learning_rate": 1.897627118644068e-05,
1415
+ "loss": 1.6689,
1416
+ "step": 2010
1417
+ },
1418
+ {
1419
+ "epoch": 0.05,
1420
+ "grad_norm": 26.5,
1421
+ "learning_rate": 1.896949152542373e-05,
1422
+ "loss": 1.6536,
1423
+ "step": 2020
1424
+ },
1425
+ {
1426
+ "epoch": 0.05,
1427
+ "grad_norm": 10.875,
1428
+ "learning_rate": 1.8962711864406783e-05,
1429
+ "loss": 1.5775,
1430
+ "step": 2030
1431
+ },
1432
+ {
1433
+ "epoch": 0.05,
1434
+ "grad_norm": 15.5,
1435
+ "learning_rate": 1.8955932203389833e-05,
1436
+ "loss": 1.5513,
1437
+ "step": 2040
1438
+ },
1439
+ {
1440
+ "epoch": 0.05,
1441
+ "grad_norm": 13.5,
1442
+ "learning_rate": 1.8949152542372882e-05,
1443
+ "loss": 1.5543,
1444
+ "step": 2050
1445
+ },
1446
+ {
1447
+ "epoch": 0.05,
1448
+ "grad_norm": 11.5625,
1449
+ "learning_rate": 1.8942372881355932e-05,
1450
+ "loss": 1.6387,
1451
+ "step": 2060
1452
+ },
1453
+ {
1454
+ "epoch": 0.05,
1455
+ "grad_norm": 10.25,
1456
+ "learning_rate": 1.8935593220338985e-05,
1457
+ "loss": 1.5928,
1458
+ "step": 2070
1459
+ },
1460
+ {
1461
+ "epoch": 0.05,
1462
+ "grad_norm": 9.5625,
1463
+ "learning_rate": 1.8928813559322035e-05,
1464
+ "loss": 1.5242,
1465
+ "step": 2080
1466
+ },
1467
+ {
1468
+ "epoch": 0.05,
1469
+ "grad_norm": 8.5,
1470
+ "learning_rate": 1.8922033898305085e-05,
1471
+ "loss": 1.5208,
1472
+ "step": 2090
1473
+ },
1474
+ {
1475
+ "epoch": 0.05,
1476
+ "grad_norm": 10.5625,
1477
+ "learning_rate": 1.8915254237288138e-05,
1478
+ "loss": 1.4693,
1479
+ "step": 2100
1480
+ },
1481
+ {
1482
+ "epoch": 0.05,
1483
+ "grad_norm": 18.625,
1484
+ "learning_rate": 1.8908474576271187e-05,
1485
+ "loss": 1.6504,
1486
+ "step": 2110
1487
+ },
1488
+ {
1489
+ "epoch": 0.05,
1490
+ "grad_norm": 15.1875,
1491
+ "learning_rate": 1.8901694915254237e-05,
1492
+ "loss": 1.3865,
1493
+ "step": 2120
1494
+ },
1495
+ {
1496
+ "epoch": 0.05,
1497
+ "grad_norm": 18.75,
1498
+ "learning_rate": 1.889491525423729e-05,
1499
+ "loss": 1.5483,
1500
+ "step": 2130
1501
+ },
1502
+ {
1503
+ "epoch": 0.05,
1504
+ "grad_norm": 10.5,
1505
+ "learning_rate": 1.888813559322034e-05,
1506
+ "loss": 1.5347,
1507
+ "step": 2140
1508
+ },
1509
+ {
1510
+ "epoch": 0.05,
1511
+ "grad_norm": 19.875,
1512
+ "learning_rate": 1.8881355932203393e-05,
1513
+ "loss": 1.4194,
1514
+ "step": 2150
1515
+ },
1516
+ {
1517
+ "epoch": 0.05,
1518
+ "grad_norm": 9.8125,
1519
+ "learning_rate": 1.8874576271186443e-05,
1520
+ "loss": 1.617,
1521
+ "step": 2160
1522
+ },
1523
+ {
1524
+ "epoch": 0.05,
1525
+ "grad_norm": 8.25,
1526
+ "learning_rate": 1.8867796610169492e-05,
1527
+ "loss": 1.4816,
1528
+ "step": 2170
1529
+ },
1530
+ {
1531
+ "epoch": 0.05,
1532
+ "grad_norm": 20.625,
1533
+ "learning_rate": 1.8861016949152545e-05,
1534
+ "loss": 1.6254,
1535
+ "step": 2180
1536
+ },
1537
+ {
1538
+ "epoch": 0.05,
1539
+ "grad_norm": 14.6875,
1540
+ "learning_rate": 1.8854237288135595e-05,
1541
+ "loss": 1.6584,
1542
+ "step": 2190
1543
+ },
1544
+ {
1545
+ "epoch": 0.06,
1546
+ "grad_norm": 6.96875,
1547
+ "learning_rate": 1.8847457627118645e-05,
1548
+ "loss": 1.6248,
1549
+ "step": 2200
1550
+ },
1551
+ {
1552
+ "epoch": 0.06,
1553
+ "grad_norm": 15.9375,
1554
+ "learning_rate": 1.8840677966101698e-05,
1555
+ "loss": 1.5322,
1556
+ "step": 2210
1557
+ },
1558
+ {
1559
+ "epoch": 0.06,
1560
+ "grad_norm": 9.625,
1561
+ "learning_rate": 1.8833898305084748e-05,
1562
+ "loss": 1.506,
1563
+ "step": 2220
1564
+ },
1565
+ {
1566
+ "epoch": 0.06,
1567
+ "grad_norm": 9.75,
1568
+ "learning_rate": 1.8827118644067797e-05,
1569
+ "loss": 1.6489,
1570
+ "step": 2230
1571
+ },
1572
+ {
1573
+ "epoch": 0.06,
1574
+ "grad_norm": 15.4375,
1575
+ "learning_rate": 1.882033898305085e-05,
1576
+ "loss": 1.4596,
1577
+ "step": 2240
1578
+ },
1579
+ {
1580
+ "epoch": 0.06,
1581
+ "grad_norm": 11.25,
1582
+ "learning_rate": 1.88135593220339e-05,
1583
+ "loss": 1.5384,
1584
+ "step": 2250
1585
+ },
1586
+ {
1587
+ "epoch": 0.06,
1588
+ "grad_norm": 9.3125,
1589
+ "learning_rate": 1.8806779661016953e-05,
1590
+ "loss": 1.5334,
1591
+ "step": 2260
1592
+ },
1593
+ {
1594
+ "epoch": 0.06,
1595
+ "grad_norm": 8.6875,
1596
+ "learning_rate": 1.88e-05,
1597
+ "loss": 1.5069,
1598
+ "step": 2270
1599
+ },
1600
+ {
1601
+ "epoch": 0.06,
1602
+ "grad_norm": 26.0,
1603
+ "learning_rate": 1.8793220338983053e-05,
1604
+ "loss": 1.4391,
1605
+ "step": 2280
1606
+ },
1607
+ {
1608
+ "epoch": 0.06,
1609
+ "grad_norm": 16.0,
1610
+ "learning_rate": 1.8786440677966102e-05,
1611
+ "loss": 1.4752,
1612
+ "step": 2290
1613
+ },
1614
+ {
1615
+ "epoch": 0.06,
1616
+ "grad_norm": 14.8125,
1617
+ "learning_rate": 1.8779661016949152e-05,
1618
+ "loss": 1.5738,
1619
+ "step": 2300
1620
+ },
1621
+ {
1622
+ "epoch": 0.06,
1623
+ "grad_norm": 12.625,
1624
+ "learning_rate": 1.8772881355932205e-05,
1625
+ "loss": 1.3754,
1626
+ "step": 2310
1627
+ },
1628
+ {
1629
+ "epoch": 0.06,
1630
+ "grad_norm": 12.875,
1631
+ "learning_rate": 1.8766101694915255e-05,
1632
+ "loss": 1.5944,
1633
+ "step": 2320
1634
+ },
1635
+ {
1636
+ "epoch": 0.06,
1637
+ "grad_norm": 15.0625,
1638
+ "learning_rate": 1.8759322033898305e-05,
1639
+ "loss": 1.4887,
1640
+ "step": 2330
1641
+ },
1642
+ {
1643
+ "epoch": 0.06,
1644
+ "grad_norm": 9.3125,
1645
+ "learning_rate": 1.8752542372881358e-05,
1646
+ "loss": 1.4891,
1647
+ "step": 2340
1648
+ },
1649
+ {
1650
+ "epoch": 0.06,
1651
+ "grad_norm": 24.25,
1652
+ "learning_rate": 1.8745762711864407e-05,
1653
+ "loss": 1.7288,
1654
+ "step": 2350
1655
+ },
1656
+ {
1657
+ "epoch": 0.06,
1658
+ "grad_norm": 12.8125,
1659
+ "learning_rate": 1.873898305084746e-05,
1660
+ "loss": 1.4783,
1661
+ "step": 2360
1662
+ },
1663
+ {
1664
+ "epoch": 0.06,
1665
+ "grad_norm": 6.5,
1666
+ "learning_rate": 1.873220338983051e-05,
1667
+ "loss": 1.6752,
1668
+ "step": 2370
1669
+ },
1670
+ {
1671
+ "epoch": 0.06,
1672
+ "grad_norm": 17.5,
1673
+ "learning_rate": 1.872542372881356e-05,
1674
+ "loss": 1.3905,
1675
+ "step": 2380
1676
+ },
1677
+ {
1678
+ "epoch": 0.06,
1679
+ "grad_norm": 13.9375,
1680
+ "learning_rate": 1.8718644067796613e-05,
1681
+ "loss": 1.4255,
1682
+ "step": 2390
1683
+ },
1684
+ {
1685
+ "epoch": 0.06,
1686
+ "grad_norm": 11.5,
1687
+ "learning_rate": 1.8711864406779663e-05,
1688
+ "loss": 1.5024,
1689
+ "step": 2400
1690
+ },
1691
+ {
1692
+ "epoch": 0.06,
1693
+ "grad_norm": 15.375,
1694
+ "learning_rate": 1.8705084745762712e-05,
1695
+ "loss": 1.4946,
1696
+ "step": 2410
1697
+ },
1698
+ {
1699
+ "epoch": 0.06,
1700
+ "grad_norm": 12.3125,
1701
+ "learning_rate": 1.8698305084745765e-05,
1702
+ "loss": 1.1831,
1703
+ "step": 2420
1704
+ },
1705
+ {
1706
+ "epoch": 0.06,
1707
+ "grad_norm": 18.625,
1708
+ "learning_rate": 1.8691525423728815e-05,
1709
+ "loss": 1.4678,
1710
+ "step": 2430
1711
+ },
1712
+ {
1713
+ "epoch": 0.06,
1714
+ "grad_norm": 18.125,
1715
+ "learning_rate": 1.8684745762711868e-05,
1716
+ "loss": 1.3618,
1717
+ "step": 2440
1718
+ },
1719
+ {
1720
+ "epoch": 0.06,
1721
+ "grad_norm": 8.125,
1722
+ "learning_rate": 1.8677966101694918e-05,
1723
+ "loss": 1.5961,
1724
+ "step": 2450
1725
+ },
1726
+ {
1727
+ "epoch": 0.06,
1728
+ "grad_norm": 11.9375,
1729
+ "learning_rate": 1.8671186440677968e-05,
1730
+ "loss": 1.5546,
1731
+ "step": 2460
1732
+ },
1733
+ {
1734
+ "epoch": 0.06,
1735
+ "grad_norm": 10.75,
1736
+ "learning_rate": 1.866440677966102e-05,
1737
+ "loss": 1.628,
1738
+ "step": 2470
1739
+ },
1740
+ {
1741
+ "epoch": 0.06,
1742
+ "grad_norm": 21.75,
1743
+ "learning_rate": 1.8657627118644067e-05,
1744
+ "loss": 1.3956,
1745
+ "step": 2480
1746
+ },
1747
+ {
1748
+ "epoch": 0.06,
1749
+ "grad_norm": 13.875,
1750
+ "learning_rate": 1.865084745762712e-05,
1751
+ "loss": 1.4657,
1752
+ "step": 2490
1753
+ },
1754
+ {
1755
+ "epoch": 0.06,
1756
+ "grad_norm": 16.5,
1757
+ "learning_rate": 1.864406779661017e-05,
1758
+ "loss": 1.6003,
1759
+ "step": 2500
1760
+ },
1761
+ {
1762
+ "epoch": 0.06,
1763
+ "grad_norm": 11.875,
1764
+ "learning_rate": 1.863728813559322e-05,
1765
+ "loss": 1.5296,
1766
+ "step": 2510
1767
+ },
1768
+ {
1769
+ "epoch": 0.06,
1770
+ "grad_norm": 8.5,
1771
+ "learning_rate": 1.8630508474576273e-05,
1772
+ "loss": 1.5027,
1773
+ "step": 2520
1774
+ },
1775
+ {
1776
+ "epoch": 0.06,
1777
+ "grad_norm": 6.875,
1778
+ "learning_rate": 1.8623728813559322e-05,
1779
+ "loss": 1.416,
1780
+ "step": 2530
1781
+ },
1782
+ {
1783
+ "epoch": 0.06,
1784
+ "grad_norm": 16.875,
1785
+ "learning_rate": 1.8616949152542376e-05,
1786
+ "loss": 1.5729,
1787
+ "step": 2540
1788
+ },
1789
+ {
1790
+ "epoch": 0.06,
1791
+ "grad_norm": 10.5,
1792
+ "learning_rate": 1.8610169491525425e-05,
1793
+ "loss": 1.6464,
1794
+ "step": 2550
1795
+ },
1796
+ {
1797
+ "epoch": 0.06,
1798
+ "grad_norm": 12.375,
1799
+ "learning_rate": 1.8603389830508475e-05,
1800
+ "loss": 1.5534,
1801
+ "step": 2560
1802
+ },
1803
+ {
1804
+ "epoch": 0.06,
1805
+ "grad_norm": 9.0,
1806
+ "learning_rate": 1.8596610169491528e-05,
1807
+ "loss": 1.6106,
1808
+ "step": 2570
1809
+ },
1810
+ {
1811
+ "epoch": 0.06,
1812
+ "grad_norm": 84.0,
1813
+ "learning_rate": 1.8589830508474578e-05,
1814
+ "loss": 1.5543,
1815
+ "step": 2580
1816
+ },
1817
+ {
1818
+ "epoch": 0.06,
1819
+ "grad_norm": 13.6875,
1820
+ "learning_rate": 1.8583050847457627e-05,
1821
+ "loss": 1.5078,
1822
+ "step": 2590
1823
+ },
1824
+ {
1825
+ "epoch": 0.07,
1826
+ "grad_norm": 12.125,
1827
+ "learning_rate": 1.857627118644068e-05,
1828
+ "loss": 1.4837,
1829
+ "step": 2600
1830
+ },
1831
+ {
1832
+ "epoch": 0.07,
1833
+ "grad_norm": 17.375,
1834
+ "learning_rate": 1.856949152542373e-05,
1835
+ "loss": 1.3,
1836
+ "step": 2610
1837
+ },
1838
+ {
1839
+ "epoch": 0.07,
1840
+ "grad_norm": 13.375,
1841
+ "learning_rate": 1.856271186440678e-05,
1842
+ "loss": 1.6079,
1843
+ "step": 2620
1844
+ },
1845
+ {
1846
+ "epoch": 0.07,
1847
+ "grad_norm": 38.0,
1848
+ "learning_rate": 1.8555932203389833e-05,
1849
+ "loss": 1.5849,
1850
+ "step": 2630
1851
+ },
1852
+ {
1853
+ "epoch": 0.07,
1854
+ "grad_norm": 17.75,
1855
+ "learning_rate": 1.8549152542372883e-05,
1856
+ "loss": 1.4636,
1857
+ "step": 2640
1858
+ },
1859
+ {
1860
+ "epoch": 0.07,
1861
+ "grad_norm": 12.5625,
1862
+ "learning_rate": 1.8542372881355936e-05,
1863
+ "loss": 1.5832,
1864
+ "step": 2650
1865
+ },
1866
+ {
1867
+ "epoch": 0.07,
1868
+ "grad_norm": 7.6875,
1869
+ "learning_rate": 1.8535593220338986e-05,
1870
+ "loss": 1.4782,
1871
+ "step": 2660
1872
+ },
1873
+ {
1874
+ "epoch": 0.07,
1875
+ "grad_norm": 10.0,
1876
+ "learning_rate": 1.8528813559322035e-05,
1877
+ "loss": 1.5614,
1878
+ "step": 2670
1879
+ },
1880
+ {
1881
+ "epoch": 0.07,
1882
+ "grad_norm": 9.375,
1883
+ "learning_rate": 1.852203389830509e-05,
1884
+ "loss": 1.4678,
1885
+ "step": 2680
1886
+ },
1887
+ {
1888
+ "epoch": 0.07,
1889
+ "grad_norm": 12.125,
1890
+ "learning_rate": 1.8515254237288135e-05,
1891
+ "loss": 1.4207,
1892
+ "step": 2690
1893
+ },
1894
+ {
1895
+ "epoch": 0.07,
1896
+ "grad_norm": 11.6875,
1897
+ "learning_rate": 1.8508474576271188e-05,
1898
+ "loss": 1.5857,
1899
+ "step": 2700
1900
+ },
1901
+ {
1902
+ "epoch": 0.07,
1903
+ "grad_norm": 11.5,
1904
+ "learning_rate": 1.8501694915254237e-05,
1905
+ "loss": 1.5383,
1906
+ "step": 2710
1907
+ },
1908
+ {
1909
+ "epoch": 0.07,
1910
+ "grad_norm": 12.9375,
1911
+ "learning_rate": 1.849491525423729e-05,
1912
+ "loss": 1.4254,
1913
+ "step": 2720
1914
+ },
1915
+ {
1916
+ "epoch": 0.07,
1917
+ "grad_norm": 15.6875,
1918
+ "learning_rate": 1.848813559322034e-05,
1919
+ "loss": 1.4762,
1920
+ "step": 2730
1921
+ },
1922
+ {
1923
+ "epoch": 0.07,
1924
+ "grad_norm": 18.625,
1925
+ "learning_rate": 1.848135593220339e-05,
1926
+ "loss": 1.4275,
1927
+ "step": 2740
1928
+ },
1929
+ {
1930
+ "epoch": 0.07,
1931
+ "grad_norm": 12.5625,
1932
+ "learning_rate": 1.8474576271186443e-05,
1933
+ "loss": 1.2928,
1934
+ "step": 2750
1935
+ },
1936
+ {
1937
+ "epoch": 0.07,
1938
+ "grad_norm": 12.6875,
1939
+ "learning_rate": 1.8467796610169493e-05,
1940
+ "loss": 1.4166,
1941
+ "step": 2760
1942
+ },
1943
+ {
1944
+ "epoch": 0.07,
1945
+ "grad_norm": 8.1875,
1946
+ "learning_rate": 1.8461016949152542e-05,
1947
+ "loss": 1.4579,
1948
+ "step": 2770
1949
+ },
1950
+ {
1951
+ "epoch": 0.07,
1952
+ "grad_norm": 8.6875,
1953
+ "learning_rate": 1.8454237288135596e-05,
1954
+ "loss": 1.4313,
1955
+ "step": 2780
1956
+ },
1957
+ {
1958
+ "epoch": 0.07,
1959
+ "grad_norm": 8.8125,
1960
+ "learning_rate": 1.8447457627118645e-05,
1961
+ "loss": 1.5309,
1962
+ "step": 2790
1963
+ },
1964
+ {
1965
+ "epoch": 0.07,
1966
+ "grad_norm": 12.5,
1967
+ "learning_rate": 1.8440677966101695e-05,
1968
+ "loss": 1.4502,
1969
+ "step": 2800
1970
+ },
1971
+ {
1972
+ "epoch": 0.07,
1973
+ "grad_norm": 8.1875,
1974
+ "learning_rate": 1.8433898305084748e-05,
1975
+ "loss": 1.4403,
1976
+ "step": 2810
1977
+ },
1978
+ {
1979
+ "epoch": 0.07,
1980
+ "grad_norm": 8.125,
1981
+ "learning_rate": 1.8427118644067798e-05,
1982
+ "loss": 1.612,
1983
+ "step": 2820
1984
+ },
1985
+ {
1986
+ "epoch": 0.07,
1987
+ "grad_norm": 10.5,
1988
+ "learning_rate": 1.842033898305085e-05,
1989
+ "loss": 1.3711,
1990
+ "step": 2830
1991
+ },
1992
+ {
1993
+ "epoch": 0.07,
1994
+ "grad_norm": 11.25,
1995
+ "learning_rate": 1.84135593220339e-05,
1996
+ "loss": 1.4374,
1997
+ "step": 2840
1998
+ },
1999
+ {
2000
+ "epoch": 0.07,
2001
+ "grad_norm": 11.125,
2002
+ "learning_rate": 1.840677966101695e-05,
2003
+ "loss": 1.4582,
2004
+ "step": 2850
2005
+ },
2006
+ {
2007
+ "epoch": 0.07,
2008
+ "grad_norm": 9.8125,
2009
+ "learning_rate": 1.8400000000000003e-05,
2010
+ "loss": 1.5342,
2011
+ "step": 2860
2012
+ },
2013
+ {
2014
+ "epoch": 0.07,
2015
+ "grad_norm": 13.1875,
2016
+ "learning_rate": 1.8393220338983053e-05,
2017
+ "loss": 1.3282,
2018
+ "step": 2870
2019
+ },
2020
+ {
2021
+ "epoch": 0.07,
2022
+ "grad_norm": 8.5,
2023
+ "learning_rate": 1.8386440677966103e-05,
2024
+ "loss": 1.5913,
2025
+ "step": 2880
2026
+ },
2027
+ {
2028
+ "epoch": 0.07,
2029
+ "grad_norm": 11.0,
2030
+ "learning_rate": 1.8379661016949153e-05,
2031
+ "loss": 1.6801,
2032
+ "step": 2890
2033
+ },
2034
+ {
2035
+ "epoch": 0.07,
2036
+ "grad_norm": 19.25,
2037
+ "learning_rate": 1.8372881355932202e-05,
2038
+ "loss": 1.4581,
2039
+ "step": 2900
2040
+ },
2041
+ {
2042
+ "epoch": 0.07,
2043
+ "grad_norm": 17.5,
2044
+ "learning_rate": 1.8366101694915255e-05,
2045
+ "loss": 1.3822,
2046
+ "step": 2910
2047
+ },
2048
+ {
2049
+ "epoch": 0.07,
2050
+ "grad_norm": 28.125,
2051
+ "learning_rate": 1.8359322033898305e-05,
2052
+ "loss": 1.3937,
2053
+ "step": 2920
2054
+ },
2055
+ {
2056
+ "epoch": 0.07,
2057
+ "grad_norm": 11.4375,
2058
+ "learning_rate": 1.8352542372881358e-05,
2059
+ "loss": 1.4311,
2060
+ "step": 2930
2061
+ },
2062
+ {
2063
+ "epoch": 0.07,
2064
+ "grad_norm": 6.4375,
2065
+ "learning_rate": 1.8345762711864408e-05,
2066
+ "loss": 1.4237,
2067
+ "step": 2940
2068
+ },
2069
+ {
2070
+ "epoch": 0.07,
2071
+ "grad_norm": 20.625,
2072
+ "learning_rate": 1.8338983050847458e-05,
2073
+ "loss": 1.7361,
2074
+ "step": 2950
2075
+ },
2076
+ {
2077
+ "epoch": 0.07,
2078
+ "grad_norm": 13.875,
2079
+ "learning_rate": 1.833220338983051e-05,
2080
+ "loss": 1.5519,
2081
+ "step": 2960
2082
+ },
2083
+ {
2084
+ "epoch": 0.07,
2085
+ "grad_norm": 8.625,
2086
+ "learning_rate": 1.832542372881356e-05,
2087
+ "loss": 1.3663,
2088
+ "step": 2970
2089
+ },
2090
+ {
2091
+ "epoch": 0.07,
2092
+ "grad_norm": 16.875,
2093
+ "learning_rate": 1.831864406779661e-05,
2094
+ "loss": 1.4183,
2095
+ "step": 2980
2096
+ },
2097
+ {
2098
+ "epoch": 0.07,
2099
+ "grad_norm": 23.375,
2100
+ "learning_rate": 1.8311864406779663e-05,
2101
+ "loss": 1.4189,
2102
+ "step": 2990
2103
+ },
2104
+ {
2105
+ "epoch": 0.07,
2106
+ "grad_norm": 16.125,
2107
+ "learning_rate": 1.8305084745762713e-05,
2108
+ "loss": 1.5008,
2109
+ "step": 3000
2110
+ },
2111
+ {
2112
+ "epoch": 0.08,
2113
+ "grad_norm": 19.0,
2114
+ "learning_rate": 1.8298305084745766e-05,
2115
+ "loss": 1.2341,
2116
+ "step": 3010
2117
+ },
2118
+ {
2119
+ "epoch": 0.08,
2120
+ "grad_norm": 9.0625,
2121
+ "learning_rate": 1.8291525423728816e-05,
2122
+ "loss": 1.3109,
2123
+ "step": 3020
2124
+ },
2125
+ {
2126
+ "epoch": 0.08,
2127
+ "grad_norm": 14.125,
2128
+ "learning_rate": 1.8284745762711865e-05,
2129
+ "loss": 1.4832,
2130
+ "step": 3030
2131
+ },
2132
+ {
2133
+ "epoch": 0.08,
2134
+ "grad_norm": 25.75,
2135
+ "learning_rate": 1.827796610169492e-05,
2136
+ "loss": 1.4074,
2137
+ "step": 3040
2138
+ },
2139
+ {
2140
+ "epoch": 0.08,
2141
+ "grad_norm": 9.5625,
2142
+ "learning_rate": 1.8271186440677968e-05,
2143
+ "loss": 1.324,
2144
+ "step": 3050
2145
+ },
2146
+ {
2147
+ "epoch": 0.08,
2148
+ "grad_norm": 19.0,
2149
+ "learning_rate": 1.8264406779661018e-05,
2150
+ "loss": 1.6133,
2151
+ "step": 3060
2152
+ },
2153
+ {
2154
+ "epoch": 0.08,
2155
+ "grad_norm": 12.75,
2156
+ "learning_rate": 1.825762711864407e-05,
2157
+ "loss": 1.4959,
2158
+ "step": 3070
2159
+ },
2160
+ {
2161
+ "epoch": 0.08,
2162
+ "grad_norm": 17.75,
2163
+ "learning_rate": 1.825084745762712e-05,
2164
+ "loss": 1.5612,
2165
+ "step": 3080
2166
+ },
2167
+ {
2168
+ "epoch": 0.08,
2169
+ "grad_norm": 11.125,
2170
+ "learning_rate": 1.824406779661017e-05,
2171
+ "loss": 1.4213,
2172
+ "step": 3090
2173
+ },
2174
+ {
2175
+ "epoch": 0.08,
2176
+ "grad_norm": 22.625,
2177
+ "learning_rate": 1.823728813559322e-05,
2178
+ "loss": 1.6306,
2179
+ "step": 3100
2180
+ },
2181
+ {
2182
+ "epoch": 0.08,
2183
+ "grad_norm": 13.375,
2184
+ "learning_rate": 1.8230508474576273e-05,
2185
+ "loss": 1.6571,
2186
+ "step": 3110
2187
+ },
2188
+ {
2189
+ "epoch": 0.08,
2190
+ "grad_norm": 11.875,
2191
+ "learning_rate": 1.8223728813559323e-05,
2192
+ "loss": 1.5212,
2193
+ "step": 3120
2194
+ },
2195
+ {
2196
+ "epoch": 0.08,
2197
+ "grad_norm": 9.0,
2198
+ "learning_rate": 1.8216949152542373e-05,
2199
+ "loss": 1.5581,
2200
+ "step": 3130
2201
+ },
2202
+ {
2203
+ "epoch": 0.08,
2204
+ "grad_norm": 7.03125,
2205
+ "learning_rate": 1.8210169491525426e-05,
2206
+ "loss": 1.4206,
2207
+ "step": 3140
2208
+ },
2209
+ {
2210
+ "epoch": 0.08,
2211
+ "grad_norm": 11.8125,
2212
+ "learning_rate": 1.8203389830508475e-05,
2213
+ "loss": 1.5661,
2214
+ "step": 3150
2215
+ },
2216
+ {
2217
+ "epoch": 0.08,
2218
+ "grad_norm": 12.625,
2219
+ "learning_rate": 1.8196610169491525e-05,
2220
+ "loss": 1.3835,
2221
+ "step": 3160
2222
+ },
2223
+ {
2224
+ "epoch": 0.08,
2225
+ "grad_norm": 15.6875,
2226
+ "learning_rate": 1.8189830508474578e-05,
2227
+ "loss": 1.5146,
2228
+ "step": 3170
2229
+ },
2230
+ {
2231
+ "epoch": 0.08,
2232
+ "grad_norm": 8.0625,
2233
+ "learning_rate": 1.8183050847457628e-05,
2234
+ "loss": 1.466,
2235
+ "step": 3180
2236
+ },
2237
+ {
2238
+ "epoch": 0.08,
2239
+ "grad_norm": 21.625,
2240
+ "learning_rate": 1.817627118644068e-05,
2241
+ "loss": 1.5828,
2242
+ "step": 3190
2243
+ },
2244
+ {
2245
+ "epoch": 0.08,
2246
+ "grad_norm": 6.90625,
2247
+ "learning_rate": 1.816949152542373e-05,
2248
+ "loss": 1.4519,
2249
+ "step": 3200
2250
+ },
2251
+ {
2252
+ "epoch": 0.08,
2253
+ "grad_norm": 14.4375,
2254
+ "learning_rate": 1.816271186440678e-05,
2255
+ "loss": 1.6093,
2256
+ "step": 3210
2257
+ },
2258
+ {
2259
+ "epoch": 0.08,
2260
+ "grad_norm": 20.625,
2261
+ "learning_rate": 1.8155932203389833e-05,
2262
+ "loss": 1.6915,
2263
+ "step": 3220
2264
+ },
2265
+ {
2266
+ "epoch": 0.08,
2267
+ "grad_norm": 10.1875,
2268
+ "learning_rate": 1.8149152542372883e-05,
2269
+ "loss": 1.5521,
2270
+ "step": 3230
2271
+ },
2272
+ {
2273
+ "epoch": 0.08,
2274
+ "grad_norm": 10.625,
2275
+ "learning_rate": 1.8142372881355933e-05,
2276
+ "loss": 1.4542,
2277
+ "step": 3240
2278
+ },
2279
+ {
2280
+ "epoch": 0.08,
2281
+ "grad_norm": 8.6875,
2282
+ "learning_rate": 1.8135593220338986e-05,
2283
+ "loss": 1.494,
2284
+ "step": 3250
2285
+ },
2286
+ {
2287
+ "epoch": 0.08,
2288
+ "grad_norm": 9.625,
2289
+ "learning_rate": 1.8128813559322036e-05,
2290
+ "loss": 1.3653,
2291
+ "step": 3260
2292
+ },
2293
+ {
2294
+ "epoch": 0.08,
2295
+ "grad_norm": 12.375,
2296
+ "learning_rate": 1.8122033898305085e-05,
2297
+ "loss": 1.3581,
2298
+ "step": 3270
2299
+ },
2300
+ {
2301
+ "epoch": 0.08,
2302
+ "grad_norm": 22.625,
2303
+ "learning_rate": 1.811525423728814e-05,
2304
+ "loss": 1.6393,
2305
+ "step": 3280
2306
+ },
2307
+ {
2308
+ "epoch": 0.08,
2309
+ "grad_norm": 19.875,
2310
+ "learning_rate": 1.8108474576271188e-05,
2311
+ "loss": 1.4813,
2312
+ "step": 3290
2313
+ },
2314
+ {
2315
+ "epoch": 0.08,
2316
+ "grad_norm": 15.3125,
2317
+ "learning_rate": 1.810169491525424e-05,
2318
+ "loss": 1.3628,
2319
+ "step": 3300
2320
+ },
2321
+ {
2322
+ "epoch": 0.08,
2323
+ "grad_norm": 10.25,
2324
+ "learning_rate": 1.8094915254237288e-05,
2325
+ "loss": 1.5978,
2326
+ "step": 3310
2327
+ },
2328
+ {
2329
+ "epoch": 0.08,
2330
+ "grad_norm": 12.5625,
2331
+ "learning_rate": 1.808813559322034e-05,
2332
+ "loss": 1.5745,
2333
+ "step": 3320
2334
+ },
2335
+ {
2336
+ "epoch": 0.08,
2337
+ "grad_norm": 20.0,
2338
+ "learning_rate": 1.808135593220339e-05,
2339
+ "loss": 1.4374,
2340
+ "step": 3330
2341
+ },
2342
+ {
2343
+ "epoch": 0.08,
2344
+ "grad_norm": 8.9375,
2345
+ "learning_rate": 1.807457627118644e-05,
2346
+ "loss": 1.4625,
2347
+ "step": 3340
2348
+ },
2349
+ {
2350
+ "epoch": 0.08,
2351
+ "grad_norm": 22.625,
2352
+ "learning_rate": 1.8067796610169493e-05,
2353
+ "loss": 1.6427,
2354
+ "step": 3350
2355
+ },
2356
+ {
2357
+ "epoch": 0.08,
2358
+ "grad_norm": 12.6875,
2359
+ "learning_rate": 1.8061016949152543e-05,
2360
+ "loss": 1.6387,
2361
+ "step": 3360
2362
+ },
2363
+ {
2364
+ "epoch": 0.08,
2365
+ "grad_norm": 13.1875,
2366
+ "learning_rate": 1.8054237288135593e-05,
2367
+ "loss": 1.405,
2368
+ "step": 3370
2369
+ },
2370
+ {
2371
+ "epoch": 0.08,
2372
+ "grad_norm": 22.875,
2373
+ "learning_rate": 1.8047457627118646e-05,
2374
+ "loss": 1.3973,
2375
+ "step": 3380
2376
+ },
2377
+ {
2378
+ "epoch": 0.08,
2379
+ "grad_norm": 8.625,
2380
+ "learning_rate": 1.8040677966101695e-05,
2381
+ "loss": 1.4561,
2382
+ "step": 3390
2383
+ },
2384
+ {
2385
+ "epoch": 0.09,
2386
+ "grad_norm": 12.625,
2387
+ "learning_rate": 1.803389830508475e-05,
2388
+ "loss": 1.4632,
2389
+ "step": 3400
2390
+ },
2391
+ {
2392
+ "epoch": 0.09,
2393
+ "grad_norm": 18.125,
2394
+ "learning_rate": 1.8027118644067798e-05,
2395
+ "loss": 1.4032,
2396
+ "step": 3410
2397
+ },
2398
+ {
2399
+ "epoch": 0.09,
2400
+ "grad_norm": 12.75,
2401
+ "learning_rate": 1.8020338983050848e-05,
2402
+ "loss": 1.5775,
2403
+ "step": 3420
2404
+ },
2405
+ {
2406
+ "epoch": 0.09,
2407
+ "grad_norm": 17.875,
2408
+ "learning_rate": 1.80135593220339e-05,
2409
+ "loss": 1.4885,
2410
+ "step": 3430
2411
+ },
2412
+ {
2413
+ "epoch": 0.09,
2414
+ "grad_norm": 17.25,
2415
+ "learning_rate": 1.800677966101695e-05,
2416
+ "loss": 1.414,
2417
+ "step": 3440
2418
+ },
2419
+ {
2420
+ "epoch": 0.09,
2421
+ "grad_norm": 10.5,
2422
+ "learning_rate": 1.8e-05,
2423
+ "loss": 1.4406,
2424
+ "step": 3450
2425
+ },
2426
+ {
2427
+ "epoch": 0.09,
2428
+ "grad_norm": 15.75,
2429
+ "learning_rate": 1.7993220338983054e-05,
2430
+ "loss": 1.4872,
2431
+ "step": 3460
2432
+ },
2433
+ {
2434
+ "epoch": 0.09,
2435
+ "grad_norm": 10.25,
2436
+ "learning_rate": 1.7986440677966103e-05,
2437
+ "loss": 1.4513,
2438
+ "step": 3470
2439
+ },
2440
+ {
2441
+ "epoch": 0.09,
2442
+ "grad_norm": 11.875,
2443
+ "learning_rate": 1.7979661016949156e-05,
2444
+ "loss": 1.5059,
2445
+ "step": 3480
2446
+ },
2447
+ {
2448
+ "epoch": 0.09,
2449
+ "grad_norm": 14.25,
2450
+ "learning_rate": 1.7972881355932206e-05,
2451
+ "loss": 1.4446,
2452
+ "step": 3490
2453
+ },
2454
+ {
2455
+ "epoch": 0.09,
2456
+ "grad_norm": 14.0,
2457
+ "learning_rate": 1.7966101694915256e-05,
2458
+ "loss": 1.4928,
2459
+ "step": 3500
2460
+ },
2461
+ {
2462
+ "epoch": 0.09,
2463
+ "grad_norm": 19.5,
2464
+ "learning_rate": 1.795932203389831e-05,
2465
+ "loss": 1.5305,
2466
+ "step": 3510
2467
+ },
2468
+ {
2469
+ "epoch": 0.09,
2470
+ "grad_norm": 10.0,
2471
+ "learning_rate": 1.7952542372881355e-05,
2472
+ "loss": 1.4932,
2473
+ "step": 3520
2474
+ },
2475
+ {
2476
+ "epoch": 0.09,
2477
+ "grad_norm": 32.0,
2478
+ "learning_rate": 1.7945762711864408e-05,
2479
+ "loss": 1.2962,
2480
+ "step": 3530
2481
+ },
2482
+ {
2483
+ "epoch": 0.09,
2484
+ "grad_norm": 12.0,
2485
+ "learning_rate": 1.7938983050847458e-05,
2486
+ "loss": 1.2428,
2487
+ "step": 3540
2488
+ },
2489
+ {
2490
+ "epoch": 0.09,
2491
+ "grad_norm": 12.4375,
2492
+ "learning_rate": 1.7932203389830508e-05,
2493
+ "loss": 1.5858,
2494
+ "step": 3550
2495
+ },
2496
+ {
2497
+ "epoch": 0.09,
2498
+ "grad_norm": 7.78125,
2499
+ "learning_rate": 1.792542372881356e-05,
2500
+ "loss": 1.4792,
2501
+ "step": 3560
2502
+ },
2503
+ {
2504
+ "epoch": 0.09,
2505
+ "grad_norm": 8.3125,
2506
+ "learning_rate": 1.791864406779661e-05,
2507
+ "loss": 1.4305,
2508
+ "step": 3570
2509
+ },
2510
+ {
2511
+ "epoch": 0.09,
2512
+ "grad_norm": 19.5,
2513
+ "learning_rate": 1.7911864406779664e-05,
2514
+ "loss": 1.5068,
2515
+ "step": 3580
2516
+ },
2517
+ {
2518
+ "epoch": 0.09,
2519
+ "grad_norm": 12.6875,
2520
+ "learning_rate": 1.7905084745762713e-05,
2521
+ "loss": 1.3162,
2522
+ "step": 3590
2523
+ },
2524
+ {
2525
+ "epoch": 0.09,
2526
+ "grad_norm": 16.875,
2527
+ "learning_rate": 1.7898305084745763e-05,
2528
+ "loss": 1.5589,
2529
+ "step": 3600
2530
+ },
2531
+ {
2532
+ "epoch": 0.09,
2533
+ "grad_norm": 12.0625,
2534
+ "learning_rate": 1.7891525423728816e-05,
2535
+ "loss": 1.3933,
2536
+ "step": 3610
2537
+ },
2538
+ {
2539
+ "epoch": 0.09,
2540
+ "grad_norm": 10.75,
2541
+ "learning_rate": 1.7884745762711866e-05,
2542
+ "loss": 1.409,
2543
+ "step": 3620
2544
+ },
2545
+ {
2546
+ "epoch": 0.09,
2547
+ "grad_norm": 10.1875,
2548
+ "learning_rate": 1.7877966101694916e-05,
2549
+ "loss": 1.5357,
2550
+ "step": 3630
2551
+ },
2552
+ {
2553
+ "epoch": 0.09,
2554
+ "grad_norm": 8.8125,
2555
+ "learning_rate": 1.787118644067797e-05,
2556
+ "loss": 1.4958,
2557
+ "step": 3640
2558
+ },
2559
+ {
2560
+ "epoch": 0.09,
2561
+ "grad_norm": 12.25,
2562
+ "learning_rate": 1.7864406779661018e-05,
2563
+ "loss": 1.4976,
2564
+ "step": 3650
2565
+ },
2566
+ {
2567
+ "epoch": 0.09,
2568
+ "grad_norm": 22.125,
2569
+ "learning_rate": 1.785762711864407e-05,
2570
+ "loss": 1.472,
2571
+ "step": 3660
2572
+ },
2573
+ {
2574
+ "epoch": 0.09,
2575
+ "grad_norm": 8.6875,
2576
+ "learning_rate": 1.785084745762712e-05,
2577
+ "loss": 1.2726,
2578
+ "step": 3670
2579
+ },
2580
+ {
2581
+ "epoch": 0.09,
2582
+ "grad_norm": 12.1875,
2583
+ "learning_rate": 1.784406779661017e-05,
2584
+ "loss": 1.3201,
2585
+ "step": 3680
2586
+ },
2587
+ {
2588
+ "epoch": 0.09,
2589
+ "grad_norm": 12.4375,
2590
+ "learning_rate": 1.7837288135593224e-05,
2591
+ "loss": 1.4802,
2592
+ "step": 3690
2593
+ },
2594
+ {
2595
+ "epoch": 0.09,
2596
+ "grad_norm": 7.0625,
2597
+ "learning_rate": 1.7830508474576274e-05,
2598
+ "loss": 1.4807,
2599
+ "step": 3700
2600
+ },
2601
+ {
2602
+ "epoch": 0.09,
2603
+ "grad_norm": 32.5,
2604
+ "learning_rate": 1.7823728813559323e-05,
2605
+ "loss": 1.7589,
2606
+ "step": 3710
2607
+ },
2608
+ {
2609
+ "epoch": 0.09,
2610
+ "grad_norm": 14.8125,
2611
+ "learning_rate": 1.7816949152542376e-05,
2612
+ "loss": 1.4768,
2613
+ "step": 3720
2614
+ },
2615
+ {
2616
+ "epoch": 0.09,
2617
+ "grad_norm": 16.5,
2618
+ "learning_rate": 1.7810169491525423e-05,
2619
+ "loss": 1.3863,
2620
+ "step": 3730
2621
+ },
2622
+ {
2623
+ "epoch": 0.09,
2624
+ "grad_norm": 10.5625,
2625
+ "learning_rate": 1.7803389830508476e-05,
2626
+ "loss": 1.4855,
2627
+ "step": 3740
2628
+ },
2629
+ {
2630
+ "epoch": 0.09,
2631
+ "grad_norm": 12.3125,
2632
+ "learning_rate": 1.7796610169491526e-05,
2633
+ "loss": 1.4989,
2634
+ "step": 3750
2635
+ },
2636
+ {
2637
+ "epoch": 0.09,
2638
+ "grad_norm": 15.375,
2639
+ "learning_rate": 1.778983050847458e-05,
2640
+ "loss": 1.4212,
2641
+ "step": 3760
2642
+ },
2643
+ {
2644
+ "epoch": 0.09,
2645
+ "grad_norm": 13.6875,
2646
+ "learning_rate": 1.778305084745763e-05,
2647
+ "loss": 1.3991,
2648
+ "step": 3770
2649
+ },
2650
+ {
2651
+ "epoch": 0.09,
2652
+ "grad_norm": 11.3125,
2653
+ "learning_rate": 1.7776271186440678e-05,
2654
+ "loss": 1.496,
2655
+ "step": 3780
2656
+ },
2657
+ {
2658
+ "epoch": 0.09,
2659
+ "grad_norm": 32.75,
2660
+ "learning_rate": 1.776949152542373e-05,
2661
+ "loss": 1.5595,
2662
+ "step": 3790
2663
+ },
2664
+ {
2665
+ "epoch": 0.1,
2666
+ "grad_norm": 11.75,
2667
+ "learning_rate": 1.776271186440678e-05,
2668
+ "loss": 1.5164,
2669
+ "step": 3800
2670
+ },
2671
+ {
2672
+ "epoch": 0.1,
2673
+ "grad_norm": 11.125,
2674
+ "learning_rate": 1.775593220338983e-05,
2675
+ "loss": 1.5812,
2676
+ "step": 3810
2677
+ },
2678
+ {
2679
+ "epoch": 0.1,
2680
+ "grad_norm": 23.125,
2681
+ "learning_rate": 1.7749152542372884e-05,
2682
+ "loss": 1.5681,
2683
+ "step": 3820
2684
+ },
2685
+ {
2686
+ "epoch": 0.1,
2687
+ "grad_norm": 11.875,
2688
+ "learning_rate": 1.7742372881355933e-05,
2689
+ "loss": 1.5163,
2690
+ "step": 3830
2691
+ },
2692
+ {
2693
+ "epoch": 0.1,
2694
+ "grad_norm": 14.875,
2695
+ "learning_rate": 1.7735593220338983e-05,
2696
+ "loss": 1.427,
2697
+ "step": 3840
2698
+ },
2699
+ {
2700
+ "epoch": 0.1,
2701
+ "grad_norm": 26.25,
2702
+ "learning_rate": 1.7728813559322036e-05,
2703
+ "loss": 1.4703,
2704
+ "step": 3850
2705
+ },
2706
+ {
2707
+ "epoch": 0.1,
2708
+ "grad_norm": 9.125,
2709
+ "learning_rate": 1.7722033898305086e-05,
2710
+ "loss": 1.3093,
2711
+ "step": 3860
2712
+ },
2713
+ {
2714
+ "epoch": 0.1,
2715
+ "grad_norm": 31.5,
2716
+ "learning_rate": 1.771525423728814e-05,
2717
+ "loss": 1.2996,
2718
+ "step": 3870
2719
+ },
2720
+ {
2721
+ "epoch": 0.1,
2722
+ "grad_norm": 18.5,
2723
+ "learning_rate": 1.770847457627119e-05,
2724
+ "loss": 1.5404,
2725
+ "step": 3880
2726
+ },
2727
+ {
2728
+ "epoch": 0.1,
2729
+ "grad_norm": 16.25,
2730
+ "learning_rate": 1.770169491525424e-05,
2731
+ "loss": 1.4538,
2732
+ "step": 3890
2733
+ },
2734
+ {
2735
+ "epoch": 0.1,
2736
+ "grad_norm": 9.6875,
2737
+ "learning_rate": 1.769491525423729e-05,
2738
+ "loss": 1.5769,
2739
+ "step": 3900
2740
+ },
2741
+ {
2742
+ "epoch": 0.1,
2743
+ "grad_norm": 19.375,
2744
+ "learning_rate": 1.768813559322034e-05,
2745
+ "loss": 1.4964,
2746
+ "step": 3910
2747
+ },
2748
+ {
2749
+ "epoch": 0.1,
2750
+ "grad_norm": 10.6875,
2751
+ "learning_rate": 1.768135593220339e-05,
2752
+ "loss": 1.5531,
2753
+ "step": 3920
2754
+ },
2755
+ {
2756
+ "epoch": 0.1,
2757
+ "grad_norm": 8.4375,
2758
+ "learning_rate": 1.767457627118644e-05,
2759
+ "loss": 1.561,
2760
+ "step": 3930
2761
+ },
2762
+ {
2763
+ "epoch": 0.1,
2764
+ "grad_norm": 12.5,
2765
+ "learning_rate": 1.7667796610169494e-05,
2766
+ "loss": 1.5124,
2767
+ "step": 3940
2768
+ },
2769
+ {
2770
+ "epoch": 0.1,
2771
+ "grad_norm": 14.9375,
2772
+ "learning_rate": 1.7661016949152543e-05,
2773
+ "loss": 1.396,
2774
+ "step": 3950
2775
+ },
2776
+ {
2777
+ "epoch": 0.1,
2778
+ "grad_norm": 6.84375,
2779
+ "learning_rate": 1.7654237288135593e-05,
2780
+ "loss": 1.507,
2781
+ "step": 3960
2782
+ },
2783
+ {
2784
+ "epoch": 0.1,
2785
+ "grad_norm": 9.4375,
2786
+ "learning_rate": 1.7647457627118646e-05,
2787
+ "loss": 1.5669,
2788
+ "step": 3970
2789
+ },
2790
+ {
2791
+ "epoch": 0.1,
2792
+ "grad_norm": 9.5625,
2793
+ "learning_rate": 1.7640677966101696e-05,
2794
+ "loss": 1.3942,
2795
+ "step": 3980
2796
+ },
2797
+ {
2798
+ "epoch": 0.1,
2799
+ "grad_norm": 19.5,
2800
+ "learning_rate": 1.7633898305084746e-05,
2801
+ "loss": 1.4645,
2802
+ "step": 3990
2803
+ },
2804
+ {
2805
+ "epoch": 0.1,
2806
+ "grad_norm": 18.375,
2807
+ "learning_rate": 1.76271186440678e-05,
2808
+ "loss": 1.3539,
2809
+ "step": 4000
2810
+ },
2811
+ {
2812
+ "epoch": 0.1,
2813
+ "grad_norm": 28.0,
2814
+ "learning_rate": 1.762033898305085e-05,
2815
+ "loss": 1.5859,
2816
+ "step": 4010
2817
+ },
2818
+ {
2819
+ "epoch": 0.1,
2820
+ "grad_norm": 11.1875,
2821
+ "learning_rate": 1.7613559322033898e-05,
2822
+ "loss": 1.3802,
2823
+ "step": 4020
2824
+ },
2825
+ {
2826
+ "epoch": 0.1,
2827
+ "grad_norm": 21.875,
2828
+ "learning_rate": 1.760677966101695e-05,
2829
+ "loss": 1.3353,
2830
+ "step": 4030
2831
+ },
2832
+ {
2833
+ "epoch": 0.1,
2834
+ "grad_norm": 20.75,
2835
+ "learning_rate": 1.76e-05,
2836
+ "loss": 1.5717,
2837
+ "step": 4040
2838
+ },
2839
+ {
2840
+ "epoch": 0.1,
2841
+ "grad_norm": 11.0,
2842
+ "learning_rate": 1.7593220338983054e-05,
2843
+ "loss": 1.5115,
2844
+ "step": 4050
2845
+ },
2846
+ {
2847
+ "epoch": 0.1,
2848
+ "grad_norm": 22.0,
2849
+ "learning_rate": 1.7586440677966104e-05,
2850
+ "loss": 1.4051,
2851
+ "step": 4060
2852
+ },
2853
+ {
2854
+ "epoch": 0.1,
2855
+ "grad_norm": 14.9375,
2856
+ "learning_rate": 1.7579661016949153e-05,
2857
+ "loss": 1.6576,
2858
+ "step": 4070
2859
+ },
2860
+ {
2861
+ "epoch": 0.1,
2862
+ "grad_norm": 11.25,
2863
+ "learning_rate": 1.7572881355932206e-05,
2864
+ "loss": 1.6336,
2865
+ "step": 4080
2866
+ },
2867
+ {
2868
+ "epoch": 0.1,
2869
+ "grad_norm": 9.75,
2870
+ "learning_rate": 1.7566101694915256e-05,
2871
+ "loss": 1.3511,
2872
+ "step": 4090
2873
+ },
2874
+ {
2875
+ "epoch": 0.1,
2876
+ "grad_norm": 13.6875,
2877
+ "learning_rate": 1.7559322033898306e-05,
2878
+ "loss": 1.4982,
2879
+ "step": 4100
2880
+ },
2881
+ {
2882
+ "epoch": 0.1,
2883
+ "grad_norm": 25.625,
2884
+ "learning_rate": 1.755254237288136e-05,
2885
+ "loss": 1.2387,
2886
+ "step": 4110
2887
+ },
2888
+ {
2889
+ "epoch": 0.1,
2890
+ "grad_norm": 13.3125,
2891
+ "learning_rate": 1.754576271186441e-05,
2892
+ "loss": 1.7004,
2893
+ "step": 4120
2894
+ },
2895
+ {
2896
+ "epoch": 0.1,
2897
+ "grad_norm": 15.875,
2898
+ "learning_rate": 1.753898305084746e-05,
2899
+ "loss": 1.5061,
2900
+ "step": 4130
2901
+ },
2902
+ {
2903
+ "epoch": 0.1,
2904
+ "grad_norm": 15.25,
2905
+ "learning_rate": 1.7532203389830508e-05,
2906
+ "loss": 1.5026,
2907
+ "step": 4140
2908
+ },
2909
+ {
2910
+ "epoch": 0.1,
2911
+ "grad_norm": 6.96875,
2912
+ "learning_rate": 1.752542372881356e-05,
2913
+ "loss": 1.5616,
2914
+ "step": 4150
2915
+ },
2916
+ {
2917
+ "epoch": 0.1,
2918
+ "grad_norm": 11.25,
2919
+ "learning_rate": 1.751864406779661e-05,
2920
+ "loss": 1.5947,
2921
+ "step": 4160
2922
+ },
2923
+ {
2924
+ "epoch": 0.1,
2925
+ "grad_norm": 12.9375,
2926
+ "learning_rate": 1.751186440677966e-05,
2927
+ "loss": 1.5071,
2928
+ "step": 4170
2929
+ },
2930
+ {
2931
+ "epoch": 0.1,
2932
+ "grad_norm": 13.6875,
2933
+ "learning_rate": 1.7505084745762714e-05,
2934
+ "loss": 1.3603,
2935
+ "step": 4180
2936
+ },
2937
+ {
2938
+ "epoch": 0.1,
2939
+ "grad_norm": 27.0,
2940
+ "learning_rate": 1.7498305084745763e-05,
2941
+ "loss": 1.413,
2942
+ "step": 4190
2943
+ },
2944
+ {
2945
+ "epoch": 0.1,
2946
+ "grad_norm": 15.5625,
2947
+ "learning_rate": 1.7491525423728813e-05,
2948
+ "loss": 1.5441,
2949
+ "step": 4200
2950
+ },
2951
+ {
2952
+ "epoch": 0.11,
2953
+ "grad_norm": 7.65625,
2954
+ "learning_rate": 1.7484745762711866e-05,
2955
+ "loss": 1.4837,
2956
+ "step": 4210
2957
+ },
2958
+ {
2959
+ "epoch": 0.11,
2960
+ "grad_norm": 18.375,
2961
+ "learning_rate": 1.7477966101694916e-05,
2962
+ "loss": 1.4118,
2963
+ "step": 4220
2964
+ },
2965
+ {
2966
+ "epoch": 0.11,
2967
+ "grad_norm": 26.5,
2968
+ "learning_rate": 1.747118644067797e-05,
2969
+ "loss": 1.3557,
2970
+ "step": 4230
2971
+ },
2972
+ {
2973
+ "epoch": 0.11,
2974
+ "grad_norm": 20.875,
2975
+ "learning_rate": 1.746440677966102e-05,
2976
+ "loss": 1.3283,
2977
+ "step": 4240
2978
+ },
2979
+ {
2980
+ "epoch": 0.11,
2981
+ "grad_norm": 13.6875,
2982
+ "learning_rate": 1.745762711864407e-05,
2983
+ "loss": 1.5952,
2984
+ "step": 4250
2985
+ },
2986
+ {
2987
+ "epoch": 0.11,
2988
+ "grad_norm": 14.4375,
2989
+ "learning_rate": 1.745084745762712e-05,
2990
+ "loss": 1.4947,
2991
+ "step": 4260
2992
+ },
2993
+ {
2994
+ "epoch": 0.11,
2995
+ "grad_norm": 10.6875,
2996
+ "learning_rate": 1.744406779661017e-05,
2997
+ "loss": 1.5308,
2998
+ "step": 4270
2999
+ },
3000
+ {
3001
+ "epoch": 0.11,
3002
+ "grad_norm": 11.75,
3003
+ "learning_rate": 1.743728813559322e-05,
3004
+ "loss": 1.5157,
3005
+ "step": 4280
3006
+ },
3007
+ {
3008
+ "epoch": 0.11,
3009
+ "grad_norm": 14.5625,
3010
+ "learning_rate": 1.7430508474576274e-05,
3011
+ "loss": 1.5048,
3012
+ "step": 4290
3013
+ },
3014
+ {
3015
+ "epoch": 0.11,
3016
+ "grad_norm": 16.75,
3017
+ "learning_rate": 1.7423728813559324e-05,
3018
+ "loss": 1.4284,
3019
+ "step": 4300
3020
+ },
3021
+ {
3022
+ "epoch": 0.11,
3023
+ "grad_norm": 9.3125,
3024
+ "learning_rate": 1.7416949152542373e-05,
3025
+ "loss": 1.5088,
3026
+ "step": 4310
3027
+ },
3028
+ {
3029
+ "epoch": 0.11,
3030
+ "grad_norm": 15.8125,
3031
+ "learning_rate": 1.7410169491525427e-05,
3032
+ "loss": 1.5715,
3033
+ "step": 4320
3034
+ },
3035
+ {
3036
+ "epoch": 0.11,
3037
+ "grad_norm": 12.4375,
3038
+ "learning_rate": 1.7403389830508476e-05,
3039
+ "loss": 1.441,
3040
+ "step": 4330
3041
+ },
3042
+ {
3043
+ "epoch": 0.11,
3044
+ "grad_norm": 11.125,
3045
+ "learning_rate": 1.739661016949153e-05,
3046
+ "loss": 1.3917,
3047
+ "step": 4340
3048
+ },
3049
+ {
3050
+ "epoch": 0.11,
3051
+ "grad_norm": 22.25,
3052
+ "learning_rate": 1.7389830508474576e-05,
3053
+ "loss": 1.354,
3054
+ "step": 4350
3055
+ },
3056
+ {
3057
+ "epoch": 0.11,
3058
+ "grad_norm": 9.1875,
3059
+ "learning_rate": 1.738305084745763e-05,
3060
+ "loss": 1.5789,
3061
+ "step": 4360
3062
+ },
3063
+ {
3064
+ "epoch": 0.11,
3065
+ "grad_norm": 8.8125,
3066
+ "learning_rate": 1.737627118644068e-05,
3067
+ "loss": 1.5641,
3068
+ "step": 4370
3069
+ },
3070
+ {
3071
+ "epoch": 0.11,
3072
+ "grad_norm": 23.125,
3073
+ "learning_rate": 1.7369491525423728e-05,
3074
+ "loss": 1.4324,
3075
+ "step": 4380
3076
+ },
3077
+ {
3078
+ "epoch": 0.11,
3079
+ "grad_norm": 27.25,
3080
+ "learning_rate": 1.736271186440678e-05,
3081
+ "loss": 1.6862,
3082
+ "step": 4390
3083
+ },
3084
+ {
3085
+ "epoch": 0.11,
3086
+ "grad_norm": 10.4375,
3087
+ "learning_rate": 1.735593220338983e-05,
3088
+ "loss": 1.5332,
3089
+ "step": 4400
3090
+ },
3091
+ {
3092
+ "epoch": 0.11,
3093
+ "grad_norm": 14.5625,
3094
+ "learning_rate": 1.734915254237288e-05,
3095
+ "loss": 1.4554,
3096
+ "step": 4410
3097
+ },
3098
+ {
3099
+ "epoch": 0.11,
3100
+ "grad_norm": 22.125,
3101
+ "learning_rate": 1.7342372881355934e-05,
3102
+ "loss": 1.5159,
3103
+ "step": 4420
3104
+ },
3105
+ {
3106
+ "epoch": 0.11,
3107
+ "grad_norm": 26.625,
3108
+ "learning_rate": 1.7335593220338983e-05,
3109
+ "loss": 1.4236,
3110
+ "step": 4430
3111
+ },
3112
+ {
3113
+ "epoch": 0.11,
3114
+ "grad_norm": 9.625,
3115
+ "learning_rate": 1.7328813559322037e-05,
3116
+ "loss": 1.5137,
3117
+ "step": 4440
3118
+ },
3119
+ {
3120
+ "epoch": 0.11,
3121
+ "grad_norm": 16.75,
3122
+ "learning_rate": 1.7322033898305086e-05,
3123
+ "loss": 1.551,
3124
+ "step": 4450
3125
+ },
3126
+ {
3127
+ "epoch": 0.11,
3128
+ "grad_norm": 11.1875,
3129
+ "learning_rate": 1.7315254237288136e-05,
3130
+ "loss": 1.5243,
3131
+ "step": 4460
3132
+ },
3133
+ {
3134
+ "epoch": 0.11,
3135
+ "grad_norm": 19.375,
3136
+ "learning_rate": 1.730847457627119e-05,
3137
+ "loss": 1.3952,
3138
+ "step": 4470
3139
+ },
3140
+ {
3141
+ "epoch": 0.11,
3142
+ "grad_norm": 10.75,
3143
+ "learning_rate": 1.730169491525424e-05,
3144
+ "loss": 1.5068,
3145
+ "step": 4480
3146
+ },
3147
+ {
3148
+ "epoch": 0.11,
3149
+ "grad_norm": 12.75,
3150
+ "learning_rate": 1.729491525423729e-05,
3151
+ "loss": 1.5121,
3152
+ "step": 4490
3153
+ },
3154
+ {
3155
+ "epoch": 0.11,
3156
+ "grad_norm": 12.1875,
3157
+ "learning_rate": 1.728813559322034e-05,
3158
+ "loss": 1.5624,
3159
+ "step": 4500
3160
+ },
3161
+ {
3162
+ "epoch": 0.11,
3163
+ "grad_norm": 5.5625,
3164
+ "learning_rate": 1.728135593220339e-05,
3165
+ "loss": 1.5784,
3166
+ "step": 4510
3167
+ },
3168
+ {
3169
+ "epoch": 0.11,
3170
+ "grad_norm": 20.625,
3171
+ "learning_rate": 1.7274576271186444e-05,
3172
+ "loss": 1.5975,
3173
+ "step": 4520
3174
+ },
3175
+ {
3176
+ "epoch": 0.11,
3177
+ "grad_norm": 16.5,
3178
+ "learning_rate": 1.7267796610169494e-05,
3179
+ "loss": 1.5058,
3180
+ "step": 4530
3181
+ },
3182
+ {
3183
+ "epoch": 0.11,
3184
+ "grad_norm": 23.125,
3185
+ "learning_rate": 1.7261016949152544e-05,
3186
+ "loss": 1.5025,
3187
+ "step": 4540
3188
+ },
3189
+ {
3190
+ "epoch": 0.11,
3191
+ "grad_norm": 8.1875,
3192
+ "learning_rate": 1.7254237288135597e-05,
3193
+ "loss": 1.5016,
3194
+ "step": 4550
3195
+ },
3196
+ {
3197
+ "epoch": 0.11,
3198
+ "grad_norm": 18.75,
3199
+ "learning_rate": 1.7247457627118643e-05,
3200
+ "loss": 1.6883,
3201
+ "step": 4560
3202
+ },
3203
+ {
3204
+ "epoch": 0.11,
3205
+ "grad_norm": 10.4375,
3206
+ "learning_rate": 1.7240677966101696e-05,
3207
+ "loss": 1.3446,
3208
+ "step": 4570
3209
+ },
3210
+ {
3211
+ "epoch": 0.11,
3212
+ "grad_norm": 10.6875,
3213
+ "learning_rate": 1.7233898305084746e-05,
3214
+ "loss": 1.4538,
3215
+ "step": 4580
3216
+ },
3217
+ {
3218
+ "epoch": 0.11,
3219
+ "grad_norm": 24.5,
3220
+ "learning_rate": 1.7227118644067796e-05,
3221
+ "loss": 1.6639,
3222
+ "step": 4590
3223
+ },
3224
+ {
3225
+ "epoch": 0.12,
3226
+ "grad_norm": 9.125,
3227
+ "learning_rate": 1.722033898305085e-05,
3228
+ "loss": 1.4906,
3229
+ "step": 4600
3230
+ },
3231
+ {
3232
+ "epoch": 0.12,
3233
+ "grad_norm": 6.53125,
3234
+ "learning_rate": 1.72135593220339e-05,
3235
+ "loss": 1.4948,
3236
+ "step": 4610
3237
+ },
3238
+ {
3239
+ "epoch": 0.12,
3240
+ "grad_norm": 11.875,
3241
+ "learning_rate": 1.720677966101695e-05,
3242
+ "loss": 1.4362,
3243
+ "step": 4620
3244
+ },
3245
+ {
3246
+ "epoch": 0.12,
3247
+ "grad_norm": 15.4375,
3248
+ "learning_rate": 1.72e-05,
3249
+ "loss": 1.5428,
3250
+ "step": 4630
3251
+ },
3252
+ {
3253
+ "epoch": 0.12,
3254
+ "grad_norm": 8.75,
3255
+ "learning_rate": 1.719322033898305e-05,
3256
+ "loss": 1.2071,
3257
+ "step": 4640
3258
+ },
3259
+ {
3260
+ "epoch": 0.12,
3261
+ "grad_norm": 21.125,
3262
+ "learning_rate": 1.7186440677966104e-05,
3263
+ "loss": 1.5102,
3264
+ "step": 4650
3265
+ },
3266
+ {
3267
+ "epoch": 0.12,
3268
+ "grad_norm": 13.8125,
3269
+ "learning_rate": 1.7179661016949154e-05,
3270
+ "loss": 1.3775,
3271
+ "step": 4660
3272
+ },
3273
+ {
3274
+ "epoch": 0.12,
3275
+ "grad_norm": 18.5,
3276
+ "learning_rate": 1.7172881355932204e-05,
3277
+ "loss": 1.4434,
3278
+ "step": 4670
3279
+ },
3280
+ {
3281
+ "epoch": 0.12,
3282
+ "grad_norm": 12.0625,
3283
+ "learning_rate": 1.7166101694915257e-05,
3284
+ "loss": 1.5176,
3285
+ "step": 4680
3286
+ },
3287
+ {
3288
+ "epoch": 0.12,
3289
+ "grad_norm": 15.875,
3290
+ "learning_rate": 1.7159322033898306e-05,
3291
+ "loss": 1.4333,
3292
+ "step": 4690
3293
+ },
3294
+ {
3295
+ "epoch": 0.12,
3296
+ "grad_norm": 18.375,
3297
+ "learning_rate": 1.715254237288136e-05,
3298
+ "loss": 1.4007,
3299
+ "step": 4700
3300
+ },
3301
+ {
3302
+ "epoch": 0.12,
3303
+ "grad_norm": 5.8125,
3304
+ "learning_rate": 1.714576271186441e-05,
3305
+ "loss": 1.5271,
3306
+ "step": 4710
3307
+ },
3308
+ {
3309
+ "epoch": 0.12,
3310
+ "grad_norm": 18.875,
3311
+ "learning_rate": 1.713898305084746e-05,
3312
+ "loss": 1.5802,
3313
+ "step": 4720
3314
+ },
3315
+ {
3316
+ "epoch": 0.12,
3317
+ "grad_norm": 19.625,
3318
+ "learning_rate": 1.7132203389830512e-05,
3319
+ "loss": 1.684,
3320
+ "step": 4730
3321
+ },
3322
+ {
3323
+ "epoch": 0.12,
3324
+ "grad_norm": 17.5,
3325
+ "learning_rate": 1.712542372881356e-05,
3326
+ "loss": 1.3561,
3327
+ "step": 4740
3328
+ },
3329
+ {
3330
+ "epoch": 0.12,
3331
+ "grad_norm": 9.125,
3332
+ "learning_rate": 1.711864406779661e-05,
3333
+ "loss": 1.4918,
3334
+ "step": 4750
3335
+ },
3336
+ {
3337
+ "epoch": 0.12,
3338
+ "grad_norm": 20.0,
3339
+ "learning_rate": 1.711186440677966e-05,
3340
+ "loss": 1.6214,
3341
+ "step": 4760
3342
+ },
3343
+ {
3344
+ "epoch": 0.12,
3345
+ "grad_norm": 18.5,
3346
+ "learning_rate": 1.710508474576271e-05,
3347
+ "loss": 1.5829,
3348
+ "step": 4770
3349
+ },
3350
+ {
3351
+ "epoch": 0.12,
3352
+ "grad_norm": 18.25,
3353
+ "learning_rate": 1.7098305084745764e-05,
3354
+ "loss": 1.3968,
3355
+ "step": 4780
3356
+ },
3357
+ {
3358
+ "epoch": 0.12,
3359
+ "grad_norm": 42.5,
3360
+ "learning_rate": 1.7091525423728814e-05,
3361
+ "loss": 1.4853,
3362
+ "step": 4790
3363
+ },
3364
+ {
3365
+ "epoch": 0.12,
3366
+ "grad_norm": 8.4375,
3367
+ "learning_rate": 1.7084745762711867e-05,
3368
+ "loss": 1.4132,
3369
+ "step": 4800
3370
+ },
3371
+ {
3372
+ "epoch": 0.12,
3373
+ "grad_norm": 10.4375,
3374
+ "learning_rate": 1.7077966101694916e-05,
3375
+ "loss": 1.3807,
3376
+ "step": 4810
3377
+ },
3378
+ {
3379
+ "epoch": 0.12,
3380
+ "grad_norm": 13.1875,
3381
+ "learning_rate": 1.7071186440677966e-05,
3382
+ "loss": 1.426,
3383
+ "step": 4820
3384
+ },
3385
+ {
3386
+ "epoch": 0.12,
3387
+ "grad_norm": 12.875,
3388
+ "learning_rate": 1.706440677966102e-05,
3389
+ "loss": 1.5588,
3390
+ "step": 4830
3391
+ },
3392
+ {
3393
+ "epoch": 0.12,
3394
+ "grad_norm": 8.875,
3395
+ "learning_rate": 1.705762711864407e-05,
3396
+ "loss": 1.5123,
3397
+ "step": 4840
3398
+ },
3399
+ {
3400
+ "epoch": 0.12,
3401
+ "grad_norm": 18.125,
3402
+ "learning_rate": 1.705084745762712e-05,
3403
+ "loss": 1.3166,
3404
+ "step": 4850
3405
+ },
3406
+ {
3407
+ "epoch": 0.12,
3408
+ "grad_norm": 12.5625,
3409
+ "learning_rate": 1.704406779661017e-05,
3410
+ "loss": 1.5191,
3411
+ "step": 4860
3412
+ },
3413
+ {
3414
+ "epoch": 0.12,
3415
+ "grad_norm": 14.75,
3416
+ "learning_rate": 1.703728813559322e-05,
3417
+ "loss": 1.4739,
3418
+ "step": 4870
3419
+ },
3420
+ {
3421
+ "epoch": 0.12,
3422
+ "grad_norm": 22.875,
3423
+ "learning_rate": 1.703050847457627e-05,
3424
+ "loss": 1.5437,
3425
+ "step": 4880
3426
+ },
3427
+ {
3428
+ "epoch": 0.12,
3429
+ "grad_norm": 14.375,
3430
+ "learning_rate": 1.7023728813559324e-05,
3431
+ "loss": 1.501,
3432
+ "step": 4890
3433
+ },
3434
+ {
3435
+ "epoch": 0.12,
3436
+ "grad_norm": 14.125,
3437
+ "learning_rate": 1.7016949152542374e-05,
3438
+ "loss": 1.6071,
3439
+ "step": 4900
3440
+ },
3441
+ {
3442
+ "epoch": 0.12,
3443
+ "grad_norm": 13.5,
3444
+ "learning_rate": 1.7010169491525427e-05,
3445
+ "loss": 1.4153,
3446
+ "step": 4910
3447
+ },
3448
+ {
3449
+ "epoch": 0.12,
3450
+ "grad_norm": 15.875,
3451
+ "learning_rate": 1.7003389830508477e-05,
3452
+ "loss": 1.5077,
3453
+ "step": 4920
3454
+ },
3455
+ {
3456
+ "epoch": 0.12,
3457
+ "grad_norm": 18.75,
3458
+ "learning_rate": 1.6996610169491526e-05,
3459
+ "loss": 1.4882,
3460
+ "step": 4930
3461
+ },
3462
+ {
3463
+ "epoch": 0.12,
3464
+ "grad_norm": 12.5625,
3465
+ "learning_rate": 1.698983050847458e-05,
3466
+ "loss": 1.3678,
3467
+ "step": 4940
3468
+ },
3469
+ {
3470
+ "epoch": 0.12,
3471
+ "grad_norm": 8.25,
3472
+ "learning_rate": 1.698305084745763e-05,
3473
+ "loss": 1.4071,
3474
+ "step": 4950
3475
+ },
3476
+ {
3477
+ "epoch": 0.12,
3478
+ "grad_norm": 13.125,
3479
+ "learning_rate": 1.697627118644068e-05,
3480
+ "loss": 1.5199,
3481
+ "step": 4960
3482
+ },
3483
+ {
3484
+ "epoch": 0.12,
3485
+ "grad_norm": 23.5,
3486
+ "learning_rate": 1.696949152542373e-05,
3487
+ "loss": 1.484,
3488
+ "step": 4970
3489
+ },
3490
+ {
3491
+ "epoch": 0.12,
3492
+ "grad_norm": 15.4375,
3493
+ "learning_rate": 1.6962711864406782e-05,
3494
+ "loss": 1.4716,
3495
+ "step": 4980
3496
+ },
3497
+ {
3498
+ "epoch": 0.12,
3499
+ "grad_norm": 18.125,
3500
+ "learning_rate": 1.695593220338983e-05,
3501
+ "loss": 1.4641,
3502
+ "step": 4990
3503
+ },
3504
+ {
3505
+ "epoch": 0.12,
3506
+ "grad_norm": 10.75,
3507
+ "learning_rate": 1.694915254237288e-05,
3508
+ "loss": 1.3656,
3509
+ "step": 5000
3510
+ },
3511
+ {
3512
+ "epoch": 0.12,
3513
+ "eval_loss": 1.4290132522583008,
3514
+ "eval_runtime": 59.1678,
3515
+ "eval_samples_per_second": 16.901,
3516
+ "eval_steps_per_second": 16.901,
3517
+ "step": 5000
3518
+ }
3519
+ ],
3520
+ "logging_steps": 10,
3521
+ "max_steps": 30000,
3522
+ "num_input_tokens_seen": 0,
3523
+ "num_train_epochs": 1,
3524
+ "save_steps": 5000,
3525
+ "total_flos": 7.866849165312e+16,
3526
+ "train_batch_size": 1,
3527
+ "trial_name": null,
3528
+ "trial_params": null
3529
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7820755b7e93a89875274fe688fed0ee277c7ddf3988ab6d83c4abe4d5d9b438
3
+ size 4856