FastPlus Backup checkpoint-1000 at global step 1000
Browse files- checkpoint-1000/config.json +19 -0
- checkpoint-1000/generation_config.json +6 -0
- checkpoint-1000/model.safetensors +3 -0
- checkpoint-1000/optimizer.pt +3 -0
- checkpoint-1000/rng_state.pth +3 -0
- checkpoint-1000/scaler.pt +3 -0
- checkpoint-1000/scheduler.pt +3 -0
- checkpoint-1000/trainer_state.json +834 -0
- checkpoint-1000/training_args.bin +3 -0
checkpoint-1000/config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"FastPlusForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"dtype": "float32",
|
| 6 |
+
"hidden_size": 768,
|
| 7 |
+
"initializer_range": 0.02,
|
| 8 |
+
"intermediate_size": 2304,
|
| 9 |
+
"kd_alpha": 0.4,
|
| 10 |
+
"kd_temperature": 2.5,
|
| 11 |
+
"max_position_embeddings": 512,
|
| 12 |
+
"model_type": "fastplus",
|
| 13 |
+
"num_attention_heads": 12,
|
| 14 |
+
"num_hidden_layers": 12,
|
| 15 |
+
"tie_word_embeddings": true,
|
| 16 |
+
"transformers_version": "5.12.0",
|
| 17 |
+
"use_cache": false,
|
| 18 |
+
"vocab_size": 31968
|
| 19 |
+
}
|
checkpoint-1000/generation_config.json
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"output_attentions": false,
|
| 4 |
+
"output_hidden_states": false,
|
| 5 |
+
"transformers_version": "5.12.0"
|
| 6 |
+
}
|
checkpoint-1000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0cb24b038c0088c9c4becb272d889c4e53775d8fec2bff25644aefce40474aaa
|
| 3 |
+
size 395782848
|
checkpoint-1000/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7fc5447b0494f3b468471b7c2dece74ab27dbb778af4ebcb5616301842324101
|
| 3 |
+
size 444577931
|
checkpoint-1000/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:61c19bab1174704a4a4441475683bf1270277af15d2e2c95e964789128e482c4
|
| 3 |
+
size 14645
|
checkpoint-1000/scaler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ef8c0bfd0bdadf0b8f987dce56db2d4b0f2a390418dd924793a8ead4195b05b8
|
| 3 |
+
size 1383
|
checkpoint-1000/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0d6eba1f40ea6a9d0d814d1347bf1f25c99d63ed7e69733c849c43b0c89528f2
|
| 3 |
+
size 6009
|
checkpoint-1000/trainer_state.json
ADDED
|
@@ -0,0 +1,834 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.03333333333333333,
|
| 6 |
+
"eval_steps": 500,
|
| 7 |
+
"global_step": 1000,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0003333333333333333,
|
| 14 |
+
"grad_norm": 246.7178192138672,
|
| 15 |
+
"learning_rate": 5.6e-07,
|
| 16 |
+
"loss": 166.2992919921875,
|
| 17 |
+
"original_ce_loss": 10.3932,
|
| 18 |
+
"step": 10
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"epoch": 0.0006666666666666666,
|
| 22 |
+
"grad_norm": 244.59628295898438,
|
| 23 |
+
"learning_rate": 1.3600000000000001e-06,
|
| 24 |
+
"loss": 166.11868896484376,
|
| 25 |
+
"original_ce_loss": 10.3819,
|
| 26 |
+
"step": 20
|
| 27 |
+
},
|
| 28 |
+
{
|
| 29 |
+
"epoch": 0.001,
|
| 30 |
+
"grad_norm": 240.88597106933594,
|
| 31 |
+
"learning_rate": 2.16e-06,
|
| 32 |
+
"loss": 165.65159912109374,
|
| 33 |
+
"original_ce_loss": 10.3527,
|
| 34 |
+
"step": 30
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"epoch": 0.0013333333333333333,
|
| 38 |
+
"grad_norm": 260.6956787109375,
|
| 39 |
+
"learning_rate": 2.96e-06,
|
| 40 |
+
"loss": 164.7892822265625,
|
| 41 |
+
"original_ce_loss": 10.2988,
|
| 42 |
+
"step": 40
|
| 43 |
+
},
|
| 44 |
+
{
|
| 45 |
+
"epoch": 0.0016666666666666668,
|
| 46 |
+
"grad_norm": 296.10198974609375,
|
| 47 |
+
"learning_rate": 3.7600000000000004e-06,
|
| 48 |
+
"loss": 163.08580322265624,
|
| 49 |
+
"original_ce_loss": 10.1923,
|
| 50 |
+
"step": 50
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"epoch": 0.002,
|
| 54 |
+
"grad_norm": 347.94500732421875,
|
| 55 |
+
"learning_rate": 4.56e-06,
|
| 56 |
+
"loss": 159.1340576171875,
|
| 57 |
+
"original_ce_loss": 9.9453,
|
| 58 |
+
"step": 60
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.0023333333333333335,
|
| 62 |
+
"grad_norm": 193.275634765625,
|
| 63 |
+
"learning_rate": 5.28e-06,
|
| 64 |
+
"loss": 150.7270263671875,
|
| 65 |
+
"original_ce_loss": 9.4199,
|
| 66 |
+
"step": 70
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 0.0026666666666666666,
|
| 70 |
+
"grad_norm": 61.063079833984375,
|
| 71 |
+
"learning_rate": 6.08e-06,
|
| 72 |
+
"loss": 141.30562744140624,
|
| 73 |
+
"original_ce_loss": 8.8311,
|
| 74 |
+
"step": 80
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.003,
|
| 78 |
+
"grad_norm": 47.50356674194336,
|
| 79 |
+
"learning_rate": 6.88e-06,
|
| 80 |
+
"loss": 135.84681396484376,
|
| 81 |
+
"original_ce_loss": 8.4899,
|
| 82 |
+
"step": 90
|
| 83 |
+
},
|
| 84 |
+
{
|
| 85 |
+
"epoch": 0.0033333333333333335,
|
| 86 |
+
"grad_norm": 48.022335052490234,
|
| 87 |
+
"learning_rate": 7.680000000000001e-06,
|
| 88 |
+
"loss": 130.2412841796875,
|
| 89 |
+
"original_ce_loss": 8.1395,
|
| 90 |
+
"step": 100
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.0036666666666666666,
|
| 94 |
+
"grad_norm": 46.11091232299805,
|
| 95 |
+
"learning_rate": 8.48e-06,
|
| 96 |
+
"loss": 124.87786865234375,
|
| 97 |
+
"original_ce_loss": 7.8043,
|
| 98 |
+
"step": 110
|
| 99 |
+
},
|
| 100 |
+
{
|
| 101 |
+
"epoch": 0.004,
|
| 102 |
+
"grad_norm": 36.11503601074219,
|
| 103 |
+
"learning_rate": 9.280000000000001e-06,
|
| 104 |
+
"loss": 119.092626953125,
|
| 105 |
+
"original_ce_loss": 7.4428,
|
| 106 |
+
"step": 120
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"epoch": 0.004333333333333333,
|
| 110 |
+
"grad_norm": 26.08365249633789,
|
| 111 |
+
"learning_rate": 1.008e-05,
|
| 112 |
+
"loss": 113.6919921875,
|
| 113 |
+
"original_ce_loss": 7.1052,
|
| 114 |
+
"step": 130
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.004666666666666667,
|
| 118 |
+
"grad_norm": 16.784683227539062,
|
| 119 |
+
"learning_rate": 1.0880000000000001e-05,
|
| 120 |
+
"loss": 109.4510498046875,
|
| 121 |
+
"original_ce_loss": 6.8402,
|
| 122 |
+
"step": 140
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"epoch": 0.005,
|
| 126 |
+
"grad_norm": 12.295577049255371,
|
| 127 |
+
"learning_rate": 1.168e-05,
|
| 128 |
+
"loss": 106.78978271484375,
|
| 129 |
+
"original_ce_loss": 6.6738,
|
| 130 |
+
"step": 150
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 0.005333333333333333,
|
| 134 |
+
"grad_norm": 181.7386474609375,
|
| 135 |
+
"learning_rate": 1.2400000000000002e-05,
|
| 136 |
+
"loss": 104.75615234375,
|
| 137 |
+
"original_ce_loss": 6.5467,
|
| 138 |
+
"step": 160
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"epoch": 0.005666666666666667,
|
| 142 |
+
"grad_norm": 17.787628173828125,
|
| 143 |
+
"learning_rate": 1.3200000000000002e-05,
|
| 144 |
+
"loss": 103.94554443359375,
|
| 145 |
+
"original_ce_loss": 6.4961,
|
| 146 |
+
"step": 170
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.006,
|
| 150 |
+
"grad_norm": 19.06531524658203,
|
| 151 |
+
"learning_rate": 1.4e-05,
|
| 152 |
+
"loss": 103.14150390625,
|
| 153 |
+
"original_ce_loss": 6.4458,
|
| 154 |
+
"step": 180
|
| 155 |
+
},
|
| 156 |
+
{
|
| 157 |
+
"epoch": 0.006333333333333333,
|
| 158 |
+
"grad_norm": 24.76767349243164,
|
| 159 |
+
"learning_rate": 1.48e-05,
|
| 160 |
+
"loss": 102.0536376953125,
|
| 161 |
+
"original_ce_loss": 6.3778,
|
| 162 |
+
"step": 190
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"epoch": 0.006666666666666667,
|
| 166 |
+
"grad_norm": 15.098638534545898,
|
| 167 |
+
"learning_rate": 1.5600000000000003e-05,
|
| 168 |
+
"loss": 100.90601806640625,
|
| 169 |
+
"original_ce_loss": 6.3061,
|
| 170 |
+
"step": 200
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.007,
|
| 174 |
+
"grad_norm": 14.419249534606934,
|
| 175 |
+
"learning_rate": 1.64e-05,
|
| 176 |
+
"loss": 100.32697143554688,
|
| 177 |
+
"original_ce_loss": 6.2699,
|
| 178 |
+
"step": 210
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"epoch": 0.007333333333333333,
|
| 182 |
+
"grad_norm": NaN,
|
| 183 |
+
"learning_rate": 1.704e-05,
|
| 184 |
+
"loss": 99.49810180664062,
|
| 185 |
+
"original_ce_loss": 6.2181,
|
| 186 |
+
"step": 220
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
"epoch": 0.007666666666666666,
|
| 190 |
+
"grad_norm": 18.77039909362793,
|
| 191 |
+
"learning_rate": 1.7760000000000003e-05,
|
| 192 |
+
"loss": 98.45899047851563,
|
| 193 |
+
"original_ce_loss": 6.1532,
|
| 194 |
+
"step": 230
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"epoch": 0.008,
|
| 198 |
+
"grad_norm": 22.13999366760254,
|
| 199 |
+
"learning_rate": 1.8560000000000002e-05,
|
| 200 |
+
"loss": 97.58836059570312,
|
| 201 |
+
"original_ce_loss": 6.0987,
|
| 202 |
+
"step": 240
|
| 203 |
+
},
|
| 204 |
+
{
|
| 205 |
+
"epoch": 0.008333333333333333,
|
| 206 |
+
"grad_norm": 15.347926139831543,
|
| 207 |
+
"learning_rate": 1.936e-05,
|
| 208 |
+
"loss": 96.4761474609375,
|
| 209 |
+
"original_ce_loss": 6.0292,
|
| 210 |
+
"step": 250
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.008666666666666666,
|
| 214 |
+
"grad_norm": 10.597929000854492,
|
| 215 |
+
"learning_rate": 2.016e-05,
|
| 216 |
+
"loss": 94.91881713867187,
|
| 217 |
+
"original_ce_loss": 5.9319,
|
| 218 |
+
"step": 260
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"epoch": 0.009,
|
| 222 |
+
"grad_norm": 29.104537963867188,
|
| 223 |
+
"learning_rate": 2.0960000000000003e-05,
|
| 224 |
+
"loss": 93.14630126953125,
|
| 225 |
+
"original_ce_loss": 5.8211,
|
| 226 |
+
"step": 270
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.009333333333333334,
|
| 230 |
+
"grad_norm": 30.951608657836914,
|
| 231 |
+
"learning_rate": 2.1680000000000002e-05,
|
| 232 |
+
"loss": 91.8282958984375,
|
| 233 |
+
"original_ce_loss": 5.7387,
|
| 234 |
+
"step": 280
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"epoch": 0.009666666666666667,
|
| 238 |
+
"grad_norm": 19.87868881225586,
|
| 239 |
+
"learning_rate": 2.2480000000000005e-05,
|
| 240 |
+
"loss": 90.33818359375,
|
| 241 |
+
"original_ce_loss": 5.6456,
|
| 242 |
+
"step": 290
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"epoch": 0.01,
|
| 246 |
+
"grad_norm": 30.976680755615234,
|
| 247 |
+
"learning_rate": 2.312e-05,
|
| 248 |
+
"loss": 89.595849609375,
|
| 249 |
+
"original_ce_loss": 5.5992,
|
| 250 |
+
"step": 300
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.010333333333333333,
|
| 254 |
+
"grad_norm": 17.352371215820312,
|
| 255 |
+
"learning_rate": 2.392e-05,
|
| 256 |
+
"loss": 88.54055786132812,
|
| 257 |
+
"original_ce_loss": 5.5333,
|
| 258 |
+
"step": 310
|
| 259 |
+
},
|
| 260 |
+
{
|
| 261 |
+
"epoch": 0.010666666666666666,
|
| 262 |
+
"grad_norm": 31.46754264831543,
|
| 263 |
+
"learning_rate": 2.472e-05,
|
| 264 |
+
"loss": 87.1218505859375,
|
| 265 |
+
"original_ce_loss": 5.4446,
|
| 266 |
+
"step": 320
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"epoch": 0.011,
|
| 270 |
+
"grad_norm": 19.462528228759766,
|
| 271 |
+
"learning_rate": 2.5520000000000003e-05,
|
| 272 |
+
"loss": 86.0860107421875,
|
| 273 |
+
"original_ce_loss": 5.3798,
|
| 274 |
+
"step": 330
|
| 275 |
+
},
|
| 276 |
+
{
|
| 277 |
+
"epoch": 0.011333333333333334,
|
| 278 |
+
"grad_norm": 17.978193283081055,
|
| 279 |
+
"learning_rate": 2.6320000000000002e-05,
|
| 280 |
+
"loss": 85.21943359375,
|
| 281 |
+
"original_ce_loss": 5.3257,
|
| 282 |
+
"step": 340
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.011666666666666667,
|
| 286 |
+
"grad_norm": 22.971702575683594,
|
| 287 |
+
"learning_rate": 2.7120000000000005e-05,
|
| 288 |
+
"loss": 83.27525634765625,
|
| 289 |
+
"original_ce_loss": 5.2042,
|
| 290 |
+
"step": 350
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"epoch": 0.012,
|
| 294 |
+
"grad_norm": 25.951129913330078,
|
| 295 |
+
"learning_rate": 2.792e-05,
|
| 296 |
+
"loss": 82.56115112304687,
|
| 297 |
+
"original_ce_loss": 5.1595,
|
| 298 |
+
"step": 360
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"epoch": 0.012333333333333333,
|
| 302 |
+
"grad_norm": 34.82793045043945,
|
| 303 |
+
"learning_rate": 2.8720000000000003e-05,
|
| 304 |
+
"loss": 80.75704345703124,
|
| 305 |
+
"original_ce_loss": 5.0468,
|
| 306 |
+
"step": 370
|
| 307 |
+
},
|
| 308 |
+
{
|
| 309 |
+
"epoch": 0.012666666666666666,
|
| 310 |
+
"grad_norm": 21.33478546142578,
|
| 311 |
+
"learning_rate": 2.9520000000000002e-05,
|
| 312 |
+
"loss": 80.074853515625,
|
| 313 |
+
"original_ce_loss": 5.0041,
|
| 314 |
+
"step": 380
|
| 315 |
+
},
|
| 316 |
+
{
|
| 317 |
+
"epoch": 0.013,
|
| 318 |
+
"grad_norm": NaN,
|
| 319 |
+
"learning_rate": 3.032e-05,
|
| 320 |
+
"loss": 78.41221313476562,
|
| 321 |
+
"original_ce_loss": 4.9002,
|
| 322 |
+
"step": 390
|
| 323 |
+
},
|
| 324 |
+
{
|
| 325 |
+
"epoch": 0.013333333333333334,
|
| 326 |
+
"grad_norm": 64.517333984375,
|
| 327 |
+
"learning_rate": 3.104e-05,
|
| 328 |
+
"loss": 76.89620361328124,
|
| 329 |
+
"original_ce_loss": 4.8055,
|
| 330 |
+
"step": 400
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.013666666666666667,
|
| 334 |
+
"grad_norm": 32.529396057128906,
|
| 335 |
+
"learning_rate": 3.184000000000001e-05,
|
| 336 |
+
"loss": 75.56863403320312,
|
| 337 |
+
"original_ce_loss": 4.7225,
|
| 338 |
+
"step": 410
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.014,
|
| 342 |
+
"grad_norm": 27.579341888427734,
|
| 343 |
+
"learning_rate": 3.264e-05,
|
| 344 |
+
"loss": 74.20821533203124,
|
| 345 |
+
"original_ce_loss": 4.6375,
|
| 346 |
+
"step": 420
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.014333333333333333,
|
| 350 |
+
"grad_norm": 35.1613655090332,
|
| 351 |
+
"learning_rate": 3.344e-05,
|
| 352 |
+
"loss": 72.89371337890626,
|
| 353 |
+
"original_ce_loss": 4.5553,
|
| 354 |
+
"step": 430
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.014666666666666666,
|
| 358 |
+
"grad_norm": 25.330854415893555,
|
| 359 |
+
"learning_rate": 3.4240000000000004e-05,
|
| 360 |
+
"loss": 71.7568603515625,
|
| 361 |
+
"original_ce_loss": 4.4843,
|
| 362 |
+
"step": 440
|
| 363 |
+
},
|
| 364 |
+
{
|
| 365 |
+
"epoch": 0.015,
|
| 366 |
+
"grad_norm": 90.75880432128906,
|
| 367 |
+
"learning_rate": 3.4960000000000004e-05,
|
| 368 |
+
"loss": 71.937451171875,
|
| 369 |
+
"original_ce_loss": 4.4956,
|
| 370 |
+
"step": 450
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"epoch": 0.015333333333333332,
|
| 374 |
+
"grad_norm": 335.9816589355469,
|
| 375 |
+
"learning_rate": 3.576e-05,
|
| 376 |
+
"loss": 72.51287231445312,
|
| 377 |
+
"original_ce_loss": 4.5315,
|
| 378 |
+
"step": 460
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"epoch": 0.015666666666666666,
|
| 382 |
+
"grad_norm": 62.99739074707031,
|
| 383 |
+
"learning_rate": 3.656e-05,
|
| 384 |
+
"loss": 70.11299438476563,
|
| 385 |
+
"original_ce_loss": 4.3815,
|
| 386 |
+
"step": 470
|
| 387 |
+
},
|
| 388 |
+
{
|
| 389 |
+
"epoch": 0.016,
|
| 390 |
+
"grad_norm": 45.36550521850586,
|
| 391 |
+
"learning_rate": 3.736000000000001e-05,
|
| 392 |
+
"loss": 68.48368530273437,
|
| 393 |
+
"original_ce_loss": 4.2797,
|
| 394 |
+
"step": 480
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.01633333333333333,
|
| 398 |
+
"grad_norm": 28.657081604003906,
|
| 399 |
+
"learning_rate": 3.808e-05,
|
| 400 |
+
"loss": 67.57816772460937,
|
| 401 |
+
"original_ce_loss": 4.2231,
|
| 402 |
+
"step": 490
|
| 403 |
+
},
|
| 404 |
+
{
|
| 405 |
+
"epoch": 0.016666666666666666,
|
| 406 |
+
"grad_norm": 58.84739303588867,
|
| 407 |
+
"learning_rate": 3.888e-05,
|
| 408 |
+
"loss": 67.36370849609375,
|
| 409 |
+
"original_ce_loss": 4.2097,
|
| 410 |
+
"step": 500
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.017,
|
| 414 |
+
"grad_norm": 21.05079460144043,
|
| 415 |
+
"learning_rate": 3.9680000000000006e-05,
|
| 416 |
+
"loss": 65.88310546875,
|
| 417 |
+
"original_ce_loss": 4.1172,
|
| 418 |
+
"step": 510
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.017333333333333333,
|
| 422 |
+
"grad_norm": 19.669618606567383,
|
| 423 |
+
"learning_rate": 3.999999591719913e-05,
|
| 424 |
+
"loss": 63.6228271484375,
|
| 425 |
+
"original_ce_loss": 3.9759,
|
| 426 |
+
"step": 520
|
| 427 |
+
},
|
| 428 |
+
{
|
| 429 |
+
"epoch": 0.017666666666666667,
|
| 430 |
+
"grad_norm": 26.515731811523438,
|
| 431 |
+
"learning_rate": 3.999997096675536e-05,
|
| 432 |
+
"loss": 62.17679443359375,
|
| 433 |
+
"original_ce_loss": 3.8855,
|
| 434 |
+
"step": 530
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"epoch": 0.018,
|
| 438 |
+
"grad_norm": 21.496002197265625,
|
| 439 |
+
"learning_rate": 3.9999923334118786e-05,
|
| 440 |
+
"loss": 61.1276611328125,
|
| 441 |
+
"original_ce_loss": 3.82,
|
| 442 |
+
"step": 540
|
| 443 |
+
},
|
| 444 |
+
{
|
| 445 |
+
"epoch": 0.018333333333333333,
|
| 446 |
+
"grad_norm": 26.728601455688477,
|
| 447 |
+
"learning_rate": 3.999985301934345e-05,
|
| 448 |
+
"loss": 60.27257080078125,
|
| 449 |
+
"original_ce_loss": 3.7665,
|
| 450 |
+
"step": 550
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"epoch": 0.018666666666666668,
|
| 454 |
+
"grad_norm": 32.353240966796875,
|
| 455 |
+
"learning_rate": 3.9999760022509074e-05,
|
| 456 |
+
"loss": 59.5458984375,
|
| 457 |
+
"original_ce_loss": 3.7211,
|
| 458 |
+
"step": 560
|
| 459 |
+
},
|
| 460 |
+
{
|
| 461 |
+
"epoch": 0.019,
|
| 462 |
+
"grad_norm": 27.575580596923828,
|
| 463 |
+
"learning_rate": 3.9999644343721135e-05,
|
| 464 |
+
"loss": 58.12828369140625,
|
| 465 |
+
"original_ce_loss": 3.6325,
|
| 466 |
+
"step": 570
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 0.019333333333333334,
|
| 470 |
+
"grad_norm": 24.331390380859375,
|
| 471 |
+
"learning_rate": 3.999950598311082e-05,
|
| 472 |
+
"loss": 57.249261474609376,
|
| 473 |
+
"original_ce_loss": 3.5775,
|
| 474 |
+
"step": 580
|
| 475 |
+
},
|
| 476 |
+
{
|
| 477 |
+
"epoch": 0.019666666666666666,
|
| 478 |
+
"grad_norm": 27.248210906982422,
|
| 479 |
+
"learning_rate": 3.999934494083506e-05,
|
| 480 |
+
"loss": 56.11417236328125,
|
| 481 |
+
"original_ce_loss": 3.5066,
|
| 482 |
+
"step": 590
|
| 483 |
+
},
|
| 484 |
+
{
|
| 485 |
+
"epoch": 0.02,
|
| 486 |
+
"grad_norm": 37.048988342285156,
|
| 487 |
+
"learning_rate": 3.999916121707648e-05,
|
| 488 |
+
"loss": 55.547882080078125,
|
| 489 |
+
"original_ce_loss": 3.4712,
|
| 490 |
+
"step": 600
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.02033333333333333,
|
| 494 |
+
"grad_norm": 39.78192901611328,
|
| 495 |
+
"learning_rate": 3.999895481204345e-05,
|
| 496 |
+
"loss": 54.420806884765625,
|
| 497 |
+
"original_ce_loss": 3.4008,
|
| 498 |
+
"step": 610
|
| 499 |
+
},
|
| 500 |
+
{
|
| 501 |
+
"epoch": 0.020666666666666667,
|
| 502 |
+
"grad_norm": 22.940998077392578,
|
| 503 |
+
"learning_rate": 3.9998725725970054e-05,
|
| 504 |
+
"loss": 54.5435302734375,
|
| 505 |
+
"original_ce_loss": 3.4084,
|
| 506 |
+
"step": 620
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.021,
|
| 510 |
+
"grad_norm": 31.371349334716797,
|
| 511 |
+
"learning_rate": 3.999847395911611e-05,
|
| 512 |
+
"loss": 54.152032470703126,
|
| 513 |
+
"original_ce_loss": 3.384,
|
| 514 |
+
"step": 630
|
| 515 |
+
},
|
| 516 |
+
{
|
| 517 |
+
"epoch": 0.021333333333333333,
|
| 518 |
+
"grad_norm": 19.122173309326172,
|
| 519 |
+
"learning_rate": 3.999819951176713e-05,
|
| 520 |
+
"loss": 53.30802001953125,
|
| 521 |
+
"original_ce_loss": 3.3312,
|
| 522 |
+
"step": 640
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 0.021666666666666667,
|
| 526 |
+
"grad_norm": 19.67505645751953,
|
| 527 |
+
"learning_rate": 3.999790238423439e-05,
|
| 528 |
+
"loss": 52.82340698242187,
|
| 529 |
+
"original_ce_loss": 3.3009,
|
| 530 |
+
"step": 650
|
| 531 |
+
},
|
| 532 |
+
{
|
| 533 |
+
"epoch": 0.022,
|
| 534 |
+
"grad_norm": 36.539642333984375,
|
| 535 |
+
"learning_rate": 3.999758257685485e-05,
|
| 536 |
+
"loss": 52.68773193359375,
|
| 537 |
+
"original_ce_loss": 3.2925,
|
| 538 |
+
"step": 660
|
| 539 |
+
},
|
| 540 |
+
{
|
| 541 |
+
"epoch": 0.022333333333333334,
|
| 542 |
+
"grad_norm": 30.446720123291016,
|
| 543 |
+
"learning_rate": 3.999724008999121e-05,
|
| 544 |
+
"loss": 52.1708984375,
|
| 545 |
+
"original_ce_loss": 3.2602,
|
| 546 |
+
"step": 670
|
| 547 |
+
},
|
| 548 |
+
{
|
| 549 |
+
"epoch": 0.02266666666666667,
|
| 550 |
+
"grad_norm": 17.4969539642334,
|
| 551 |
+
"learning_rate": 3.99968749240319e-05,
|
| 552 |
+
"loss": 51.11235656738281,
|
| 553 |
+
"original_ce_loss": 3.194,
|
| 554 |
+
"step": 680
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.023,
|
| 558 |
+
"grad_norm": 23.769330978393555,
|
| 559 |
+
"learning_rate": 3.999648707939104e-05,
|
| 560 |
+
"loss": 50.4834228515625,
|
| 561 |
+
"original_ce_loss": 3.1547,
|
| 562 |
+
"step": 690
|
| 563 |
+
},
|
| 564 |
+
{
|
| 565 |
+
"epoch": 0.023333333333333334,
|
| 566 |
+
"grad_norm": 18.74240493774414,
|
| 567 |
+
"learning_rate": 3.99960765565085e-05,
|
| 568 |
+
"loss": 50.2678955078125,
|
| 569 |
+
"original_ce_loss": 3.1412,
|
| 570 |
+
"step": 700
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 0.023666666666666666,
|
| 574 |
+
"grad_norm": 18.66421890258789,
|
| 575 |
+
"learning_rate": 3.999564335584986e-05,
|
| 576 |
+
"loss": 50.34214782714844,
|
| 577 |
+
"original_ce_loss": 3.1459,
|
| 578 |
+
"step": 710
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 0.024,
|
| 582 |
+
"grad_norm": 16.374645233154297,
|
| 583 |
+
"learning_rate": 3.999518747790642e-05,
|
| 584 |
+
"loss": 49.75227355957031,
|
| 585 |
+
"original_ce_loss": 3.109,
|
| 586 |
+
"step": 720
|
| 587 |
+
},
|
| 588 |
+
{
|
| 589 |
+
"epoch": 0.024333333333333332,
|
| 590 |
+
"grad_norm": 69.73137664794922,
|
| 591 |
+
"learning_rate": 3.999470892319519e-05,
|
| 592 |
+
"loss": 49.799835205078125,
|
| 593 |
+
"original_ce_loss": 3.112,
|
| 594 |
+
"step": 730
|
| 595 |
+
},
|
| 596 |
+
{
|
| 597 |
+
"epoch": 0.024666666666666667,
|
| 598 |
+
"grad_norm": 21.611284255981445,
|
| 599 |
+
"learning_rate": 3.9994207692258896e-05,
|
| 600 |
+
"loss": 49.9420654296875,
|
| 601 |
+
"original_ce_loss": 3.1209,
|
| 602 |
+
"step": 740
|
| 603 |
+
},
|
| 604 |
+
{
|
| 605 |
+
"epoch": 0.025,
|
| 606 |
+
"grad_norm": 21.31991195678711,
|
| 607 |
+
"learning_rate": 3.9993683785666e-05,
|
| 608 |
+
"loss": 50.11099853515625,
|
| 609 |
+
"original_ce_loss": 3.1314,
|
| 610 |
+
"step": 750
|
| 611 |
+
},
|
| 612 |
+
{
|
| 613 |
+
"epoch": 0.025333333333333333,
|
| 614 |
+
"grad_norm": 19.646024703979492,
|
| 615 |
+
"learning_rate": 3.999313720401068e-05,
|
| 616 |
+
"loss": 49.42765502929687,
|
| 617 |
+
"original_ce_loss": 3.0887,
|
| 618 |
+
"step": 760
|
| 619 |
+
},
|
| 620 |
+
{
|
| 621 |
+
"epoch": 0.025666666666666667,
|
| 622 |
+
"grad_norm": 21.394819259643555,
|
| 623 |
+
"learning_rate": 3.99925679479128e-05,
|
| 624 |
+
"loss": 49.35731201171875,
|
| 625 |
+
"original_ce_loss": 3.0843,
|
| 626 |
+
"step": 770
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.026,
|
| 630 |
+
"grad_norm": 17.190773010253906,
|
| 631 |
+
"learning_rate": 3.999203623130939e-05,
|
| 632 |
+
"loss": 49.1930419921875,
|
| 633 |
+
"original_ce_loss": 3.074,
|
| 634 |
+
"step": 780
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 0.026333333333333334,
|
| 638 |
+
"grad_norm": 15.881075859069824,
|
| 639 |
+
"learning_rate": 3.999142389557055e-05,
|
| 640 |
+
"loss": 48.067745971679685,
|
| 641 |
+
"original_ce_loss": 3.0037,
|
| 642 |
+
"step": 790
|
| 643 |
+
},
|
| 644 |
+
{
|
| 645 |
+
"epoch": 0.02666666666666667,
|
| 646 |
+
"grad_norm": Infinity,
|
| 647 |
+
"learning_rate": 3.9990788887332245e-05,
|
| 648 |
+
"loss": 47.95630798339844,
|
| 649 |
+
"original_ce_loss": 2.9967,
|
| 650 |
+
"step": 800
|
| 651 |
+
},
|
| 652 |
+
{
|
| 653 |
+
"epoch": 0.027,
|
| 654 |
+
"grad_norm": 18.853967666625977,
|
| 655 |
+
"learning_rate": 3.9990197995525406e-05,
|
| 656 |
+
"loss": 47.47606201171875,
|
| 657 |
+
"original_ce_loss": 2.9667,
|
| 658 |
+
"step": 810
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"epoch": 0.027333333333333334,
|
| 662 |
+
"grad_norm": 19.68868637084961,
|
| 663 |
+
"learning_rate": 3.998951991154343e-05,
|
| 664 |
+
"loss": 47.4399169921875,
|
| 665 |
+
"original_ce_loss": 2.9645,
|
| 666 |
+
"step": 820
|
| 667 |
+
},
|
| 668 |
+
{
|
| 669 |
+
"epoch": 0.027666666666666666,
|
| 670 |
+
"grad_norm": 32.64214324951172,
|
| 671 |
+
"learning_rate": 3.9988819157221313e-05,
|
| 672 |
+
"loss": 46.914346313476564,
|
| 673 |
+
"original_ce_loss": 2.9316,
|
| 674 |
+
"step": 830
|
| 675 |
+
},
|
| 676 |
+
{
|
| 677 |
+
"epoch": 0.028,
|
| 678 |
+
"grad_norm": 14.900030136108398,
|
| 679 |
+
"learning_rate": 3.99880957333538e-05,
|
| 680 |
+
"loss": 46.99378967285156,
|
| 681 |
+
"original_ce_loss": 2.9366,
|
| 682 |
+
"step": 840
|
| 683 |
+
},
|
| 684 |
+
{
|
| 685 |
+
"epoch": 0.028333333333333332,
|
| 686 |
+
"grad_norm": 47.476219177246094,
|
| 687 |
+
"learning_rate": 3.998734964076132e-05,
|
| 688 |
+
"loss": 48.06187744140625,
|
| 689 |
+
"original_ce_loss": 3.0033,
|
| 690 |
+
"step": 850
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.028666666666666667,
|
| 694 |
+
"grad_norm": 15.936322212219238,
|
| 695 |
+
"learning_rate": 3.998658088029003e-05,
|
| 696 |
+
"loss": 48.313577270507814,
|
| 697 |
+
"original_ce_loss": 3.0191,
|
| 698 |
+
"step": 860
|
| 699 |
+
},
|
| 700 |
+
{
|
| 701 |
+
"epoch": 0.029,
|
| 702 |
+
"grad_norm": 39.59748458862305,
|
| 703 |
+
"learning_rate": 3.998578945281179e-05,
|
| 704 |
+
"loss": 49.42246704101562,
|
| 705 |
+
"original_ce_loss": 3.0884,
|
| 706 |
+
"step": 870
|
| 707 |
+
},
|
| 708 |
+
{
|
| 709 |
+
"epoch": 0.029333333333333333,
|
| 710 |
+
"grad_norm": 38.61612319946289,
|
| 711 |
+
"learning_rate": 3.998497535922417e-05,
|
| 712 |
+
"loss": 51.2785400390625,
|
| 713 |
+
"original_ce_loss": 3.2044,
|
| 714 |
+
"step": 880
|
| 715 |
+
},
|
| 716 |
+
{
|
| 717 |
+
"epoch": 0.029666666666666668,
|
| 718 |
+
"grad_norm": 61.205810546875,
|
| 719 |
+
"learning_rate": 3.9984138600450434e-05,
|
| 720 |
+
"loss": 52.58193359375,
|
| 721 |
+
"original_ce_loss": 3.2858,
|
| 722 |
+
"step": 890
|
| 723 |
+
},
|
| 724 |
+
{
|
| 725 |
+
"epoch": 0.03,
|
| 726 |
+
"grad_norm": 33.091522216796875,
|
| 727 |
+
"learning_rate": 3.9983279177439575e-05,
|
| 728 |
+
"loss": 59.30086059570313,
|
| 729 |
+
"original_ce_loss": 3.7058,
|
| 730 |
+
"step": 900
|
| 731 |
+
},
|
| 732 |
+
{
|
| 733 |
+
"epoch": 0.030333333333333334,
|
| 734 |
+
"grad_norm": 18.325098037719727,
|
| 735 |
+
"learning_rate": 3.998239709116626e-05,
|
| 736 |
+
"loss": 62.75926513671875,
|
| 737 |
+
"original_ce_loss": 3.9219,
|
| 738 |
+
"step": 910
|
| 739 |
+
},
|
| 740 |
+
{
|
| 741 |
+
"epoch": 0.030666666666666665,
|
| 742 |
+
"grad_norm": 76.63507080078125,
|
| 743 |
+
"learning_rate": 3.998149234263088e-05,
|
| 744 |
+
"loss": 72.96146850585937,
|
| 745 |
+
"original_ce_loss": 4.5596,
|
| 746 |
+
"step": 920
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 0.031,
|
| 750 |
+
"grad_norm": 29.289752960205078,
|
| 751 |
+
"learning_rate": 3.998056493285952e-05,
|
| 752 |
+
"loss": 70.06895751953125,
|
| 753 |
+
"original_ce_loss": 4.3788,
|
| 754 |
+
"step": 930
|
| 755 |
+
},
|
| 756 |
+
{
|
| 757 |
+
"epoch": 0.03133333333333333,
|
| 758 |
+
"grad_norm": 46.0661735534668,
|
| 759 |
+
"learning_rate": 3.997961486290396e-05,
|
| 760 |
+
"loss": 70.70223999023438,
|
| 761 |
+
"original_ce_loss": 4.4184,
|
| 762 |
+
"step": 940
|
| 763 |
+
},
|
| 764 |
+
{
|
| 765 |
+
"epoch": 0.03166666666666667,
|
| 766 |
+
"grad_norm": 21.264270782470703,
|
| 767 |
+
"learning_rate": 3.9978642133841694e-05,
|
| 768 |
+
"loss": 70.68958740234375,
|
| 769 |
+
"original_ce_loss": 4.4176,
|
| 770 |
+
"step": 950
|
| 771 |
+
},
|
| 772 |
+
{
|
| 773 |
+
"epoch": 0.032,
|
| 774 |
+
"grad_norm": 81.46379089355469,
|
| 775 |
+
"learning_rate": 3.99776467467759e-05,
|
| 776 |
+
"loss": 73.30789794921876,
|
| 777 |
+
"original_ce_loss": 4.5812,
|
| 778 |
+
"step": 960
|
| 779 |
+
},
|
| 780 |
+
{
|
| 781 |
+
"epoch": 0.03233333333333333,
|
| 782 |
+
"grad_norm": 20.64461326599121,
|
| 783 |
+
"learning_rate": 3.997662870283545e-05,
|
| 784 |
+
"loss": 76.69268188476562,
|
| 785 |
+
"original_ce_loss": 4.7928,
|
| 786 |
+
"step": 970
|
| 787 |
+
},
|
| 788 |
+
{
|
| 789 |
+
"epoch": 0.03266666666666666,
|
| 790 |
+
"grad_norm": 36.781532287597656,
|
| 791 |
+
"learning_rate": 3.997558800317494e-05,
|
| 792 |
+
"loss": 73.93734130859374,
|
| 793 |
+
"original_ce_loss": 4.6206,
|
| 794 |
+
"step": 980
|
| 795 |
+
},
|
| 796 |
+
{
|
| 797 |
+
"epoch": 0.033,
|
| 798 |
+
"grad_norm": 56.953147888183594,
|
| 799 |
+
"learning_rate": 3.997452464897462e-05,
|
| 800 |
+
"loss": 73.31669311523437,
|
| 801 |
+
"original_ce_loss": 4.5818,
|
| 802 |
+
"step": 990
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 0.03333333333333333,
|
| 806 |
+
"grad_norm": 40.67665481567383,
|
| 807 |
+
"learning_rate": 3.997343864144046e-05,
|
| 808 |
+
"loss": 73.99210205078126,
|
| 809 |
+
"original_ce_loss": 4.624,
|
| 810 |
+
"step": 1000
|
| 811 |
+
}
|
| 812 |
+
],
|
| 813 |
+
"logging_steps": 10,
|
| 814 |
+
"max_steps": 30000,
|
| 815 |
+
"num_input_tokens_seen": 0,
|
| 816 |
+
"num_train_epochs": 9223372036854775807,
|
| 817 |
+
"save_steps": 200,
|
| 818 |
+
"stateful_callbacks": {
|
| 819 |
+
"TrainerControl": {
|
| 820 |
+
"args": {
|
| 821 |
+
"should_epoch_stop": false,
|
| 822 |
+
"should_evaluate": false,
|
| 823 |
+
"should_log": false,
|
| 824 |
+
"should_save": true,
|
| 825 |
+
"should_training_stop": false
|
| 826 |
+
},
|
| 827 |
+
"attributes": {}
|
| 828 |
+
}
|
| 829 |
+
},
|
| 830 |
+
"total_flos": 7.3686459088896e+16,
|
| 831 |
+
"train_batch_size": 16,
|
| 832 |
+
"trial_name": null,
|
| 833 |
+
"trial_params": null
|
| 834 |
+
}
|
checkpoint-1000/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e02c2c50360797f77395ba65fdebdf9248982951d642cf2ffe29ee47f19498f5
|
| 3 |
+
size 5137
|