FastPlus40m 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/scaler.pt +3 -0
- checkpoint-1000/scheduler.pt +3 -0
- checkpoint-1000/tokenizer.json +0 -0
- checkpoint-1000/tokenizer_config.json +13 -0
- checkpoint-1000/trainer_state.json +834 -0
checkpoint-1000/config.json
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"FastPlus40mForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"dtype": "float32",
|
| 6 |
+
"hidden_size": 512,
|
| 7 |
+
"initializer_range": 0.02,
|
| 8 |
+
"intermediate_size": 1536,
|
| 9 |
+
"kd_alpha": 0.4,
|
| 10 |
+
"kd_temperature": 2.5,
|
| 11 |
+
"max_position_embeddings": 512,
|
| 12 |
+
"model_type": "fastplus_40m",
|
| 13 |
+
"num_attention_heads": 8,
|
| 14 |
+
"num_hidden_layers": 11,
|
| 15 |
+
"tie_word_embeddings": true,
|
| 16 |
+
"transformers_version": "5.12.0",
|
| 17 |
+
"use_cache": false,
|
| 18 |
+
"vocab_size": 8160
|
| 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 sha256:6001a8be4760144d3685770b32c504bacba3cbd1c2fae7f20b5fa934e0eb891a
|
| 3 |
+
size 159784280
|
checkpoint-1000/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:17a650718163295513bb5ccee84f1ddb472eb6bd2c7cc88fe23dc47f3b90dc8d
|
| 3 |
+
size 179195723
|
checkpoint-1000/scaler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0f273ffa7a52c4903a4fd8b5d1ce5b574e09dcb68d37e82bc7e6190585d79e3b
|
| 3 |
+
size 1383
|
checkpoint-1000/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:64b398265895e919020e5b01c5796f4a4efaa4dad6b41ff9aa57078430f244dd
|
| 3 |
+
size 5625
|
checkpoint-1000/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
checkpoint-1000/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"backend": "tokenizers",
|
| 3 |
+
"bos_token": "<s>",
|
| 4 |
+
"eos_token": "</s>",
|
| 5 |
+
"extra_special_tokens": [
|
| 6 |
+
"<|user|>",
|
| 7 |
+
"<|assistant|>"
|
| 8 |
+
],
|
| 9 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 10 |
+
"pad_token": "<pad>",
|
| 11 |
+
"tokenizer_class": "TokenizersBackend",
|
| 12 |
+
"unk_token": "<unk>"
|
| 13 |
+
}
|
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": 0.5185881437785629,
|
| 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": 0.005185881437785629,
|
| 14 |
+
"grad_norm": 266.6409606933594,
|
| 15 |
+
"learning_rate": 2.5499999999999996e-05,
|
| 16 |
+
"loss": 144.2415283203125,
|
| 17 |
+
"original_ce_loss": 9.0144,
|
| 18 |
+
"step": 10
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"epoch": 0.010371762875571258,
|
| 22 |
+
"grad_norm": 272.673095703125,
|
| 23 |
+
"learning_rate": 6.8e-05,
|
| 24 |
+
"loss": 143.50185546875,
|
| 25 |
+
"original_ce_loss": 8.9682,
|
| 26 |
+
"step": 20
|
| 27 |
+
},
|
| 28 |
+
{
|
| 29 |
+
"epoch": 0.015557644313356886,
|
| 30 |
+
"grad_norm": 437.08270263671875,
|
| 31 |
+
"learning_rate": 0.00010625,
|
| 32 |
+
"loss": 140.732373046875,
|
| 33 |
+
"original_ce_loss": 8.7951,
|
| 34 |
+
"step": 30
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"epoch": 0.020743525751142515,
|
| 38 |
+
"grad_norm": 268.7778625488281,
|
| 39 |
+
"learning_rate": 0.00014450000000000002,
|
| 40 |
+
"loss": 128.16519775390626,
|
| 41 |
+
"original_ce_loss": 8.0096,
|
| 42 |
+
"step": 40
|
| 43 |
+
},
|
| 44 |
+
{
|
| 45 |
+
"epoch": 0.025929407188928142,
|
| 46 |
+
"grad_norm": 38.690486907958984,
|
| 47 |
+
"learning_rate": 0.000187,
|
| 48 |
+
"loss": 117.56219482421875,
|
| 49 |
+
"original_ce_loss": 7.347,
|
| 50 |
+
"step": 50
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"epoch": 0.031115288626713773,
|
| 54 |
+
"grad_norm": 31.22739028930664,
|
| 55 |
+
"learning_rate": 0.0002295,
|
| 56 |
+
"loss": 111.48564453125,
|
| 57 |
+
"original_ce_loss": 6.9672,
|
| 58 |
+
"step": 60
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.0363011700644994,
|
| 62 |
+
"grad_norm": 17.76206398010254,
|
| 63 |
+
"learning_rate": 0.000272,
|
| 64 |
+
"loss": 104.34268798828126,
|
| 65 |
+
"original_ce_loss": 6.5207,
|
| 66 |
+
"step": 70
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 0.04148705150228503,
|
| 70 |
+
"grad_norm": 101.92194366455078,
|
| 71 |
+
"learning_rate": 0.0003145,
|
| 72 |
+
"loss": 100.91859130859375,
|
| 73 |
+
"original_ce_loss": 6.3067,
|
| 74 |
+
"step": 80
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.04667293294007066,
|
| 78 |
+
"grad_norm": 28.36432647705078,
|
| 79 |
+
"learning_rate": 0.00035699999999999995,
|
| 80 |
+
"loss": 100.308935546875,
|
| 81 |
+
"original_ce_loss": 6.2686,
|
| 82 |
+
"step": 90
|
| 83 |
+
},
|
| 84 |
+
{
|
| 85 |
+
"epoch": 0.051858814377856284,
|
| 86 |
+
"grad_norm": 17.224742889404297,
|
| 87 |
+
"learning_rate": 0.00039949999999999995,
|
| 88 |
+
"loss": 100.14071655273438,
|
| 89 |
+
"original_ce_loss": 6.2581,
|
| 90 |
+
"step": 100
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.057044695815641915,
|
| 94 |
+
"grad_norm": 52.153076171875,
|
| 95 |
+
"learning_rate": 0.0004249971787418828,
|
| 96 |
+
"loss": 99.686962890625,
|
| 97 |
+
"original_ce_loss": 6.2298,
|
| 98 |
+
"step": 110
|
| 99 |
+
},
|
| 100 |
+
{
|
| 101 |
+
"epoch": 0.062230577253427545,
|
| 102 |
+
"grad_norm": 16.022294998168945,
|
| 103 |
+
"learning_rate": 0.00042494702512599277,
|
| 104 |
+
"loss": 99.31576538085938,
|
| 105 |
+
"original_ce_loss": 6.2066,
|
| 106 |
+
"step": 120
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"epoch": 0.06741645869121317,
|
| 110 |
+
"grad_norm": 20.817411422729492,
|
| 111 |
+
"learning_rate": 0.00042483419391697364,
|
| 112 |
+
"loss": 97.84152221679688,
|
| 113 |
+
"original_ce_loss": 6.1144,
|
| 114 |
+
"step": 130
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.0726023401289988,
|
| 118 |
+
"grad_norm": 13.969531059265137,
|
| 119 |
+
"learning_rate": 0.00042465871840307033,
|
| 120 |
+
"loss": 97.0799072265625,
|
| 121 |
+
"original_ce_loss": 6.0668,
|
| 122 |
+
"step": 140
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"epoch": 0.07778822156678443,
|
| 126 |
+
"grad_norm": 12.968173027038574,
|
| 127 |
+
"learning_rate": 0.00042442065035428465,
|
| 128 |
+
"loss": 95.65609741210938,
|
| 129 |
+
"original_ce_loss": 5.9779,
|
| 130 |
+
"step": 150
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 0.08297410300457006,
|
| 134 |
+
"grad_norm": 16.16962242126465,
|
| 135 |
+
"learning_rate": 0.00042412006000710166,
|
| 136 |
+
"loss": 94.4610107421875,
|
| 137 |
+
"original_ce_loss": 5.9032,
|
| 138 |
+
"step": 160
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"epoch": 0.08815998444235569,
|
| 142 |
+
"grad_norm": 15.63354206085205,
|
| 143 |
+
"learning_rate": 0.0004237570360437679,
|
| 144 |
+
"loss": 93.08035888671876,
|
| 145 |
+
"original_ce_loss": 5.8169,
|
| 146 |
+
"step": 170
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.09334586588014132,
|
| 150 |
+
"grad_norm": 19.45943260192871,
|
| 151 |
+
"learning_rate": 0.00042333168556612805,
|
| 152 |
+
"loss": 91.3424560546875,
|
| 153 |
+
"original_ce_loss": 5.7083,
|
| 154 |
+
"step": 180
|
| 155 |
+
},
|
| 156 |
+
{
|
| 157 |
+
"epoch": 0.09853174731792694,
|
| 158 |
+
"grad_norm": 12.314143180847168,
|
| 159 |
+
"learning_rate": 0.0004228441340640267,
|
| 160 |
+
"loss": 89.93196411132813,
|
| 161 |
+
"original_ce_loss": 5.6201,
|
| 162 |
+
"step": 190
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"epoch": 0.10371762875571257,
|
| 166 |
+
"grad_norm": 10.550477981567383,
|
| 167 |
+
"learning_rate": 0.00042229452537828567,
|
| 168 |
+
"loss": 87.97509765625,
|
| 169 |
+
"original_ce_loss": 5.4978,
|
| 170 |
+
"step": 200
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.1089035101934982,
|
| 174 |
+
"grad_norm": 18.28213882446289,
|
| 175 |
+
"learning_rate": 0.00042168302165826693,
|
| 176 |
+
"loss": 86.10928344726562,
|
| 177 |
+
"original_ce_loss": 5.3812,
|
| 178 |
+
"step": 210
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"epoch": 0.11408939163128383,
|
| 182 |
+
"grad_norm": 19.054182052612305,
|
| 183 |
+
"learning_rate": 0.0004210098033140343,
|
| 184 |
+
"loss": 84.20543212890625,
|
| 185 |
+
"original_ce_loss": 5.2622,
|
| 186 |
+
"step": 220
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
"epoch": 0.11927527306906946,
|
| 190 |
+
"grad_norm": 11.32401180267334,
|
| 191 |
+
"learning_rate": 0.00042035130458263533,
|
| 192 |
+
"loss": 82.33561401367187,
|
| 193 |
+
"original_ce_loss": 5.1454,
|
| 194 |
+
"step": 230
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"epoch": 0.12446115450685509,
|
| 198 |
+
"grad_norm": 22.777610778808594,
|
| 199 |
+
"learning_rate": 0.00041956139064506354,
|
| 200 |
+
"loss": 82.01093139648438,
|
| 201 |
+
"original_ce_loss": 5.1251,
|
| 202 |
+
"step": 240
|
| 203 |
+
},
|
| 204 |
+
{
|
| 205 |
+
"epoch": 0.12964703594464072,
|
| 206 |
+
"grad_norm": 14.207376480102539,
|
| 207 |
+
"learning_rate": 0.00041879823026024296,
|
| 208 |
+
"loss": 86.503125,
|
| 209 |
+
"original_ce_loss": 5.4058,
|
| 210 |
+
"step": 250
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.13483291738242634,
|
| 214 |
+
"grad_norm": NaN,
|
| 215 |
+
"learning_rate": 0.00041798577018333747,
|
| 216 |
+
"loss": 89.17313232421876,
|
| 217 |
+
"original_ce_loss": 5.5727,
|
| 218 |
+
"step": 260
|
| 219 |
+
},
|
| 220 |
+
{
|
| 221 |
+
"epoch": 0.14001879882021198,
|
| 222 |
+
"grad_norm": 25.01464080810547,
|
| 223 |
+
"learning_rate": 0.00041741683681028474,
|
| 224 |
+
"loss": 89.0274169921875,
|
| 225 |
+
"original_ce_loss": 5.5636,
|
| 226 |
+
"step": 270
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.1452046802579976,
|
| 230 |
+
"grad_norm": 19.480703353881836,
|
| 231 |
+
"learning_rate": 0.0004164202798601907,
|
| 232 |
+
"loss": 88.31423950195312,
|
| 233 |
+
"original_ce_loss": 5.519,
|
| 234 |
+
"step": 280
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"epoch": 0.15039056169578324,
|
| 238 |
+
"grad_norm": 12.254472732543945,
|
| 239 |
+
"learning_rate": 0.0004155797029694326,
|
| 240 |
+
"loss": 82.24807739257812,
|
| 241 |
+
"original_ce_loss": 5.1399,
|
| 242 |
+
"step": 290
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"epoch": 0.15557644313356886,
|
| 246 |
+
"grad_norm": 25.232437133789062,
|
| 247 |
+
"learning_rate": 0.0004144750780853972,
|
| 248 |
+
"loss": 76.16737060546875,
|
| 249 |
+
"original_ce_loss": 4.7599,
|
| 250 |
+
"step": 300
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.16076232457135448,
|
| 254 |
+
"grad_norm": 19.2685546875,
|
| 255 |
+
"learning_rate": 0.00041331086511459537,
|
| 256 |
+
"loss": 73.167529296875,
|
| 257 |
+
"original_ce_loss": 4.5724,
|
| 258 |
+
"step": 310
|
| 259 |
+
},
|
| 260 |
+
{
|
| 261 |
+
"epoch": 0.16594820600914012,
|
| 262 |
+
"grad_norm": 18.996484756469727,
|
| 263 |
+
"learning_rate": 0.0004120874075312033,
|
| 264 |
+
"loss": 71.38079223632812,
|
| 265 |
+
"original_ce_loss": 4.4607,
|
| 266 |
+
"step": 320
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"epoch": 0.17113408744692574,
|
| 270 |
+
"grad_norm": 16.110937118530273,
|
| 271 |
+
"learning_rate": 0.00041080506628815295,
|
| 272 |
+
"loss": 69.8762939453125,
|
| 273 |
+
"original_ce_loss": 4.3667,
|
| 274 |
+
"step": 330
|
| 275 |
+
},
|
| 276 |
+
{
|
| 277 |
+
"epoch": 0.17631996888471138,
|
| 278 |
+
"grad_norm": 19.29561424255371,
|
| 279 |
+
"learning_rate": 0.00040946421971064156,
|
| 280 |
+
"loss": 68.62410888671874,
|
| 281 |
+
"original_ce_loss": 4.2884,
|
| 282 |
+
"step": 340
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.181505850322497,
|
| 286 |
+
"grad_norm": 24.205785751342773,
|
| 287 |
+
"learning_rate": 0.0004080652633845155,
|
| 288 |
+
"loss": 67.31400756835937,
|
| 289 |
+
"original_ce_loss": 4.2065,
|
| 290 |
+
"step": 350
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"epoch": 0.18669173176028264,
|
| 294 |
+
"grad_norm": 22.90777587890625,
|
| 295 |
+
"learning_rate": 0.00040660861003956165,
|
| 296 |
+
"loss": 66.45149536132813,
|
| 297 |
+
"original_ce_loss": 4.1526,
|
| 298 |
+
"step": 360
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"epoch": 0.19187761319806826,
|
| 302 |
+
"grad_norm": 23.12751007080078,
|
| 303 |
+
"learning_rate": 0.0004050946894277412,
|
| 304 |
+
"loss": 65.15361938476562,
|
| 305 |
+
"original_ce_loss": 4.0715,
|
| 306 |
+
"step": 370
|
| 307 |
+
},
|
| 308 |
+
{
|
| 309 |
+
"epoch": 0.19706349463585388,
|
| 310 |
+
"grad_norm": 21.39095687866211,
|
| 311 |
+
"learning_rate": 0.0004035239481964011,
|
| 312 |
+
"loss": 64.37594604492188,
|
| 313 |
+
"original_ce_loss": 4.0229,
|
| 314 |
+
"step": 380
|
| 315 |
+
},
|
| 316 |
+
{
|
| 317 |
+
"epoch": 0.20224937607363952,
|
| 318 |
+
"grad_norm": 21.83388900756836,
|
| 319 |
+
"learning_rate": 0.000401896849756501,
|
| 320 |
+
"loss": 63.70396728515625,
|
| 321 |
+
"original_ce_loss": 3.9809,
|
| 322 |
+
"step": 390
|
| 323 |
+
},
|
| 324 |
+
{
|
| 325 |
+
"epoch": 0.20743525751142514,
|
| 326 |
+
"grad_norm": 20.973325729370117,
|
| 327 |
+
"learning_rate": 0.0004002138741458948,
|
| 328 |
+
"loss": 62.312738037109376,
|
| 329 |
+
"original_ce_loss": 3.894,
|
| 330 |
+
"step": 400
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.21262113894921078,
|
| 334 |
+
"grad_norm": 16.469514846801758,
|
| 335 |
+
"learning_rate": 0.0003984755178877064,
|
| 336 |
+
"loss": 61.592352294921874,
|
| 337 |
+
"original_ce_loss": 3.849,
|
| 338 |
+
"step": 410
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.2178070203869964,
|
| 342 |
+
"grad_norm": 17.492990493774414,
|
| 343 |
+
"learning_rate": 0.00039668229384384187,
|
| 344 |
+
"loss": 60.990655517578126,
|
| 345 |
+
"original_ce_loss": 3.8114,
|
| 346 |
+
"step": 420
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.22299290182478204,
|
| 350 |
+
"grad_norm": 22.369491577148438,
|
| 351 |
+
"learning_rate": 0.00039483473106368165,
|
| 352 |
+
"loss": 60.037548828125,
|
| 353 |
+
"original_ce_loss": 3.7518,
|
| 354 |
+
"step": 430
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.22817878326256766,
|
| 358 |
+
"grad_norm": 18.716108322143555,
|
| 359 |
+
"learning_rate": 0.0003929333746279965,
|
| 360 |
+
"loss": 58.94083862304687,
|
| 361 |
+
"original_ce_loss": 3.6832,
|
| 362 |
+
"step": 440
|
| 363 |
+
},
|
| 364 |
+
{
|
| 365 |
+
"epoch": 0.23336466470035327,
|
| 366 |
+
"grad_norm": 20.013795852661133,
|
| 367 |
+
"learning_rate": 0.00039097878548813433,
|
| 368 |
+
"loss": 58.7142578125,
|
| 369 |
+
"original_ce_loss": 3.6691,
|
| 370 |
+
"step": 450
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"epoch": 0.23855054613813892,
|
| 374 |
+
"grad_norm": 21.67531967163086,
|
| 375 |
+
"learning_rate": 0.0003891746175849475,
|
| 376 |
+
"loss": 57.764068603515625,
|
| 377 |
+
"original_ce_loss": 3.6097,
|
| 378 |
+
"step": 460
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"epoch": 0.24373642757592454,
|
| 382 |
+
"grad_norm": 16.39697265625,
|
| 383 |
+
"learning_rate": 0.00038712048783883754,
|
| 384 |
+
"loss": 57.28966064453125,
|
| 385 |
+
"original_ce_loss": 3.5801,
|
| 386 |
+
"step": 470
|
| 387 |
+
},
|
| 388 |
+
{
|
| 389 |
+
"epoch": 0.24892230901371018,
|
| 390 |
+
"grad_norm": 23.23872184753418,
|
| 391 |
+
"learning_rate": 0.00038501484034668117,
|
| 392 |
+
"loss": 56.9778564453125,
|
| 393 |
+
"original_ce_loss": 3.5606,
|
| 394 |
+
"step": 480
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 0.2541081904514958,
|
| 398 |
+
"grad_norm": 20.47743797302246,
|
| 399 |
+
"learning_rate": 0.0003828582963311888,
|
| 400 |
+
"loss": 56.27522583007813,
|
| 401 |
+
"original_ce_loss": 3.5167,
|
| 402 |
+
"step": 490
|
| 403 |
+
},
|
| 404 |
+
{
|
| 405 |
+
"epoch": 0.25929407188928144,
|
| 406 |
+
"grad_norm": 17.760643005371094,
|
| 407 |
+
"learning_rate": 0.00038065149203091593,
|
| 408 |
+
"loss": 55.25050048828125,
|
| 409 |
+
"original_ce_loss": 3.4526,
|
| 410 |
+
"step": 500
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.2644799533270671,
|
| 414 |
+
"grad_norm": 15.715126037597656,
|
| 415 |
+
"learning_rate": 0.00037839507851255536,
|
| 416 |
+
"loss": 54.68311767578125,
|
| 417 |
+
"original_ce_loss": 3.4172,
|
| 418 |
+
"step": 510
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.2696658347648527,
|
| 422 |
+
"grad_norm": 17.068340301513672,
|
| 423 |
+
"learning_rate": 0.00037608972147885504,
|
| 424 |
+
"loss": 54.7534423828125,
|
| 425 |
+
"original_ce_loss": 3.4216,
|
| 426 |
+
"step": 520
|
| 427 |
+
},
|
| 428 |
+
{
|
| 429 |
+
"epoch": 0.2748517162026383,
|
| 430 |
+
"grad_norm": 17.919513702392578,
|
| 431 |
+
"learning_rate": 0.00037373610107221803,
|
| 432 |
+
"loss": 54.63709716796875,
|
| 433 |
+
"original_ce_loss": 3.4143,
|
| 434 |
+
"step": 530
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"epoch": 0.28003759764042396,
|
| 438 |
+
"grad_norm": 19.53166389465332,
|
| 439 |
+
"learning_rate": 0.00037133491167404194,
|
| 440 |
+
"loss": 53.847723388671874,
|
| 441 |
+
"original_ce_loss": 3.3649,
|
| 442 |
+
"step": 540
|
| 443 |
+
},
|
| 444 |
+
{
|
| 445 |
+
"epoch": 0.28522347907820955,
|
| 446 |
+
"grad_norm": 23.71944236755371,
|
| 447 |
+
"learning_rate": 0.0003688868616998578,
|
| 448 |
+
"loss": 53.34617919921875,
|
| 449 |
+
"original_ce_loss": 3.3336,
|
| 450 |
+
"step": 550
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"epoch": 0.2904093605159952,
|
| 454 |
+
"grad_norm": 19.73995590209961,
|
| 455 |
+
"learning_rate": 0.0003663926733903284,
|
| 456 |
+
"loss": 52.91886596679687,
|
| 457 |
+
"original_ce_loss": 3.3069,
|
| 458 |
+
"step": 560
|
| 459 |
+
},
|
| 460 |
+
{
|
| 461 |
+
"epoch": 0.29559524195378084,
|
| 462 |
+
"grad_norm": 18.594099044799805,
|
| 463 |
+
"learning_rate": 0.00036385308259816797,
|
| 464 |
+
"loss": 52.07973022460938,
|
| 465 |
+
"original_ce_loss": 3.2545,
|
| 466 |
+
"step": 570
|
| 467 |
+
},
|
| 468 |
+
{
|
| 469 |
+
"epoch": 0.3007811233915665,
|
| 470 |
+
"grad_norm": 19.95176887512207,
|
| 471 |
+
"learning_rate": 0.00036126883857104554,
|
| 472 |
+
"loss": 51.62211303710937,
|
| 473 |
+
"original_ce_loss": 3.2259,
|
| 474 |
+
"step": 580
|
| 475 |
+
},
|
| 476 |
+
{
|
| 477 |
+
"epoch": 0.3059670048293521,
|
| 478 |
+
"grad_norm": 19.418689727783203,
|
| 479 |
+
"learning_rate": 0.00035864070373053703,
|
| 480 |
+
"loss": 51.4861328125,
|
| 481 |
+
"original_ce_loss": 3.2174,
|
| 482 |
+
"step": 590
|
| 483 |
+
},
|
| 484 |
+
{
|
| 485 |
+
"epoch": 0.3111528862671377,
|
| 486 |
+
"grad_norm": 17.515121459960938,
|
| 487 |
+
"learning_rate": 0.00035596945344719004,
|
| 488 |
+
"loss": 50.923562622070314,
|
| 489 |
+
"original_ce_loss": 3.1822,
|
| 490 |
+
"step": 600
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.31633876770492336,
|
| 494 |
+
"grad_norm": 17.70184326171875,
|
| 495 |
+
"learning_rate": 0.0003532558758117689,
|
| 496 |
+
"loss": 50.95531311035156,
|
| 497 |
+
"original_ce_loss": 3.1842,
|
| 498 |
+
"step": 610
|
| 499 |
+
},
|
| 500 |
+
{
|
| 501 |
+
"epoch": 0.32152464914270895,
|
| 502 |
+
"grad_norm": 19.828022003173828,
|
| 503 |
+
"learning_rate": 0.00035050077140274654,
|
| 504 |
+
"loss": 50.60614929199219,
|
| 505 |
+
"original_ce_loss": 3.1624,
|
| 506 |
+
"step": 620
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.3267105305804946,
|
| 510 |
+
"grad_norm": 17.798105239868164,
|
| 511 |
+
"learning_rate": 0.0003477049530501123,
|
| 512 |
+
"loss": 49.87000732421875,
|
| 513 |
+
"original_ce_loss": 3.1164,
|
| 514 |
+
"step": 630
|
| 515 |
+
},
|
| 516 |
+
{
|
| 517 |
+
"epoch": 0.33189641201828024,
|
| 518 |
+
"grad_norm": 16.208608627319336,
|
| 519 |
+
"learning_rate": 0.0003448692455955654,
|
| 520 |
+
"loss": 49.59539794921875,
|
| 521 |
+
"original_ce_loss": 3.0992,
|
| 522 |
+
"step": 640
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 0.3370822934560659,
|
| 526 |
+
"grad_norm": 17.62627410888672,
|
| 527 |
+
"learning_rate": 0.000341994485649164,
|
| 528 |
+
"loss": 49.53982238769531,
|
| 529 |
+
"original_ce_loss": 3.0957,
|
| 530 |
+
"step": 650
|
| 531 |
+
},
|
| 532 |
+
{
|
| 533 |
+
"epoch": 0.3422681748938515,
|
| 534 |
+
"grad_norm": 17.039188385009766,
|
| 535 |
+
"learning_rate": 0.00033908152134250305,
|
| 536 |
+
"loss": 49.455194091796876,
|
| 537 |
+
"original_ce_loss": 3.0904,
|
| 538 |
+
"step": 660
|
| 539 |
+
},
|
| 540 |
+
{
|
| 541 |
+
"epoch": 0.3474540563316371,
|
| 542 |
+
"grad_norm": 16.977937698364258,
|
| 543 |
+
"learning_rate": 0.00033613121207849256,
|
| 544 |
+
"loss": 49.12623596191406,
|
| 545 |
+
"original_ce_loss": 3.0699,
|
| 546 |
+
"step": 670
|
| 547 |
+
},
|
| 548 |
+
{
|
| 549 |
+
"epoch": 0.35263993776942276,
|
| 550 |
+
"grad_norm": 16.655452728271484,
|
| 551 |
+
"learning_rate": 0.0003331444282778105,
|
| 552 |
+
"loss": 49.13233337402344,
|
| 553 |
+
"original_ce_loss": 3.0703,
|
| 554 |
+
"step": 680
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.35782581920720835,
|
| 558 |
+
"grad_norm": 19.1124267578125,
|
| 559 |
+
"learning_rate": 0.00033012205112210554,
|
| 560 |
+
"loss": 48.60373229980469,
|
| 561 |
+
"original_ce_loss": 3.0372,
|
| 562 |
+
"step": 690
|
| 563 |
+
},
|
| 564 |
+
{
|
| 565 |
+
"epoch": 0.363011700644994,
|
| 566 |
+
"grad_norm": 15.127534866333008,
|
| 567 |
+
"learning_rate": 0.00032706497229402433,
|
| 568 |
+
"loss": 48.820004272460935,
|
| 569 |
+
"original_ce_loss": 3.0508,
|
| 570 |
+
"step": 700
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 0.36819758208277964,
|
| 574 |
+
"grad_norm": 14.931536674499512,
|
| 575 |
+
"learning_rate": 0.00032397409371414114,
|
| 576 |
+
"loss": 48.08933715820312,
|
| 577 |
+
"original_ce_loss": 3.0051,
|
| 578 |
+
"step": 710
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 0.3733834635205653,
|
| 582 |
+
"grad_norm": 14.94162368774414,
|
| 583 |
+
"learning_rate": 0.0003208503272748677,
|
| 584 |
+
"loss": 47.9984619140625,
|
| 585 |
+
"original_ce_loss": 2.9994,
|
| 586 |
+
"step": 720
|
| 587 |
+
},
|
| 588 |
+
{
|
| 589 |
+
"epoch": 0.3785693449583509,
|
| 590 |
+
"grad_norm": 15.836956977844238,
|
| 591 |
+
"learning_rate": 0.0003176945945714194,
|
| 592 |
+
"loss": 47.84305114746094,
|
| 593 |
+
"original_ce_loss": 2.9897,
|
| 594 |
+
"step": 730
|
| 595 |
+
},
|
| 596 |
+
{
|
| 597 |
+
"epoch": 0.3837552263961365,
|
| 598 |
+
"grad_norm": 16.20482635498047,
|
| 599 |
+
"learning_rate": 0.00031450782662992055,
|
| 600 |
+
"loss": 47.57247314453125,
|
| 601 |
+
"original_ce_loss": 2.9728,
|
| 602 |
+
"step": 740
|
| 603 |
+
},
|
| 604 |
+
{
|
| 605 |
+
"epoch": 0.38894110783392216,
|
| 606 |
+
"grad_norm": 18.379453659057617,
|
| 607 |
+
"learning_rate": 0.00031129096363272617,
|
| 608 |
+
"loss": 47.402618408203125,
|
| 609 |
+
"original_ce_loss": 2.9622,
|
| 610 |
+
"step": 750
|
| 611 |
+
},
|
| 612 |
+
{
|
| 613 |
+
"epoch": 0.39412698927170775,
|
| 614 |
+
"grad_norm": 14.954010963439941,
|
| 615 |
+
"learning_rate": 0.0003080449546410435,
|
| 616 |
+
"loss": 47.014581298828126,
|
| 617 |
+
"original_ce_loss": 2.9379,
|
| 618 |
+
"step": 760
|
| 619 |
+
},
|
| 620 |
+
{
|
| 621 |
+
"epoch": 0.3993128707094934,
|
| 622 |
+
"grad_norm": 21.57929229736328,
|
| 623 |
+
"learning_rate": 0.0003047707573149325,
|
| 624 |
+
"loss": 46.799008178710935,
|
| 625 |
+
"original_ce_loss": 2.9245,
|
| 626 |
+
"step": 770
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.40449875214727904,
|
| 630 |
+
"grad_norm": 15.74538803100586,
|
| 631 |
+
"learning_rate": 0.0003014693376307713,
|
| 632 |
+
"loss": 46.99599609375,
|
| 633 |
+
"original_ce_loss": 2.9368,
|
| 634 |
+
"step": 780
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 0.4096846335850647,
|
| 638 |
+
"grad_norm": 13.851374626159668,
|
| 639 |
+
"learning_rate": 0.0002981416695962664,
|
| 640 |
+
"loss": 46.543948364257815,
|
| 641 |
+
"original_ce_loss": 2.9085,
|
| 642 |
+
"step": 790
|
| 643 |
+
},
|
| 644 |
+
{
|
| 645 |
+
"epoch": 0.4148705150228503,
|
| 646 |
+
"grad_norm": 16.17242431640625,
|
| 647 |
+
"learning_rate": 0.0002947887349630945,
|
| 648 |
+
"loss": 46.3514892578125,
|
| 649 |
+
"original_ce_loss": 2.8965,
|
| 650 |
+
"step": 800
|
| 651 |
+
},
|
| 652 |
+
{
|
| 653 |
+
"epoch": 0.4200563964606359,
|
| 654 |
+
"grad_norm": 15.615081787109375,
|
| 655 |
+
"learning_rate": 0.0002914115229372593,
|
| 656 |
+
"loss": 46.29792175292969,
|
| 657 |
+
"original_ce_loss": 2.8931,
|
| 658 |
+
"step": 810
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"epoch": 0.42524227789842156,
|
| 662 |
+
"grad_norm": 37.81509017944336,
|
| 663 |
+
"learning_rate": 0.0002880110298872492,
|
| 664 |
+
"loss": 46.14377136230469,
|
| 665 |
+
"original_ce_loss": 2.8835,
|
| 666 |
+
"step": 820
|
| 667 |
+
},
|
| 668 |
+
{
|
| 669 |
+
"epoch": 0.43042815933620715,
|
| 670 |
+
"grad_norm": 17.120006561279297,
|
| 671 |
+
"learning_rate": 0.00028458825905008136,
|
| 672 |
+
"loss": 45.960760498046874,
|
| 673 |
+
"original_ce_loss": 2.8721,
|
| 674 |
+
"step": 830
|
| 675 |
+
},
|
| 676 |
+
{
|
| 677 |
+
"epoch": 0.4356140407739928,
|
| 678 |
+
"grad_norm": 18.173112869262695,
|
| 679 |
+
"learning_rate": 0.0002811442202353204,
|
| 680 |
+
"loss": 45.85289306640625,
|
| 681 |
+
"original_ce_loss": 2.8653,
|
| 682 |
+
"step": 840
|
| 683 |
+
},
|
| 684 |
+
{
|
| 685 |
+
"epoch": 0.44079992221177844,
|
| 686 |
+
"grad_norm": 14.53779125213623,
|
| 687 |
+
"learning_rate": 0.00027767992952715677,
|
| 688 |
+
"loss": 45.891665649414065,
|
| 689 |
+
"original_ce_loss": 2.8678,
|
| 690 |
+
"step": 850
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.4459858036495641,
|
| 694 |
+
"grad_norm": 16.2801570892334,
|
| 695 |
+
"learning_rate": 0.00027419640898463483,
|
| 696 |
+
"loss": 45.703085327148436,
|
| 697 |
+
"original_ce_loss": 2.856,
|
| 698 |
+
"step": 860
|
| 699 |
+
},
|
| 700 |
+
{
|
| 701 |
+
"epoch": 0.45117168508734967,
|
| 702 |
+
"grad_norm": 16.596567153930664,
|
| 703 |
+
"learning_rate": 0.0002706946863401179,
|
| 704 |
+
"loss": 45.31675415039062,
|
| 705 |
+
"original_ce_loss": 2.8318,
|
| 706 |
+
"step": 870
|
| 707 |
+
},
|
| 708 |
+
{
|
| 709 |
+
"epoch": 0.4563575665251353,
|
| 710 |
+
"grad_norm": 15.852386474609375,
|
| 711 |
+
"learning_rate": 0.00026752842691598927,
|
| 712 |
+
"loss": 44.9458740234375,
|
| 713 |
+
"original_ce_loss": 2.8086,
|
| 714 |
+
"step": 880
|
| 715 |
+
},
|
| 716 |
+
{
|
| 717 |
+
"epoch": 0.46154344796292096,
|
| 718 |
+
"grad_norm": 16.00336265563965,
|
| 719 |
+
"learning_rate": 0.00026399497066780355,
|
| 720 |
+
"loss": 45.187741088867185,
|
| 721 |
+
"original_ce_loss": 2.8238,
|
| 722 |
+
"step": 890
|
| 723 |
+
},
|
| 724 |
+
{
|
| 725 |
+
"epoch": 0.46672932940070655,
|
| 726 |
+
"grad_norm": 13.984482765197754,
|
| 727 |
+
"learning_rate": 0.00026044632201662794,
|
| 728 |
+
"loss": 45.00270385742188,
|
| 729 |
+
"original_ce_loss": 2.8122,
|
| 730 |
+
"step": 900
|
| 731 |
+
},
|
| 732 |
+
{
|
| 733 |
+
"epoch": 0.4719152108384922,
|
| 734 |
+
"grad_norm": 17.663005828857422,
|
| 735 |
+
"learning_rate": 0.0002568835279093716,
|
| 736 |
+
"loss": 44.76839599609375,
|
| 737 |
+
"original_ce_loss": 2.7976,
|
| 738 |
+
"step": 910
|
| 739 |
+
},
|
| 740 |
+
{
|
| 741 |
+
"epoch": 0.47710109227627784,
|
| 742 |
+
"grad_norm": 15.069225311279297,
|
| 743 |
+
"learning_rate": 0.0002533076394662341,
|
| 744 |
+
"loss": 44.91849975585937,
|
| 745 |
+
"original_ce_loss": 2.8069,
|
| 746 |
+
"step": 920
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 0.4822869737140635,
|
| 750 |
+
"grad_norm": 112.49778747558594,
|
| 751 |
+
"learning_rate": 0.00024971971167059694,
|
| 752 |
+
"loss": 44.654718017578126,
|
| 753 |
+
"original_ce_loss": 2.7905,
|
| 754 |
+
"step": 930
|
| 755 |
+
},
|
| 756 |
+
{
|
| 757 |
+
"epoch": 0.48747285515184907,
|
| 758 |
+
"grad_norm": 16.59328269958496,
|
| 759 |
+
"learning_rate": 0.0002461208030577744,
|
| 760 |
+
"loss": 44.622955322265625,
|
| 761 |
+
"original_ce_loss": 2.7885,
|
| 762 |
+
"step": 940
|
| 763 |
+
},
|
| 764 |
+
{
|
| 765 |
+
"epoch": 0.4926587365896347,
|
| 766 |
+
"grad_norm": 16.046436309814453,
|
| 767 |
+
"learning_rate": 0.00024251197540271787,
|
| 768 |
+
"loss": 44.73280639648438,
|
| 769 |
+
"original_ce_loss": 2.7953,
|
| 770 |
+
"step": 950
|
| 771 |
+
},
|
| 772 |
+
{
|
| 773 |
+
"epoch": 0.49784461802742036,
|
| 774 |
+
"grad_norm": 21.73870086669922,
|
| 775 |
+
"learning_rate": 0.000238894293406763,
|
| 776 |
+
"loss": 44.69492797851562,
|
| 777 |
+
"original_ce_loss": 2.793,
|
| 778 |
+
"step": 960
|
| 779 |
+
},
|
| 780 |
+
{
|
| 781 |
+
"epoch": 0.503030499465206,
|
| 782 |
+
"grad_norm": 25.98779296875,
|
| 783 |
+
"learning_rate": 0.00023526882438351462,
|
| 784 |
+
"loss": 44.65166015625,
|
| 785 |
+
"original_ce_loss": 2.7903,
|
| 786 |
+
"step": 970
|
| 787 |
+
},
|
| 788 |
+
{
|
| 789 |
+
"epoch": 0.5082163809029916,
|
| 790 |
+
"grad_norm": 20.669517517089844,
|
| 791 |
+
"learning_rate": 0.00023163663794396023,
|
| 792 |
+
"loss": 44.669866943359374,
|
| 793 |
+
"original_ce_loss": 2.7914,
|
| 794 |
+
"step": 980
|
| 795 |
+
},
|
| 796 |
+
{
|
| 797 |
+
"epoch": 0.5134022623407772,
|
| 798 |
+
"grad_norm": 28.18373680114746,
|
| 799 |
+
"learning_rate": 0.00022799880568090614,
|
| 800 |
+
"loss": 44.25715942382813,
|
| 801 |
+
"original_ce_loss": 2.7656,
|
| 802 |
+
"step": 990
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 0.5185881437785629,
|
| 806 |
+
"grad_norm": 28.256511688232422,
|
| 807 |
+
"learning_rate": 0.00022435640085282846,
|
| 808 |
+
"loss": 44.33907775878906,
|
| 809 |
+
"original_ce_loss": 2.7707,
|
| 810 |
+
"step": 1000
|
| 811 |
+
}
|
| 812 |
+
],
|
| 813 |
+
"logging_steps": 10,
|
| 814 |
+
"max_steps": 1929,
|
| 815 |
+
"num_input_tokens_seen": 0,
|
| 816 |
+
"num_train_epochs": 1,
|
| 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": 1.0595084938051584e+16,
|
| 831 |
+
"train_batch_size": 16,
|
| 832 |
+
"trial_name": null,
|
| 833 |
+
"trial_params": null
|
| 834 |
+
}
|