Hugging Face's logo

CMU-AIR2
/
math-phi-1-5-FULL-Arithmetic-4k-MWP-6k

Text Generation
Transformers
Safetensors
phi
text-generation-inference
Inference Endpoints
Model card Files Community
UltimoUno commited on
Commit
e478d83
•
1 Parent(s): 70ccce0

Uploaded checkpoint-6000

Browse files
Files changed (8) hide show
  1. config.json +30 -0
  2. generation_config.json +4 -0
  3. model.safetensors +3 -0
  4. optimizer.pt +3 -0
  5. rng_state.pth +3 -0
  6. scheduler.pt +3 -0
  7. trainer_state.json +4245 -0
  8. training_args.bin +3 -0
config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "CMU-AIR2/math-phi-1-5-FULL-Arithmetic-4k",
3
+ "architectures": [
4
+ "PhiForCausalLM"
5
+ ],
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": null,
8
+ "embd_pdrop": 0.0,
9
+ "eos_token_id": null,
10
+ "hidden_act": "gelu_new",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 8192,
14
+ "layer_norm_eps": 1e-05,
15
+ "max_position_embeddings": 2048,
16
+ "model_type": "phi",
17
+ "num_attention_heads": 32,
18
+ "num_hidden_layers": 24,
19
+ "num_key_value_heads": 32,
20
+ "partial_rotary_factor": 0.5,
21
+ "qk_layernorm": false,
22
+ "resid_pdrop": 0.0,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": false,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.38.2",
28
+ "use_cache": true,
29
+ "vocab_size": 51200
30
+ }
generation_config.json ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "transformers_version": "4.38.2"
4
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f3f0917f22378d500d93480cb10d969827f5396c9357952ab7fc22bd3aa14d56
3
+ size 2836579040
optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:130511d831cd584f958344fba5e4cb7a9e26a41e631a2379506be094359ac3f3
3
+ size 5673376169
rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ac8af3b52ac2d9369a13333b931634a9039769d2792e069f75c077dd9ed80f0
3
+ size 14244
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fb25ce47f6518a30396a3e2b69291ef0a8952319b40bf48c48e3e42a2b8686bb
3
+ size 1064
trainer_state.json ADDED
@@ -0,0 +1,4245 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 2.151694459386767,
5
+ "eval_steps": 2000,
6
+ "global_step": 6000,
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": 5.90625,
14
+ "learning_rate": 3.0000000000000004e-07,
15
+ "loss": 1.4893,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.01,
20
+ "grad_norm": 10.0625,
21
+ "learning_rate": 6.000000000000001e-07,
22
+ "loss": 1.6671,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.01,
27
+ "grad_norm": 7.03125,
28
+ "learning_rate": 9e-07,
29
+ "loss": 1.3866,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.01,
34
+ "grad_norm": 5.9375,
35
+ "learning_rate": 1.2000000000000002e-06,
36
+ "loss": 1.433,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.02,
41
+ "grad_norm": 8.1875,
42
+ "learning_rate": 1.5e-06,
43
+ "loss": 1.4135,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.02,
48
+ "grad_norm": 8.375,
49
+ "learning_rate": 1.8e-06,
50
+ "loss": 1.3779,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.03,
55
+ "grad_norm": 7.15625,
56
+ "learning_rate": 2.1000000000000002e-06,
57
+ "loss": 1.4444,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.03,
62
+ "grad_norm": 7.71875,
63
+ "learning_rate": 2.4000000000000003e-06,
64
+ "loss": 1.3717,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.03,
69
+ "grad_norm": 6.65625,
70
+ "learning_rate": 2.7e-06,
71
+ "loss": 1.3646,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.04,
76
+ "grad_norm": 6.03125,
77
+ "learning_rate": 3e-06,
78
+ "loss": 1.4607,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.04,
83
+ "grad_norm": 7.125,
84
+ "learning_rate": 3.3e-06,
85
+ "loss": 1.509,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.04,
90
+ "grad_norm": 6.90625,
91
+ "learning_rate": 3.6e-06,
92
+ "loss": 1.2552,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.05,
97
+ "grad_norm": 7.78125,
98
+ "learning_rate": 3.9e-06,
99
+ "loss": 1.3142,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.05,
104
+ "grad_norm": 6.8125,
105
+ "learning_rate": 4.2000000000000004e-06,
106
+ "loss": 1.3774,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.05,
111
+ "grad_norm": 8.375,
112
+ "learning_rate": 4.5e-06,
113
+ "loss": 1.3869,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.06,
118
+ "grad_norm": 5.1875,
119
+ "learning_rate": 4.800000000000001e-06,
120
+ "loss": 1.2843,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.06,
125
+ "grad_norm": 5.71875,
126
+ "learning_rate": 5.1e-06,
127
+ "loss": 1.3458,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.06,
132
+ "grad_norm": 6.21875,
133
+ "learning_rate": 5.4e-06,
134
+ "loss": 1.2259,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.07,
139
+ "grad_norm": 5.71875,
140
+ "learning_rate": 5.7000000000000005e-06,
141
+ "loss": 1.1647,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.07,
146
+ "grad_norm": 5.875,
147
+ "learning_rate": 6e-06,
148
+ "loss": 1.0721,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.08,
153
+ "grad_norm": 5.09375,
154
+ "learning_rate": 6.3e-06,
155
+ "loss": 1.0382,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.08,
160
+ "grad_norm": 5.1875,
161
+ "learning_rate": 6.6e-06,
162
+ "loss": 1.1419,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.08,
167
+ "grad_norm": 5.1875,
168
+ "learning_rate": 6.900000000000001e-06,
169
+ "loss": 1.0656,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.09,
174
+ "grad_norm": 4.8125,
175
+ "learning_rate": 7.2e-06,
176
+ "loss": 1.1806,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.09,
181
+ "grad_norm": 3.84375,
182
+ "learning_rate": 7.5e-06,
183
+ "loss": 0.9791,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.09,
188
+ "grad_norm": 6.5625,
189
+ "learning_rate": 7.8e-06,
190
+ "loss": 1.1894,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.1,
195
+ "grad_norm": 3.65625,
196
+ "learning_rate": 8.1e-06,
197
+ "loss": 0.91,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.1,
202
+ "grad_norm": 4.90625,
203
+ "learning_rate": 8.400000000000001e-06,
204
+ "loss": 1.0708,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.1,
209
+ "grad_norm": 6.53125,
210
+ "learning_rate": 8.7e-06,
211
+ "loss": 0.9747,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.11,
216
+ "grad_norm": 4.28125,
217
+ "learning_rate": 9e-06,
218
+ "loss": 1.0158,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.11,
223
+ "grad_norm": 4.34375,
224
+ "learning_rate": 9.3e-06,
225
+ "loss": 1.1078,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.11,
230
+ "grad_norm": 3.171875,
231
+ "learning_rate": 9.600000000000001e-06,
232
+ "loss": 1.0437,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.12,
237
+ "grad_norm": 4.25,
238
+ "learning_rate": 9.9e-06,
239
+ "loss": 0.9891,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.12,
244
+ "grad_norm": 4.875,
245
+ "learning_rate": 1.02e-05,
246
+ "loss": 0.8667,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.13,
251
+ "grad_norm": 3.703125,
252
+ "learning_rate": 1.05e-05,
253
+ "loss": 1.0903,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.13,
258
+ "grad_norm": 3.390625,
259
+ "learning_rate": 1.08e-05,
260
+ "loss": 0.9947,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.13,
265
+ "grad_norm": 5.6875,
266
+ "learning_rate": 1.11e-05,
267
+ "loss": 1.043,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.14,
272
+ "grad_norm": 3.03125,
273
+ "learning_rate": 1.1400000000000001e-05,
274
+ "loss": 0.8305,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.14,
279
+ "grad_norm": 6.0625,
280
+ "learning_rate": 1.1700000000000001e-05,
281
+ "loss": 0.9741,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.14,
286
+ "grad_norm": 4.34375,
287
+ "learning_rate": 1.2e-05,
288
+ "loss": 1.219,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.15,
293
+ "grad_norm": 5.53125,
294
+ "learning_rate": 1.2299999999999999e-05,
295
+ "loss": 1.0167,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.15,
300
+ "grad_norm": 3.578125,
301
+ "learning_rate": 1.26e-05,
302
+ "loss": 0.87,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.15,
307
+ "grad_norm": 5.59375,
308
+ "learning_rate": 1.29e-05,
309
+ "loss": 1.0169,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.16,
314
+ "grad_norm": 3.453125,
315
+ "learning_rate": 1.32e-05,
316
+ "loss": 0.9178,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.16,
321
+ "grad_norm": 4.21875,
322
+ "learning_rate": 1.3500000000000001e-05,
323
+ "loss": 1.0044,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.16,
328
+ "grad_norm": 4.875,
329
+ "learning_rate": 1.3800000000000002e-05,
330
+ "loss": 0.897,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.17,
335
+ "grad_norm": 3.5,
336
+ "learning_rate": 1.4099999999999999e-05,
337
+ "loss": 0.8968,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.17,
342
+ "grad_norm": 5.15625,
343
+ "learning_rate": 1.44e-05,
344
+ "loss": 0.9446,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.18,
349
+ "grad_norm": 2.375,
350
+ "learning_rate": 1.47e-05,
351
+ "loss": 0.8887,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.18,
356
+ "grad_norm": 3.84375,
357
+ "learning_rate": 1.5e-05,
358
+ "loss": 1.0239,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.18,
363
+ "grad_norm": 2.390625,
364
+ "learning_rate": 1.4984210526315789e-05,
365
+ "loss": 0.9251,
366
+ "step": 510
367
+ },
368
+ {
369
+ "epoch": 0.19,
370
+ "grad_norm": 4.9375,
371
+ "learning_rate": 1.496842105263158e-05,
372
+ "loss": 0.9681,
373
+ "step": 520
374
+ },
375
+ {
376
+ "epoch": 0.19,
377
+ "grad_norm": 4.6875,
378
+ "learning_rate": 1.4952631578947368e-05,
379
+ "loss": 0.8625,
380
+ "step": 530
381
+ },
382
+ {
383
+ "epoch": 0.19,
384
+ "grad_norm": 3.734375,
385
+ "learning_rate": 1.4936842105263158e-05,
386
+ "loss": 0.9131,
387
+ "step": 540
388
+ },
389
+ {
390
+ "epoch": 0.2,
391
+ "grad_norm": 3.1875,
392
+ "learning_rate": 1.4921052631578947e-05,
393
+ "loss": 0.8903,
394
+ "step": 550
395
+ },
396
+ {
397
+ "epoch": 0.2,
398
+ "grad_norm": 4.09375,
399
+ "learning_rate": 1.4905263157894737e-05,
400
+ "loss": 0.9482,
401
+ "step": 560
402
+ },
403
+ {
404
+ "epoch": 0.2,
405
+ "grad_norm": 2.921875,
406
+ "learning_rate": 1.4889473684210526e-05,
407
+ "loss": 0.8134,
408
+ "step": 570
409
+ },
410
+ {
411
+ "epoch": 0.21,
412
+ "grad_norm": 4.15625,
413
+ "learning_rate": 1.4873684210526315e-05,
414
+ "loss": 0.896,
415
+ "step": 580
416
+ },
417
+ {
418
+ "epoch": 0.21,
419
+ "grad_norm": 5.46875,
420
+ "learning_rate": 1.4857894736842107e-05,
421
+ "loss": 0.9682,
422
+ "step": 590
423
+ },
424
+ {
425
+ "epoch": 0.22,
426
+ "grad_norm": 3.90625,
427
+ "learning_rate": 1.4842105263157895e-05,
428
+ "loss": 0.8046,
429
+ "step": 600
430
+ },
431
+ {
432
+ "epoch": 0.22,
433
+ "grad_norm": 3.96875,
434
+ "learning_rate": 1.4826315789473686e-05,
435
+ "loss": 0.9797,
436
+ "step": 610
437
+ },
438
+ {
439
+ "epoch": 0.22,
440
+ "grad_norm": 3.171875,
441
+ "learning_rate": 1.4810526315789474e-05,
442
+ "loss": 0.8386,
443
+ "step": 620
444
+ },
445
+ {
446
+ "epoch": 0.23,
447
+ "grad_norm": 4.46875,
448
+ "learning_rate": 1.4794736842105265e-05,
449
+ "loss": 0.9006,
450
+ "step": 630
451
+ },
452
+ {
453
+ "epoch": 0.23,
454
+ "grad_norm": 4.0,
455
+ "learning_rate": 1.4778947368421053e-05,
456
+ "loss": 0.8116,
457
+ "step": 640
458
+ },
459
+ {
460
+ "epoch": 0.23,
461
+ "grad_norm": 4.03125,
462
+ "learning_rate": 1.4763157894736842e-05,
463
+ "loss": 0.8465,
464
+ "step": 650
465
+ },
466
+ {
467
+ "epoch": 0.24,
468
+ "grad_norm": 4.09375,
469
+ "learning_rate": 1.4747368421052632e-05,
470
+ "loss": 0.843,
471
+ "step": 660
472
+ },
473
+ {
474
+ "epoch": 0.24,
475
+ "grad_norm": 3.21875,
476
+ "learning_rate": 1.4731578947368421e-05,
477
+ "loss": 0.9078,
478
+ "step": 670
479
+ },
480
+ {
481
+ "epoch": 0.24,
482
+ "grad_norm": 3.53125,
483
+ "learning_rate": 1.4715789473684211e-05,
484
+ "loss": 0.855,
485
+ "step": 680
486
+ },
487
+ {
488
+ "epoch": 0.25,
489
+ "grad_norm": 3.546875,
490
+ "learning_rate": 1.47e-05,
491
+ "loss": 0.9115,
492
+ "step": 690
493
+ },
494
+ {
495
+ "epoch": 0.25,
496
+ "grad_norm": 4.5625,
497
+ "learning_rate": 1.468421052631579e-05,
498
+ "loss": 0.8013,
499
+ "step": 700
500
+ },
501
+ {
502
+ "epoch": 0.25,
503
+ "grad_norm": 3.515625,
504
+ "learning_rate": 1.4668421052631579e-05,
505
+ "loss": 0.8253,
506
+ "step": 710
507
+ },
508
+ {
509
+ "epoch": 0.26,
510
+ "grad_norm": 3.4375,
511
+ "learning_rate": 1.4652631578947367e-05,
512
+ "loss": 0.8778,
513
+ "step": 720
514
+ },
515
+ {
516
+ "epoch": 0.26,
517
+ "grad_norm": 6.4375,
518
+ "learning_rate": 1.4636842105263158e-05,
519
+ "loss": 0.9265,
520
+ "step": 730
521
+ },
522
+ {
523
+ "epoch": 0.27,
524
+ "grad_norm": 4.0625,
525
+ "learning_rate": 1.4621052631578946e-05,
526
+ "loss": 0.8955,
527
+ "step": 740
528
+ },
529
+ {
530
+ "epoch": 0.27,
531
+ "grad_norm": 3.6875,
532
+ "learning_rate": 1.4605263157894737e-05,
533
+ "loss": 0.8791,
534
+ "step": 750
535
+ },
536
+ {
537
+ "epoch": 0.27,
538
+ "grad_norm": 3.703125,
539
+ "learning_rate": 1.4589473684210527e-05,
540
+ "loss": 0.8493,
541
+ "step": 760
542
+ },
543
+ {
544
+ "epoch": 0.28,
545
+ "grad_norm": 3.984375,
546
+ "learning_rate": 1.4573684210526317e-05,
547
+ "loss": 0.9307,
548
+ "step": 770
549
+ },
550
+ {
551
+ "epoch": 0.28,
552
+ "grad_norm": 3.125,
553
+ "learning_rate": 1.4557894736842106e-05,
554
+ "loss": 0.9093,
555
+ "step": 780
556
+ },
557
+ {
558
+ "epoch": 0.28,
559
+ "grad_norm": 4.40625,
560
+ "learning_rate": 1.4542105263157895e-05,
561
+ "loss": 0.873,
562
+ "step": 790
563
+ },
564
+ {
565
+ "epoch": 0.29,
566
+ "grad_norm": 4.21875,
567
+ "learning_rate": 1.4526315789473685e-05,
568
+ "loss": 0.9238,
569
+ "step": 800
570
+ },
571
+ {
572
+ "epoch": 0.29,
573
+ "grad_norm": 3.25,
574
+ "learning_rate": 1.4510526315789474e-05,
575
+ "loss": 0.8425,
576
+ "step": 810
577
+ },
578
+ {
579
+ "epoch": 0.29,
580
+ "grad_norm": 3.125,
581
+ "learning_rate": 1.4494736842105264e-05,
582
+ "loss": 0.9208,
583
+ "step": 820
584
+ },
585
+ {
586
+ "epoch": 0.3,
587
+ "grad_norm": 5.625,
588
+ "learning_rate": 1.4478947368421053e-05,
589
+ "loss": 0.8656,
590
+ "step": 830
591
+ },
592
+ {
593
+ "epoch": 0.3,
594
+ "grad_norm": 2.890625,
595
+ "learning_rate": 1.4463157894736843e-05,
596
+ "loss": 0.9557,
597
+ "step": 840
598
+ },
599
+ {
600
+ "epoch": 0.3,
601
+ "grad_norm": 4.15625,
602
+ "learning_rate": 1.4447368421052632e-05,
603
+ "loss": 0.859,
604
+ "step": 850
605
+ },
606
+ {
607
+ "epoch": 0.31,
608
+ "grad_norm": 3.984375,
609
+ "learning_rate": 1.443157894736842e-05,
610
+ "loss": 0.9057,
611
+ "step": 860
612
+ },
613
+ {
614
+ "epoch": 0.31,
615
+ "grad_norm": 3.890625,
616
+ "learning_rate": 1.441578947368421e-05,
617
+ "loss": 0.9119,
618
+ "step": 870
619
+ },
620
+ {
621
+ "epoch": 0.32,
622
+ "grad_norm": 4.34375,
623
+ "learning_rate": 1.44e-05,
624
+ "loss": 0.8758,
625
+ "step": 880
626
+ },
627
+ {
628
+ "epoch": 0.32,
629
+ "grad_norm": 3.828125,
630
+ "learning_rate": 1.438421052631579e-05,
631
+ "loss": 0.8534,
632
+ "step": 890
633
+ },
634
+ {
635
+ "epoch": 0.32,
636
+ "grad_norm": 2.59375,
637
+ "learning_rate": 1.4368421052631578e-05,
638
+ "loss": 0.8624,
639
+ "step": 900
640
+ },
641
+ {
642
+ "epoch": 0.33,
643
+ "grad_norm": 3.828125,
644
+ "learning_rate": 1.4352631578947369e-05,
645
+ "loss": 0.8776,
646
+ "step": 910
647
+ },
648
+ {
649
+ "epoch": 0.33,
650
+ "grad_norm": 4.5625,
651
+ "learning_rate": 1.4336842105263159e-05,
652
+ "loss": 0.7811,
653
+ "step": 920
654
+ },
655
+ {
656
+ "epoch": 0.33,
657
+ "grad_norm": 3.71875,
658
+ "learning_rate": 1.4321052631578948e-05,
659
+ "loss": 0.9363,
660
+ "step": 930
661
+ },
662
+ {
663
+ "epoch": 0.34,
664
+ "grad_norm": 3.359375,
665
+ "learning_rate": 1.4305263157894738e-05,
666
+ "loss": 0.87,
667
+ "step": 940
668
+ },
669
+ {
670
+ "epoch": 0.34,
671
+ "grad_norm": 3.875,
672
+ "learning_rate": 1.4289473684210527e-05,
673
+ "loss": 0.8802,
674
+ "step": 950
675
+ },
676
+ {
677
+ "epoch": 0.34,
678
+ "grad_norm": 3.5625,
679
+ "learning_rate": 1.4273684210526317e-05,
680
+ "loss": 0.8649,
681
+ "step": 960
682
+ },
683
+ {
684
+ "epoch": 0.35,
685
+ "grad_norm": 5.1875,
686
+ "learning_rate": 1.4257894736842106e-05,
687
+ "loss": 0.8778,
688
+ "step": 970
689
+ },
690
+ {
691
+ "epoch": 0.35,
692
+ "grad_norm": 5.1875,
693
+ "learning_rate": 1.4242105263157896e-05,
694
+ "loss": 0.9465,
695
+ "step": 980
696
+ },
697
+ {
698
+ "epoch": 0.36,
699
+ "grad_norm": 3.40625,
700
+ "learning_rate": 1.4226315789473685e-05,
701
+ "loss": 0.7351,
702
+ "step": 990
703
+ },
704
+ {
705
+ "epoch": 0.36,
706
+ "grad_norm": 3.65625,
707
+ "learning_rate": 1.4210526315789473e-05,
708
+ "loss": 0.8345,
709
+ "step": 1000
710
+ },
711
+ {
712
+ "epoch": 0.36,
713
+ "grad_norm": 4.34375,
714
+ "learning_rate": 1.4194736842105264e-05,
715
+ "loss": 0.844,
716
+ "step": 1010
717
+ },
718
+ {
719
+ "epoch": 0.37,
720
+ "grad_norm": 3.546875,
721
+ "learning_rate": 1.4178947368421052e-05,
722
+ "loss": 0.8772,
723
+ "step": 1020
724
+ },
725
+ {
726
+ "epoch": 0.37,
727
+ "grad_norm": 3.859375,
728
+ "learning_rate": 1.4163157894736843e-05,
729
+ "loss": 0.8421,
730
+ "step": 1030
731
+ },
732
+ {
733
+ "epoch": 0.37,
734
+ "grad_norm": 4.1875,
735
+ "learning_rate": 1.4147368421052631e-05,
736
+ "loss": 0.8795,
737
+ "step": 1040
738
+ },
739
+ {
740
+ "epoch": 0.38,
741
+ "grad_norm": 3.28125,
742
+ "learning_rate": 1.4131578947368422e-05,
743
+ "loss": 0.7592,
744
+ "step": 1050
745
+ },
746
+ {
747
+ "epoch": 0.38,
748
+ "grad_norm": 5.375,
749
+ "learning_rate": 1.411578947368421e-05,
750
+ "loss": 0.8124,
751
+ "step": 1060
752
+ },
753
+ {
754
+ "epoch": 0.38,
755
+ "grad_norm": 2.84375,
756
+ "learning_rate": 1.4099999999999999e-05,
757
+ "loss": 0.7272,
758
+ "step": 1070
759
+ },
760
+ {
761
+ "epoch": 0.39,
762
+ "grad_norm": 3.125,
763
+ "learning_rate": 1.408421052631579e-05,
764
+ "loss": 0.7565,
765
+ "step": 1080
766
+ },
767
+ {
768
+ "epoch": 0.39,
769
+ "grad_norm": 3.34375,
770
+ "learning_rate": 1.406842105263158e-05,
771
+ "loss": 0.957,
772
+ "step": 1090
773
+ },
774
+ {
775
+ "epoch": 0.39,
776
+ "grad_norm": 4.78125,
777
+ "learning_rate": 1.405263157894737e-05,
778
+ "loss": 0.8601,
779
+ "step": 1100
780
+ },
781
+ {
782
+ "epoch": 0.4,
783
+ "grad_norm": 2.828125,
784
+ "learning_rate": 1.4036842105263158e-05,
785
+ "loss": 0.9521,
786
+ "step": 1110
787
+ },
788
+ {
789
+ "epoch": 0.4,
790
+ "grad_norm": 3.3125,
791
+ "learning_rate": 1.4021052631578949e-05,
792
+ "loss": 0.8389,
793
+ "step": 1120
794
+ },
795
+ {
796
+ "epoch": 0.41,
797
+ "grad_norm": 5.40625,
798
+ "learning_rate": 1.4005263157894737e-05,
799
+ "loss": 0.7895,
800
+ "step": 1130
801
+ },
802
+ {
803
+ "epoch": 0.41,
804
+ "grad_norm": 3.546875,
805
+ "learning_rate": 1.3989473684210526e-05,
806
+ "loss": 0.9063,
807
+ "step": 1140
808
+ },
809
+ {
810
+ "epoch": 0.41,
811
+ "grad_norm": 3.015625,
812
+ "learning_rate": 1.3973684210526316e-05,
813
+ "loss": 0.8427,
814
+ "step": 1150
815
+ },
816
+ {
817
+ "epoch": 0.42,
818
+ "grad_norm": 4.25,
819
+ "learning_rate": 1.3957894736842105e-05,
820
+ "loss": 0.8178,
821
+ "step": 1160
822
+ },
823
+ {
824
+ "epoch": 0.42,
825
+ "grad_norm": 3.546875,
826
+ "learning_rate": 1.3942105263157895e-05,
827
+ "loss": 0.8688,
828
+ "step": 1170
829
+ },
830
+ {
831
+ "epoch": 0.42,
832
+ "grad_norm": 3.4375,
833
+ "learning_rate": 1.3926315789473684e-05,
834
+ "loss": 0.8471,
835
+ "step": 1180
836
+ },
837
+ {
838
+ "epoch": 0.43,
839
+ "grad_norm": 4.28125,
840
+ "learning_rate": 1.3910526315789474e-05,
841
+ "loss": 0.7473,
842
+ "step": 1190
843
+ },
844
+ {
845
+ "epoch": 0.43,
846
+ "grad_norm": 3.609375,
847
+ "learning_rate": 1.3894736842105263e-05,
848
+ "loss": 0.903,
849
+ "step": 1200
850
+ },
851
+ {
852
+ "epoch": 0.43,
853
+ "grad_norm": 3.46875,
854
+ "learning_rate": 1.3878947368421052e-05,
855
+ "loss": 0.9214,
856
+ "step": 1210
857
+ },
858
+ {
859
+ "epoch": 0.44,
860
+ "grad_norm": 3.75,
861
+ "learning_rate": 1.3863157894736842e-05,
862
+ "loss": 0.9448,
863
+ "step": 1220
864
+ },
865
+ {
866
+ "epoch": 0.44,
867
+ "grad_norm": 2.609375,
868
+ "learning_rate": 1.384736842105263e-05,
869
+ "loss": 0.7912,
870
+ "step": 1230
871
+ },
872
+ {
873
+ "epoch": 0.44,
874
+ "grad_norm": 5.71875,
875
+ "learning_rate": 1.3831578947368421e-05,
876
+ "loss": 0.887,
877
+ "step": 1240
878
+ },
879
+ {
880
+ "epoch": 0.45,
881
+ "grad_norm": 3.15625,
882
+ "learning_rate": 1.3815789473684211e-05,
883
+ "loss": 0.8479,
884
+ "step": 1250
885
+ },
886
+ {
887
+ "epoch": 0.45,
888
+ "grad_norm": 3.125,
889
+ "learning_rate": 1.3800000000000002e-05,
890
+ "loss": 0.7985,
891
+ "step": 1260
892
+ },
893
+ {
894
+ "epoch": 0.46,
895
+ "grad_norm": 4.03125,
896
+ "learning_rate": 1.378421052631579e-05,
897
+ "loss": 0.7114,
898
+ "step": 1270
899
+ },
900
+ {
901
+ "epoch": 0.46,
902
+ "grad_norm": 3.765625,
903
+ "learning_rate": 1.3768421052631579e-05,
904
+ "loss": 0.8742,
905
+ "step": 1280
906
+ },
907
+ {
908
+ "epoch": 0.46,
909
+ "grad_norm": 3.984375,
910
+ "learning_rate": 1.375263157894737e-05,
911
+ "loss": 0.8908,
912
+ "step": 1290
913
+ },
914
+ {
915
+ "epoch": 0.47,
916
+ "grad_norm": 3.640625,
917
+ "learning_rate": 1.3736842105263158e-05,
918
+ "loss": 0.8237,
919
+ "step": 1300
920
+ },
921
+ {
922
+ "epoch": 0.47,
923
+ "grad_norm": 3.78125,
924
+ "learning_rate": 1.3721052631578948e-05,
925
+ "loss": 0.7752,
926
+ "step": 1310
927
+ },
928
+ {
929
+ "epoch": 0.47,
930
+ "grad_norm": 4.71875,
931
+ "learning_rate": 1.3705263157894737e-05,
932
+ "loss": 0.8437,
933
+ "step": 1320
934
+ },
935
+ {
936
+ "epoch": 0.48,
937
+ "grad_norm": 2.875,
938
+ "learning_rate": 1.3689473684210527e-05,
939
+ "loss": 0.7395,
940
+ "step": 1330
941
+ },
942
+ {
943
+ "epoch": 0.48,
944
+ "grad_norm": 3.375,
945
+ "learning_rate": 1.3673684210526316e-05,
946
+ "loss": 0.9191,
947
+ "step": 1340
948
+ },
949
+ {
950
+ "epoch": 0.48,
951
+ "grad_norm": 4.65625,
952
+ "learning_rate": 1.3657894736842106e-05,
953
+ "loss": 0.8401,
954
+ "step": 1350
955
+ },
956
+ {
957
+ "epoch": 0.49,
958
+ "grad_norm": 3.625,
959
+ "learning_rate": 1.3642105263157895e-05,
960
+ "loss": 0.7785,
961
+ "step": 1360
962
+ },
963
+ {
964
+ "epoch": 0.49,
965
+ "grad_norm": 4.25,
966
+ "learning_rate": 1.3626315789473684e-05,
967
+ "loss": 0.9515,
968
+ "step": 1370
969
+ },
970
+ {
971
+ "epoch": 0.49,
972
+ "grad_norm": 2.78125,
973
+ "learning_rate": 1.3610526315789474e-05,
974
+ "loss": 0.8812,
975
+ "step": 1380
976
+ },
977
+ {
978
+ "epoch": 0.5,
979
+ "grad_norm": 2.296875,
980
+ "learning_rate": 1.3594736842105263e-05,
981
+ "loss": 0.8685,
982
+ "step": 1390
983
+ },
984
+ {
985
+ "epoch": 0.5,
986
+ "grad_norm": 4.03125,
987
+ "learning_rate": 1.3578947368421053e-05,
988
+ "loss": 0.9704,
989
+ "step": 1400
990
+ },
991
+ {
992
+ "epoch": 0.51,
993
+ "grad_norm": 5.375,
994
+ "learning_rate": 1.3563157894736842e-05,
995
+ "loss": 0.7972,
996
+ "step": 1410
997
+ },
998
+ {
999
+ "epoch": 0.51,
1000
+ "grad_norm": 2.3125,
1001
+ "learning_rate": 1.3547368421052634e-05,
1002
+ "loss": 0.8166,
1003
+ "step": 1420
1004
+ },
1005
+ {
1006
+ "epoch": 0.51,
1007
+ "grad_norm": 4.0625,
1008
+ "learning_rate": 1.3531578947368422e-05,
1009
+ "loss": 0.8718,
1010
+ "step": 1430
1011
+ },
1012
+ {
1013
+ "epoch": 0.52,
1014
+ "grad_norm": 3.75,
1015
+ "learning_rate": 1.3515789473684211e-05,
1016
+ "loss": 0.8607,
1017
+ "step": 1440
1018
+ },
1019
+ {
1020
+ "epoch": 0.52,
1021
+ "grad_norm": 4.0625,
1022
+ "learning_rate": 1.3500000000000001e-05,
1023
+ "loss": 0.8328,
1024
+ "step": 1450
1025
+ },
1026
+ {
1027
+ "epoch": 0.52,
1028
+ "grad_norm": 3.96875,
1029
+ "learning_rate": 1.348421052631579e-05,
1030
+ "loss": 0.925,
1031
+ "step": 1460
1032
+ },
1033
+ {
1034
+ "epoch": 0.53,
1035
+ "grad_norm": 4.71875,
1036
+ "learning_rate": 1.346842105263158e-05,
1037
+ "loss": 0.9331,
1038
+ "step": 1470
1039
+ },
1040
+ {
1041
+ "epoch": 0.53,
1042
+ "grad_norm": 3.421875,
1043
+ "learning_rate": 1.3452631578947369e-05,
1044
+ "loss": 0.8584,
1045
+ "step": 1480
1046
+ },
1047
+ {
1048
+ "epoch": 0.53,
1049
+ "grad_norm": 3.578125,
1050
+ "learning_rate": 1.343684210526316e-05,
1051
+ "loss": 0.8312,
1052
+ "step": 1490
1053
+ },
1054
+ {
1055
+ "epoch": 0.54,
1056
+ "grad_norm": 3.0,
1057
+ "learning_rate": 1.3421052631578948e-05,
1058
+ "loss": 0.8311,
1059
+ "step": 1500
1060
+ },
1061
+ {
1062
+ "epoch": 0.54,
1063
+ "grad_norm": 3.859375,
1064
+ "learning_rate": 1.3405263157894736e-05,
1065
+ "loss": 0.8207,
1066
+ "step": 1510
1067
+ },
1068
+ {
1069
+ "epoch": 0.55,
1070
+ "grad_norm": 3.640625,
1071
+ "learning_rate": 1.3389473684210527e-05,
1072
+ "loss": 0.9551,
1073
+ "step": 1520
1074
+ },
1075
+ {
1076
+ "epoch": 0.55,
1077
+ "grad_norm": 4.6875,
1078
+ "learning_rate": 1.3373684210526315e-05,
1079
+ "loss": 0.8096,
1080
+ "step": 1530
1081
+ },
1082
+ {
1083
+ "epoch": 0.55,
1084
+ "grad_norm": 3.90625,
1085
+ "learning_rate": 1.3357894736842106e-05,
1086
+ "loss": 0.8908,
1087
+ "step": 1540
1088
+ },
1089
+ {
1090
+ "epoch": 0.56,
1091
+ "grad_norm": 3.796875,
1092
+ "learning_rate": 1.3342105263157894e-05,
1093
+ "loss": 0.7875,
1094
+ "step": 1550
1095
+ },
1096
+ {
1097
+ "epoch": 0.56,
1098
+ "grad_norm": 3.234375,
1099
+ "learning_rate": 1.3326315789473685e-05,
1100
+ "loss": 0.7913,
1101
+ "step": 1560
1102
+ },
1103
+ {
1104
+ "epoch": 0.56,
1105
+ "grad_norm": 3.890625,
1106
+ "learning_rate": 1.3310526315789473e-05,
1107
+ "loss": 0.7862,
1108
+ "step": 1570
1109
+ },
1110
+ {
1111
+ "epoch": 0.57,
1112
+ "grad_norm": 5.1875,
1113
+ "learning_rate": 1.3294736842105262e-05,
1114
+ "loss": 0.9223,
1115
+ "step": 1580
1116
+ },
1117
+ {
1118
+ "epoch": 0.57,
1119
+ "grad_norm": 3.453125,
1120
+ "learning_rate": 1.3278947368421054e-05,
1121
+ "loss": 0.8516,
1122
+ "step": 1590
1123
+ },
1124
+ {
1125
+ "epoch": 0.57,
1126
+ "grad_norm": 2.78125,
1127
+ "learning_rate": 1.3263157894736843e-05,
1128
+ "loss": 0.8969,
1129
+ "step": 1600
1130
+ },
1131
+ {
1132
+ "epoch": 0.58,
1133
+ "grad_norm": 2.984375,
1134
+ "learning_rate": 1.3247368421052633e-05,
1135
+ "loss": 0.8309,
1136
+ "step": 1610
1137
+ },
1138
+ {
1139
+ "epoch": 0.58,
1140
+ "grad_norm": 3.734375,
1141
+ "learning_rate": 1.3231578947368422e-05,
1142
+ "loss": 0.7816,
1143
+ "step": 1620
1144
+ },
1145
+ {
1146
+ "epoch": 0.58,
1147
+ "grad_norm": 5.0625,
1148
+ "learning_rate": 1.3215789473684212e-05,
1149
+ "loss": 0.8899,
1150
+ "step": 1630
1151
+ },
1152
+ {
1153
+ "epoch": 0.59,
1154
+ "grad_norm": 4.21875,
1155
+ "learning_rate": 1.32e-05,
1156
+ "loss": 0.9195,
1157
+ "step": 1640
1158
+ },
1159
+ {
1160
+ "epoch": 0.59,
1161
+ "grad_norm": 3.71875,
1162
+ "learning_rate": 1.318421052631579e-05,
1163
+ "loss": 0.8773,
1164
+ "step": 1650
1165
+ },
1166
+ {
1167
+ "epoch": 0.6,
1168
+ "grad_norm": 3.75,
1169
+ "learning_rate": 1.316842105263158e-05,
1170
+ "loss": 0.8922,
1171
+ "step": 1660
1172
+ },
1173
+ {
1174
+ "epoch": 0.6,
1175
+ "grad_norm": 4.0,
1176
+ "learning_rate": 1.3152631578947368e-05,
1177
+ "loss": 0.8647,
1178
+ "step": 1670
1179
+ },
1180
+ {
1181
+ "epoch": 0.6,
1182
+ "grad_norm": 3.296875,
1183
+ "learning_rate": 1.3136842105263159e-05,
1184
+ "loss": 0.8625,
1185
+ "step": 1680
1186
+ },
1187
+ {
1188
+ "epoch": 0.61,
1189
+ "grad_norm": 4.21875,
1190
+ "learning_rate": 1.3121052631578947e-05,
1191
+ "loss": 0.89,
1192
+ "step": 1690
1193
+ },
1194
+ {
1195
+ "epoch": 0.61,
1196
+ "grad_norm": 3.546875,
1197
+ "learning_rate": 1.3105263157894738e-05,
1198
+ "loss": 0.8426,
1199
+ "step": 1700
1200
+ },
1201
+ {
1202
+ "epoch": 0.61,
1203
+ "grad_norm": 3.734375,
1204
+ "learning_rate": 1.3089473684210526e-05,
1205
+ "loss": 0.8909,
1206
+ "step": 1710
1207
+ },
1208
+ {
1209
+ "epoch": 0.62,
1210
+ "grad_norm": 4.25,
1211
+ "learning_rate": 1.3073684210526315e-05,
1212
+ "loss": 0.8788,
1213
+ "step": 1720
1214
+ },
1215
+ {
1216
+ "epoch": 0.62,
1217
+ "grad_norm": 3.875,
1218
+ "learning_rate": 1.3057894736842105e-05,
1219
+ "loss": 0.8895,
1220
+ "step": 1730
1221
+ },
1222
+ {
1223
+ "epoch": 0.62,
1224
+ "grad_norm": 4.0,
1225
+ "learning_rate": 1.3042105263157894e-05,
1226
+ "loss": 0.883,
1227
+ "step": 1740
1228
+ },
1229
+ {
1230
+ "epoch": 0.63,
1231
+ "grad_norm": 4.78125,
1232
+ "learning_rate": 1.3026315789473684e-05,
1233
+ "loss": 0.8837,
1234
+ "step": 1750
1235
+ },
1236
+ {
1237
+ "epoch": 0.63,
1238
+ "grad_norm": 4.34375,
1239
+ "learning_rate": 1.3010526315789475e-05,
1240
+ "loss": 0.8173,
1241
+ "step": 1760
1242
+ },
1243
+ {
1244
+ "epoch": 0.63,
1245
+ "grad_norm": 3.96875,
1246
+ "learning_rate": 1.2994736842105265e-05,
1247
+ "loss": 0.8328,
1248
+ "step": 1770
1249
+ },
1250
+ {
1251
+ "epoch": 0.64,
1252
+ "grad_norm": 2.84375,
1253
+ "learning_rate": 1.2978947368421054e-05,
1254
+ "loss": 0.8218,
1255
+ "step": 1780
1256
+ },
1257
+ {
1258
+ "epoch": 0.64,
1259
+ "grad_norm": 4.625,
1260
+ "learning_rate": 1.2963157894736842e-05,
1261
+ "loss": 0.89,
1262
+ "step": 1790
1263
+ },
1264
+ {
1265
+ "epoch": 0.65,
1266
+ "grad_norm": 3.640625,
1267
+ "learning_rate": 1.2947368421052633e-05,
1268
+ "loss": 0.9269,
1269
+ "step": 1800
1270
+ },
1271
+ {
1272
+ "epoch": 0.65,
1273
+ "grad_norm": 4.1875,
1274
+ "learning_rate": 1.2931578947368421e-05,
1275
+ "loss": 0.7664,
1276
+ "step": 1810
1277
+ },
1278
+ {
1279
+ "epoch": 0.65,
1280
+ "grad_norm": 5.34375,
1281
+ "learning_rate": 1.2915789473684212e-05,
1282
+ "loss": 0.9444,
1283
+ "step": 1820
1284
+ },
1285
+ {
1286
+ "epoch": 0.66,
1287
+ "grad_norm": 3.375,
1288
+ "learning_rate": 1.29e-05,
1289
+ "loss": 0.8094,
1290
+ "step": 1830
1291
+ },
1292
+ {
1293
+ "epoch": 0.66,
1294
+ "grad_norm": 3.90625,
1295
+ "learning_rate": 1.288421052631579e-05,
1296
+ "loss": 0.7694,
1297
+ "step": 1840
1298
+ },
1299
+ {
1300
+ "epoch": 0.66,
1301
+ "grad_norm": 4.59375,
1302
+ "learning_rate": 1.2868421052631579e-05,
1303
+ "loss": 0.8341,
1304
+ "step": 1850
1305
+ },
1306
+ {
1307
+ "epoch": 0.67,
1308
+ "grad_norm": 3.90625,
1309
+ "learning_rate": 1.2852631578947368e-05,
1310
+ "loss": 0.811,
1311
+ "step": 1860
1312
+ },
1313
+ {
1314
+ "epoch": 0.67,
1315
+ "grad_norm": 2.8125,
1316
+ "learning_rate": 1.2836842105263158e-05,
1317
+ "loss": 0.8261,
1318
+ "step": 1870
1319
+ },
1320
+ {
1321
+ "epoch": 0.67,
1322
+ "grad_norm": 3.65625,
1323
+ "learning_rate": 1.2821052631578947e-05,
1324
+ "loss": 0.8654,
1325
+ "step": 1880
1326
+ },
1327
+ {
1328
+ "epoch": 0.68,
1329
+ "grad_norm": 4.1875,
1330
+ "learning_rate": 1.2805263157894737e-05,
1331
+ "loss": 0.9158,
1332
+ "step": 1890
1333
+ },
1334
+ {
1335
+ "epoch": 0.68,
1336
+ "grad_norm": 4.125,
1337
+ "learning_rate": 1.2789473684210526e-05,
1338
+ "loss": 0.7241,
1339
+ "step": 1900
1340
+ },
1341
+ {
1342
+ "epoch": 0.68,
1343
+ "grad_norm": 4.59375,
1344
+ "learning_rate": 1.2773684210526316e-05,
1345
+ "loss": 0.84,
1346
+ "step": 1910
1347
+ },
1348
+ {
1349
+ "epoch": 0.69,
1350
+ "grad_norm": 2.765625,
1351
+ "learning_rate": 1.2757894736842106e-05,
1352
+ "loss": 0.9277,
1353
+ "step": 1920
1354
+ },
1355
+ {
1356
+ "epoch": 0.69,
1357
+ "grad_norm": 3.390625,
1358
+ "learning_rate": 1.2742105263157895e-05,
1359
+ "loss": 0.8024,
1360
+ "step": 1930
1361
+ },
1362
+ {
1363
+ "epoch": 0.7,
1364
+ "grad_norm": 6.34375,
1365
+ "learning_rate": 1.2726315789473685e-05,
1366
+ "loss": 0.7964,
1367
+ "step": 1940
1368
+ },
1369
+ {
1370
+ "epoch": 0.7,
1371
+ "grad_norm": 2.875,
1372
+ "learning_rate": 1.2710526315789474e-05,
1373
+ "loss": 0.8335,
1374
+ "step": 1950
1375
+ },
1376
+ {
1377
+ "epoch": 0.7,
1378
+ "grad_norm": 3.0,
1379
+ "learning_rate": 1.2694736842105264e-05,
1380
+ "loss": 0.8329,
1381
+ "step": 1960
1382
+ },
1383
+ {
1384
+ "epoch": 0.71,
1385
+ "grad_norm": 3.4375,
1386
+ "learning_rate": 1.2678947368421053e-05,
1387
+ "loss": 0.8553,
1388
+ "step": 1970
1389
+ },
1390
+ {
1391
+ "epoch": 0.71,
1392
+ "grad_norm": 4.25,
1393
+ "learning_rate": 1.2663157894736843e-05,
1394
+ "loss": 0.8801,
1395
+ "step": 1980
1396
+ },
1397
+ {
1398
+ "epoch": 0.71,
1399
+ "grad_norm": 4.03125,
1400
+ "learning_rate": 1.2647368421052632e-05,
1401
+ "loss": 0.813,
1402
+ "step": 1990
1403
+ },
1404
+ {
1405
+ "epoch": 0.72,
1406
+ "grad_norm": 4.8125,
1407
+ "learning_rate": 1.263157894736842e-05,
1408
+ "loss": 0.7953,
1409
+ "step": 2000
1410
+ },
1411
+ {
1412
+ "epoch": 0.72,
1413
+ "eval_loss": 0.965668797492981,
1414
+ "eval_runtime": 63.3755,
1415
+ "eval_samples_per_second": 15.779,
1416
+ "eval_steps_per_second": 15.779,
1417
+ "step": 2000
1418
+ },
1419
+ {
1420
+ "epoch": 0.72,
1421
+ "grad_norm": 3.9375,
1422
+ "learning_rate": 1.2615789473684211e-05,
1423
+ "loss": 0.7854,
1424
+ "step": 2010
1425
+ },
1426
+ {
1427
+ "epoch": 0.72,
1428
+ "grad_norm": 4.28125,
1429
+ "learning_rate": 1.26e-05,
1430
+ "loss": 0.9323,
1431
+ "step": 2020
1432
+ },
1433
+ {
1434
+ "epoch": 0.73,
1435
+ "grad_norm": 3.1875,
1436
+ "learning_rate": 1.258421052631579e-05,
1437
+ "loss": 0.8514,
1438
+ "step": 2030
1439
+ },
1440
+ {
1441
+ "epoch": 0.73,
1442
+ "grad_norm": 3.09375,
1443
+ "learning_rate": 1.2568421052631579e-05,
1444
+ "loss": 0.7566,
1445
+ "step": 2040
1446
+ },
1447
+ {
1448
+ "epoch": 0.74,
1449
+ "grad_norm": 6.90625,
1450
+ "learning_rate": 1.2552631578947369e-05,
1451
+ "loss": 0.8933,
1452
+ "step": 2050
1453
+ },
1454
+ {
1455
+ "epoch": 0.74,
1456
+ "grad_norm": 4.59375,
1457
+ "learning_rate": 1.2536842105263158e-05,
1458
+ "loss": 0.7971,
1459
+ "step": 2060
1460
+ },
1461
+ {
1462
+ "epoch": 0.74,
1463
+ "grad_norm": 2.859375,
1464
+ "learning_rate": 1.2521052631578946e-05,
1465
+ "loss": 0.7583,
1466
+ "step": 2070
1467
+ },
1468
+ {
1469
+ "epoch": 0.75,
1470
+ "grad_norm": 3.1875,
1471
+ "learning_rate": 1.2505263157894737e-05,
1472
+ "loss": 0.844,
1473
+ "step": 2080
1474
+ },
1475
+ {
1476
+ "epoch": 0.75,
1477
+ "grad_norm": 3.921875,
1478
+ "learning_rate": 1.2489473684210527e-05,
1479
+ "loss": 0.8764,
1480
+ "step": 2090
1481
+ },
1482
+ {
1483
+ "epoch": 0.75,
1484
+ "grad_norm": 5.09375,
1485
+ "learning_rate": 1.2473684210526317e-05,
1486
+ "loss": 0.8263,
1487
+ "step": 2100
1488
+ },
1489
+ {
1490
+ "epoch": 0.76,
1491
+ "grad_norm": 3.796875,
1492
+ "learning_rate": 1.2457894736842106e-05,
1493
+ "loss": 0.876,
1494
+ "step": 2110
1495
+ },
1496
+ {
1497
+ "epoch": 0.76,
1498
+ "grad_norm": 2.625,
1499
+ "learning_rate": 1.2442105263157896e-05,
1500
+ "loss": 0.8852,
1501
+ "step": 2120
1502
+ },
1503
+ {
1504
+ "epoch": 0.76,
1505
+ "grad_norm": 3.375,
1506
+ "learning_rate": 1.2426315789473685e-05,
1507
+ "loss": 0.8555,
1508
+ "step": 2130
1509
+ },
1510
+ {
1511
+ "epoch": 0.77,
1512
+ "grad_norm": 3.328125,
1513
+ "learning_rate": 1.2410526315789474e-05,
1514
+ "loss": 0.8724,
1515
+ "step": 2140
1516
+ },
1517
+ {
1518
+ "epoch": 0.77,
1519
+ "grad_norm": 3.265625,
1520
+ "learning_rate": 1.2394736842105264e-05,
1521
+ "loss": 0.8052,
1522
+ "step": 2150
1523
+ },
1524
+ {
1525
+ "epoch": 0.77,
1526
+ "grad_norm": 4.375,
1527
+ "learning_rate": 1.2378947368421053e-05,
1528
+ "loss": 0.6949,
1529
+ "step": 2160
1530
+ },
1531
+ {
1532
+ "epoch": 0.78,
1533
+ "grad_norm": 4.4375,
1534
+ "learning_rate": 1.2363157894736843e-05,
1535
+ "loss": 0.8075,
1536
+ "step": 2170
1537
+ },
1538
+ {
1539
+ "epoch": 0.78,
1540
+ "grad_norm": 3.09375,
1541
+ "learning_rate": 1.2347368421052631e-05,
1542
+ "loss": 0.8539,
1543
+ "step": 2180
1544
+ },
1545
+ {
1546
+ "epoch": 0.79,
1547
+ "grad_norm": 3.640625,
1548
+ "learning_rate": 1.2331578947368422e-05,
1549
+ "loss": 0.7066,
1550
+ "step": 2190
1551
+ },
1552
+ {
1553
+ "epoch": 0.79,
1554
+ "grad_norm": 4.125,
1555
+ "learning_rate": 1.231578947368421e-05,
1556
+ "loss": 0.8651,
1557
+ "step": 2200
1558
+ },
1559
+ {
1560
+ "epoch": 0.79,
1561
+ "grad_norm": 4.75,
1562
+ "learning_rate": 1.2299999999999999e-05,
1563
+ "loss": 0.7921,
1564
+ "step": 2210
1565
+ },
1566
+ {
1567
+ "epoch": 0.8,
1568
+ "grad_norm": 3.875,
1569
+ "learning_rate": 1.228421052631579e-05,
1570
+ "loss": 0.7711,
1571
+ "step": 2220
1572
+ },
1573
+ {
1574
+ "epoch": 0.8,
1575
+ "grad_norm": 4.21875,
1576
+ "learning_rate": 1.2268421052631578e-05,
1577
+ "loss": 0.8328,
1578
+ "step": 2230
1579
+ },
1580
+ {
1581
+ "epoch": 0.8,
1582
+ "grad_norm": 2.96875,
1583
+ "learning_rate": 1.2252631578947368e-05,
1584
+ "loss": 0.872,
1585
+ "step": 2240
1586
+ },
1587
+ {
1588
+ "epoch": 0.81,
1589
+ "grad_norm": 3.203125,
1590
+ "learning_rate": 1.2236842105263159e-05,
1591
+ "loss": 0.7755,
1592
+ "step": 2250
1593
+ },
1594
+ {
1595
+ "epoch": 0.81,
1596
+ "grad_norm": 3.15625,
1597
+ "learning_rate": 1.2221052631578949e-05,
1598
+ "loss": 0.7647,
1599
+ "step": 2260
1600
+ },
1601
+ {
1602
+ "epoch": 0.81,
1603
+ "grad_norm": 3.0,
1604
+ "learning_rate": 1.2205263157894738e-05,
1605
+ "loss": 0.8639,
1606
+ "step": 2270
1607
+ },
1608
+ {
1609
+ "epoch": 0.82,
1610
+ "grad_norm": 4.625,
1611
+ "learning_rate": 1.2189473684210526e-05,
1612
+ "loss": 0.8007,
1613
+ "step": 2280
1614
+ },
1615
+ {
1616
+ "epoch": 0.82,
1617
+ "grad_norm": 4.78125,
1618
+ "learning_rate": 1.2173684210526317e-05,
1619
+ "loss": 0.8344,
1620
+ "step": 2290
1621
+ },
1622
+ {
1623
+ "epoch": 0.82,
1624
+ "grad_norm": 3.140625,
1625
+ "learning_rate": 1.2157894736842105e-05,
1626
+ "loss": 0.9327,
1627
+ "step": 2300
1628
+ },
1629
+ {
1630
+ "epoch": 0.83,
1631
+ "grad_norm": 4.09375,
1632
+ "learning_rate": 1.2142105263157896e-05,
1633
+ "loss": 0.8617,
1634
+ "step": 2310
1635
+ },
1636
+ {
1637
+ "epoch": 0.83,
1638
+ "grad_norm": 4.1875,
1639
+ "learning_rate": 1.2126315789473684e-05,
1640
+ "loss": 0.8129,
1641
+ "step": 2320
1642
+ },
1643
+ {
1644
+ "epoch": 0.84,
1645
+ "grad_norm": 3.171875,
1646
+ "learning_rate": 1.2110526315789475e-05,
1647
+ "loss": 0.9715,
1648
+ "step": 2330
1649
+ },
1650
+ {
1651
+ "epoch": 0.84,
1652
+ "grad_norm": 4.8125,
1653
+ "learning_rate": 1.2094736842105263e-05,
1654
+ "loss": 0.8432,
1655
+ "step": 2340
1656
+ },
1657
+ {
1658
+ "epoch": 0.84,
1659
+ "grad_norm": 3.71875,
1660
+ "learning_rate": 1.2078947368421052e-05,
1661
+ "loss": 0.7863,
1662
+ "step": 2350
1663
+ },
1664
+ {
1665
+ "epoch": 0.85,
1666
+ "grad_norm": 3.671875,
1667
+ "learning_rate": 1.2063157894736842e-05,
1668
+ "loss": 0.8806,
1669
+ "step": 2360
1670
+ },
1671
+ {
1672
+ "epoch": 0.85,
1673
+ "grad_norm": 3.390625,
1674
+ "learning_rate": 1.2047368421052631e-05,
1675
+ "loss": 0.8683,
1676
+ "step": 2370
1677
+ },
1678
+ {
1679
+ "epoch": 0.85,
1680
+ "grad_norm": 3.5625,
1681
+ "learning_rate": 1.2031578947368421e-05,
1682
+ "loss": 0.8558,
1683
+ "step": 2380
1684
+ },
1685
+ {
1686
+ "epoch": 0.86,
1687
+ "grad_norm": 3.328125,
1688
+ "learning_rate": 1.201578947368421e-05,
1689
+ "loss": 0.7749,
1690
+ "step": 2390
1691
+ },
1692
+ {
1693
+ "epoch": 0.86,
1694
+ "grad_norm": 5.0,
1695
+ "learning_rate": 1.2e-05,
1696
+ "loss": 0.8283,
1697
+ "step": 2400
1698
+ },
1699
+ {
1700
+ "epoch": 0.86,
1701
+ "grad_norm": 4.40625,
1702
+ "learning_rate": 1.1984210526315789e-05,
1703
+ "loss": 0.8553,
1704
+ "step": 2410
1705
+ },
1706
+ {
1707
+ "epoch": 0.87,
1708
+ "grad_norm": 3.546875,
1709
+ "learning_rate": 1.196842105263158e-05,
1710
+ "loss": 0.8862,
1711
+ "step": 2420
1712
+ },
1713
+ {
1714
+ "epoch": 0.87,
1715
+ "grad_norm": 3.59375,
1716
+ "learning_rate": 1.195263157894737e-05,
1717
+ "loss": 0.8069,
1718
+ "step": 2430
1719
+ },
1720
+ {
1721
+ "epoch": 0.88,
1722
+ "grad_norm": 4.28125,
1723
+ "learning_rate": 1.1936842105263158e-05,
1724
+ "loss": 0.9325,
1725
+ "step": 2440
1726
+ },
1727
+ {
1728
+ "epoch": 0.88,
1729
+ "grad_norm": 3.546875,
1730
+ "learning_rate": 1.1921052631578949e-05,
1731
+ "loss": 0.8751,
1732
+ "step": 2450
1733
+ },
1734
+ {
1735
+ "epoch": 0.88,
1736
+ "grad_norm": 5.71875,
1737
+ "learning_rate": 1.1905263157894737e-05,
1738
+ "loss": 0.9068,
1739
+ "step": 2460
1740
+ },
1741
+ {
1742
+ "epoch": 0.89,
1743
+ "grad_norm": 3.5,
1744
+ "learning_rate": 1.1889473684210528e-05,
1745
+ "loss": 0.8121,
1746
+ "step": 2470
1747
+ },
1748
+ {
1749
+ "epoch": 0.89,
1750
+ "grad_norm": 3.125,
1751
+ "learning_rate": 1.1873684210526316e-05,
1752
+ "loss": 0.8152,
1753
+ "step": 2480
1754
+ },
1755
+ {
1756
+ "epoch": 0.89,
1757
+ "grad_norm": 3.5625,
1758
+ "learning_rate": 1.1857894736842105e-05,
1759
+ "loss": 0.7132,
1760
+ "step": 2490
1761
+ },
1762
+ {
1763
+ "epoch": 0.9,
1764
+ "grad_norm": 4.8125,
1765
+ "learning_rate": 1.1842105263157895e-05,
1766
+ "loss": 0.8085,
1767
+ "step": 2500
1768
+ },
1769
+ {
1770
+ "epoch": 0.9,
1771
+ "grad_norm": 3.171875,
1772
+ "learning_rate": 1.1826315789473684e-05,
1773
+ "loss": 0.8593,
1774
+ "step": 2510
1775
+ },
1776
+ {
1777
+ "epoch": 0.9,
1778
+ "grad_norm": 3.984375,
1779
+ "learning_rate": 1.1810526315789474e-05,
1780
+ "loss": 0.7501,
1781
+ "step": 2520
1782
+ },
1783
+ {
1784
+ "epoch": 0.91,
1785
+ "grad_norm": 2.609375,
1786
+ "learning_rate": 1.1794736842105263e-05,
1787
+ "loss": 0.8399,
1788
+ "step": 2530
1789
+ },
1790
+ {
1791
+ "epoch": 0.91,
1792
+ "grad_norm": 5.09375,
1793
+ "learning_rate": 1.1778947368421053e-05,
1794
+ "loss": 0.6944,
1795
+ "step": 2540
1796
+ },
1797
+ {
1798
+ "epoch": 0.91,
1799
+ "grad_norm": 3.796875,
1800
+ "learning_rate": 1.1763157894736842e-05,
1801
+ "loss": 0.8347,
1802
+ "step": 2550
1803
+ },
1804
+ {
1805
+ "epoch": 0.92,
1806
+ "grad_norm": 3.734375,
1807
+ "learning_rate": 1.174736842105263e-05,
1808
+ "loss": 0.8156,
1809
+ "step": 2560
1810
+ },
1811
+ {
1812
+ "epoch": 0.92,
1813
+ "grad_norm": 4.25,
1814
+ "learning_rate": 1.173157894736842e-05,
1815
+ "loss": 0.712,
1816
+ "step": 2570
1817
+ },
1818
+ {
1819
+ "epoch": 0.93,
1820
+ "grad_norm": 4.125,
1821
+ "learning_rate": 1.171578947368421e-05,
1822
+ "loss": 0.8281,
1823
+ "step": 2580
1824
+ },
1825
+ {
1826
+ "epoch": 0.93,
1827
+ "grad_norm": 3.109375,
1828
+ "learning_rate": 1.1700000000000001e-05,
1829
+ "loss": 0.964,
1830
+ "step": 2590
1831
+ },
1832
+ {
1833
+ "epoch": 0.93,
1834
+ "grad_norm": 3.890625,
1835
+ "learning_rate": 1.168421052631579e-05,
1836
+ "loss": 0.783,
1837
+ "step": 2600
1838
+ },
1839
+ {
1840
+ "epoch": 0.94,
1841
+ "grad_norm": 2.9375,
1842
+ "learning_rate": 1.166842105263158e-05,
1843
+ "loss": 0.7572,
1844
+ "step": 2610
1845
+ },
1846
+ {
1847
+ "epoch": 0.94,
1848
+ "grad_norm": 3.09375,
1849
+ "learning_rate": 1.1652631578947369e-05,
1850
+ "loss": 0.8726,
1851
+ "step": 2620
1852
+ },
1853
+ {
1854
+ "epoch": 0.94,
1855
+ "grad_norm": 3.6875,
1856
+ "learning_rate": 1.1636842105263158e-05,
1857
+ "loss": 0.8499,
1858
+ "step": 2630
1859
+ },
1860
+ {
1861
+ "epoch": 0.95,
1862
+ "grad_norm": 4.09375,
1863
+ "learning_rate": 1.1621052631578948e-05,
1864
+ "loss": 0.7741,
1865
+ "step": 2640
1866
+ },
1867
+ {
1868
+ "epoch": 0.95,
1869
+ "grad_norm": 2.703125,
1870
+ "learning_rate": 1.1605263157894737e-05,
1871
+ "loss": 0.7522,
1872
+ "step": 2650
1873
+ },
1874
+ {
1875
+ "epoch": 0.95,
1876
+ "grad_norm": 3.5,
1877
+ "learning_rate": 1.1589473684210527e-05,
1878
+ "loss": 0.8632,
1879
+ "step": 2660
1880
+ },
1881
+ {
1882
+ "epoch": 0.96,
1883
+ "grad_norm": 3.875,
1884
+ "learning_rate": 1.1573684210526316e-05,
1885
+ "loss": 0.7483,
1886
+ "step": 2670
1887
+ },
1888
+ {
1889
+ "epoch": 0.96,
1890
+ "grad_norm": 3.078125,
1891
+ "learning_rate": 1.1557894736842106e-05,
1892
+ "loss": 0.7606,
1893
+ "step": 2680
1894
+ },
1895
+ {
1896
+ "epoch": 0.96,
1897
+ "grad_norm": 4.875,
1898
+ "learning_rate": 1.1542105263157895e-05,
1899
+ "loss": 0.7057,
1900
+ "step": 2690
1901
+ },
1902
+ {
1903
+ "epoch": 0.97,
1904
+ "grad_norm": 3.015625,
1905
+ "learning_rate": 1.1526315789473683e-05,
1906
+ "loss": 0.7932,
1907
+ "step": 2700
1908
+ },
1909
+ {
1910
+ "epoch": 0.97,
1911
+ "grad_norm": 3.234375,
1912
+ "learning_rate": 1.1510526315789474e-05,
1913
+ "loss": 0.7906,
1914
+ "step": 2710
1915
+ },
1916
+ {
1917
+ "epoch": 0.98,
1918
+ "grad_norm": 4.03125,
1919
+ "learning_rate": 1.1494736842105262e-05,
1920
+ "loss": 0.8328,
1921
+ "step": 2720
1922
+ },
1923
+ {
1924
+ "epoch": 0.98,
1925
+ "grad_norm": 3.234375,
1926
+ "learning_rate": 1.1478947368421053e-05,
1927
+ "loss": 0.7279,
1928
+ "step": 2730
1929
+ },
1930
+ {
1931
+ "epoch": 0.98,
1932
+ "grad_norm": 4.09375,
1933
+ "learning_rate": 1.1463157894736841e-05,
1934
+ "loss": 0.8951,
1935
+ "step": 2740
1936
+ },
1937
+ {
1938
+ "epoch": 0.99,
1939
+ "grad_norm": 2.875,
1940
+ "learning_rate": 1.1447368421052632e-05,
1941
+ "loss": 0.8575,
1942
+ "step": 2750
1943
+ },
1944
+ {
1945
+ "epoch": 0.99,
1946
+ "grad_norm": 3.90625,
1947
+ "learning_rate": 1.1431578947368422e-05,
1948
+ "loss": 0.8459,
1949
+ "step": 2760
1950
+ },
1951
+ {
1952
+ "epoch": 0.99,
1953
+ "grad_norm": 3.0625,
1954
+ "learning_rate": 1.141578947368421e-05,
1955
+ "loss": 0.7296,
1956
+ "step": 2770
1957
+ },
1958
+ {
1959
+ "epoch": 1.0,
1960
+ "grad_norm": 4.5625,
1961
+ "learning_rate": 1.1400000000000001e-05,
1962
+ "loss": 0.8719,
1963
+ "step": 2780
1964
+ },
1965
+ {
1966
+ "epoch": 1.0,
1967
+ "grad_norm": 3.71875,
1968
+ "learning_rate": 1.138421052631579e-05,
1969
+ "loss": 0.7355,
1970
+ "step": 2790
1971
+ },
1972
+ {
1973
+ "epoch": 1.0,
1974
+ "grad_norm": 2.875,
1975
+ "learning_rate": 1.136842105263158e-05,
1976
+ "loss": 0.9747,
1977
+ "step": 2800
1978
+ },
1979
+ {
1980
+ "epoch": 1.01,
1981
+ "grad_norm": 3.875,
1982
+ "learning_rate": 1.1352631578947369e-05,
1983
+ "loss": 0.7992,
1984
+ "step": 2810
1985
+ },
1986
+ {
1987
+ "epoch": 1.01,
1988
+ "grad_norm": 3.390625,
1989
+ "learning_rate": 1.1336842105263159e-05,
1990
+ "loss": 0.7168,
1991
+ "step": 2820
1992
+ },
1993
+ {
1994
+ "epoch": 1.01,
1995
+ "grad_norm": 3.75,
1996
+ "learning_rate": 1.1321052631578948e-05,
1997
+ "loss": 0.6836,
1998
+ "step": 2830
1999
+ },
2000
+ {
2001
+ "epoch": 1.02,
2002
+ "grad_norm": 4.0,
2003
+ "learning_rate": 1.1305263157894736e-05,
2004
+ "loss": 0.8253,
2005
+ "step": 2840
2006
+ },
2007
+ {
2008
+ "epoch": 1.02,
2009
+ "grad_norm": 3.0625,
2010
+ "learning_rate": 1.1289473684210527e-05,
2011
+ "loss": 0.8678,
2012
+ "step": 2850
2013
+ },
2014
+ {
2015
+ "epoch": 1.03,
2016
+ "grad_norm": 3.234375,
2017
+ "learning_rate": 1.1273684210526315e-05,
2018
+ "loss": 0.7433,
2019
+ "step": 2860
2020
+ },
2021
+ {
2022
+ "epoch": 1.03,
2023
+ "grad_norm": 4.28125,
2024
+ "learning_rate": 1.1257894736842106e-05,
2025
+ "loss": 0.8963,
2026
+ "step": 2870
2027
+ },
2028
+ {
2029
+ "epoch": 1.03,
2030
+ "grad_norm": 3.796875,
2031
+ "learning_rate": 1.1242105263157894e-05,
2032
+ "loss": 0.9059,
2033
+ "step": 2880
2034
+ },
2035
+ {
2036
+ "epoch": 1.04,
2037
+ "grad_norm": 4.21875,
2038
+ "learning_rate": 1.1226315789473685e-05,
2039
+ "loss": 0.7148,
2040
+ "step": 2890
2041
+ },
2042
+ {
2043
+ "epoch": 1.04,
2044
+ "grad_norm": 4.03125,
2045
+ "learning_rate": 1.1210526315789473e-05,
2046
+ "loss": 0.8227,
2047
+ "step": 2900
2048
+ },
2049
+ {
2050
+ "epoch": 1.04,
2051
+ "grad_norm": 3.421875,
2052
+ "learning_rate": 1.1194736842105264e-05,
2053
+ "loss": 0.7934,
2054
+ "step": 2910
2055
+ },
2056
+ {
2057
+ "epoch": 1.05,
2058
+ "grad_norm": 4.03125,
2059
+ "learning_rate": 1.1178947368421054e-05,
2060
+ "loss": 0.8192,
2061
+ "step": 2920
2062
+ },
2063
+ {
2064
+ "epoch": 1.05,
2065
+ "grad_norm": 5.1875,
2066
+ "learning_rate": 1.1163157894736842e-05,
2067
+ "loss": 0.7516,
2068
+ "step": 2930
2069
+ },
2070
+ {
2071
+ "epoch": 1.05,
2072
+ "grad_norm": 4.46875,
2073
+ "learning_rate": 1.1147368421052633e-05,
2074
+ "loss": 0.8524,
2075
+ "step": 2940
2076
+ },
2077
+ {
2078
+ "epoch": 1.06,
2079
+ "grad_norm": 3.265625,
2080
+ "learning_rate": 1.1131578947368421e-05,
2081
+ "loss": 0.7609,
2082
+ "step": 2950
2083
+ },
2084
+ {
2085
+ "epoch": 1.06,
2086
+ "grad_norm": 2.625,
2087
+ "learning_rate": 1.1115789473684212e-05,
2088
+ "loss": 0.7896,
2089
+ "step": 2960
2090
+ },
2091
+ {
2092
+ "epoch": 1.07,
2093
+ "grad_norm": 3.4375,
2094
+ "learning_rate": 1.11e-05,
2095
+ "loss": 0.7638,
2096
+ "step": 2970
2097
+ },
2098
+ {
2099
+ "epoch": 1.07,
2100
+ "grad_norm": 4.125,
2101
+ "learning_rate": 1.108421052631579e-05,
2102
+ "loss": 0.7804,
2103
+ "step": 2980
2104
+ },
2105
+ {
2106
+ "epoch": 1.07,
2107
+ "grad_norm": 3.75,
2108
+ "learning_rate": 1.106842105263158e-05,
2109
+ "loss": 0.7906,
2110
+ "step": 2990
2111
+ },
2112
+ {
2113
+ "epoch": 1.08,
2114
+ "grad_norm": 4.0,
2115
+ "learning_rate": 1.1052631578947368e-05,
2116
+ "loss": 0.8006,
2117
+ "step": 3000
2118
+ },
2119
+ {
2120
+ "epoch": 1.08,
2121
+ "grad_norm": 3.609375,
2122
+ "learning_rate": 1.1036842105263158e-05,
2123
+ "loss": 0.8677,
2124
+ "step": 3010
2125
+ },
2126
+ {
2127
+ "epoch": 1.08,
2128
+ "grad_norm": 3.953125,
2129
+ "learning_rate": 1.1021052631578947e-05,
2130
+ "loss": 0.9164,
2131
+ "step": 3020
2132
+ },
2133
+ {
2134
+ "epoch": 1.09,
2135
+ "grad_norm": 4.34375,
2136
+ "learning_rate": 1.1005263157894737e-05,
2137
+ "loss": 0.8206,
2138
+ "step": 3030
2139
+ },
2140
+ {
2141
+ "epoch": 1.09,
2142
+ "grad_norm": 3.953125,
2143
+ "learning_rate": 1.0989473684210526e-05,
2144
+ "loss": 0.8203,
2145
+ "step": 3040
2146
+ },
2147
+ {
2148
+ "epoch": 1.09,
2149
+ "grad_norm": 3.8125,
2150
+ "learning_rate": 1.0973684210526316e-05,
2151
+ "loss": 0.7283,
2152
+ "step": 3050
2153
+ },
2154
+ {
2155
+ "epoch": 1.1,
2156
+ "grad_norm": 4.46875,
2157
+ "learning_rate": 1.0957894736842105e-05,
2158
+ "loss": 0.7454,
2159
+ "step": 3060
2160
+ },
2161
+ {
2162
+ "epoch": 1.1,
2163
+ "grad_norm": 4.09375,
2164
+ "learning_rate": 1.0942105263157894e-05,
2165
+ "loss": 0.7764,
2166
+ "step": 3070
2167
+ },
2168
+ {
2169
+ "epoch": 1.1,
2170
+ "grad_norm": 3.15625,
2171
+ "learning_rate": 1.0926315789473684e-05,
2172
+ "loss": 0.7876,
2173
+ "step": 3080
2174
+ },
2175
+ {
2176
+ "epoch": 1.11,
2177
+ "grad_norm": 2.953125,
2178
+ "learning_rate": 1.0910526315789474e-05,
2179
+ "loss": 0.7466,
2180
+ "step": 3090
2181
+ },
2182
+ {
2183
+ "epoch": 1.11,
2184
+ "grad_norm": 3.859375,
2185
+ "learning_rate": 1.0894736842105265e-05,
2186
+ "loss": 0.7939,
2187
+ "step": 3100
2188
+ },
2189
+ {
2190
+ "epoch": 1.12,
2191
+ "grad_norm": 3.734375,
2192
+ "learning_rate": 1.0878947368421053e-05,
2193
+ "loss": 0.8303,
2194
+ "step": 3110
2195
+ },
2196
+ {
2197
+ "epoch": 1.12,
2198
+ "grad_norm": 4.34375,
2199
+ "learning_rate": 1.0863157894736844e-05,
2200
+ "loss": 0.7779,
2201
+ "step": 3120
2202
+ },
2203
+ {
2204
+ "epoch": 1.12,
2205
+ "grad_norm": 4.9375,
2206
+ "learning_rate": 1.0847368421052632e-05,
2207
+ "loss": 0.9279,
2208
+ "step": 3130
2209
+ },
2210
+ {
2211
+ "epoch": 1.13,
2212
+ "grad_norm": 3.5625,
2213
+ "learning_rate": 1.0831578947368421e-05,
2214
+ "loss": 0.8196,
2215
+ "step": 3140
2216
+ },
2217
+ {
2218
+ "epoch": 1.13,
2219
+ "grad_norm": 3.375,
2220
+ "learning_rate": 1.0815789473684211e-05,
2221
+ "loss": 0.7357,
2222
+ "step": 3150
2223
+ },
2224
+ {
2225
+ "epoch": 1.13,
2226
+ "grad_norm": 4.5625,
2227
+ "learning_rate": 1.08e-05,
2228
+ "loss": 0.8071,
2229
+ "step": 3160
2230
+ },
2231
+ {
2232
+ "epoch": 1.14,
2233
+ "grad_norm": 4.125,
2234
+ "learning_rate": 1.078421052631579e-05,
2235
+ "loss": 0.8873,
2236
+ "step": 3170
2237
+ },
2238
+ {
2239
+ "epoch": 1.14,
2240
+ "grad_norm": 3.796875,
2241
+ "learning_rate": 1.0768421052631579e-05,
2242
+ "loss": 0.8192,
2243
+ "step": 3180
2244
+ },
2245
+ {
2246
+ "epoch": 1.14,
2247
+ "grad_norm": 3.453125,
2248
+ "learning_rate": 1.075263157894737e-05,
2249
+ "loss": 0.7588,
2250
+ "step": 3190
2251
+ },
2252
+ {
2253
+ "epoch": 1.15,
2254
+ "grad_norm": 4.03125,
2255
+ "learning_rate": 1.0736842105263158e-05,
2256
+ "loss": 0.7912,
2257
+ "step": 3200
2258
+ },
2259
+ {
2260
+ "epoch": 1.15,
2261
+ "grad_norm": 4.25,
2262
+ "learning_rate": 1.0721052631578947e-05,
2263
+ "loss": 0.8396,
2264
+ "step": 3210
2265
+ },
2266
+ {
2267
+ "epoch": 1.15,
2268
+ "grad_norm": 3.15625,
2269
+ "learning_rate": 1.0705263157894737e-05,
2270
+ "loss": 0.8006,
2271
+ "step": 3220
2272
+ },
2273
+ {
2274
+ "epoch": 1.16,
2275
+ "grad_norm": 4.96875,
2276
+ "learning_rate": 1.0689473684210526e-05,
2277
+ "loss": 0.8191,
2278
+ "step": 3230
2279
+ },
2280
+ {
2281
+ "epoch": 1.16,
2282
+ "grad_norm": 3.859375,
2283
+ "learning_rate": 1.0673684210526316e-05,
2284
+ "loss": 0.7822,
2285
+ "step": 3240
2286
+ },
2287
+ {
2288
+ "epoch": 1.17,
2289
+ "grad_norm": 3.09375,
2290
+ "learning_rate": 1.0657894736842106e-05,
2291
+ "loss": 0.6842,
2292
+ "step": 3250
2293
+ },
2294
+ {
2295
+ "epoch": 1.17,
2296
+ "grad_norm": 3.953125,
2297
+ "learning_rate": 1.0642105263157897e-05,
2298
+ "loss": 0.6702,
2299
+ "step": 3260
2300
+ },
2301
+ {
2302
+ "epoch": 1.17,
2303
+ "grad_norm": 4.125,
2304
+ "learning_rate": 1.0626315789473685e-05,
2305
+ "loss": 0.943,
2306
+ "step": 3270
2307
+ },
2308
+ {
2309
+ "epoch": 1.18,
2310
+ "grad_norm": 3.9375,
2311
+ "learning_rate": 1.0610526315789474e-05,
2312
+ "loss": 0.7917,
2313
+ "step": 3280
2314
+ },
2315
+ {
2316
+ "epoch": 1.18,
2317
+ "grad_norm": 4.375,
2318
+ "learning_rate": 1.0594736842105264e-05,
2319
+ "loss": 0.7856,
2320
+ "step": 3290
2321
+ },
2322
+ {
2323
+ "epoch": 1.18,
2324
+ "grad_norm": 5.1875,
2325
+ "learning_rate": 1.0578947368421053e-05,
2326
+ "loss": 0.7467,
2327
+ "step": 3300
2328
+ },
2329
+ {
2330
+ "epoch": 1.19,
2331
+ "grad_norm": 4.375,
2332
+ "learning_rate": 1.0563157894736843e-05,
2333
+ "loss": 0.8295,
2334
+ "step": 3310
2335
+ },
2336
+ {
2337
+ "epoch": 1.19,
2338
+ "grad_norm": 4.34375,
2339
+ "learning_rate": 1.0547368421052632e-05,
2340
+ "loss": 0.8781,
2341
+ "step": 3320
2342
+ },
2343
+ {
2344
+ "epoch": 1.19,
2345
+ "grad_norm": 2.390625,
2346
+ "learning_rate": 1.0531578947368422e-05,
2347
+ "loss": 0.7586,
2348
+ "step": 3330
2349
+ },
2350
+ {
2351
+ "epoch": 1.2,
2352
+ "grad_norm": 3.984375,
2353
+ "learning_rate": 1.051578947368421e-05,
2354
+ "loss": 0.8071,
2355
+ "step": 3340
2356
+ },
2357
+ {
2358
+ "epoch": 1.2,
2359
+ "grad_norm": 3.5625,
2360
+ "learning_rate": 1.05e-05,
2361
+ "loss": 0.7813,
2362
+ "step": 3350
2363
+ },
2364
+ {
2365
+ "epoch": 1.2,
2366
+ "grad_norm": 4.09375,
2367
+ "learning_rate": 1.048421052631579e-05,
2368
+ "loss": 0.7212,
2369
+ "step": 3360
2370
+ },
2371
+ {
2372
+ "epoch": 1.21,
2373
+ "grad_norm": 3.921875,
2374
+ "learning_rate": 1.0468421052631578e-05,
2375
+ "loss": 0.8672,
2376
+ "step": 3370
2377
+ },
2378
+ {
2379
+ "epoch": 1.21,
2380
+ "grad_norm": 4.15625,
2381
+ "learning_rate": 1.0452631578947369e-05,
2382
+ "loss": 0.7315,
2383
+ "step": 3380
2384
+ },
2385
+ {
2386
+ "epoch": 1.22,
2387
+ "grad_norm": 4.03125,
2388
+ "learning_rate": 1.0436842105263157e-05,
2389
+ "loss": 0.7626,
2390
+ "step": 3390
2391
+ },
2392
+ {
2393
+ "epoch": 1.22,
2394
+ "grad_norm": 3.890625,
2395
+ "learning_rate": 1.0421052631578948e-05,
2396
+ "loss": 0.7502,
2397
+ "step": 3400
2398
+ },
2399
+ {
2400
+ "epoch": 1.22,
2401
+ "grad_norm": 3.203125,
2402
+ "learning_rate": 1.0405263157894736e-05,
2403
+ "loss": 0.813,
2404
+ "step": 3410
2405
+ },
2406
+ {
2407
+ "epoch": 1.23,
2408
+ "grad_norm": 3.796875,
2409
+ "learning_rate": 1.0389473684210527e-05,
2410
+ "loss": 0.703,
2411
+ "step": 3420
2412
+ },
2413
+ {
2414
+ "epoch": 1.23,
2415
+ "grad_norm": 3.421875,
2416
+ "learning_rate": 1.0373684210526317e-05,
2417
+ "loss": 0.7764,
2418
+ "step": 3430
2419
+ },
2420
+ {
2421
+ "epoch": 1.23,
2422
+ "grad_norm": 3.765625,
2423
+ "learning_rate": 1.0357894736842106e-05,
2424
+ "loss": 0.7738,
2425
+ "step": 3440
2426
+ },
2427
+ {
2428
+ "epoch": 1.24,
2429
+ "grad_norm": 3.40625,
2430
+ "learning_rate": 1.0342105263157896e-05,
2431
+ "loss": 0.8006,
2432
+ "step": 3450
2433
+ },
2434
+ {
2435
+ "epoch": 1.24,
2436
+ "grad_norm": 3.046875,
2437
+ "learning_rate": 1.0326315789473685e-05,
2438
+ "loss": 0.6495,
2439
+ "step": 3460
2440
+ },
2441
+ {
2442
+ "epoch": 1.24,
2443
+ "grad_norm": 3.359375,
2444
+ "learning_rate": 1.0310526315789475e-05,
2445
+ "loss": 0.8163,
2446
+ "step": 3470
2447
+ },
2448
+ {
2449
+ "epoch": 1.25,
2450
+ "grad_norm": 2.921875,
2451
+ "learning_rate": 1.0294736842105264e-05,
2452
+ "loss": 0.7458,
2453
+ "step": 3480
2454
+ },
2455
+ {
2456
+ "epoch": 1.25,
2457
+ "grad_norm": 4.21875,
2458
+ "learning_rate": 1.0278947368421052e-05,
2459
+ "loss": 0.8422,
2460
+ "step": 3490
2461
+ },
2462
+ {
2463
+ "epoch": 1.26,
2464
+ "grad_norm": 4.03125,
2465
+ "learning_rate": 1.0263157894736843e-05,
2466
+ "loss": 0.8339,
2467
+ "step": 3500
2468
+ },
2469
+ {
2470
+ "epoch": 1.26,
2471
+ "grad_norm": 2.96875,
2472
+ "learning_rate": 1.0247368421052631e-05,
2473
+ "loss": 0.7214,
2474
+ "step": 3510
2475
+ },
2476
+ {
2477
+ "epoch": 1.26,
2478
+ "grad_norm": 4.0625,
2479
+ "learning_rate": 1.0231578947368422e-05,
2480
+ "loss": 0.7303,
2481
+ "step": 3520
2482
+ },
2483
+ {
2484
+ "epoch": 1.27,
2485
+ "grad_norm": 4.0,
2486
+ "learning_rate": 1.021578947368421e-05,
2487
+ "loss": 0.9404,
2488
+ "step": 3530
2489
+ },
2490
+ {
2491
+ "epoch": 1.27,
2492
+ "grad_norm": 4.0,
2493
+ "learning_rate": 1.02e-05,
2494
+ "loss": 0.657,
2495
+ "step": 3540
2496
+ },
2497
+ {
2498
+ "epoch": 1.27,
2499
+ "grad_norm": 4.65625,
2500
+ "learning_rate": 1.018421052631579e-05,
2501
+ "loss": 0.7938,
2502
+ "step": 3550
2503
+ },
2504
+ {
2505
+ "epoch": 1.28,
2506
+ "grad_norm": 3.671875,
2507
+ "learning_rate": 1.0168421052631578e-05,
2508
+ "loss": 0.769,
2509
+ "step": 3560
2510
+ },
2511
+ {
2512
+ "epoch": 1.28,
2513
+ "grad_norm": 3.765625,
2514
+ "learning_rate": 1.0152631578947368e-05,
2515
+ "loss": 0.8255,
2516
+ "step": 3570
2517
+ },
2518
+ {
2519
+ "epoch": 1.28,
2520
+ "grad_norm": 3.78125,
2521
+ "learning_rate": 1.0136842105263157e-05,
2522
+ "loss": 0.7263,
2523
+ "step": 3580
2524
+ },
2525
+ {
2526
+ "epoch": 1.29,
2527
+ "grad_norm": 4.59375,
2528
+ "learning_rate": 1.0121052631578949e-05,
2529
+ "loss": 0.669,
2530
+ "step": 3590
2531
+ },
2532
+ {
2533
+ "epoch": 1.29,
2534
+ "grad_norm": 4.3125,
2535
+ "learning_rate": 1.0105263157894738e-05,
2536
+ "loss": 0.85,
2537
+ "step": 3600
2538
+ },
2539
+ {
2540
+ "epoch": 1.29,
2541
+ "grad_norm": 4.875,
2542
+ "learning_rate": 1.0089473684210528e-05,
2543
+ "loss": 0.7775,
2544
+ "step": 3610
2545
+ },
2546
+ {
2547
+ "epoch": 1.3,
2548
+ "grad_norm": 4.0625,
2549
+ "learning_rate": 1.0073684210526317e-05,
2550
+ "loss": 0.7895,
2551
+ "step": 3620
2552
+ },
2553
+ {
2554
+ "epoch": 1.3,
2555
+ "grad_norm": 4.25,
2556
+ "learning_rate": 1.0057894736842105e-05,
2557
+ "loss": 0.7614,
2558
+ "step": 3630
2559
+ },
2560
+ {
2561
+ "epoch": 1.31,
2562
+ "grad_norm": 3.515625,
2563
+ "learning_rate": 1.0042105263157896e-05,
2564
+ "loss": 0.7206,
2565
+ "step": 3640
2566
+ },
2567
+ {
2568
+ "epoch": 1.31,
2569
+ "grad_norm": 4.75,
2570
+ "learning_rate": 1.0026315789473684e-05,
2571
+ "loss": 0.8636,
2572
+ "step": 3650
2573
+ },
2574
+ {
2575
+ "epoch": 1.31,
2576
+ "grad_norm": 3.390625,
2577
+ "learning_rate": 1.0010526315789474e-05,
2578
+ "loss": 0.8654,
2579
+ "step": 3660
2580
+ },
2581
+ {
2582
+ "epoch": 1.32,
2583
+ "grad_norm": 4.1875,
2584
+ "learning_rate": 9.994736842105263e-06,
2585
+ "loss": 0.8104,
2586
+ "step": 3670
2587
+ },
2588
+ {
2589
+ "epoch": 1.32,
2590
+ "grad_norm": 3.890625,
2591
+ "learning_rate": 9.978947368421053e-06,
2592
+ "loss": 0.7288,
2593
+ "step": 3680
2594
+ },
2595
+ {
2596
+ "epoch": 1.32,
2597
+ "grad_norm": 4.25,
2598
+ "learning_rate": 9.963157894736842e-06,
2599
+ "loss": 0.7018,
2600
+ "step": 3690
2601
+ },
2602
+ {
2603
+ "epoch": 1.33,
2604
+ "grad_norm": 2.984375,
2605
+ "learning_rate": 9.94736842105263e-06,
2606
+ "loss": 0.8822,
2607
+ "step": 3700
2608
+ },
2609
+ {
2610
+ "epoch": 1.33,
2611
+ "grad_norm": 3.234375,
2612
+ "learning_rate": 9.931578947368421e-06,
2613
+ "loss": 0.8066,
2614
+ "step": 3710
2615
+ },
2616
+ {
2617
+ "epoch": 1.33,
2618
+ "grad_norm": 4.0625,
2619
+ "learning_rate": 9.91578947368421e-06,
2620
+ "loss": 0.8029,
2621
+ "step": 3720
2622
+ },
2623
+ {
2624
+ "epoch": 1.34,
2625
+ "grad_norm": 4.625,
2626
+ "learning_rate": 9.9e-06,
2627
+ "loss": 0.8314,
2628
+ "step": 3730
2629
+ },
2630
+ {
2631
+ "epoch": 1.34,
2632
+ "grad_norm": 3.1875,
2633
+ "learning_rate": 9.884210526315789e-06,
2634
+ "loss": 0.7684,
2635
+ "step": 3740
2636
+ },
2637
+ {
2638
+ "epoch": 1.34,
2639
+ "grad_norm": 3.734375,
2640
+ "learning_rate": 9.868421052631579e-06,
2641
+ "loss": 0.8704,
2642
+ "step": 3750
2643
+ },
2644
+ {
2645
+ "epoch": 1.35,
2646
+ "grad_norm": 3.0625,
2647
+ "learning_rate": 9.85263157894737e-06,
2648
+ "loss": 0.6407,
2649
+ "step": 3760
2650
+ },
2651
+ {
2652
+ "epoch": 1.35,
2653
+ "grad_norm": 3.09375,
2654
+ "learning_rate": 9.836842105263158e-06,
2655
+ "loss": 0.8706,
2656
+ "step": 3770
2657
+ },
2658
+ {
2659
+ "epoch": 1.36,
2660
+ "grad_norm": 3.484375,
2661
+ "learning_rate": 9.821052631578948e-06,
2662
+ "loss": 0.6614,
2663
+ "step": 3780
2664
+ },
2665
+ {
2666
+ "epoch": 1.36,
2667
+ "grad_norm": 6.25,
2668
+ "learning_rate": 9.805263157894737e-06,
2669
+ "loss": 0.9533,
2670
+ "step": 3790
2671
+ },
2672
+ {
2673
+ "epoch": 1.36,
2674
+ "grad_norm": 3.546875,
2675
+ "learning_rate": 9.789473684210527e-06,
2676
+ "loss": 0.7741,
2677
+ "step": 3800
2678
+ },
2679
+ {
2680
+ "epoch": 1.37,
2681
+ "grad_norm": 4.46875,
2682
+ "learning_rate": 9.773684210526316e-06,
2683
+ "loss": 0.8563,
2684
+ "step": 3810
2685
+ },
2686
+ {
2687
+ "epoch": 1.37,
2688
+ "grad_norm": 3.71875,
2689
+ "learning_rate": 9.757894736842106e-06,
2690
+ "loss": 0.7783,
2691
+ "step": 3820
2692
+ },
2693
+ {
2694
+ "epoch": 1.37,
2695
+ "grad_norm": 3.65625,
2696
+ "learning_rate": 9.742105263157895e-06,
2697
+ "loss": 0.8157,
2698
+ "step": 3830
2699
+ },
2700
+ {
2701
+ "epoch": 1.38,
2702
+ "grad_norm": 3.484375,
2703
+ "learning_rate": 9.726315789473684e-06,
2704
+ "loss": 0.8049,
2705
+ "step": 3840
2706
+ },
2707
+ {
2708
+ "epoch": 1.38,
2709
+ "grad_norm": 4.03125,
2710
+ "learning_rate": 9.710526315789474e-06,
2711
+ "loss": 0.7587,
2712
+ "step": 3850
2713
+ },
2714
+ {
2715
+ "epoch": 1.38,
2716
+ "grad_norm": 3.59375,
2717
+ "learning_rate": 9.694736842105263e-06,
2718
+ "loss": 0.7946,
2719
+ "step": 3860
2720
+ },
2721
+ {
2722
+ "epoch": 1.39,
2723
+ "grad_norm": 4.46875,
2724
+ "learning_rate": 9.678947368421053e-06,
2725
+ "loss": 0.8342,
2726
+ "step": 3870
2727
+ },
2728
+ {
2729
+ "epoch": 1.39,
2730
+ "grad_norm": 3.28125,
2731
+ "learning_rate": 9.663157894736842e-06,
2732
+ "loss": 0.7517,
2733
+ "step": 3880
2734
+ },
2735
+ {
2736
+ "epoch": 1.4,
2737
+ "grad_norm": 3.34375,
2738
+ "learning_rate": 9.647368421052632e-06,
2739
+ "loss": 0.8783,
2740
+ "step": 3890
2741
+ },
2742
+ {
2743
+ "epoch": 1.4,
2744
+ "grad_norm": 3.40625,
2745
+ "learning_rate": 9.63157894736842e-06,
2746
+ "loss": 0.8379,
2747
+ "step": 3900
2748
+ },
2749
+ {
2750
+ "epoch": 1.4,
2751
+ "grad_norm": 3.09375,
2752
+ "learning_rate": 9.61578947368421e-06,
2753
+ "loss": 0.7126,
2754
+ "step": 3910
2755
+ },
2756
+ {
2757
+ "epoch": 1.41,
2758
+ "grad_norm": 3.8125,
2759
+ "learning_rate": 9.600000000000001e-06,
2760
+ "loss": 0.8612,
2761
+ "step": 3920
2762
+ },
2763
+ {
2764
+ "epoch": 1.41,
2765
+ "grad_norm": 4.90625,
2766
+ "learning_rate": 9.58421052631579e-06,
2767
+ "loss": 0.729,
2768
+ "step": 3930
2769
+ },
2770
+ {
2771
+ "epoch": 1.41,
2772
+ "grad_norm": 4.21875,
2773
+ "learning_rate": 9.56842105263158e-06,
2774
+ "loss": 0.7619,
2775
+ "step": 3940
2776
+ },
2777
+ {
2778
+ "epoch": 1.42,
2779
+ "grad_norm": 4.625,
2780
+ "learning_rate": 9.552631578947369e-06,
2781
+ "loss": 0.8175,
2782
+ "step": 3950
2783
+ },
2784
+ {
2785
+ "epoch": 1.42,
2786
+ "grad_norm": 3.1875,
2787
+ "learning_rate": 9.53684210526316e-06,
2788
+ "loss": 0.6401,
2789
+ "step": 3960
2790
+ },
2791
+ {
2792
+ "epoch": 1.42,
2793
+ "grad_norm": 3.953125,
2794
+ "learning_rate": 9.521052631578948e-06,
2795
+ "loss": 0.8253,
2796
+ "step": 3970
2797
+ },
2798
+ {
2799
+ "epoch": 1.43,
2800
+ "grad_norm": 4.25,
2801
+ "learning_rate": 9.505263157894737e-06,
2802
+ "loss": 0.7396,
2803
+ "step": 3980
2804
+ },
2805
+ {
2806
+ "epoch": 1.43,
2807
+ "grad_norm": 2.96875,
2808
+ "learning_rate": 9.489473684210527e-06,
2809
+ "loss": 0.7403,
2810
+ "step": 3990
2811
+ },
2812
+ {
2813
+ "epoch": 1.43,
2814
+ "grad_norm": 3.46875,
2815
+ "learning_rate": 9.473684210526315e-06,
2816
+ "loss": 0.7155,
2817
+ "step": 4000
2818
+ },
2819
+ {
2820
+ "epoch": 1.43,
2821
+ "eval_loss": 0.9205163717269897,
2822
+ "eval_runtime": 63.4494,
2823
+ "eval_samples_per_second": 15.761,
2824
+ "eval_steps_per_second": 15.761,
2825
+ "step": 4000
2826
+ },
2827
+ {
2828
+ "epoch": 1.44,
2829
+ "grad_norm": 3.796875,
2830
+ "learning_rate": 9.457894736842106e-06,
2831
+ "loss": 0.7251,
2832
+ "step": 4010
2833
+ },
2834
+ {
2835
+ "epoch": 1.44,
2836
+ "grad_norm": 3.859375,
2837
+ "learning_rate": 9.442105263157894e-06,
2838
+ "loss": 0.728,
2839
+ "step": 4020
2840
+ },
2841
+ {
2842
+ "epoch": 1.45,
2843
+ "grad_norm": 4.1875,
2844
+ "learning_rate": 9.426315789473685e-06,
2845
+ "loss": 0.9233,
2846
+ "step": 4030
2847
+ },
2848
+ {
2849
+ "epoch": 1.45,
2850
+ "grad_norm": 3.4375,
2851
+ "learning_rate": 9.410526315789473e-06,
2852
+ "loss": 0.854,
2853
+ "step": 4040
2854
+ },
2855
+ {
2856
+ "epoch": 1.45,
2857
+ "grad_norm": 4.09375,
2858
+ "learning_rate": 9.394736842105262e-06,
2859
+ "loss": 0.826,
2860
+ "step": 4050
2861
+ },
2862
+ {
2863
+ "epoch": 1.46,
2864
+ "grad_norm": 3.8125,
2865
+ "learning_rate": 9.378947368421052e-06,
2866
+ "loss": 0.7484,
2867
+ "step": 4060
2868
+ },
2869
+ {
2870
+ "epoch": 1.46,
2871
+ "grad_norm": 2.953125,
2872
+ "learning_rate": 9.363157894736841e-06,
2873
+ "loss": 0.818,
2874
+ "step": 4070
2875
+ },
2876
+ {
2877
+ "epoch": 1.46,
2878
+ "grad_norm": 5.09375,
2879
+ "learning_rate": 9.347368421052631e-06,
2880
+ "loss": 0.8191,
2881
+ "step": 4080
2882
+ },
2883
+ {
2884
+ "epoch": 1.47,
2885
+ "grad_norm": 2.546875,
2886
+ "learning_rate": 9.331578947368422e-06,
2887
+ "loss": 0.8531,
2888
+ "step": 4090
2889
+ },
2890
+ {
2891
+ "epoch": 1.47,
2892
+ "grad_norm": 3.078125,
2893
+ "learning_rate": 9.315789473684212e-06,
2894
+ "loss": 0.715,
2895
+ "step": 4100
2896
+ },
2897
+ {
2898
+ "epoch": 1.47,
2899
+ "grad_norm": 3.921875,
2900
+ "learning_rate": 9.3e-06,
2901
+ "loss": 0.6001,
2902
+ "step": 4110
2903
+ },
2904
+ {
2905
+ "epoch": 1.48,
2906
+ "grad_norm": 3.90625,
2907
+ "learning_rate": 9.28421052631579e-06,
2908
+ "loss": 0.7272,
2909
+ "step": 4120
2910
+ },
2911
+ {
2912
+ "epoch": 1.48,
2913
+ "grad_norm": 3.5625,
2914
+ "learning_rate": 9.26842105263158e-06,
2915
+ "loss": 0.7937,
2916
+ "step": 4130
2917
+ },
2918
+ {
2919
+ "epoch": 1.48,
2920
+ "grad_norm": 3.1875,
2921
+ "learning_rate": 9.252631578947368e-06,
2922
+ "loss": 0.8779,
2923
+ "step": 4140
2924
+ },
2925
+ {
2926
+ "epoch": 1.49,
2927
+ "grad_norm": 4.4375,
2928
+ "learning_rate": 9.236842105263159e-06,
2929
+ "loss": 0.7326,
2930
+ "step": 4150
2931
+ },
2932
+ {
2933
+ "epoch": 1.49,
2934
+ "grad_norm": 3.8125,
2935
+ "learning_rate": 9.221052631578947e-06,
2936
+ "loss": 0.6901,
2937
+ "step": 4160
2938
+ },
2939
+ {
2940
+ "epoch": 1.5,
2941
+ "grad_norm": 4.5,
2942
+ "learning_rate": 9.205263157894738e-06,
2943
+ "loss": 0.6585,
2944
+ "step": 4170
2945
+ },
2946
+ {
2947
+ "epoch": 1.5,
2948
+ "grad_norm": 3.59375,
2949
+ "learning_rate": 9.189473684210526e-06,
2950
+ "loss": 0.7615,
2951
+ "step": 4180
2952
+ },
2953
+ {
2954
+ "epoch": 1.5,
2955
+ "grad_norm": 3.640625,
2956
+ "learning_rate": 9.173684210526315e-06,
2957
+ "loss": 0.7954,
2958
+ "step": 4190
2959
+ },
2960
+ {
2961
+ "epoch": 1.51,
2962
+ "grad_norm": 3.625,
2963
+ "learning_rate": 9.157894736842105e-06,
2964
+ "loss": 0.793,
2965
+ "step": 4200
2966
+ },
2967
+ {
2968
+ "epoch": 1.51,
2969
+ "grad_norm": 4.5,
2970
+ "learning_rate": 9.142105263157894e-06,
2971
+ "loss": 0.8207,
2972
+ "step": 4210
2973
+ },
2974
+ {
2975
+ "epoch": 1.51,
2976
+ "grad_norm": 4.3125,
2977
+ "learning_rate": 9.126315789473684e-06,
2978
+ "loss": 0.8392,
2979
+ "step": 4220
2980
+ },
2981
+ {
2982
+ "epoch": 1.52,
2983
+ "grad_norm": 3.953125,
2984
+ "learning_rate": 9.110526315789473e-06,
2985
+ "loss": 0.9122,
2986
+ "step": 4230
2987
+ },
2988
+ {
2989
+ "epoch": 1.52,
2990
+ "grad_norm": 3.96875,
2991
+ "learning_rate": 9.094736842105263e-06,
2992
+ "loss": 0.7102,
2993
+ "step": 4240
2994
+ },
2995
+ {
2996
+ "epoch": 1.52,
2997
+ "grad_norm": 3.203125,
2998
+ "learning_rate": 9.078947368421054e-06,
2999
+ "loss": 0.8345,
3000
+ "step": 4250
3001
+ },
3002
+ {
3003
+ "epoch": 1.53,
3004
+ "grad_norm": 3.8125,
3005
+ "learning_rate": 9.063157894736842e-06,
3006
+ "loss": 0.7776,
3007
+ "step": 4260
3008
+ },
3009
+ {
3010
+ "epoch": 1.53,
3011
+ "grad_norm": 3.21875,
3012
+ "learning_rate": 9.047368421052633e-06,
3013
+ "loss": 0.8157,
3014
+ "step": 4270
3015
+ },
3016
+ {
3017
+ "epoch": 1.53,
3018
+ "grad_norm": 3.125,
3019
+ "learning_rate": 9.031578947368421e-06,
3020
+ "loss": 0.804,
3021
+ "step": 4280
3022
+ },
3023
+ {
3024
+ "epoch": 1.54,
3025
+ "grad_norm": 2.90625,
3026
+ "learning_rate": 9.015789473684212e-06,
3027
+ "loss": 0.8333,
3028
+ "step": 4290
3029
+ },
3030
+ {
3031
+ "epoch": 1.54,
3032
+ "grad_norm": 2.65625,
3033
+ "learning_rate": 9e-06,
3034
+ "loss": 0.7909,
3035
+ "step": 4300
3036
+ },
3037
+ {
3038
+ "epoch": 1.55,
3039
+ "grad_norm": 5.15625,
3040
+ "learning_rate": 8.98421052631579e-06,
3041
+ "loss": 0.737,
3042
+ "step": 4310
3043
+ },
3044
+ {
3045
+ "epoch": 1.55,
3046
+ "grad_norm": 4.3125,
3047
+ "learning_rate": 8.96842105263158e-06,
3048
+ "loss": 0.8604,
3049
+ "step": 4320
3050
+ },
3051
+ {
3052
+ "epoch": 1.55,
3053
+ "grad_norm": 4.21875,
3054
+ "learning_rate": 8.952631578947368e-06,
3055
+ "loss": 0.7934,
3056
+ "step": 4330
3057
+ },
3058
+ {
3059
+ "epoch": 1.56,
3060
+ "grad_norm": 3.296875,
3061
+ "learning_rate": 8.936842105263158e-06,
3062
+ "loss": 0.6818,
3063
+ "step": 4340
3064
+ },
3065
+ {
3066
+ "epoch": 1.56,
3067
+ "grad_norm": 3.40625,
3068
+ "learning_rate": 8.921052631578947e-06,
3069
+ "loss": 0.7706,
3070
+ "step": 4350
3071
+ },
3072
+ {
3073
+ "epoch": 1.56,
3074
+ "grad_norm": 4.8125,
3075
+ "learning_rate": 8.905263157894737e-06,
3076
+ "loss": 0.8421,
3077
+ "step": 4360
3078
+ },
3079
+ {
3080
+ "epoch": 1.57,
3081
+ "grad_norm": 4.375,
3082
+ "learning_rate": 8.889473684210526e-06,
3083
+ "loss": 0.7885,
3084
+ "step": 4370
3085
+ },
3086
+ {
3087
+ "epoch": 1.57,
3088
+ "grad_norm": 2.59375,
3089
+ "learning_rate": 8.873684210526316e-06,
3090
+ "loss": 0.7816,
3091
+ "step": 4380
3092
+ },
3093
+ {
3094
+ "epoch": 1.57,
3095
+ "grad_norm": 4.0625,
3096
+ "learning_rate": 8.857894736842105e-06,
3097
+ "loss": 0.7497,
3098
+ "step": 4390
3099
+ },
3100
+ {
3101
+ "epoch": 1.58,
3102
+ "grad_norm": 3.140625,
3103
+ "learning_rate": 8.842105263157893e-06,
3104
+ "loss": 0.6982,
3105
+ "step": 4400
3106
+ },
3107
+ {
3108
+ "epoch": 1.58,
3109
+ "grad_norm": 3.578125,
3110
+ "learning_rate": 8.826315789473684e-06,
3111
+ "loss": 0.899,
3112
+ "step": 4410
3113
+ },
3114
+ {
3115
+ "epoch": 1.59,
3116
+ "grad_norm": 3.484375,
3117
+ "learning_rate": 8.810526315789474e-06,
3118
+ "loss": 0.7436,
3119
+ "step": 4420
3120
+ },
3121
+ {
3122
+ "epoch": 1.59,
3123
+ "grad_norm": 5.03125,
3124
+ "learning_rate": 8.794736842105264e-06,
3125
+ "loss": 0.8648,
3126
+ "step": 4430
3127
+ },
3128
+ {
3129
+ "epoch": 1.59,
3130
+ "grad_norm": 4.46875,
3131
+ "learning_rate": 8.778947368421053e-06,
3132
+ "loss": 0.7769,
3133
+ "step": 4440
3134
+ },
3135
+ {
3136
+ "epoch": 1.6,
3137
+ "grad_norm": 4.1875,
3138
+ "learning_rate": 8.763157894736843e-06,
3139
+ "loss": 0.8222,
3140
+ "step": 4450
3141
+ },
3142
+ {
3143
+ "epoch": 1.6,
3144
+ "grad_norm": 4.40625,
3145
+ "learning_rate": 8.747368421052632e-06,
3146
+ "loss": 0.8141,
3147
+ "step": 4460
3148
+ },
3149
+ {
3150
+ "epoch": 1.6,
3151
+ "grad_norm": 3.984375,
3152
+ "learning_rate": 8.731578947368422e-06,
3153
+ "loss": 0.8907,
3154
+ "step": 4470
3155
+ },
3156
+ {
3157
+ "epoch": 1.61,
3158
+ "grad_norm": 4.71875,
3159
+ "learning_rate": 8.715789473684211e-06,
3160
+ "loss": 0.6802,
3161
+ "step": 4480
3162
+ },
3163
+ {
3164
+ "epoch": 1.61,
3165
+ "grad_norm": 4.09375,
3166
+ "learning_rate": 8.7e-06,
3167
+ "loss": 0.7409,
3168
+ "step": 4490
3169
+ },
3170
+ {
3171
+ "epoch": 1.61,
3172
+ "grad_norm": 3.0625,
3173
+ "learning_rate": 8.68421052631579e-06,
3174
+ "loss": 0.6993,
3175
+ "step": 4500
3176
+ },
3177
+ {
3178
+ "epoch": 1.62,
3179
+ "grad_norm": 4.15625,
3180
+ "learning_rate": 8.668421052631579e-06,
3181
+ "loss": 0.696,
3182
+ "step": 4510
3183
+ },
3184
+ {
3185
+ "epoch": 1.62,
3186
+ "grad_norm": 2.640625,
3187
+ "learning_rate": 8.652631578947369e-06,
3188
+ "loss": 0.8095,
3189
+ "step": 4520
3190
+ },
3191
+ {
3192
+ "epoch": 1.62,
3193
+ "grad_norm": 3.0625,
3194
+ "learning_rate": 8.636842105263158e-06,
3195
+ "loss": 0.6222,
3196
+ "step": 4530
3197
+ },
3198
+ {
3199
+ "epoch": 1.63,
3200
+ "grad_norm": 3.28125,
3201
+ "learning_rate": 8.621052631578948e-06,
3202
+ "loss": 0.7821,
3203
+ "step": 4540
3204
+ },
3205
+ {
3206
+ "epoch": 1.63,
3207
+ "grad_norm": 3.890625,
3208
+ "learning_rate": 8.605263157894737e-06,
3209
+ "loss": 0.7144,
3210
+ "step": 4550
3211
+ },
3212
+ {
3213
+ "epoch": 1.64,
3214
+ "grad_norm": 2.90625,
3215
+ "learning_rate": 8.589473684210525e-06,
3216
+ "loss": 0.762,
3217
+ "step": 4560
3218
+ },
3219
+ {
3220
+ "epoch": 1.64,
3221
+ "grad_norm": 4.5625,
3222
+ "learning_rate": 8.573684210526316e-06,
3223
+ "loss": 0.8258,
3224
+ "step": 4570
3225
+ },
3226
+ {
3227
+ "epoch": 1.64,
3228
+ "grad_norm": 3.671875,
3229
+ "learning_rate": 8.557894736842104e-06,
3230
+ "loss": 0.853,
3231
+ "step": 4580
3232
+ },
3233
+ {
3234
+ "epoch": 1.65,
3235
+ "grad_norm": 3.65625,
3236
+ "learning_rate": 8.542105263157896e-06,
3237
+ "loss": 0.8365,
3238
+ "step": 4590
3239
+ },
3240
+ {
3241
+ "epoch": 1.65,
3242
+ "grad_norm": 3.9375,
3243
+ "learning_rate": 8.526315789473685e-06,
3244
+ "loss": 0.8248,
3245
+ "step": 4600
3246
+ },
3247
+ {
3248
+ "epoch": 1.65,
3249
+ "grad_norm": 4.5,
3250
+ "learning_rate": 8.510526315789475e-06,
3251
+ "loss": 0.7747,
3252
+ "step": 4610
3253
+ },
3254
+ {
3255
+ "epoch": 1.66,
3256
+ "grad_norm": 3.328125,
3257
+ "learning_rate": 8.494736842105264e-06,
3258
+ "loss": 0.8395,
3259
+ "step": 4620
3260
+ },
3261
+ {
3262
+ "epoch": 1.66,
3263
+ "grad_norm": 3.46875,
3264
+ "learning_rate": 8.478947368421053e-06,
3265
+ "loss": 0.789,
3266
+ "step": 4630
3267
+ },
3268
+ {
3269
+ "epoch": 1.66,
3270
+ "grad_norm": 3.65625,
3271
+ "learning_rate": 8.463157894736843e-06,
3272
+ "loss": 0.7156,
3273
+ "step": 4640
3274
+ },
3275
+ {
3276
+ "epoch": 1.67,
3277
+ "grad_norm": 3.375,
3278
+ "learning_rate": 8.447368421052632e-06,
3279
+ "loss": 0.7587,
3280
+ "step": 4650
3281
+ },
3282
+ {
3283
+ "epoch": 1.67,
3284
+ "grad_norm": 3.578125,
3285
+ "learning_rate": 8.431578947368422e-06,
3286
+ "loss": 0.7462,
3287
+ "step": 4660
3288
+ },
3289
+ {
3290
+ "epoch": 1.67,
3291
+ "grad_norm": 3.046875,
3292
+ "learning_rate": 8.41578947368421e-06,
3293
+ "loss": 0.7426,
3294
+ "step": 4670
3295
+ },
3296
+ {
3297
+ "epoch": 1.68,
3298
+ "grad_norm": 2.6875,
3299
+ "learning_rate": 8.400000000000001e-06,
3300
+ "loss": 0.823,
3301
+ "step": 4680
3302
+ },
3303
+ {
3304
+ "epoch": 1.68,
3305
+ "grad_norm": 4.5,
3306
+ "learning_rate": 8.38421052631579e-06,
3307
+ "loss": 0.7004,
3308
+ "step": 4690
3309
+ },
3310
+ {
3311
+ "epoch": 1.69,
3312
+ "grad_norm": 3.390625,
3313
+ "learning_rate": 8.368421052631578e-06,
3314
+ "loss": 0.7149,
3315
+ "step": 4700
3316
+ },
3317
+ {
3318
+ "epoch": 1.69,
3319
+ "grad_norm": 3.96875,
3320
+ "learning_rate": 8.352631578947369e-06,
3321
+ "loss": 0.7712,
3322
+ "step": 4710
3323
+ },
3324
+ {
3325
+ "epoch": 1.69,
3326
+ "grad_norm": 4.0625,
3327
+ "learning_rate": 8.336842105263157e-06,
3328
+ "loss": 0.785,
3329
+ "step": 4720
3330
+ },
3331
+ {
3332
+ "epoch": 1.7,
3333
+ "grad_norm": 4.375,
3334
+ "learning_rate": 8.321052631578947e-06,
3335
+ "loss": 0.7706,
3336
+ "step": 4730
3337
+ },
3338
+ {
3339
+ "epoch": 1.7,
3340
+ "grad_norm": 4.46875,
3341
+ "learning_rate": 8.305263157894736e-06,
3342
+ "loss": 0.8951,
3343
+ "step": 4740
3344
+ },
3345
+ {
3346
+ "epoch": 1.7,
3347
+ "grad_norm": 4.6875,
3348
+ "learning_rate": 8.289473684210526e-06,
3349
+ "loss": 0.876,
3350
+ "step": 4750
3351
+ },
3352
+ {
3353
+ "epoch": 1.71,
3354
+ "grad_norm": 3.515625,
3355
+ "learning_rate": 8.273684210526317e-06,
3356
+ "loss": 0.7448,
3357
+ "step": 4760
3358
+ },
3359
+ {
3360
+ "epoch": 1.71,
3361
+ "grad_norm": 4.78125,
3362
+ "learning_rate": 8.257894736842105e-06,
3363
+ "loss": 0.7204,
3364
+ "step": 4770
3365
+ },
3366
+ {
3367
+ "epoch": 1.71,
3368
+ "grad_norm": 4.375,
3369
+ "learning_rate": 8.242105263157896e-06,
3370
+ "loss": 0.7273,
3371
+ "step": 4780
3372
+ },
3373
+ {
3374
+ "epoch": 1.72,
3375
+ "grad_norm": 4.34375,
3376
+ "learning_rate": 8.226315789473684e-06,
3377
+ "loss": 0.796,
3378
+ "step": 4790
3379
+ },
3380
+ {
3381
+ "epoch": 1.72,
3382
+ "grad_norm": 4.1875,
3383
+ "learning_rate": 8.210526315789475e-06,
3384
+ "loss": 0.685,
3385
+ "step": 4800
3386
+ },
3387
+ {
3388
+ "epoch": 1.72,
3389
+ "grad_norm": 4.25,
3390
+ "learning_rate": 8.194736842105263e-06,
3391
+ "loss": 0.9413,
3392
+ "step": 4810
3393
+ },
3394
+ {
3395
+ "epoch": 1.73,
3396
+ "grad_norm": 3.953125,
3397
+ "learning_rate": 8.178947368421054e-06,
3398
+ "loss": 0.8169,
3399
+ "step": 4820
3400
+ },
3401
+ {
3402
+ "epoch": 1.73,
3403
+ "grad_norm": 4.34375,
3404
+ "learning_rate": 8.163157894736842e-06,
3405
+ "loss": 0.7446,
3406
+ "step": 4830
3407
+ },
3408
+ {
3409
+ "epoch": 1.74,
3410
+ "grad_norm": 4.4375,
3411
+ "learning_rate": 8.147368421052631e-06,
3412
+ "loss": 0.7173,
3413
+ "step": 4840
3414
+ },
3415
+ {
3416
+ "epoch": 1.74,
3417
+ "grad_norm": 3.6875,
3418
+ "learning_rate": 8.131578947368421e-06,
3419
+ "loss": 0.7899,
3420
+ "step": 4850
3421
+ },
3422
+ {
3423
+ "epoch": 1.74,
3424
+ "grad_norm": 5.90625,
3425
+ "learning_rate": 8.11578947368421e-06,
3426
+ "loss": 0.7022,
3427
+ "step": 4860
3428
+ },
3429
+ {
3430
+ "epoch": 1.75,
3431
+ "grad_norm": 4.625,
3432
+ "learning_rate": 8.1e-06,
3433
+ "loss": 0.7862,
3434
+ "step": 4870
3435
+ },
3436
+ {
3437
+ "epoch": 1.75,
3438
+ "grad_norm": 3.046875,
3439
+ "learning_rate": 8.084210526315789e-06,
3440
+ "loss": 0.6982,
3441
+ "step": 4880
3442
+ },
3443
+ {
3444
+ "epoch": 1.75,
3445
+ "grad_norm": 3.75,
3446
+ "learning_rate": 8.06842105263158e-06,
3447
+ "loss": 0.816,
3448
+ "step": 4890
3449
+ },
3450
+ {
3451
+ "epoch": 1.76,
3452
+ "grad_norm": 3.828125,
3453
+ "learning_rate": 8.052631578947368e-06,
3454
+ "loss": 0.7501,
3455
+ "step": 4900
3456
+ },
3457
+ {
3458
+ "epoch": 1.76,
3459
+ "grad_norm": 3.078125,
3460
+ "learning_rate": 8.036842105263157e-06,
3461
+ "loss": 0.7459,
3462
+ "step": 4910
3463
+ },
3464
+ {
3465
+ "epoch": 1.76,
3466
+ "grad_norm": 3.34375,
3467
+ "learning_rate": 8.021052631578949e-06,
3468
+ "loss": 0.7931,
3469
+ "step": 4920
3470
+ },
3471
+ {
3472
+ "epoch": 1.77,
3473
+ "grad_norm": 5.46875,
3474
+ "learning_rate": 8.005263157894737e-06,
3475
+ "loss": 0.86,
3476
+ "step": 4930
3477
+ },
3478
+ {
3479
+ "epoch": 1.77,
3480
+ "grad_norm": 5.15625,
3481
+ "learning_rate": 7.989473684210528e-06,
3482
+ "loss": 0.7316,
3483
+ "step": 4940
3484
+ },
3485
+ {
3486
+ "epoch": 1.78,
3487
+ "grad_norm": 2.75,
3488
+ "learning_rate": 7.973684210526316e-06,
3489
+ "loss": 0.7452,
3490
+ "step": 4950
3491
+ },
3492
+ {
3493
+ "epoch": 1.78,
3494
+ "grad_norm": 5.1875,
3495
+ "learning_rate": 7.957894736842107e-06,
3496
+ "loss": 0.7412,
3497
+ "step": 4960
3498
+ },
3499
+ {
3500
+ "epoch": 1.78,
3501
+ "grad_norm": 3.609375,
3502
+ "learning_rate": 7.942105263157895e-06,
3503
+ "loss": 0.8649,
3504
+ "step": 4970
3505
+ },
3506
+ {
3507
+ "epoch": 1.79,
3508
+ "grad_norm": 4.375,
3509
+ "learning_rate": 7.926315789473684e-06,
3510
+ "loss": 0.6906,
3511
+ "step": 4980
3512
+ },
3513
+ {
3514
+ "epoch": 1.79,
3515
+ "grad_norm": 3.109375,
3516
+ "learning_rate": 7.910526315789474e-06,
3517
+ "loss": 0.8417,
3518
+ "step": 4990
3519
+ },
3520
+ {
3521
+ "epoch": 1.79,
3522
+ "grad_norm": 3.734375,
3523
+ "learning_rate": 7.894736842105263e-06,
3524
+ "loss": 0.7587,
3525
+ "step": 5000
3526
+ },
3527
+ {
3528
+ "epoch": 1.8,
3529
+ "grad_norm": 3.53125,
3530
+ "learning_rate": 7.878947368421053e-06,
3531
+ "loss": 0.7634,
3532
+ "step": 5010
3533
+ },
3534
+ {
3535
+ "epoch": 1.8,
3536
+ "grad_norm": 3.71875,
3537
+ "learning_rate": 7.863157894736842e-06,
3538
+ "loss": 0.6928,
3539
+ "step": 5020
3540
+ },
3541
+ {
3542
+ "epoch": 1.8,
3543
+ "grad_norm": 3.515625,
3544
+ "learning_rate": 7.847368421052632e-06,
3545
+ "loss": 0.7879,
3546
+ "step": 5030
3547
+ },
3548
+ {
3549
+ "epoch": 1.81,
3550
+ "grad_norm": 3.515625,
3551
+ "learning_rate": 7.831578947368421e-06,
3552
+ "loss": 0.7295,
3553
+ "step": 5040
3554
+ },
3555
+ {
3556
+ "epoch": 1.81,
3557
+ "grad_norm": 4.53125,
3558
+ "learning_rate": 7.81578947368421e-06,
3559
+ "loss": 0.8336,
3560
+ "step": 5050
3561
+ },
3562
+ {
3563
+ "epoch": 1.81,
3564
+ "grad_norm": 4.9375,
3565
+ "learning_rate": 7.8e-06,
3566
+ "loss": 0.8038,
3567
+ "step": 5060
3568
+ },
3569
+ {
3570
+ "epoch": 1.82,
3571
+ "grad_norm": 3.828125,
3572
+ "learning_rate": 7.784210526315789e-06,
3573
+ "loss": 0.7216,
3574
+ "step": 5070
3575
+ },
3576
+ {
3577
+ "epoch": 1.82,
3578
+ "grad_norm": 2.9375,
3579
+ "learning_rate": 7.768421052631579e-06,
3580
+ "loss": 0.8983,
3581
+ "step": 5080
3582
+ },
3583
+ {
3584
+ "epoch": 1.83,
3585
+ "grad_norm": 4.375,
3586
+ "learning_rate": 7.75263157894737e-06,
3587
+ "loss": 0.8195,
3588
+ "step": 5090
3589
+ },
3590
+ {
3591
+ "epoch": 1.83,
3592
+ "grad_norm": 3.59375,
3593
+ "learning_rate": 7.73684210526316e-06,
3594
+ "loss": 0.7921,
3595
+ "step": 5100
3596
+ },
3597
+ {
3598
+ "epoch": 1.83,
3599
+ "grad_norm": 3.328125,
3600
+ "learning_rate": 7.721052631578948e-06,
3601
+ "loss": 0.8048,
3602
+ "step": 5110
3603
+ },
3604
+ {
3605
+ "epoch": 1.84,
3606
+ "grad_norm": 4.1875,
3607
+ "learning_rate": 7.705263157894737e-06,
3608
+ "loss": 0.8693,
3609
+ "step": 5120
3610
+ },
3611
+ {
3612
+ "epoch": 1.84,
3613
+ "grad_norm": 3.546875,
3614
+ "learning_rate": 7.689473684210527e-06,
3615
+ "loss": 0.9209,
3616
+ "step": 5130
3617
+ },
3618
+ {
3619
+ "epoch": 1.84,
3620
+ "grad_norm": 4.5625,
3621
+ "learning_rate": 7.673684210526316e-06,
3622
+ "loss": 0.7321,
3623
+ "step": 5140
3624
+ },
3625
+ {
3626
+ "epoch": 1.85,
3627
+ "grad_norm": 6.1875,
3628
+ "learning_rate": 7.657894736842106e-06,
3629
+ "loss": 0.8093,
3630
+ "step": 5150
3631
+ },
3632
+ {
3633
+ "epoch": 1.85,
3634
+ "grad_norm": 3.515625,
3635
+ "learning_rate": 7.642105263157895e-06,
3636
+ "loss": 0.7525,
3637
+ "step": 5160
3638
+ },
3639
+ {
3640
+ "epoch": 1.85,
3641
+ "grad_norm": 3.15625,
3642
+ "learning_rate": 7.626315789473685e-06,
3643
+ "loss": 0.7801,
3644
+ "step": 5170
3645
+ },
3646
+ {
3647
+ "epoch": 1.86,
3648
+ "grad_norm": 4.28125,
3649
+ "learning_rate": 7.610526315789474e-06,
3650
+ "loss": 0.7285,
3651
+ "step": 5180
3652
+ },
3653
+ {
3654
+ "epoch": 1.86,
3655
+ "grad_norm": 3.890625,
3656
+ "learning_rate": 7.594736842105262e-06,
3657
+ "loss": 0.8266,
3658
+ "step": 5190
3659
+ },
3660
+ {
3661
+ "epoch": 1.86,
3662
+ "grad_norm": 3.703125,
3663
+ "learning_rate": 7.578947368421053e-06,
3664
+ "loss": 0.8848,
3665
+ "step": 5200
3666
+ },
3667
+ {
3668
+ "epoch": 1.87,
3669
+ "grad_norm": 4.15625,
3670
+ "learning_rate": 7.563157894736842e-06,
3671
+ "loss": 0.7602,
3672
+ "step": 5210
3673
+ },
3674
+ {
3675
+ "epoch": 1.87,
3676
+ "grad_norm": 2.78125,
3677
+ "learning_rate": 7.5473684210526326e-06,
3678
+ "loss": 0.5791,
3679
+ "step": 5220
3680
+ },
3681
+ {
3682
+ "epoch": 1.88,
3683
+ "grad_norm": 4.78125,
3684
+ "learning_rate": 7.531578947368421e-06,
3685
+ "loss": 0.8423,
3686
+ "step": 5230
3687
+ },
3688
+ {
3689
+ "epoch": 1.88,
3690
+ "grad_norm": 5.375,
3691
+ "learning_rate": 7.5157894736842115e-06,
3692
+ "loss": 0.7297,
3693
+ "step": 5240
3694
+ },
3695
+ {
3696
+ "epoch": 1.88,
3697
+ "grad_norm": 3.140625,
3698
+ "learning_rate": 7.5e-06,
3699
+ "loss": 0.7256,
3700
+ "step": 5250
3701
+ },
3702
+ {
3703
+ "epoch": 1.89,
3704
+ "grad_norm": 3.5625,
3705
+ "learning_rate": 7.48421052631579e-06,
3706
+ "loss": 0.7688,
3707
+ "step": 5260
3708
+ },
3709
+ {
3710
+ "epoch": 1.89,
3711
+ "grad_norm": 3.015625,
3712
+ "learning_rate": 7.468421052631579e-06,
3713
+ "loss": 0.7772,
3714
+ "step": 5270
3715
+ },
3716
+ {
3717
+ "epoch": 1.89,
3718
+ "grad_norm": 3.984375,
3719
+ "learning_rate": 7.452631578947369e-06,
3720
+ "loss": 0.8327,
3721
+ "step": 5280
3722
+ },
3723
+ {
3724
+ "epoch": 1.9,
3725
+ "grad_norm": 2.4375,
3726
+ "learning_rate": 7.436842105263157e-06,
3727
+ "loss": 0.7026,
3728
+ "step": 5290
3729
+ },
3730
+ {
3731
+ "epoch": 1.9,
3732
+ "grad_norm": 5.625,
3733
+ "learning_rate": 7.421052631578948e-06,
3734
+ "loss": 0.7857,
3735
+ "step": 5300
3736
+ },
3737
+ {
3738
+ "epoch": 1.9,
3739
+ "grad_norm": 3.171875,
3740
+ "learning_rate": 7.405263157894737e-06,
3741
+ "loss": 0.821,
3742
+ "step": 5310
3743
+ },
3744
+ {
3745
+ "epoch": 1.91,
3746
+ "grad_norm": 4.875,
3747
+ "learning_rate": 7.389473684210527e-06,
3748
+ "loss": 0.8284,
3749
+ "step": 5320
3750
+ },
3751
+ {
3752
+ "epoch": 1.91,
3753
+ "grad_norm": 3.5625,
3754
+ "learning_rate": 7.373684210526316e-06,
3755
+ "loss": 0.8545,
3756
+ "step": 5330
3757
+ },
3758
+ {
3759
+ "epoch": 1.92,
3760
+ "grad_norm": 3.984375,
3761
+ "learning_rate": 7.357894736842106e-06,
3762
+ "loss": 0.773,
3763
+ "step": 5340
3764
+ },
3765
+ {
3766
+ "epoch": 1.92,
3767
+ "grad_norm": 3.515625,
3768
+ "learning_rate": 7.342105263157895e-06,
3769
+ "loss": 0.777,
3770
+ "step": 5350
3771
+ },
3772
+ {
3773
+ "epoch": 1.92,
3774
+ "grad_norm": 4.78125,
3775
+ "learning_rate": 7.326315789473684e-06,
3776
+ "loss": 0.7814,
3777
+ "step": 5360
3778
+ },
3779
+ {
3780
+ "epoch": 1.93,
3781
+ "grad_norm": 4.5,
3782
+ "learning_rate": 7.310526315789473e-06,
3783
+ "loss": 0.7729,
3784
+ "step": 5370
3785
+ },
3786
+ {
3787
+ "epoch": 1.93,
3788
+ "grad_norm": 3.296875,
3789
+ "learning_rate": 7.2947368421052636e-06,
3790
+ "loss": 0.7087,
3791
+ "step": 5380
3792
+ },
3793
+ {
3794
+ "epoch": 1.93,
3795
+ "grad_norm": 5.28125,
3796
+ "learning_rate": 7.278947368421053e-06,
3797
+ "loss": 0.7536,
3798
+ "step": 5390
3799
+ },
3800
+ {
3801
+ "epoch": 1.94,
3802
+ "grad_norm": 4.375,
3803
+ "learning_rate": 7.2631578947368426e-06,
3804
+ "loss": 0.7356,
3805
+ "step": 5400
3806
+ },
3807
+ {
3808
+ "epoch": 1.94,
3809
+ "grad_norm": 4.125,
3810
+ "learning_rate": 7.247368421052632e-06,
3811
+ "loss": 0.7866,
3812
+ "step": 5410
3813
+ },
3814
+ {
3815
+ "epoch": 1.94,
3816
+ "grad_norm": 3.234375,
3817
+ "learning_rate": 7.2315789473684215e-06,
3818
+ "loss": 0.7174,
3819
+ "step": 5420
3820
+ },
3821
+ {
3822
+ "epoch": 1.95,
3823
+ "grad_norm": 4.0625,
3824
+ "learning_rate": 7.21578947368421e-06,
3825
+ "loss": 0.7692,
3826
+ "step": 5430
3827
+ },
3828
+ {
3829
+ "epoch": 1.95,
3830
+ "grad_norm": 4.65625,
3831
+ "learning_rate": 7.2e-06,
3832
+ "loss": 0.7995,
3833
+ "step": 5440
3834
+ },
3835
+ {
3836
+ "epoch": 1.95,
3837
+ "grad_norm": 4.28125,
3838
+ "learning_rate": 7.184210526315789e-06,
3839
+ "loss": 0.6847,
3840
+ "step": 5450
3841
+ },
3842
+ {
3843
+ "epoch": 1.96,
3844
+ "grad_norm": 3.453125,
3845
+ "learning_rate": 7.1684210526315795e-06,
3846
+ "loss": 0.7299,
3847
+ "step": 5460
3848
+ },
3849
+ {
3850
+ "epoch": 1.96,
3851
+ "grad_norm": 4.25,
3852
+ "learning_rate": 7.152631578947369e-06,
3853
+ "loss": 0.898,
3854
+ "step": 5470
3855
+ },
3856
+ {
3857
+ "epoch": 1.97,
3858
+ "grad_norm": 2.875,
3859
+ "learning_rate": 7.1368421052631585e-06,
3860
+ "loss": 0.7486,
3861
+ "step": 5480
3862
+ },
3863
+ {
3864
+ "epoch": 1.97,
3865
+ "grad_norm": 4.09375,
3866
+ "learning_rate": 7.121052631578948e-06,
3867
+ "loss": 0.7697,
3868
+ "step": 5490
3869
+ },
3870
+ {
3871
+ "epoch": 1.97,
3872
+ "grad_norm": 3.34375,
3873
+ "learning_rate": 7.105263157894737e-06,
3874
+ "loss": 0.7652,
3875
+ "step": 5500
3876
+ },
3877
+ {
3878
+ "epoch": 1.98,
3879
+ "grad_norm": 3.84375,
3880
+ "learning_rate": 7.089473684210526e-06,
3881
+ "loss": 0.7781,
3882
+ "step": 5510
3883
+ },
3884
+ {
3885
+ "epoch": 1.98,
3886
+ "grad_norm": 3.515625,
3887
+ "learning_rate": 7.073684210526316e-06,
3888
+ "loss": 0.8173,
3889
+ "step": 5520
3890
+ },
3891
+ {
3892
+ "epoch": 1.98,
3893
+ "grad_norm": 4.9375,
3894
+ "learning_rate": 7.057894736842105e-06,
3895
+ "loss": 0.7466,
3896
+ "step": 5530
3897
+ },
3898
+ {
3899
+ "epoch": 1.99,
3900
+ "grad_norm": 4.53125,
3901
+ "learning_rate": 7.042105263157895e-06,
3902
+ "loss": 0.8411,
3903
+ "step": 5540
3904
+ },
3905
+ {
3906
+ "epoch": 1.99,
3907
+ "grad_norm": 3.921875,
3908
+ "learning_rate": 7.026315789473685e-06,
3909
+ "loss": 0.8425,
3910
+ "step": 5550
3911
+ },
3912
+ {
3913
+ "epoch": 1.99,
3914
+ "grad_norm": 4.625,
3915
+ "learning_rate": 7.010526315789474e-06,
3916
+ "loss": 0.732,
3917
+ "step": 5560
3918
+ },
3919
+ {
3920
+ "epoch": 2.0,
3921
+ "grad_norm": 4.0,
3922
+ "learning_rate": 6.994736842105263e-06,
3923
+ "loss": 0.7609,
3924
+ "step": 5570
3925
+ },
3926
+ {
3927
+ "epoch": 2.0,
3928
+ "grad_norm": 3.828125,
3929
+ "learning_rate": 6.9789473684210525e-06,
3930
+ "loss": 0.7138,
3931
+ "step": 5580
3932
+ },
3933
+ {
3934
+ "epoch": 2.0,
3935
+ "grad_norm": 3.1875,
3936
+ "learning_rate": 6.963157894736842e-06,
3937
+ "loss": 0.7042,
3938
+ "step": 5590
3939
+ },
3940
+ {
3941
+ "epoch": 2.01,
3942
+ "grad_norm": 3.46875,
3943
+ "learning_rate": 6.9473684210526315e-06,
3944
+ "loss": 0.7703,
3945
+ "step": 5600
3946
+ },
3947
+ {
3948
+ "epoch": 2.01,
3949
+ "grad_norm": 3.203125,
3950
+ "learning_rate": 6.931578947368421e-06,
3951
+ "loss": 0.7245,
3952
+ "step": 5610
3953
+ },
3954
+ {
3955
+ "epoch": 2.02,
3956
+ "grad_norm": 4.46875,
3957
+ "learning_rate": 6.9157894736842105e-06,
3958
+ "loss": 0.8364,
3959
+ "step": 5620
3960
+ },
3961
+ {
3962
+ "epoch": 2.02,
3963
+ "grad_norm": 3.65625,
3964
+ "learning_rate": 6.900000000000001e-06,
3965
+ "loss": 0.7138,
3966
+ "step": 5630
3967
+ },
3968
+ {
3969
+ "epoch": 2.02,
3970
+ "grad_norm": 4.6875,
3971
+ "learning_rate": 6.8842105263157895e-06,
3972
+ "loss": 0.7988,
3973
+ "step": 5640
3974
+ },
3975
+ {
3976
+ "epoch": 2.03,
3977
+ "grad_norm": 4.59375,
3978
+ "learning_rate": 6.868421052631579e-06,
3979
+ "loss": 0.7449,
3980
+ "step": 5650
3981
+ },
3982
+ {
3983
+ "epoch": 2.03,
3984
+ "grad_norm": 3.65625,
3985
+ "learning_rate": 6.8526315789473685e-06,
3986
+ "loss": 0.7205,
3987
+ "step": 5660
3988
+ },
3989
+ {
3990
+ "epoch": 2.03,
3991
+ "grad_norm": 3.484375,
3992
+ "learning_rate": 6.836842105263158e-06,
3993
+ "loss": 0.7968,
3994
+ "step": 5670
3995
+ },
3996
+ {
3997
+ "epoch": 2.04,
3998
+ "grad_norm": 3.28125,
3999
+ "learning_rate": 6.8210526315789475e-06,
4000
+ "loss": 0.7845,
4001
+ "step": 5680
4002
+ },
4003
+ {
4004
+ "epoch": 2.04,
4005
+ "grad_norm": 4.90625,
4006
+ "learning_rate": 6.805263157894737e-06,
4007
+ "loss": 0.824,
4008
+ "step": 5690
4009
+ },
4010
+ {
4011
+ "epoch": 2.04,
4012
+ "grad_norm": 4.84375,
4013
+ "learning_rate": 6.7894736842105264e-06,
4014
+ "loss": 0.7712,
4015
+ "step": 5700
4016
+ },
4017
+ {
4018
+ "epoch": 2.05,
4019
+ "grad_norm": 3.390625,
4020
+ "learning_rate": 6.773684210526317e-06,
4021
+ "loss": 0.8114,
4022
+ "step": 5710
4023
+ },
4024
+ {
4025
+ "epoch": 2.05,
4026
+ "grad_norm": 3.75,
4027
+ "learning_rate": 6.7578947368421054e-06,
4028
+ "loss": 0.7015,
4029
+ "step": 5720
4030
+ },
4031
+ {
4032
+ "epoch": 2.05,
4033
+ "grad_norm": 5.71875,
4034
+ "learning_rate": 6.742105263157895e-06,
4035
+ "loss": 0.6591,
4036
+ "step": 5730
4037
+ },
4038
+ {
4039
+ "epoch": 2.06,
4040
+ "grad_norm": 4.1875,
4041
+ "learning_rate": 6.726315789473684e-06,
4042
+ "loss": 0.7993,
4043
+ "step": 5740
4044
+ },
4045
+ {
4046
+ "epoch": 2.06,
4047
+ "grad_norm": 3.09375,
4048
+ "learning_rate": 6.710526315789474e-06,
4049
+ "loss": 0.8858,
4050
+ "step": 5750
4051
+ },
4052
+ {
4053
+ "epoch": 2.07,
4054
+ "grad_norm": 7.1875,
4055
+ "learning_rate": 6.694736842105263e-06,
4056
+ "loss": 0.8687,
4057
+ "step": 5760
4058
+ },
4059
+ {
4060
+ "epoch": 2.07,
4061
+ "grad_norm": 4.6875,
4062
+ "learning_rate": 6.678947368421053e-06,
4063
+ "loss": 0.7612,
4064
+ "step": 5770
4065
+ },
4066
+ {
4067
+ "epoch": 2.07,
4068
+ "grad_norm": 3.65625,
4069
+ "learning_rate": 6.663157894736842e-06,
4070
+ "loss": 0.7144,
4071
+ "step": 5780
4072
+ },
4073
+ {
4074
+ "epoch": 2.08,
4075
+ "grad_norm": 3.9375,
4076
+ "learning_rate": 6.647368421052631e-06,
4077
+ "loss": 0.78,
4078
+ "step": 5790
4079
+ },
4080
+ {
4081
+ "epoch": 2.08,
4082
+ "grad_norm": 4.3125,
4083
+ "learning_rate": 6.631578947368421e-06,
4084
+ "loss": 0.7489,
4085
+ "step": 5800
4086
+ },
4087
+ {
4088
+ "epoch": 2.08,
4089
+ "grad_norm": 3.453125,
4090
+ "learning_rate": 6.615789473684211e-06,
4091
+ "loss": 0.7635,
4092
+ "step": 5810
4093
+ },
4094
+ {
4095
+ "epoch": 2.09,
4096
+ "grad_norm": 2.90625,
4097
+ "learning_rate": 6.6e-06,
4098
+ "loss": 0.7358,
4099
+ "step": 5820
4100
+ },
4101
+ {
4102
+ "epoch": 2.09,
4103
+ "grad_norm": 2.96875,
4104
+ "learning_rate": 6.58421052631579e-06,
4105
+ "loss": 0.7606,
4106
+ "step": 5830
4107
+ },
4108
+ {
4109
+ "epoch": 2.09,
4110
+ "grad_norm": 4.4375,
4111
+ "learning_rate": 6.568421052631579e-06,
4112
+ "loss": 0.8625,
4113
+ "step": 5840
4114
+ },
4115
+ {
4116
+ "epoch": 2.1,
4117
+ "grad_norm": 3.625,
4118
+ "learning_rate": 6.552631578947369e-06,
4119
+ "loss": 0.7742,
4120
+ "step": 5850
4121
+ },
4122
+ {
4123
+ "epoch": 2.1,
4124
+ "grad_norm": 3.109375,
4125
+ "learning_rate": 6.5368421052631575e-06,
4126
+ "loss": 0.7725,
4127
+ "step": 5860
4128
+ },
4129
+ {
4130
+ "epoch": 2.11,
4131
+ "grad_norm": 3.96875,
4132
+ "learning_rate": 6.521052631578947e-06,
4133
+ "loss": 0.808,
4134
+ "step": 5870
4135
+ },
4136
+ {
4137
+ "epoch": 2.11,
4138
+ "grad_norm": 3.890625,
4139
+ "learning_rate": 6.505263157894737e-06,
4140
+ "loss": 0.756,
4141
+ "step": 5880
4142
+ },
4143
+ {
4144
+ "epoch": 2.11,
4145
+ "grad_norm": 3.734375,
4146
+ "learning_rate": 6.489473684210527e-06,
4147
+ "loss": 0.7694,
4148
+ "step": 5890
4149
+ },
4150
+ {
4151
+ "epoch": 2.12,
4152
+ "grad_norm": 4.21875,
4153
+ "learning_rate": 6.473684210526316e-06,
4154
+ "loss": 0.7441,
4155
+ "step": 5900
4156
+ },
4157
+ {
4158
+ "epoch": 2.12,
4159
+ "grad_norm": 3.671875,
4160
+ "learning_rate": 6.457894736842106e-06,
4161
+ "loss": 0.7455,
4162
+ "step": 5910
4163
+ },
4164
+ {
4165
+ "epoch": 2.12,
4166
+ "grad_norm": 4.46875,
4167
+ "learning_rate": 6.442105263157895e-06,
4168
+ "loss": 0.8083,
4169
+ "step": 5920
4170
+ },
4171
+ {
4172
+ "epoch": 2.13,
4173
+ "grad_norm": 3.84375,
4174
+ "learning_rate": 6.426315789473684e-06,
4175
+ "loss": 0.7362,
4176
+ "step": 5930
4177
+ },
4178
+ {
4179
+ "epoch": 2.13,
4180
+ "grad_norm": 4.53125,
4181
+ "learning_rate": 6.410526315789473e-06,
4182
+ "loss": 0.8266,
4183
+ "step": 5940
4184
+ },
4185
+ {
4186
+ "epoch": 2.13,
4187
+ "grad_norm": 5.28125,
4188
+ "learning_rate": 6.394736842105263e-06,
4189
+ "loss": 0.779,
4190
+ "step": 5950
4191
+ },
4192
+ {
4193
+ "epoch": 2.14,
4194
+ "grad_norm": 3.71875,
4195
+ "learning_rate": 6.378947368421053e-06,
4196
+ "loss": 0.7819,
4197
+ "step": 5960
4198
+ },
4199
+ {
4200
+ "epoch": 2.14,
4201
+ "grad_norm": 3.875,
4202
+ "learning_rate": 6.363157894736843e-06,
4203
+ "loss": 0.774,
4204
+ "step": 5970
4205
+ },
4206
+ {
4207
+ "epoch": 2.14,
4208
+ "grad_norm": 3.265625,
4209
+ "learning_rate": 6.347368421052632e-06,
4210
+ "loss": 0.7117,
4211
+ "step": 5980
4212
+ },
4213
+ {
4214
+ "epoch": 2.15,
4215
+ "grad_norm": 3.015625,
4216
+ "learning_rate": 6.331578947368422e-06,
4217
+ "loss": 0.71,
4218
+ "step": 5990
4219
+ },
4220
+ {
4221
+ "epoch": 2.15,
4222
+ "grad_norm": 3.984375,
4223
+ "learning_rate": 6.31578947368421e-06,
4224
+ "loss": 0.7329,
4225
+ "step": 6000
4226
+ },
4227
+ {
4228
+ "epoch": 2.15,
4229
+ "eval_loss": 0.8978841304779053,
4230
+ "eval_runtime": 63.4016,
4231
+ "eval_samples_per_second": 15.772,
4232
+ "eval_steps_per_second": 15.772,
4233
+ "step": 6000
4234
+ }
4235
+ ],
4236
+ "logging_steps": 10,
4237
+ "max_steps": 10000,
4238
+ "num_input_tokens_seen": 0,
4239
+ "num_train_epochs": 4,
4240
+ "save_steps": 2000,
4241
+ "total_flos": 9.683532251136e+16,
4242
+ "train_batch_size": 1,
4243
+ "trial_name": null,
4244
+ "trial_params": null
4245
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d602b2c99d2658ac27305bba445001533af6893069ff3fd80cf173fc5e71cf17
3
+ size 4920