Uploaded checkpoint-1000
Browse files- config.json +30 -0
- generation_config.json +4 -0
- model.safetensors +3 -0
- optimizer.pt +3 -0
- rng_state.pth +3 -0
- scheduler.pt +3 -0
- trainer_state.json +729 -0
- 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-Curriculum-Subjects-6-to-7",
|
| 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 sha256:2b1fd8884f5b91c6a101024d81fb6446b3b0c1792daf26799032d1c8a41cedd6
|
| 3 |
+
size 2836579040
|
optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:723d36768c71157889cc9d960f42d72528d0e7fc57d5b3178248578c45213bfc
|
| 3 |
+
size 5673376169
|
rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cc42c52a1f904d5f80e51646f04f0cafaec70013c88aebd27b8446a73c004742
|
| 3 |
+
size 14244
|
scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:606e0975e746bdba2a559c41750c66492e3ca938517218ba62e2e53ed4a011f7
|
| 3 |
+
size 1064
|
trainer_state.json
ADDED
|
@@ -0,0 +1,729 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 0.16666666666666666,
|
| 5 |
+
"eval_steps": 1000,
|
| 6 |
+
"global_step": 1000,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.0,
|
| 13 |
+
"grad_norm": 21.75,
|
| 14 |
+
"learning_rate": 3.7125e-06,
|
| 15 |
+
"loss": 1.6333,
|
| 16 |
+
"step": 10
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.0,
|
| 20 |
+
"grad_norm": 11.75,
|
| 21 |
+
"learning_rate": 3.675e-06,
|
| 22 |
+
"loss": 1.8863,
|
| 23 |
+
"step": 20
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.01,
|
| 27 |
+
"grad_norm": 30.25,
|
| 28 |
+
"learning_rate": 3.6375e-06,
|
| 29 |
+
"loss": 2.1646,
|
| 30 |
+
"step": 30
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.01,
|
| 34 |
+
"grad_norm": 20.0,
|
| 35 |
+
"learning_rate": 3.6e-06,
|
| 36 |
+
"loss": 2.0098,
|
| 37 |
+
"step": 40
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.01,
|
| 41 |
+
"grad_norm": 21.125,
|
| 42 |
+
"learning_rate": 3.5624999999999998e-06,
|
| 43 |
+
"loss": 1.7331,
|
| 44 |
+
"step": 50
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.01,
|
| 48 |
+
"grad_norm": 18.125,
|
| 49 |
+
"learning_rate": 3.5249999999999997e-06,
|
| 50 |
+
"loss": 1.8913,
|
| 51 |
+
"step": 60
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.01,
|
| 55 |
+
"grad_norm": 17.5,
|
| 56 |
+
"learning_rate": 3.4875e-06,
|
| 57 |
+
"loss": 1.8686,
|
| 58 |
+
"step": 70
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.01,
|
| 62 |
+
"grad_norm": 37.0,
|
| 63 |
+
"learning_rate": 3.4500000000000004e-06,
|
| 64 |
+
"loss": 1.6756,
|
| 65 |
+
"step": 80
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.01,
|
| 69 |
+
"grad_norm": 21.125,
|
| 70 |
+
"learning_rate": 3.4125000000000004e-06,
|
| 71 |
+
"loss": 2.0029,
|
| 72 |
+
"step": 90
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.02,
|
| 76 |
+
"grad_norm": 14.0,
|
| 77 |
+
"learning_rate": 3.3750000000000003e-06,
|
| 78 |
+
"loss": 1.4412,
|
| 79 |
+
"step": 100
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.02,
|
| 83 |
+
"grad_norm": 12.75,
|
| 84 |
+
"learning_rate": 3.3375000000000002e-06,
|
| 85 |
+
"loss": 2.0556,
|
| 86 |
+
"step": 110
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.02,
|
| 90 |
+
"grad_norm": 31.625,
|
| 91 |
+
"learning_rate": 3.3e-06,
|
| 92 |
+
"loss": 1.9298,
|
| 93 |
+
"step": 120
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.02,
|
| 97 |
+
"grad_norm": 15.0625,
|
| 98 |
+
"learning_rate": 3.2625e-06,
|
| 99 |
+
"loss": 2.0686,
|
| 100 |
+
"step": 130
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.02,
|
| 104 |
+
"grad_norm": 15.75,
|
| 105 |
+
"learning_rate": 3.225e-06,
|
| 106 |
+
"loss": 1.6107,
|
| 107 |
+
"step": 140
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.03,
|
| 111 |
+
"grad_norm": 28.5,
|
| 112 |
+
"learning_rate": 3.1875e-06,
|
| 113 |
+
"loss": 2.4047,
|
| 114 |
+
"step": 150
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.03,
|
| 118 |
+
"grad_norm": 17.375,
|
| 119 |
+
"learning_rate": 3.15e-06,
|
| 120 |
+
"loss": 1.5916,
|
| 121 |
+
"step": 160
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.03,
|
| 125 |
+
"grad_norm": 23.0,
|
| 126 |
+
"learning_rate": 3.1125e-06,
|
| 127 |
+
"loss": 1.9109,
|
| 128 |
+
"step": 170
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.03,
|
| 132 |
+
"grad_norm": 27.5,
|
| 133 |
+
"learning_rate": 3.0749999999999998e-06,
|
| 134 |
+
"loss": 2.1359,
|
| 135 |
+
"step": 180
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.03,
|
| 139 |
+
"grad_norm": 18.0,
|
| 140 |
+
"learning_rate": 3.0375e-06,
|
| 141 |
+
"loss": 2.1144,
|
| 142 |
+
"step": 190
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 0.03,
|
| 146 |
+
"grad_norm": 14.4375,
|
| 147 |
+
"learning_rate": 3e-06,
|
| 148 |
+
"loss": 1.6672,
|
| 149 |
+
"step": 200
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 0.04,
|
| 153 |
+
"grad_norm": 22.125,
|
| 154 |
+
"learning_rate": 2.9625e-06,
|
| 155 |
+
"loss": 1.6704,
|
| 156 |
+
"step": 210
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 0.04,
|
| 160 |
+
"grad_norm": 28.5,
|
| 161 |
+
"learning_rate": 2.9250000000000004e-06,
|
| 162 |
+
"loss": 1.9675,
|
| 163 |
+
"step": 220
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.04,
|
| 167 |
+
"grad_norm": 13.0,
|
| 168 |
+
"learning_rate": 2.8875000000000003e-06,
|
| 169 |
+
"loss": 1.9873,
|
| 170 |
+
"step": 230
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.04,
|
| 174 |
+
"grad_norm": 13.8125,
|
| 175 |
+
"learning_rate": 2.8500000000000002e-06,
|
| 176 |
+
"loss": 1.7557,
|
| 177 |
+
"step": 240
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.04,
|
| 181 |
+
"grad_norm": 19.125,
|
| 182 |
+
"learning_rate": 2.8125e-06,
|
| 183 |
+
"loss": 1.7556,
|
| 184 |
+
"step": 250
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 0.04,
|
| 188 |
+
"grad_norm": 16.5,
|
| 189 |
+
"learning_rate": 2.775e-06,
|
| 190 |
+
"loss": 2.0693,
|
| 191 |
+
"step": 260
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 0.04,
|
| 195 |
+
"grad_norm": 18.25,
|
| 196 |
+
"learning_rate": 2.7375e-06,
|
| 197 |
+
"loss": 1.6011,
|
| 198 |
+
"step": 270
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 0.05,
|
| 202 |
+
"grad_norm": 21.875,
|
| 203 |
+
"learning_rate": 2.7e-06,
|
| 204 |
+
"loss": 1.6058,
|
| 205 |
+
"step": 280
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.05,
|
| 209 |
+
"grad_norm": 18.0,
|
| 210 |
+
"learning_rate": 2.6625e-06,
|
| 211 |
+
"loss": 1.8085,
|
| 212 |
+
"step": 290
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 0.05,
|
| 216 |
+
"grad_norm": 14.875,
|
| 217 |
+
"learning_rate": 2.625e-06,
|
| 218 |
+
"loss": 1.777,
|
| 219 |
+
"step": 300
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 0.05,
|
| 223 |
+
"grad_norm": 20.75,
|
| 224 |
+
"learning_rate": 2.5875e-06,
|
| 225 |
+
"loss": 1.8708,
|
| 226 |
+
"step": 310
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.05,
|
| 230 |
+
"grad_norm": 14.625,
|
| 231 |
+
"learning_rate": 2.55e-06,
|
| 232 |
+
"loss": 1.5629,
|
| 233 |
+
"step": 320
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 0.06,
|
| 237 |
+
"grad_norm": 16.125,
|
| 238 |
+
"learning_rate": 2.5125e-06,
|
| 239 |
+
"loss": 1.7416,
|
| 240 |
+
"step": 330
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 0.06,
|
| 244 |
+
"grad_norm": 19.5,
|
| 245 |
+
"learning_rate": 2.475e-06,
|
| 246 |
+
"loss": 1.9123,
|
| 247 |
+
"step": 340
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 0.06,
|
| 251 |
+
"grad_norm": 17.375,
|
| 252 |
+
"learning_rate": 2.4375e-06,
|
| 253 |
+
"loss": 1.6956,
|
| 254 |
+
"step": 350
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 0.06,
|
| 258 |
+
"grad_norm": 30.125,
|
| 259 |
+
"learning_rate": 2.4000000000000003e-06,
|
| 260 |
+
"loss": 1.863,
|
| 261 |
+
"step": 360
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 0.06,
|
| 265 |
+
"grad_norm": 24.25,
|
| 266 |
+
"learning_rate": 2.3625000000000003e-06,
|
| 267 |
+
"loss": 1.7432,
|
| 268 |
+
"step": 370
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.06,
|
| 272 |
+
"grad_norm": 15.5625,
|
| 273 |
+
"learning_rate": 2.325e-06,
|
| 274 |
+
"loss": 1.6194,
|
| 275 |
+
"step": 380
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.07,
|
| 279 |
+
"grad_norm": 20.75,
|
| 280 |
+
"learning_rate": 2.2875e-06,
|
| 281 |
+
"loss": 2.1294,
|
| 282 |
+
"step": 390
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.07,
|
| 286 |
+
"grad_norm": 24.125,
|
| 287 |
+
"learning_rate": 2.25e-06,
|
| 288 |
+
"loss": 1.8663,
|
| 289 |
+
"step": 400
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.07,
|
| 293 |
+
"grad_norm": 22.625,
|
| 294 |
+
"learning_rate": 2.2125e-06,
|
| 295 |
+
"loss": 1.8445,
|
| 296 |
+
"step": 410
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.07,
|
| 300 |
+
"grad_norm": 23.375,
|
| 301 |
+
"learning_rate": 2.175e-06,
|
| 302 |
+
"loss": 2.1122,
|
| 303 |
+
"step": 420
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.07,
|
| 307 |
+
"grad_norm": 20.875,
|
| 308 |
+
"learning_rate": 2.1375e-06,
|
| 309 |
+
"loss": 1.8364,
|
| 310 |
+
"step": 430
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.07,
|
| 314 |
+
"grad_norm": 20.0,
|
| 315 |
+
"learning_rate": 2.1000000000000002e-06,
|
| 316 |
+
"loss": 1.6375,
|
| 317 |
+
"step": 440
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 0.07,
|
| 321 |
+
"grad_norm": 18.25,
|
| 322 |
+
"learning_rate": 2.0625e-06,
|
| 323 |
+
"loss": 1.5882,
|
| 324 |
+
"step": 450
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 0.08,
|
| 328 |
+
"grad_norm": 34.25,
|
| 329 |
+
"learning_rate": 2.025e-06,
|
| 330 |
+
"loss": 2.0043,
|
| 331 |
+
"step": 460
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 0.08,
|
| 335 |
+
"grad_norm": 28.75,
|
| 336 |
+
"learning_rate": 1.9875e-06,
|
| 337 |
+
"loss": 1.9231,
|
| 338 |
+
"step": 470
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.08,
|
| 342 |
+
"grad_norm": 17.5,
|
| 343 |
+
"learning_rate": 1.95e-06,
|
| 344 |
+
"loss": 2.0631,
|
| 345 |
+
"step": 480
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.08,
|
| 349 |
+
"grad_norm": 17.125,
|
| 350 |
+
"learning_rate": 1.9125e-06,
|
| 351 |
+
"loss": 2.0359,
|
| 352 |
+
"step": 490
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.08,
|
| 356 |
+
"grad_norm": 21.0,
|
| 357 |
+
"learning_rate": 1.875e-06,
|
| 358 |
+
"loss": 1.8175,
|
| 359 |
+
"step": 500
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.09,
|
| 363 |
+
"grad_norm": 17.5,
|
| 364 |
+
"learning_rate": 1.8375e-06,
|
| 365 |
+
"loss": 2.0844,
|
| 366 |
+
"step": 510
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 0.09,
|
| 370 |
+
"grad_norm": 20.25,
|
| 371 |
+
"learning_rate": 1.8e-06,
|
| 372 |
+
"loss": 1.6329,
|
| 373 |
+
"step": 520
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 0.09,
|
| 377 |
+
"grad_norm": 12.375,
|
| 378 |
+
"learning_rate": 1.7624999999999999e-06,
|
| 379 |
+
"loss": 1.9352,
|
| 380 |
+
"step": 530
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 0.09,
|
| 384 |
+
"grad_norm": 21.375,
|
| 385 |
+
"learning_rate": 1.7250000000000002e-06,
|
| 386 |
+
"loss": 1.8842,
|
| 387 |
+
"step": 540
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 0.09,
|
| 391 |
+
"grad_norm": 16.375,
|
| 392 |
+
"learning_rate": 1.6875000000000001e-06,
|
| 393 |
+
"loss": 2.1543,
|
| 394 |
+
"step": 550
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.09,
|
| 398 |
+
"grad_norm": 15.625,
|
| 399 |
+
"learning_rate": 1.65e-06,
|
| 400 |
+
"loss": 2.0122,
|
| 401 |
+
"step": 560
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 0.1,
|
| 405 |
+
"grad_norm": 16.0,
|
| 406 |
+
"learning_rate": 1.6125e-06,
|
| 407 |
+
"loss": 2.3606,
|
| 408 |
+
"step": 570
|
| 409 |
+
},
|
| 410 |
+
{
|
| 411 |
+
"epoch": 0.1,
|
| 412 |
+
"grad_norm": 22.375,
|
| 413 |
+
"learning_rate": 1.575e-06,
|
| 414 |
+
"loss": 1.8354,
|
| 415 |
+
"step": 580
|
| 416 |
+
},
|
| 417 |
+
{
|
| 418 |
+
"epoch": 0.1,
|
| 419 |
+
"grad_norm": 18.75,
|
| 420 |
+
"learning_rate": 1.5374999999999999e-06,
|
| 421 |
+
"loss": 1.962,
|
| 422 |
+
"step": 590
|
| 423 |
+
},
|
| 424 |
+
{
|
| 425 |
+
"epoch": 0.1,
|
| 426 |
+
"grad_norm": 22.625,
|
| 427 |
+
"learning_rate": 1.5e-06,
|
| 428 |
+
"loss": 1.6487,
|
| 429 |
+
"step": 600
|
| 430 |
+
},
|
| 431 |
+
{
|
| 432 |
+
"epoch": 0.1,
|
| 433 |
+
"grad_norm": 17.875,
|
| 434 |
+
"learning_rate": 1.4625000000000002e-06,
|
| 435 |
+
"loss": 1.5496,
|
| 436 |
+
"step": 610
|
| 437 |
+
},
|
| 438 |
+
{
|
| 439 |
+
"epoch": 0.1,
|
| 440 |
+
"grad_norm": 20.25,
|
| 441 |
+
"learning_rate": 1.4250000000000001e-06,
|
| 442 |
+
"loss": 2.0923,
|
| 443 |
+
"step": 620
|
| 444 |
+
},
|
| 445 |
+
{
|
| 446 |
+
"epoch": 0.1,
|
| 447 |
+
"grad_norm": 20.125,
|
| 448 |
+
"learning_rate": 1.3875e-06,
|
| 449 |
+
"loss": 2.0688,
|
| 450 |
+
"step": 630
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"epoch": 0.11,
|
| 454 |
+
"grad_norm": 14.875,
|
| 455 |
+
"learning_rate": 1.35e-06,
|
| 456 |
+
"loss": 2.1762,
|
| 457 |
+
"step": 640
|
| 458 |
+
},
|
| 459 |
+
{
|
| 460 |
+
"epoch": 0.11,
|
| 461 |
+
"grad_norm": 51.0,
|
| 462 |
+
"learning_rate": 1.3125e-06,
|
| 463 |
+
"loss": 1.7076,
|
| 464 |
+
"step": 650
|
| 465 |
+
},
|
| 466 |
+
{
|
| 467 |
+
"epoch": 0.11,
|
| 468 |
+
"grad_norm": 27.0,
|
| 469 |
+
"learning_rate": 1.275e-06,
|
| 470 |
+
"loss": 2.1918,
|
| 471 |
+
"step": 660
|
| 472 |
+
},
|
| 473 |
+
{
|
| 474 |
+
"epoch": 0.11,
|
| 475 |
+
"grad_norm": 16.125,
|
| 476 |
+
"learning_rate": 1.2375e-06,
|
| 477 |
+
"loss": 2.0706,
|
| 478 |
+
"step": 670
|
| 479 |
+
},
|
| 480 |
+
{
|
| 481 |
+
"epoch": 0.11,
|
| 482 |
+
"grad_norm": 21.0,
|
| 483 |
+
"learning_rate": 1.2000000000000002e-06,
|
| 484 |
+
"loss": 1.8599,
|
| 485 |
+
"step": 680
|
| 486 |
+
},
|
| 487 |
+
{
|
| 488 |
+
"epoch": 0.12,
|
| 489 |
+
"grad_norm": 19.375,
|
| 490 |
+
"learning_rate": 1.1625e-06,
|
| 491 |
+
"loss": 2.1312,
|
| 492 |
+
"step": 690
|
| 493 |
+
},
|
| 494 |
+
{
|
| 495 |
+
"epoch": 0.12,
|
| 496 |
+
"grad_norm": 22.5,
|
| 497 |
+
"learning_rate": 1.125e-06,
|
| 498 |
+
"loss": 1.9219,
|
| 499 |
+
"step": 700
|
| 500 |
+
},
|
| 501 |
+
{
|
| 502 |
+
"epoch": 0.12,
|
| 503 |
+
"grad_norm": 24.75,
|
| 504 |
+
"learning_rate": 1.0875e-06,
|
| 505 |
+
"loss": 2.0628,
|
| 506 |
+
"step": 710
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.12,
|
| 510 |
+
"grad_norm": 16.875,
|
| 511 |
+
"learning_rate": 1.0500000000000001e-06,
|
| 512 |
+
"loss": 1.999,
|
| 513 |
+
"step": 720
|
| 514 |
+
},
|
| 515 |
+
{
|
| 516 |
+
"epoch": 0.12,
|
| 517 |
+
"grad_norm": 20.875,
|
| 518 |
+
"learning_rate": 1.0125e-06,
|
| 519 |
+
"loss": 1.6709,
|
| 520 |
+
"step": 730
|
| 521 |
+
},
|
| 522 |
+
{
|
| 523 |
+
"epoch": 0.12,
|
| 524 |
+
"grad_norm": 20.125,
|
| 525 |
+
"learning_rate": 9.75e-07,
|
| 526 |
+
"loss": 1.9329,
|
| 527 |
+
"step": 740
|
| 528 |
+
},
|
| 529 |
+
{
|
| 530 |
+
"epoch": 0.12,
|
| 531 |
+
"grad_norm": 15.125,
|
| 532 |
+
"learning_rate": 9.375e-07,
|
| 533 |
+
"loss": 1.8998,
|
| 534 |
+
"step": 750
|
| 535 |
+
},
|
| 536 |
+
{
|
| 537 |
+
"epoch": 0.13,
|
| 538 |
+
"grad_norm": 20.0,
|
| 539 |
+
"learning_rate": 9e-07,
|
| 540 |
+
"loss": 1.4411,
|
| 541 |
+
"step": 760
|
| 542 |
+
},
|
| 543 |
+
{
|
| 544 |
+
"epoch": 0.13,
|
| 545 |
+
"grad_norm": 18.125,
|
| 546 |
+
"learning_rate": 8.625000000000001e-07,
|
| 547 |
+
"loss": 1.758,
|
| 548 |
+
"step": 770
|
| 549 |
+
},
|
| 550 |
+
{
|
| 551 |
+
"epoch": 0.13,
|
| 552 |
+
"grad_norm": 17.0,
|
| 553 |
+
"learning_rate": 8.25e-07,
|
| 554 |
+
"loss": 2.0136,
|
| 555 |
+
"step": 780
|
| 556 |
+
},
|
| 557 |
+
{
|
| 558 |
+
"epoch": 0.13,
|
| 559 |
+
"grad_norm": 19.25,
|
| 560 |
+
"learning_rate": 7.875e-07,
|
| 561 |
+
"loss": 1.8072,
|
| 562 |
+
"step": 790
|
| 563 |
+
},
|
| 564 |
+
{
|
| 565 |
+
"epoch": 0.13,
|
| 566 |
+
"grad_norm": 14.625,
|
| 567 |
+
"learning_rate": 7.5e-07,
|
| 568 |
+
"loss": 1.9614,
|
| 569 |
+
"step": 800
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 0.14,
|
| 573 |
+
"grad_norm": 17.5,
|
| 574 |
+
"learning_rate": 7.125000000000001e-07,
|
| 575 |
+
"loss": 1.9077,
|
| 576 |
+
"step": 810
|
| 577 |
+
},
|
| 578 |
+
{
|
| 579 |
+
"epoch": 0.14,
|
| 580 |
+
"grad_norm": 25.625,
|
| 581 |
+
"learning_rate": 6.75e-07,
|
| 582 |
+
"loss": 1.6237,
|
| 583 |
+
"step": 820
|
| 584 |
+
},
|
| 585 |
+
{
|
| 586 |
+
"epoch": 0.14,
|
| 587 |
+
"grad_norm": 23.875,
|
| 588 |
+
"learning_rate": 6.375e-07,
|
| 589 |
+
"loss": 1.875,
|
| 590 |
+
"step": 830
|
| 591 |
+
},
|
| 592 |
+
{
|
| 593 |
+
"epoch": 0.14,
|
| 594 |
+
"grad_norm": 29.5,
|
| 595 |
+
"learning_rate": 6.000000000000001e-07,
|
| 596 |
+
"loss": 2.0392,
|
| 597 |
+
"step": 840
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 0.14,
|
| 601 |
+
"grad_norm": 15.6875,
|
| 602 |
+
"learning_rate": 5.625e-07,
|
| 603 |
+
"loss": 1.9254,
|
| 604 |
+
"step": 850
|
| 605 |
+
},
|
| 606 |
+
{
|
| 607 |
+
"epoch": 0.14,
|
| 608 |
+
"grad_norm": 11.8125,
|
| 609 |
+
"learning_rate": 5.250000000000001e-07,
|
| 610 |
+
"loss": 1.4531,
|
| 611 |
+
"step": 860
|
| 612 |
+
},
|
| 613 |
+
{
|
| 614 |
+
"epoch": 0.14,
|
| 615 |
+
"grad_norm": 17.875,
|
| 616 |
+
"learning_rate": 4.875e-07,
|
| 617 |
+
"loss": 1.8028,
|
| 618 |
+
"step": 870
|
| 619 |
+
},
|
| 620 |
+
{
|
| 621 |
+
"epoch": 0.15,
|
| 622 |
+
"grad_norm": 15.875,
|
| 623 |
+
"learning_rate": 4.5e-07,
|
| 624 |
+
"loss": 1.7695,
|
| 625 |
+
"step": 880
|
| 626 |
+
},
|
| 627 |
+
{
|
| 628 |
+
"epoch": 0.15,
|
| 629 |
+
"grad_norm": 20.375,
|
| 630 |
+
"learning_rate": 4.125e-07,
|
| 631 |
+
"loss": 1.8477,
|
| 632 |
+
"step": 890
|
| 633 |
+
},
|
| 634 |
+
{
|
| 635 |
+
"epoch": 0.15,
|
| 636 |
+
"grad_norm": 16.25,
|
| 637 |
+
"learning_rate": 3.75e-07,
|
| 638 |
+
"loss": 2.0242,
|
| 639 |
+
"step": 900
|
| 640 |
+
},
|
| 641 |
+
{
|
| 642 |
+
"epoch": 0.15,
|
| 643 |
+
"grad_norm": 30.5,
|
| 644 |
+
"learning_rate": 3.375e-07,
|
| 645 |
+
"loss": 1.79,
|
| 646 |
+
"step": 910
|
| 647 |
+
},
|
| 648 |
+
{
|
| 649 |
+
"epoch": 0.15,
|
| 650 |
+
"grad_norm": 21.875,
|
| 651 |
+
"learning_rate": 3.0000000000000004e-07,
|
| 652 |
+
"loss": 2.0096,
|
| 653 |
+
"step": 920
|
| 654 |
+
},
|
| 655 |
+
{
|
| 656 |
+
"epoch": 0.15,
|
| 657 |
+
"grad_norm": 18.75,
|
| 658 |
+
"learning_rate": 2.6250000000000003e-07,
|
| 659 |
+
"loss": 1.8059,
|
| 660 |
+
"step": 930
|
| 661 |
+
},
|
| 662 |
+
{
|
| 663 |
+
"epoch": 0.16,
|
| 664 |
+
"grad_norm": 15.9375,
|
| 665 |
+
"learning_rate": 2.25e-07,
|
| 666 |
+
"loss": 1.4159,
|
| 667 |
+
"step": 940
|
| 668 |
+
},
|
| 669 |
+
{
|
| 670 |
+
"epoch": 0.16,
|
| 671 |
+
"grad_norm": 21.25,
|
| 672 |
+
"learning_rate": 1.875e-07,
|
| 673 |
+
"loss": 2.0914,
|
| 674 |
+
"step": 950
|
| 675 |
+
},
|
| 676 |
+
{
|
| 677 |
+
"epoch": 0.16,
|
| 678 |
+
"grad_norm": 19.25,
|
| 679 |
+
"learning_rate": 1.5000000000000002e-07,
|
| 680 |
+
"loss": 1.9735,
|
| 681 |
+
"step": 960
|
| 682 |
+
},
|
| 683 |
+
{
|
| 684 |
+
"epoch": 0.16,
|
| 685 |
+
"grad_norm": 18.0,
|
| 686 |
+
"learning_rate": 1.125e-07,
|
| 687 |
+
"loss": 1.7791,
|
| 688 |
+
"step": 970
|
| 689 |
+
},
|
| 690 |
+
{
|
| 691 |
+
"epoch": 0.16,
|
| 692 |
+
"grad_norm": 20.25,
|
| 693 |
+
"learning_rate": 7.500000000000001e-08,
|
| 694 |
+
"loss": 2.0176,
|
| 695 |
+
"step": 980
|
| 696 |
+
},
|
| 697 |
+
{
|
| 698 |
+
"epoch": 0.17,
|
| 699 |
+
"grad_norm": 17.5,
|
| 700 |
+
"learning_rate": 3.7500000000000005e-08,
|
| 701 |
+
"loss": 2.1275,
|
| 702 |
+
"step": 990
|
| 703 |
+
},
|
| 704 |
+
{
|
| 705 |
+
"epoch": 0.17,
|
| 706 |
+
"grad_norm": 15.875,
|
| 707 |
+
"learning_rate": 0.0,
|
| 708 |
+
"loss": 1.9038,
|
| 709 |
+
"step": 1000
|
| 710 |
+
},
|
| 711 |
+
{
|
| 712 |
+
"epoch": 0.17,
|
| 713 |
+
"eval_loss": 1.9000076055526733,
|
| 714 |
+
"eval_runtime": 30.02,
|
| 715 |
+
"eval_samples_per_second": 33.311,
|
| 716 |
+
"eval_steps_per_second": 33.311,
|
| 717 |
+
"step": 1000
|
| 718 |
+
}
|
| 719 |
+
],
|
| 720 |
+
"logging_steps": 10,
|
| 721 |
+
"max_steps": 1000,
|
| 722 |
+
"num_input_tokens_seen": 0,
|
| 723 |
+
"num_train_epochs": 1,
|
| 724 |
+
"save_steps": 1000,
|
| 725 |
+
"total_flos": 8069610209280000.0,
|
| 726 |
+
"train_batch_size": 1,
|
| 727 |
+
"trial_name": null,
|
| 728 |
+
"trial_params": null
|
| 729 |
+
}
|
training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:27b9819942707cb73a703d163fa22417d1d82c41f5e212c800155a49ab103ee3
|
| 3 |
+
size 4984
|