Upload 9 files
Browse files- config.json +31 -0
- generation_config.json +6 -0
- model.safetensors +3 -0
- rng_state.pth +3 -0
- scheduler.pt +3 -0
- tokenizer.json +0 -0
- tokenizer_config.json +1 -0
- trainer_state.json +2090 -0
- training_args.bin +3 -0
config.json
ADDED
|
@@ -0,0 +1,31 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "../MistralModels/models/Mixtral-8x7B-v0.1-small-4096",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"MixtralForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 1,
|
| 8 |
+
"eos_token_id": 2,
|
| 9 |
+
"hidden_act": "silu",
|
| 10 |
+
"hidden_size": 768,
|
| 11 |
+
"initializer_range": 0.02,
|
| 12 |
+
"intermediate_size": 768,
|
| 13 |
+
"max_position_embeddings": 512,
|
| 14 |
+
"model_type": "mixtral",
|
| 15 |
+
"num_attention_heads": 8,
|
| 16 |
+
"num_experts_per_tok": 1,
|
| 17 |
+
"num_hidden_layers": 8,
|
| 18 |
+
"num_key_value_heads": 8,
|
| 19 |
+
"num_local_experts": 8,
|
| 20 |
+
"output_router_logits": false,
|
| 21 |
+
"rms_norm_eps": 1e-05,
|
| 22 |
+
"rope_theta": 1000000.0,
|
| 23 |
+
"router_aux_loss_coef": 0.02,
|
| 24 |
+
"router_jitter_noise": 0.0,
|
| 25 |
+
"sliding_window": null,
|
| 26 |
+
"tie_word_embeddings": false,
|
| 27 |
+
"torch_dtype": "bfloat16",
|
| 28 |
+
"transformers_version": "4.43.3",
|
| 29 |
+
"use_cache": true,
|
| 30 |
+
"vocab_size": 4096
|
| 31 |
+
}
|
generation_config.json
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"bos_token_id": 1,
|
| 4 |
+
"eos_token_id": 2,
|
| 5 |
+
"transformers_version": "4.43.3"
|
| 6 |
+
}
|
model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6355a1c4e91fcba689a8bdc46b285b55614c70e7284c086c7074fd6b73a2f3b5
|
| 3 |
+
size 276979168
|
rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f86f749dc73bc16a2502fef2f98f5c00b4400cb2c67fbe62653b7ed104d13779
|
| 3 |
+
size 14244
|
scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e020fcef19fdb2ed8b130f1961d42d4f3d32869aedf58e153ed40f1d0ef0385f
|
| 3 |
+
size 1064
|
tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"tokenizer_class": "PreTrainedTokenizerFast", "unk_token": "[UNK]", "cls_token": "[CLS]", "sep_token": "[SEP]", "pad_token": "[PAD]", "mask_token": "[MASK]"}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,2090 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": 2.4190404415130615,
|
| 3 |
+
"best_model_checkpoint": "./results/models/checkpoint-136458",
|
| 4 |
+
"epoch": 18.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 136458,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.06595435958316845,
|
| 13 |
+
"grad_norm": 0.63671875,
|
| 14 |
+
"learning_rate": 0.0009986809128083366,
|
| 15 |
+
"loss": 5.3104,
|
| 16 |
+
"step": 500
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.1319087191663369,
|
| 20 |
+
"grad_norm": 0.640625,
|
| 21 |
+
"learning_rate": 0.0009973618256166733,
|
| 22 |
+
"loss": 4.3028,
|
| 23 |
+
"step": 1000
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.19786307874950534,
|
| 27 |
+
"grad_norm": 0.61328125,
|
| 28 |
+
"learning_rate": 0.0009960427384250099,
|
| 29 |
+
"loss": 3.9369,
|
| 30 |
+
"step": 1500
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.2638174383326738,
|
| 34 |
+
"grad_norm": 0.5390625,
|
| 35 |
+
"learning_rate": 0.0009947236512333464,
|
| 36 |
+
"loss": 3.7115,
|
| 37 |
+
"step": 2000
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.3297717979158422,
|
| 41 |
+
"grad_norm": 0.59375,
|
| 42 |
+
"learning_rate": 0.0009934045640416832,
|
| 43 |
+
"loss": 3.5589,
|
| 44 |
+
"step": 2500
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.3957261574990107,
|
| 48 |
+
"grad_norm": 0.62890625,
|
| 49 |
+
"learning_rate": 0.0009920854768500197,
|
| 50 |
+
"loss": 3.4636,
|
| 51 |
+
"step": 3000
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.4616805170821791,
|
| 55 |
+
"grad_norm": 0.53515625,
|
| 56 |
+
"learning_rate": 0.0009907663896583565,
|
| 57 |
+
"loss": 3.3838,
|
| 58 |
+
"step": 3500
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.5276348766653476,
|
| 62 |
+
"grad_norm": 0.5859375,
|
| 63 |
+
"learning_rate": 0.000989447302466693,
|
| 64 |
+
"loss": 3.3023,
|
| 65 |
+
"step": 4000
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.593589236248516,
|
| 69 |
+
"grad_norm": 0.6171875,
|
| 70 |
+
"learning_rate": 0.0009881282152750298,
|
| 71 |
+
"loss": 3.2445,
|
| 72 |
+
"step": 4500
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.6595435958316844,
|
| 76 |
+
"grad_norm": 0.76171875,
|
| 77 |
+
"learning_rate": 0.0009868091280833663,
|
| 78 |
+
"loss": 3.205,
|
| 79 |
+
"step": 5000
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.725497955414853,
|
| 83 |
+
"grad_norm": 0.5390625,
|
| 84 |
+
"learning_rate": 0.0009854900408917029,
|
| 85 |
+
"loss": 3.157,
|
| 86 |
+
"step": 5500
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.7914523149980214,
|
| 90 |
+
"grad_norm": 0.51953125,
|
| 91 |
+
"learning_rate": 0.0009841709537000397,
|
| 92 |
+
"loss": 3.126,
|
| 93 |
+
"step": 6000
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.8574066745811898,
|
| 97 |
+
"grad_norm": 0.59765625,
|
| 98 |
+
"learning_rate": 0.0009828518665083762,
|
| 99 |
+
"loss": 3.0838,
|
| 100 |
+
"step": 6500
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.9233610341643582,
|
| 104 |
+
"grad_norm": 0.5546875,
|
| 105 |
+
"learning_rate": 0.0009815327793167127,
|
| 106 |
+
"loss": 3.0584,
|
| 107 |
+
"step": 7000
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.9893153937475268,
|
| 111 |
+
"grad_norm": 0.5234375,
|
| 112 |
+
"learning_rate": 0.0009802136921250495,
|
| 113 |
+
"loss": 3.0325,
|
| 114 |
+
"step": 7500
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 1.0,
|
| 118 |
+
"eval_loss": 3.0351455211639404,
|
| 119 |
+
"eval_runtime": 4.9415,
|
| 120 |
+
"eval_samples_per_second": 404.738,
|
| 121 |
+
"eval_steps_per_second": 1.619,
|
| 122 |
+
"step": 7581
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"epoch": 1.0552697533306952,
|
| 126 |
+
"grad_norm": 0.55859375,
|
| 127 |
+
"learning_rate": 0.000978894604933386,
|
| 128 |
+
"loss": 2.9581,
|
| 129 |
+
"step": 8000
|
| 130 |
+
},
|
| 131 |
+
{
|
| 132 |
+
"epoch": 1.1212241129138636,
|
| 133 |
+
"grad_norm": 0.625,
|
| 134 |
+
"learning_rate": 0.0009775755177417228,
|
| 135 |
+
"loss": 2.9426,
|
| 136 |
+
"step": 8500
|
| 137 |
+
},
|
| 138 |
+
{
|
| 139 |
+
"epoch": 1.187178472497032,
|
| 140 |
+
"grad_norm": 0.53125,
|
| 141 |
+
"learning_rate": 0.0009762564305500594,
|
| 142 |
+
"loss": 2.9122,
|
| 143 |
+
"step": 9000
|
| 144 |
+
},
|
| 145 |
+
{
|
| 146 |
+
"epoch": 1.2531328320802004,
|
| 147 |
+
"grad_norm": 0.58203125,
|
| 148 |
+
"learning_rate": 0.000974937343358396,
|
| 149 |
+
"loss": 2.8982,
|
| 150 |
+
"step": 9500
|
| 151 |
+
},
|
| 152 |
+
{
|
| 153 |
+
"epoch": 1.3190871916633689,
|
| 154 |
+
"grad_norm": 0.55859375,
|
| 155 |
+
"learning_rate": 0.0009736182561667326,
|
| 156 |
+
"loss": 2.891,
|
| 157 |
+
"step": 10000
|
| 158 |
+
},
|
| 159 |
+
{
|
| 160 |
+
"epoch": 1.3850415512465375,
|
| 161 |
+
"grad_norm": 0.53515625,
|
| 162 |
+
"learning_rate": 0.0009722991689750693,
|
| 163 |
+
"loss": 2.8751,
|
| 164 |
+
"step": 10500
|
| 165 |
+
},
|
| 166 |
+
{
|
| 167 |
+
"epoch": 1.450995910829706,
|
| 168 |
+
"grad_norm": 0.609375,
|
| 169 |
+
"learning_rate": 0.0009709800817834059,
|
| 170 |
+
"loss": 2.853,
|
| 171 |
+
"step": 11000
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"epoch": 1.5169502704128743,
|
| 175 |
+
"grad_norm": 0.56640625,
|
| 176 |
+
"learning_rate": 0.0009696609945917425,
|
| 177 |
+
"loss": 2.842,
|
| 178 |
+
"step": 11500
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"epoch": 1.5829046299960428,
|
| 182 |
+
"grad_norm": 0.5390625,
|
| 183 |
+
"learning_rate": 0.0009683419074000792,
|
| 184 |
+
"loss": 2.831,
|
| 185 |
+
"step": 12000
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 1.6488589895792112,
|
| 189 |
+
"grad_norm": 0.546875,
|
| 190 |
+
"learning_rate": 0.0009670228202084158,
|
| 191 |
+
"loss": 2.8257,
|
| 192 |
+
"step": 12500
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 1.7148133491623796,
|
| 196 |
+
"grad_norm": 0.58984375,
|
| 197 |
+
"learning_rate": 0.0009657037330167524,
|
| 198 |
+
"loss": 2.8069,
|
| 199 |
+
"step": 13000
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 1.780767708745548,
|
| 203 |
+
"grad_norm": 0.51171875,
|
| 204 |
+
"learning_rate": 0.0009643846458250891,
|
| 205 |
+
"loss": 2.7999,
|
| 206 |
+
"step": 13500
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 1.8467220683287167,
|
| 210 |
+
"grad_norm": 0.53125,
|
| 211 |
+
"learning_rate": 0.0009630655586334257,
|
| 212 |
+
"loss": 2.7877,
|
| 213 |
+
"step": 14000
|
| 214 |
+
},
|
| 215 |
+
{
|
| 216 |
+
"epoch": 1.9126764279118849,
|
| 217 |
+
"grad_norm": 0.625,
|
| 218 |
+
"learning_rate": 0.0009617464714417623,
|
| 219 |
+
"loss": 2.7914,
|
| 220 |
+
"step": 14500
|
| 221 |
+
},
|
| 222 |
+
{
|
| 223 |
+
"epoch": 1.9786307874950535,
|
| 224 |
+
"grad_norm": 0.546875,
|
| 225 |
+
"learning_rate": 0.0009604273842500989,
|
| 226 |
+
"loss": 2.7667,
|
| 227 |
+
"step": 15000
|
| 228 |
+
},
|
| 229 |
+
{
|
| 230 |
+
"epoch": 2.0,
|
| 231 |
+
"eval_loss": 2.802887201309204,
|
| 232 |
+
"eval_runtime": 5.0226,
|
| 233 |
+
"eval_samples_per_second": 398.199,
|
| 234 |
+
"eval_steps_per_second": 1.593,
|
| 235 |
+
"step": 15162
|
| 236 |
+
},
|
| 237 |
+
{
|
| 238 |
+
"epoch": 2.0445851470782217,
|
| 239 |
+
"grad_norm": 0.57421875,
|
| 240 |
+
"learning_rate": 0.0009591082970584356,
|
| 241 |
+
"loss": 2.6997,
|
| 242 |
+
"step": 15500
|
| 243 |
+
},
|
| 244 |
+
{
|
| 245 |
+
"epoch": 2.1105395066613903,
|
| 246 |
+
"grad_norm": 0.53125,
|
| 247 |
+
"learning_rate": 0.0009577892098667722,
|
| 248 |
+
"loss": 2.688,
|
| 249 |
+
"step": 16000
|
| 250 |
+
},
|
| 251 |
+
{
|
| 252 |
+
"epoch": 2.1764938662445585,
|
| 253 |
+
"grad_norm": 0.54296875,
|
| 254 |
+
"learning_rate": 0.0009564701226751088,
|
| 255 |
+
"loss": 2.6977,
|
| 256 |
+
"step": 16500
|
| 257 |
+
},
|
| 258 |
+
{
|
| 259 |
+
"epoch": 2.242448225827727,
|
| 260 |
+
"grad_norm": 0.578125,
|
| 261 |
+
"learning_rate": 0.0009551510354834455,
|
| 262 |
+
"loss": 2.6954,
|
| 263 |
+
"step": 17000
|
| 264 |
+
},
|
| 265 |
+
{
|
| 266 |
+
"epoch": 2.308402585410896,
|
| 267 |
+
"grad_norm": 0.53125,
|
| 268 |
+
"learning_rate": 0.000953831948291782,
|
| 269 |
+
"loss": 2.6841,
|
| 270 |
+
"step": 17500
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"epoch": 2.374356944994064,
|
| 274 |
+
"grad_norm": 0.56640625,
|
| 275 |
+
"learning_rate": 0.0009525128611001187,
|
| 276 |
+
"loss": 2.6851,
|
| 277 |
+
"step": 18000
|
| 278 |
+
},
|
| 279 |
+
{
|
| 280 |
+
"epoch": 2.4403113045772327,
|
| 281 |
+
"grad_norm": 0.546875,
|
| 282 |
+
"learning_rate": 0.0009511937739084554,
|
| 283 |
+
"loss": 2.6696,
|
| 284 |
+
"step": 18500
|
| 285 |
+
},
|
| 286 |
+
{
|
| 287 |
+
"epoch": 2.506265664160401,
|
| 288 |
+
"grad_norm": 0.54296875,
|
| 289 |
+
"learning_rate": 0.000949874686716792,
|
| 290 |
+
"loss": 2.6717,
|
| 291 |
+
"step": 19000
|
| 292 |
+
},
|
| 293 |
+
{
|
| 294 |
+
"epoch": 2.5722200237435695,
|
| 295 |
+
"grad_norm": 0.53125,
|
| 296 |
+
"learning_rate": 0.0009485555995251286,
|
| 297 |
+
"loss": 2.6728,
|
| 298 |
+
"step": 19500
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"epoch": 2.6381743833267377,
|
| 302 |
+
"grad_norm": 0.578125,
|
| 303 |
+
"learning_rate": 0.0009472365123334653,
|
| 304 |
+
"loss": 2.6487,
|
| 305 |
+
"step": 20000
|
| 306 |
+
},
|
| 307 |
+
{
|
| 308 |
+
"epoch": 2.7041287429099063,
|
| 309 |
+
"grad_norm": 0.51171875,
|
| 310 |
+
"learning_rate": 0.0009459174251418019,
|
| 311 |
+
"loss": 2.6549,
|
| 312 |
+
"step": 20500
|
| 313 |
+
},
|
| 314 |
+
{
|
| 315 |
+
"epoch": 2.770083102493075,
|
| 316 |
+
"grad_norm": 0.62109375,
|
| 317 |
+
"learning_rate": 0.0009445983379501385,
|
| 318 |
+
"loss": 2.6441,
|
| 319 |
+
"step": 21000
|
| 320 |
+
},
|
| 321 |
+
{
|
| 322 |
+
"epoch": 2.836037462076243,
|
| 323 |
+
"grad_norm": 0.57421875,
|
| 324 |
+
"learning_rate": 0.0009432792507584752,
|
| 325 |
+
"loss": 2.6374,
|
| 326 |
+
"step": 21500
|
| 327 |
+
},
|
| 328 |
+
{
|
| 329 |
+
"epoch": 2.901991821659412,
|
| 330 |
+
"grad_norm": 0.546875,
|
| 331 |
+
"learning_rate": 0.0009419601635668118,
|
| 332 |
+
"loss": 2.6291,
|
| 333 |
+
"step": 22000
|
| 334 |
+
},
|
| 335 |
+
{
|
| 336 |
+
"epoch": 2.96794618124258,
|
| 337 |
+
"grad_norm": 0.51953125,
|
| 338 |
+
"learning_rate": 0.0009406410763751484,
|
| 339 |
+
"loss": 2.6326,
|
| 340 |
+
"step": 22500
|
| 341 |
+
},
|
| 342 |
+
{
|
| 343 |
+
"epoch": 3.0,
|
| 344 |
+
"eval_loss": 2.696772575378418,
|
| 345 |
+
"eval_runtime": 5.0551,
|
| 346 |
+
"eval_samples_per_second": 395.639,
|
| 347 |
+
"eval_steps_per_second": 1.583,
|
| 348 |
+
"step": 22743
|
| 349 |
+
},
|
| 350 |
+
{
|
| 351 |
+
"epoch": 3.0339005408257487,
|
| 352 |
+
"grad_norm": 0.68359375,
|
| 353 |
+
"learning_rate": 0.000939321989183485,
|
| 354 |
+
"loss": 2.5858,
|
| 355 |
+
"step": 23000
|
| 356 |
+
},
|
| 357 |
+
{
|
| 358 |
+
"epoch": 3.099854900408917,
|
| 359 |
+
"grad_norm": 0.58984375,
|
| 360 |
+
"learning_rate": 0.0009380029019918217,
|
| 361 |
+
"loss": 2.5525,
|
| 362 |
+
"step": 23500
|
| 363 |
+
},
|
| 364 |
+
{
|
| 365 |
+
"epoch": 3.1658092599920855,
|
| 366 |
+
"grad_norm": 0.515625,
|
| 367 |
+
"learning_rate": 0.0009366838148001583,
|
| 368 |
+
"loss": 2.559,
|
| 369 |
+
"step": 24000
|
| 370 |
+
},
|
| 371 |
+
{
|
| 372 |
+
"epoch": 3.231763619575254,
|
| 373 |
+
"grad_norm": 0.53125,
|
| 374 |
+
"learning_rate": 0.0009353647276084949,
|
| 375 |
+
"loss": 2.5623,
|
| 376 |
+
"step": 24500
|
| 377 |
+
},
|
| 378 |
+
{
|
| 379 |
+
"epoch": 3.2977179791584224,
|
| 380 |
+
"grad_norm": 0.546875,
|
| 381 |
+
"learning_rate": 0.0009340456404168316,
|
| 382 |
+
"loss": 2.5558,
|
| 383 |
+
"step": 25000
|
| 384 |
+
},
|
| 385 |
+
{
|
| 386 |
+
"epoch": 3.363672338741591,
|
| 387 |
+
"grad_norm": 0.72265625,
|
| 388 |
+
"learning_rate": 0.0009327265532251682,
|
| 389 |
+
"loss": 2.5571,
|
| 390 |
+
"step": 25500
|
| 391 |
+
},
|
| 392 |
+
{
|
| 393 |
+
"epoch": 3.429626698324759,
|
| 394 |
+
"grad_norm": 0.55859375,
|
| 395 |
+
"learning_rate": 0.0009314074660335047,
|
| 396 |
+
"loss": 2.561,
|
| 397 |
+
"step": 26000
|
| 398 |
+
},
|
| 399 |
+
{
|
| 400 |
+
"epoch": 3.495581057907928,
|
| 401 |
+
"grad_norm": 0.55859375,
|
| 402 |
+
"learning_rate": 0.0009300883788418415,
|
| 403 |
+
"loss": 2.5557,
|
| 404 |
+
"step": 26500
|
| 405 |
+
},
|
| 406 |
+
{
|
| 407 |
+
"epoch": 3.561535417491096,
|
| 408 |
+
"grad_norm": 0.58203125,
|
| 409 |
+
"learning_rate": 0.000928769291650178,
|
| 410 |
+
"loss": 2.5473,
|
| 411 |
+
"step": 27000
|
| 412 |
+
},
|
| 413 |
+
{
|
| 414 |
+
"epoch": 3.6274897770742647,
|
| 415 |
+
"grad_norm": 0.578125,
|
| 416 |
+
"learning_rate": 0.0009274502044585147,
|
| 417 |
+
"loss": 2.5441,
|
| 418 |
+
"step": 27500
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 3.6934441366574333,
|
| 422 |
+
"grad_norm": 0.73046875,
|
| 423 |
+
"learning_rate": 0.0009261311172668514,
|
| 424 |
+
"loss": 2.5392,
|
| 425 |
+
"step": 28000
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 3.7593984962406015,
|
| 429 |
+
"grad_norm": 0.53515625,
|
| 430 |
+
"learning_rate": 0.000924812030075188,
|
| 431 |
+
"loss": 2.5421,
|
| 432 |
+
"step": 28500
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 3.8253528558237697,
|
| 436 |
+
"grad_norm": 0.59375,
|
| 437 |
+
"learning_rate": 0.0009234929428835246,
|
| 438 |
+
"loss": 2.5356,
|
| 439 |
+
"step": 29000
|
| 440 |
+
},
|
| 441 |
+
{
|
| 442 |
+
"epoch": 3.8913072154069384,
|
| 443 |
+
"grad_norm": 0.66015625,
|
| 444 |
+
"learning_rate": 0.0009221738556918613,
|
| 445 |
+
"loss": 2.5418,
|
| 446 |
+
"step": 29500
|
| 447 |
+
},
|
| 448 |
+
{
|
| 449 |
+
"epoch": 3.957261574990107,
|
| 450 |
+
"grad_norm": 0.55078125,
|
| 451 |
+
"learning_rate": 0.0009208547685001979,
|
| 452 |
+
"loss": 2.5229,
|
| 453 |
+
"step": 30000
|
| 454 |
+
},
|
| 455 |
+
{
|
| 456 |
+
"epoch": 4.0,
|
| 457 |
+
"eval_loss": 2.629683017730713,
|
| 458 |
+
"eval_runtime": 4.5326,
|
| 459 |
+
"eval_samples_per_second": 441.249,
|
| 460 |
+
"eval_steps_per_second": 1.765,
|
| 461 |
+
"step": 30324
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 4.023215934573275,
|
| 465 |
+
"grad_norm": 0.6953125,
|
| 466 |
+
"learning_rate": 0.0009195356813085345,
|
| 467 |
+
"loss": 2.4939,
|
| 468 |
+
"step": 30500
|
| 469 |
+
},
|
| 470 |
+
{
|
| 471 |
+
"epoch": 4.089170294156443,
|
| 472 |
+
"grad_norm": 0.546875,
|
| 473 |
+
"learning_rate": 0.0009182165941168712,
|
| 474 |
+
"loss": 2.4514,
|
| 475 |
+
"step": 31000
|
| 476 |
+
},
|
| 477 |
+
{
|
| 478 |
+
"epoch": 4.1551246537396125,
|
| 479 |
+
"grad_norm": 0.578125,
|
| 480 |
+
"learning_rate": 0.0009168975069252078,
|
| 481 |
+
"loss": 2.4666,
|
| 482 |
+
"step": 31500
|
| 483 |
+
},
|
| 484 |
+
{
|
| 485 |
+
"epoch": 4.221079013322781,
|
| 486 |
+
"grad_norm": 0.53125,
|
| 487 |
+
"learning_rate": 0.0009155784197335444,
|
| 488 |
+
"loss": 2.4649,
|
| 489 |
+
"step": 32000
|
| 490 |
+
},
|
| 491 |
+
{
|
| 492 |
+
"epoch": 4.287033372905949,
|
| 493 |
+
"grad_norm": 0.57421875,
|
| 494 |
+
"learning_rate": 0.000914259332541881,
|
| 495 |
+
"loss": 2.4642,
|
| 496 |
+
"step": 32500
|
| 497 |
+
},
|
| 498 |
+
{
|
| 499 |
+
"epoch": 4.352987732489117,
|
| 500 |
+
"grad_norm": 0.60546875,
|
| 501 |
+
"learning_rate": 0.0009129402453502177,
|
| 502 |
+
"loss": 2.4669,
|
| 503 |
+
"step": 33000
|
| 504 |
+
},
|
| 505 |
+
{
|
| 506 |
+
"epoch": 4.418942092072286,
|
| 507 |
+
"grad_norm": 0.5546875,
|
| 508 |
+
"learning_rate": 0.0009116211581585543,
|
| 509 |
+
"loss": 2.4686,
|
| 510 |
+
"step": 33500
|
| 511 |
+
},
|
| 512 |
+
{
|
| 513 |
+
"epoch": 4.484896451655454,
|
| 514 |
+
"grad_norm": 0.53125,
|
| 515 |
+
"learning_rate": 0.0009103020709668909,
|
| 516 |
+
"loss": 2.4809,
|
| 517 |
+
"step": 34000
|
| 518 |
+
},
|
| 519 |
+
{
|
| 520 |
+
"epoch": 4.550850811238623,
|
| 521 |
+
"grad_norm": 0.5546875,
|
| 522 |
+
"learning_rate": 0.0009089829837752276,
|
| 523 |
+
"loss": 2.4708,
|
| 524 |
+
"step": 34500
|
| 525 |
+
},
|
| 526 |
+
{
|
| 527 |
+
"epoch": 4.616805170821792,
|
| 528 |
+
"grad_norm": 0.5625,
|
| 529 |
+
"learning_rate": 0.0009076638965835642,
|
| 530 |
+
"loss": 2.4701,
|
| 531 |
+
"step": 35000
|
| 532 |
+
},
|
| 533 |
+
{
|
| 534 |
+
"epoch": 4.68275953040496,
|
| 535 |
+
"grad_norm": 0.56640625,
|
| 536 |
+
"learning_rate": 0.0009063448093919007,
|
| 537 |
+
"loss": 2.4714,
|
| 538 |
+
"step": 35500
|
| 539 |
+
},
|
| 540 |
+
{
|
| 541 |
+
"epoch": 4.748713889988128,
|
| 542 |
+
"grad_norm": 0.58203125,
|
| 543 |
+
"learning_rate": 0.0009050257222002375,
|
| 544 |
+
"loss": 2.4626,
|
| 545 |
+
"step": 36000
|
| 546 |
+
},
|
| 547 |
+
{
|
| 548 |
+
"epoch": 4.814668249571296,
|
| 549 |
+
"grad_norm": 0.57421875,
|
| 550 |
+
"learning_rate": 0.000903706635008574,
|
| 551 |
+
"loss": 2.4696,
|
| 552 |
+
"step": 36500
|
| 553 |
+
},
|
| 554 |
+
{
|
| 555 |
+
"epoch": 4.880622609154465,
|
| 556 |
+
"grad_norm": 0.55078125,
|
| 557 |
+
"learning_rate": 0.0009023875478169107,
|
| 558 |
+
"loss": 2.4614,
|
| 559 |
+
"step": 37000
|
| 560 |
+
},
|
| 561 |
+
{
|
| 562 |
+
"epoch": 4.9465769687376335,
|
| 563 |
+
"grad_norm": 0.5625,
|
| 564 |
+
"learning_rate": 0.0009010684606252473,
|
| 565 |
+
"loss": 2.4544,
|
| 566 |
+
"step": 37500
|
| 567 |
+
},
|
| 568 |
+
{
|
| 569 |
+
"epoch": 5.0,
|
| 570 |
+
"eval_loss": 2.593965768814087,
|
| 571 |
+
"eval_runtime": 4.0976,
|
| 572 |
+
"eval_samples_per_second": 488.092,
|
| 573 |
+
"eval_steps_per_second": 1.952,
|
| 574 |
+
"step": 37905
|
| 575 |
+
},
|
| 576 |
+
{
|
| 577 |
+
"epoch": 5.012531328320802,
|
| 578 |
+
"grad_norm": 0.55078125,
|
| 579 |
+
"learning_rate": 0.000899749373433584,
|
| 580 |
+
"loss": 2.4455,
|
| 581 |
+
"step": 38000
|
| 582 |
+
},
|
| 583 |
+
{
|
| 584 |
+
"epoch": 5.078485687903971,
|
| 585 |
+
"grad_norm": 0.60546875,
|
| 586 |
+
"learning_rate": 0.0008984302862419205,
|
| 587 |
+
"loss": 2.3874,
|
| 588 |
+
"step": 38500
|
| 589 |
+
},
|
| 590 |
+
{
|
| 591 |
+
"epoch": 5.144440047487139,
|
| 592 |
+
"grad_norm": 0.609375,
|
| 593 |
+
"learning_rate": 0.0008971111990502573,
|
| 594 |
+
"loss": 2.3907,
|
| 595 |
+
"step": 39000
|
| 596 |
+
},
|
| 597 |
+
{
|
| 598 |
+
"epoch": 5.210394407070307,
|
| 599 |
+
"grad_norm": 0.5703125,
|
| 600 |
+
"learning_rate": 0.0008957921118585939,
|
| 601 |
+
"loss": 2.3965,
|
| 602 |
+
"step": 39500
|
| 603 |
+
},
|
| 604 |
+
{
|
| 605 |
+
"epoch": 5.276348766653475,
|
| 606 |
+
"grad_norm": 0.5234375,
|
| 607 |
+
"learning_rate": 0.0008944730246669305,
|
| 608 |
+
"loss": 2.4055,
|
| 609 |
+
"step": 40000
|
| 610 |
+
},
|
| 611 |
+
{
|
| 612 |
+
"epoch": 5.3423031262366445,
|
| 613 |
+
"grad_norm": 0.6171875,
|
| 614 |
+
"learning_rate": 0.0008931539374752672,
|
| 615 |
+
"loss": 2.4024,
|
| 616 |
+
"step": 40500
|
| 617 |
+
},
|
| 618 |
+
{
|
| 619 |
+
"epoch": 5.408257485819813,
|
| 620 |
+
"grad_norm": 0.58203125,
|
| 621 |
+
"learning_rate": 0.0008918348502836038,
|
| 622 |
+
"loss": 2.402,
|
| 623 |
+
"step": 41000
|
| 624 |
+
},
|
| 625 |
+
{
|
| 626 |
+
"epoch": 5.474211845402981,
|
| 627 |
+
"grad_norm": 0.5390625,
|
| 628 |
+
"learning_rate": 0.0008905157630919404,
|
| 629 |
+
"loss": 2.4073,
|
| 630 |
+
"step": 41500
|
| 631 |
+
},
|
| 632 |
+
{
|
| 633 |
+
"epoch": 5.54016620498615,
|
| 634 |
+
"grad_norm": 0.58203125,
|
| 635 |
+
"learning_rate": 0.000889196675900277,
|
| 636 |
+
"loss": 2.4046,
|
| 637 |
+
"step": 42000
|
| 638 |
+
},
|
| 639 |
+
{
|
| 640 |
+
"epoch": 5.606120564569318,
|
| 641 |
+
"grad_norm": 0.57421875,
|
| 642 |
+
"learning_rate": 0.0008878775887086137,
|
| 643 |
+
"loss": 2.4127,
|
| 644 |
+
"step": 42500
|
| 645 |
+
},
|
| 646 |
+
{
|
| 647 |
+
"epoch": 5.672074924152486,
|
| 648 |
+
"grad_norm": 0.57421875,
|
| 649 |
+
"learning_rate": 0.0008865585015169503,
|
| 650 |
+
"loss": 2.4102,
|
| 651 |
+
"step": 43000
|
| 652 |
+
},
|
| 653 |
+
{
|
| 654 |
+
"epoch": 5.738029283735655,
|
| 655 |
+
"grad_norm": 0.546875,
|
| 656 |
+
"learning_rate": 0.0008852394143252869,
|
| 657 |
+
"loss": 2.4021,
|
| 658 |
+
"step": 43500
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"epoch": 5.803983643318824,
|
| 662 |
+
"grad_norm": 0.7109375,
|
| 663 |
+
"learning_rate": 0.0008839203271336235,
|
| 664 |
+
"loss": 2.4041,
|
| 665 |
+
"step": 44000
|
| 666 |
+
},
|
| 667 |
+
{
|
| 668 |
+
"epoch": 5.869938002901992,
|
| 669 |
+
"grad_norm": 0.59375,
|
| 670 |
+
"learning_rate": 0.0008826012399419602,
|
| 671 |
+
"loss": 2.4034,
|
| 672 |
+
"step": 44500
|
| 673 |
+
},
|
| 674 |
+
{
|
| 675 |
+
"epoch": 5.93589236248516,
|
| 676 |
+
"grad_norm": 0.63671875,
|
| 677 |
+
"learning_rate": 0.0008812821527502967,
|
| 678 |
+
"loss": 2.4059,
|
| 679 |
+
"step": 45000
|
| 680 |
+
},
|
| 681 |
+
{
|
| 682 |
+
"epoch": 6.0,
|
| 683 |
+
"eval_loss": 2.5455453395843506,
|
| 684 |
+
"eval_runtime": 4.3171,
|
| 685 |
+
"eval_samples_per_second": 463.276,
|
| 686 |
+
"eval_steps_per_second": 1.853,
|
| 687 |
+
"step": 45486
|
| 688 |
+
},
|
| 689 |
+
{
|
| 690 |
+
"epoch": 6.001846722068329,
|
| 691 |
+
"grad_norm": 0.5234375,
|
| 692 |
+
"learning_rate": 0.0008799630655586335,
|
| 693 |
+
"loss": 2.3998,
|
| 694 |
+
"step": 45500
|
| 695 |
+
},
|
| 696 |
+
{
|
| 697 |
+
"epoch": 6.067801081651497,
|
| 698 |
+
"grad_norm": 0.546875,
|
| 699 |
+
"learning_rate": 0.00087864397836697,
|
| 700 |
+
"loss": 2.3343,
|
| 701 |
+
"step": 46000
|
| 702 |
+
},
|
| 703 |
+
{
|
| 704 |
+
"epoch": 6.1337554412346655,
|
| 705 |
+
"grad_norm": 0.55859375,
|
| 706 |
+
"learning_rate": 0.0008773248911753067,
|
| 707 |
+
"loss": 2.3341,
|
| 708 |
+
"step": 46500
|
| 709 |
+
},
|
| 710 |
+
{
|
| 711 |
+
"epoch": 6.199709800817834,
|
| 712 |
+
"grad_norm": 0.5546875,
|
| 713 |
+
"learning_rate": 0.0008760058039836433,
|
| 714 |
+
"loss": 2.3527,
|
| 715 |
+
"step": 47000
|
| 716 |
+
},
|
| 717 |
+
{
|
| 718 |
+
"epoch": 6.265664160401003,
|
| 719 |
+
"grad_norm": 0.58203125,
|
| 720 |
+
"learning_rate": 0.00087468671679198,
|
| 721 |
+
"loss": 2.3633,
|
| 722 |
+
"step": 47500
|
| 723 |
+
},
|
| 724 |
+
{
|
| 725 |
+
"epoch": 6.331618519984171,
|
| 726 |
+
"grad_norm": 0.55859375,
|
| 727 |
+
"learning_rate": 0.0008733676296003165,
|
| 728 |
+
"loss": 2.3615,
|
| 729 |
+
"step": 48000
|
| 730 |
+
},
|
| 731 |
+
{
|
| 732 |
+
"epoch": 6.397572879567339,
|
| 733 |
+
"grad_norm": 0.546875,
|
| 734 |
+
"learning_rate": 0.0008720485424086533,
|
| 735 |
+
"loss": 2.3654,
|
| 736 |
+
"step": 48500
|
| 737 |
+
},
|
| 738 |
+
{
|
| 739 |
+
"epoch": 6.463527239150508,
|
| 740 |
+
"grad_norm": 0.6015625,
|
| 741 |
+
"learning_rate": 0.0008707294552169899,
|
| 742 |
+
"loss": 2.3695,
|
| 743 |
+
"step": 49000
|
| 744 |
+
},
|
| 745 |
+
{
|
| 746 |
+
"epoch": 6.5294815987336765,
|
| 747 |
+
"grad_norm": 0.5546875,
|
| 748 |
+
"learning_rate": 0.0008694103680253265,
|
| 749 |
+
"loss": 2.3698,
|
| 750 |
+
"step": 49500
|
| 751 |
+
},
|
| 752 |
+
{
|
| 753 |
+
"epoch": 6.595435958316845,
|
| 754 |
+
"grad_norm": 0.65234375,
|
| 755 |
+
"learning_rate": 0.0008680912808336632,
|
| 756 |
+
"loss": 2.3671,
|
| 757 |
+
"step": 50000
|
| 758 |
+
},
|
| 759 |
+
{
|
| 760 |
+
"epoch": 6.661390317900013,
|
| 761 |
+
"grad_norm": 0.62109375,
|
| 762 |
+
"learning_rate": 0.0008667721936419998,
|
| 763 |
+
"loss": 2.3611,
|
| 764 |
+
"step": 50500
|
| 765 |
+
},
|
| 766 |
+
{
|
| 767 |
+
"epoch": 6.727344677483182,
|
| 768 |
+
"grad_norm": 0.55078125,
|
| 769 |
+
"learning_rate": 0.0008654531064503364,
|
| 770 |
+
"loss": 2.371,
|
| 771 |
+
"step": 51000
|
| 772 |
+
},
|
| 773 |
+
{
|
| 774 |
+
"epoch": 6.79329903706635,
|
| 775 |
+
"grad_norm": 0.6484375,
|
| 776 |
+
"learning_rate": 0.000864134019258673,
|
| 777 |
+
"loss": 2.3693,
|
| 778 |
+
"step": 51500
|
| 779 |
+
},
|
| 780 |
+
{
|
| 781 |
+
"epoch": 6.859253396649518,
|
| 782 |
+
"grad_norm": 0.59375,
|
| 783 |
+
"learning_rate": 0.0008628149320670097,
|
| 784 |
+
"loss": 2.3575,
|
| 785 |
+
"step": 52000
|
| 786 |
+
},
|
| 787 |
+
{
|
| 788 |
+
"epoch": 6.9252077562326875,
|
| 789 |
+
"grad_norm": 0.5546875,
|
| 790 |
+
"learning_rate": 0.0008614958448753462,
|
| 791 |
+
"loss": 2.3638,
|
| 792 |
+
"step": 52500
|
| 793 |
+
},
|
| 794 |
+
{
|
| 795 |
+
"epoch": 6.991162115815856,
|
| 796 |
+
"grad_norm": 0.55859375,
|
| 797 |
+
"learning_rate": 0.0008601767576836829,
|
| 798 |
+
"loss": 2.3601,
|
| 799 |
+
"step": 53000
|
| 800 |
+
},
|
| 801 |
+
{
|
| 802 |
+
"epoch": 7.0,
|
| 803 |
+
"eval_loss": 2.5313472747802734,
|
| 804 |
+
"eval_runtime": 4.3987,
|
| 805 |
+
"eval_samples_per_second": 454.683,
|
| 806 |
+
"eval_steps_per_second": 1.819,
|
| 807 |
+
"step": 53067
|
| 808 |
+
},
|
| 809 |
+
{
|
| 810 |
+
"epoch": 7.057116475399024,
|
| 811 |
+
"grad_norm": 0.62890625,
|
| 812 |
+
"learning_rate": 0.0008588576704920195,
|
| 813 |
+
"loss": 2.3004,
|
| 814 |
+
"step": 53500
|
| 815 |
+
},
|
| 816 |
+
{
|
| 817 |
+
"epoch": 7.123070834982192,
|
| 818 |
+
"grad_norm": 0.5859375,
|
| 819 |
+
"learning_rate": 0.0008575385833003562,
|
| 820 |
+
"loss": 2.2926,
|
| 821 |
+
"step": 54000
|
| 822 |
+
},
|
| 823 |
+
{
|
| 824 |
+
"epoch": 7.189025194565361,
|
| 825 |
+
"grad_norm": 0.734375,
|
| 826 |
+
"learning_rate": 0.0008562194961086927,
|
| 827 |
+
"loss": 2.2966,
|
| 828 |
+
"step": 54500
|
| 829 |
+
},
|
| 830 |
+
{
|
| 831 |
+
"epoch": 7.254979554148529,
|
| 832 |
+
"grad_norm": 0.6171875,
|
| 833 |
+
"learning_rate": 0.0008549004089170295,
|
| 834 |
+
"loss": 2.3004,
|
| 835 |
+
"step": 55000
|
| 836 |
+
},
|
| 837 |
+
{
|
| 838 |
+
"epoch": 7.3209339137316976,
|
| 839 |
+
"grad_norm": 0.5859375,
|
| 840 |
+
"learning_rate": 0.000853581321725366,
|
| 841 |
+
"loss": 2.3168,
|
| 842 |
+
"step": 55500
|
| 843 |
+
},
|
| 844 |
+
{
|
| 845 |
+
"epoch": 7.386888273314866,
|
| 846 |
+
"grad_norm": 0.58203125,
|
| 847 |
+
"learning_rate": 0.0008522622345337027,
|
| 848 |
+
"loss": 2.3264,
|
| 849 |
+
"step": 56000
|
| 850 |
+
},
|
| 851 |
+
{
|
| 852 |
+
"epoch": 7.452842632898035,
|
| 853 |
+
"grad_norm": 0.60546875,
|
| 854 |
+
"learning_rate": 0.0008509431473420393,
|
| 855 |
+
"loss": 2.3132,
|
| 856 |
+
"step": 56500
|
| 857 |
+
},
|
| 858 |
+
{
|
| 859 |
+
"epoch": 7.518796992481203,
|
| 860 |
+
"grad_norm": 0.625,
|
| 861 |
+
"learning_rate": 0.000849624060150376,
|
| 862 |
+
"loss": 2.3263,
|
| 863 |
+
"step": 57000
|
| 864 |
+
},
|
| 865 |
+
{
|
| 866 |
+
"epoch": 7.584751352064371,
|
| 867 |
+
"grad_norm": 0.60546875,
|
| 868 |
+
"learning_rate": 0.0008483049729587125,
|
| 869 |
+
"loss": 2.3318,
|
| 870 |
+
"step": 57500
|
| 871 |
+
},
|
| 872 |
+
{
|
| 873 |
+
"epoch": 7.650705711647539,
|
| 874 |
+
"grad_norm": 0.5703125,
|
| 875 |
+
"learning_rate": 0.0008469858857670493,
|
| 876 |
+
"loss": 2.3313,
|
| 877 |
+
"step": 58000
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 7.7166600712307085,
|
| 881 |
+
"grad_norm": 0.5859375,
|
| 882 |
+
"learning_rate": 0.0008456667985753858,
|
| 883 |
+
"loss": 2.3265,
|
| 884 |
+
"step": 58500
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 7.782614430813877,
|
| 888 |
+
"grad_norm": 0.5859375,
|
| 889 |
+
"learning_rate": 0.0008443477113837225,
|
| 890 |
+
"loss": 2.3401,
|
| 891 |
+
"step": 59000
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 7.848568790397045,
|
| 895 |
+
"grad_norm": 0.63671875,
|
| 896 |
+
"learning_rate": 0.0008430286241920592,
|
| 897 |
+
"loss": 2.3385,
|
| 898 |
+
"step": 59500
|
| 899 |
+
},
|
| 900 |
+
{
|
| 901 |
+
"epoch": 7.914523149980214,
|
| 902 |
+
"grad_norm": 0.578125,
|
| 903 |
+
"learning_rate": 0.0008417095370003958,
|
| 904 |
+
"loss": 2.3224,
|
| 905 |
+
"step": 60000
|
| 906 |
+
},
|
| 907 |
+
{
|
| 908 |
+
"epoch": 7.980477509563382,
|
| 909 |
+
"grad_norm": 0.6328125,
|
| 910 |
+
"learning_rate": 0.0008403904498087324,
|
| 911 |
+
"loss": 2.3394,
|
| 912 |
+
"step": 60500
|
| 913 |
+
},
|
| 914 |
+
{
|
| 915 |
+
"epoch": 8.0,
|
| 916 |
+
"eval_loss": 2.508885383605957,
|
| 917 |
+
"eval_runtime": 4.5718,
|
| 918 |
+
"eval_samples_per_second": 437.467,
|
| 919 |
+
"eval_steps_per_second": 1.75,
|
| 920 |
+
"step": 60648
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 8.04643186914655,
|
| 924 |
+
"grad_norm": 0.6171875,
|
| 925 |
+
"learning_rate": 0.0008390713626170689,
|
| 926 |
+
"loss": 2.2768,
|
| 927 |
+
"step": 61000
|
| 928 |
+
},
|
| 929 |
+
{
|
| 930 |
+
"epoch": 8.112386228729719,
|
| 931 |
+
"grad_norm": 0.6015625,
|
| 932 |
+
"learning_rate": 0.0008377522754254057,
|
| 933 |
+
"loss": 2.2642,
|
| 934 |
+
"step": 61500
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 8.178340588312887,
|
| 938 |
+
"grad_norm": 0.5859375,
|
| 939 |
+
"learning_rate": 0.0008364331882337422,
|
| 940 |
+
"loss": 2.2747,
|
| 941 |
+
"step": 62000
|
| 942 |
+
},
|
| 943 |
+
{
|
| 944 |
+
"epoch": 8.244294947896057,
|
| 945 |
+
"grad_norm": 0.64453125,
|
| 946 |
+
"learning_rate": 0.0008351141010420789,
|
| 947 |
+
"loss": 2.2778,
|
| 948 |
+
"step": 62500
|
| 949 |
+
},
|
| 950 |
+
{
|
| 951 |
+
"epoch": 8.310249307479225,
|
| 952 |
+
"grad_norm": 0.65234375,
|
| 953 |
+
"learning_rate": 0.0008337950138504155,
|
| 954 |
+
"loss": 2.2876,
|
| 955 |
+
"step": 63000
|
| 956 |
+
},
|
| 957 |
+
{
|
| 958 |
+
"epoch": 8.376203667062393,
|
| 959 |
+
"grad_norm": 0.6796875,
|
| 960 |
+
"learning_rate": 0.0008324759266587522,
|
| 961 |
+
"loss": 2.2983,
|
| 962 |
+
"step": 63500
|
| 963 |
+
},
|
| 964 |
+
{
|
| 965 |
+
"epoch": 8.442158026645561,
|
| 966 |
+
"grad_norm": 0.6171875,
|
| 967 |
+
"learning_rate": 0.0008311568394670887,
|
| 968 |
+
"loss": 2.3014,
|
| 969 |
+
"step": 64000
|
| 970 |
+
},
|
| 971 |
+
{
|
| 972 |
+
"epoch": 8.50811238622873,
|
| 973 |
+
"grad_norm": 0.67578125,
|
| 974 |
+
"learning_rate": 0.0008298377522754255,
|
| 975 |
+
"loss": 2.2997,
|
| 976 |
+
"step": 64500
|
| 977 |
+
},
|
| 978 |
+
{
|
| 979 |
+
"epoch": 8.574066745811898,
|
| 980 |
+
"grad_norm": 0.81640625,
|
| 981 |
+
"learning_rate": 0.000828518665083762,
|
| 982 |
+
"loss": 2.2962,
|
| 983 |
+
"step": 65000
|
| 984 |
+
},
|
| 985 |
+
{
|
| 986 |
+
"epoch": 8.640021105395066,
|
| 987 |
+
"grad_norm": 0.6171875,
|
| 988 |
+
"learning_rate": 0.0008271995778920987,
|
| 989 |
+
"loss": 2.2889,
|
| 990 |
+
"step": 65500
|
| 991 |
+
},
|
| 992 |
+
{
|
| 993 |
+
"epoch": 8.705975464978234,
|
| 994 |
+
"grad_norm": 0.59765625,
|
| 995 |
+
"learning_rate": 0.0008258804907004353,
|
| 996 |
+
"loss": 2.2984,
|
| 997 |
+
"step": 66000
|
| 998 |
+
},
|
| 999 |
+
{
|
| 1000 |
+
"epoch": 8.771929824561404,
|
| 1001 |
+
"grad_norm": 0.59375,
|
| 1002 |
+
"learning_rate": 0.000824561403508772,
|
| 1003 |
+
"loss": 2.3166,
|
| 1004 |
+
"step": 66500
|
| 1005 |
+
},
|
| 1006 |
+
{
|
| 1007 |
+
"epoch": 8.837884184144572,
|
| 1008 |
+
"grad_norm": 0.73828125,
|
| 1009 |
+
"learning_rate": 0.0008232423163171085,
|
| 1010 |
+
"loss": 2.3091,
|
| 1011 |
+
"step": 67000
|
| 1012 |
+
},
|
| 1013 |
+
{
|
| 1014 |
+
"epoch": 8.90383854372774,
|
| 1015 |
+
"grad_norm": 0.63671875,
|
| 1016 |
+
"learning_rate": 0.0008219232291254453,
|
| 1017 |
+
"loss": 2.2962,
|
| 1018 |
+
"step": 67500
|
| 1019 |
+
},
|
| 1020 |
+
{
|
| 1021 |
+
"epoch": 8.969792903310909,
|
| 1022 |
+
"grad_norm": 0.64453125,
|
| 1023 |
+
"learning_rate": 0.0008206041419337818,
|
| 1024 |
+
"loss": 2.306,
|
| 1025 |
+
"step": 68000
|
| 1026 |
+
},
|
| 1027 |
+
{
|
| 1028 |
+
"epoch": 9.0,
|
| 1029 |
+
"eval_loss": 2.5000669956207275,
|
| 1030 |
+
"eval_runtime": 4.9297,
|
| 1031 |
+
"eval_samples_per_second": 405.707,
|
| 1032 |
+
"eval_steps_per_second": 1.623,
|
| 1033 |
+
"step": 68229
|
| 1034 |
+
},
|
| 1035 |
+
{
|
| 1036 |
+
"epoch": 9.035747262894077,
|
| 1037 |
+
"grad_norm": 0.62890625,
|
| 1038 |
+
"learning_rate": 0.0008192850547421185,
|
| 1039 |
+
"loss": 2.2644,
|
| 1040 |
+
"step": 68500
|
| 1041 |
+
},
|
| 1042 |
+
{
|
| 1043 |
+
"epoch": 9.101701622477245,
|
| 1044 |
+
"grad_norm": 0.6015625,
|
| 1045 |
+
"learning_rate": 0.000817965967550455,
|
| 1046 |
+
"loss": 2.2261,
|
| 1047 |
+
"step": 69000
|
| 1048 |
+
},
|
| 1049 |
+
{
|
| 1050 |
+
"epoch": 9.167655982060413,
|
| 1051 |
+
"grad_norm": 0.6328125,
|
| 1052 |
+
"learning_rate": 0.0008166468803587918,
|
| 1053 |
+
"loss": 2.2395,
|
| 1054 |
+
"step": 69500
|
| 1055 |
+
},
|
| 1056 |
+
{
|
| 1057 |
+
"epoch": 9.233610341643583,
|
| 1058 |
+
"grad_norm": 0.625,
|
| 1059 |
+
"learning_rate": 0.0008153277931671284,
|
| 1060 |
+
"loss": 2.2461,
|
| 1061 |
+
"step": 70000
|
| 1062 |
+
},
|
| 1063 |
+
{
|
| 1064 |
+
"epoch": 9.299564701226751,
|
| 1065 |
+
"grad_norm": 0.66796875,
|
| 1066 |
+
"learning_rate": 0.0008140087059754649,
|
| 1067 |
+
"loss": 2.256,
|
| 1068 |
+
"step": 70500
|
| 1069 |
+
},
|
| 1070 |
+
{
|
| 1071 |
+
"epoch": 9.36551906080992,
|
| 1072 |
+
"grad_norm": 0.6015625,
|
| 1073 |
+
"learning_rate": 0.0008126896187838017,
|
| 1074 |
+
"loss": 2.2607,
|
| 1075 |
+
"step": 71000
|
| 1076 |
+
},
|
| 1077 |
+
{
|
| 1078 |
+
"epoch": 9.431473420393088,
|
| 1079 |
+
"grad_norm": 0.62109375,
|
| 1080 |
+
"learning_rate": 0.0008113705315921382,
|
| 1081 |
+
"loss": 2.2576,
|
| 1082 |
+
"step": 71500
|
| 1083 |
+
},
|
| 1084 |
+
{
|
| 1085 |
+
"epoch": 9.497427779976256,
|
| 1086 |
+
"grad_norm": 0.57421875,
|
| 1087 |
+
"learning_rate": 0.0008100514444004749,
|
| 1088 |
+
"loss": 2.2616,
|
| 1089 |
+
"step": 72000
|
| 1090 |
+
},
|
| 1091 |
+
{
|
| 1092 |
+
"epoch": 9.563382139559424,
|
| 1093 |
+
"grad_norm": 0.7578125,
|
| 1094 |
+
"learning_rate": 0.0008087323572088115,
|
| 1095 |
+
"loss": 2.2711,
|
| 1096 |
+
"step": 72500
|
| 1097 |
+
},
|
| 1098 |
+
{
|
| 1099 |
+
"epoch": 9.629336499142592,
|
| 1100 |
+
"grad_norm": 0.6328125,
|
| 1101 |
+
"learning_rate": 0.0008074132700171482,
|
| 1102 |
+
"loss": 2.2697,
|
| 1103 |
+
"step": 73000
|
| 1104 |
+
},
|
| 1105 |
+
{
|
| 1106 |
+
"epoch": 9.695290858725762,
|
| 1107 |
+
"grad_norm": 0.609375,
|
| 1108 |
+
"learning_rate": 0.0008060941828254847,
|
| 1109 |
+
"loss": 2.2548,
|
| 1110 |
+
"step": 73500
|
| 1111 |
+
},
|
| 1112 |
+
{
|
| 1113 |
+
"epoch": 9.76124521830893,
|
| 1114 |
+
"grad_norm": 0.75390625,
|
| 1115 |
+
"learning_rate": 0.0008047750956338215,
|
| 1116 |
+
"loss": 2.2583,
|
| 1117 |
+
"step": 74000
|
| 1118 |
+
},
|
| 1119 |
+
{
|
| 1120 |
+
"epoch": 9.827199577892099,
|
| 1121 |
+
"grad_norm": 0.62109375,
|
| 1122 |
+
"learning_rate": 0.000803456008442158,
|
| 1123 |
+
"loss": 2.2598,
|
| 1124 |
+
"step": 74500
|
| 1125 |
+
},
|
| 1126 |
+
{
|
| 1127 |
+
"epoch": 9.893153937475267,
|
| 1128 |
+
"grad_norm": 0.60546875,
|
| 1129 |
+
"learning_rate": 0.0008021369212504947,
|
| 1130 |
+
"loss": 2.2696,
|
| 1131 |
+
"step": 75000
|
| 1132 |
+
},
|
| 1133 |
+
{
|
| 1134 |
+
"epoch": 9.959108297058435,
|
| 1135 |
+
"grad_norm": 0.59765625,
|
| 1136 |
+
"learning_rate": 0.0008008178340588313,
|
| 1137 |
+
"loss": 2.2692,
|
| 1138 |
+
"step": 75500
|
| 1139 |
+
},
|
| 1140 |
+
{
|
| 1141 |
+
"epoch": 10.0,
|
| 1142 |
+
"eval_loss": 2.4777557849884033,
|
| 1143 |
+
"eval_runtime": 3.8042,
|
| 1144 |
+
"eval_samples_per_second": 525.734,
|
| 1145 |
+
"eval_steps_per_second": 2.103,
|
| 1146 |
+
"step": 75810
|
| 1147 |
+
},
|
| 1148 |
+
{
|
| 1149 |
+
"epoch": 10.025062656641603,
|
| 1150 |
+
"grad_norm": 0.625,
|
| 1151 |
+
"learning_rate": 0.000799498746867168,
|
| 1152 |
+
"loss": 2.2365,
|
| 1153 |
+
"step": 76000
|
| 1154 |
+
},
|
| 1155 |
+
{
|
| 1156 |
+
"epoch": 10.091017016224772,
|
| 1157 |
+
"grad_norm": 0.5859375,
|
| 1158 |
+
"learning_rate": 0.0007981796596755045,
|
| 1159 |
+
"loss": 2.1978,
|
| 1160 |
+
"step": 76500
|
| 1161 |
+
},
|
| 1162 |
+
{
|
| 1163 |
+
"epoch": 10.156971375807942,
|
| 1164 |
+
"grad_norm": 0.63671875,
|
| 1165 |
+
"learning_rate": 0.0007968605724838413,
|
| 1166 |
+
"loss": 2.2063,
|
| 1167 |
+
"step": 77000
|
| 1168 |
+
},
|
| 1169 |
+
{
|
| 1170 |
+
"epoch": 10.22292573539111,
|
| 1171 |
+
"grad_norm": 0.61328125,
|
| 1172 |
+
"learning_rate": 0.0007955414852921778,
|
| 1173 |
+
"loss": 2.215,
|
| 1174 |
+
"step": 77500
|
| 1175 |
+
},
|
| 1176 |
+
{
|
| 1177 |
+
"epoch": 10.288880094974278,
|
| 1178 |
+
"grad_norm": 0.70703125,
|
| 1179 |
+
"learning_rate": 0.0007942223981005145,
|
| 1180 |
+
"loss": 2.222,
|
| 1181 |
+
"step": 78000
|
| 1182 |
+
},
|
| 1183 |
+
{
|
| 1184 |
+
"epoch": 10.354834454557446,
|
| 1185 |
+
"grad_norm": 0.75,
|
| 1186 |
+
"learning_rate": 0.000792903310908851,
|
| 1187 |
+
"loss": 2.2226,
|
| 1188 |
+
"step": 78500
|
| 1189 |
+
},
|
| 1190 |
+
{
|
| 1191 |
+
"epoch": 10.420788814140614,
|
| 1192 |
+
"grad_norm": 0.6015625,
|
| 1193 |
+
"learning_rate": 0.0007915842237171877,
|
| 1194 |
+
"loss": 2.2292,
|
| 1195 |
+
"step": 79000
|
| 1196 |
+
},
|
| 1197 |
+
{
|
| 1198 |
+
"epoch": 10.486743173723783,
|
| 1199 |
+
"grad_norm": 0.609375,
|
| 1200 |
+
"learning_rate": 0.0007902651365255243,
|
| 1201 |
+
"loss": 2.2305,
|
| 1202 |
+
"step": 79500
|
| 1203 |
+
},
|
| 1204 |
+
{
|
| 1205 |
+
"epoch": 10.55269753330695,
|
| 1206 |
+
"grad_norm": 0.625,
|
| 1207 |
+
"learning_rate": 0.0007889460493338609,
|
| 1208 |
+
"loss": 2.2233,
|
| 1209 |
+
"step": 80000
|
| 1210 |
+
},
|
| 1211 |
+
{
|
| 1212 |
+
"epoch": 10.61865189289012,
|
| 1213 |
+
"grad_norm": 0.65625,
|
| 1214 |
+
"learning_rate": 0.0007876269621421977,
|
| 1215 |
+
"loss": 2.2378,
|
| 1216 |
+
"step": 80500
|
| 1217 |
+
},
|
| 1218 |
+
{
|
| 1219 |
+
"epoch": 10.684606252473289,
|
| 1220 |
+
"grad_norm": 0.66015625,
|
| 1221 |
+
"learning_rate": 0.0007863078749505342,
|
| 1222 |
+
"loss": 2.2273,
|
| 1223 |
+
"step": 81000
|
| 1224 |
+
},
|
| 1225 |
+
{
|
| 1226 |
+
"epoch": 10.750560612056457,
|
| 1227 |
+
"grad_norm": 0.69140625,
|
| 1228 |
+
"learning_rate": 0.0007849887877588709,
|
| 1229 |
+
"loss": 2.2363,
|
| 1230 |
+
"step": 81500
|
| 1231 |
+
},
|
| 1232 |
+
{
|
| 1233 |
+
"epoch": 10.816514971639625,
|
| 1234 |
+
"grad_norm": 0.63671875,
|
| 1235 |
+
"learning_rate": 0.0007836697005672075,
|
| 1236 |
+
"loss": 2.2386,
|
| 1237 |
+
"step": 82000
|
| 1238 |
+
},
|
| 1239 |
+
{
|
| 1240 |
+
"epoch": 10.882469331222794,
|
| 1241 |
+
"grad_norm": 0.9453125,
|
| 1242 |
+
"learning_rate": 0.0007823506133755442,
|
| 1243 |
+
"loss": 2.2348,
|
| 1244 |
+
"step": 82500
|
| 1245 |
+
},
|
| 1246 |
+
{
|
| 1247 |
+
"epoch": 10.948423690805962,
|
| 1248 |
+
"grad_norm": 0.64453125,
|
| 1249 |
+
"learning_rate": 0.0007810315261838807,
|
| 1250 |
+
"loss": 2.2359,
|
| 1251 |
+
"step": 83000
|
| 1252 |
+
},
|
| 1253 |
+
{
|
| 1254 |
+
"epoch": 11.0,
|
| 1255 |
+
"eval_loss": 2.4718410968780518,
|
| 1256 |
+
"eval_runtime": 4.0713,
|
| 1257 |
+
"eval_samples_per_second": 491.242,
|
| 1258 |
+
"eval_steps_per_second": 1.965,
|
| 1259 |
+
"step": 83391
|
| 1260 |
+
},
|
| 1261 |
+
{
|
| 1262 |
+
"epoch": 11.01437805038913,
|
| 1263 |
+
"grad_norm": 0.6796875,
|
| 1264 |
+
"learning_rate": 0.0007797124389922175,
|
| 1265 |
+
"loss": 2.2184,
|
| 1266 |
+
"step": 83500
|
| 1267 |
+
},
|
| 1268 |
+
{
|
| 1269 |
+
"epoch": 11.0803324099723,
|
| 1270 |
+
"grad_norm": 0.6015625,
|
| 1271 |
+
"learning_rate": 0.000778393351800554,
|
| 1272 |
+
"loss": 2.1719,
|
| 1273 |
+
"step": 84000
|
| 1274 |
+
},
|
| 1275 |
+
{
|
| 1276 |
+
"epoch": 11.146286769555468,
|
| 1277 |
+
"grad_norm": 0.67578125,
|
| 1278 |
+
"learning_rate": 0.0007770742646088907,
|
| 1279 |
+
"loss": 2.1817,
|
| 1280 |
+
"step": 84500
|
| 1281 |
+
},
|
| 1282 |
+
{
|
| 1283 |
+
"epoch": 11.212241129138636,
|
| 1284 |
+
"grad_norm": 0.8125,
|
| 1285 |
+
"learning_rate": 0.0007757551774172273,
|
| 1286 |
+
"loss": 2.185,
|
| 1287 |
+
"step": 85000
|
| 1288 |
+
},
|
| 1289 |
+
{
|
| 1290 |
+
"epoch": 11.278195488721805,
|
| 1291 |
+
"grad_norm": 0.62890625,
|
| 1292 |
+
"learning_rate": 0.000774436090225564,
|
| 1293 |
+
"loss": 2.1905,
|
| 1294 |
+
"step": 85500
|
| 1295 |
+
},
|
| 1296 |
+
{
|
| 1297 |
+
"epoch": 11.344149848304973,
|
| 1298 |
+
"grad_norm": 0.86328125,
|
| 1299 |
+
"learning_rate": 0.0007731170030339005,
|
| 1300 |
+
"loss": 2.1953,
|
| 1301 |
+
"step": 86000
|
| 1302 |
+
},
|
| 1303 |
+
{
|
| 1304 |
+
"epoch": 11.410104207888141,
|
| 1305 |
+
"grad_norm": 0.6015625,
|
| 1306 |
+
"learning_rate": 0.0007717979158422373,
|
| 1307 |
+
"loss": 2.1873,
|
| 1308 |
+
"step": 86500
|
| 1309 |
+
},
|
| 1310 |
+
{
|
| 1311 |
+
"epoch": 11.47605856747131,
|
| 1312 |
+
"grad_norm": 0.63671875,
|
| 1313 |
+
"learning_rate": 0.0007704788286505738,
|
| 1314 |
+
"loss": 2.2014,
|
| 1315 |
+
"step": 87000
|
| 1316 |
+
},
|
| 1317 |
+
{
|
| 1318 |
+
"epoch": 11.54201292705448,
|
| 1319 |
+
"grad_norm": 0.703125,
|
| 1320 |
+
"learning_rate": 0.0007691597414589104,
|
| 1321 |
+
"loss": 2.2029,
|
| 1322 |
+
"step": 87500
|
| 1323 |
+
},
|
| 1324 |
+
{
|
| 1325 |
+
"epoch": 11.607967286637647,
|
| 1326 |
+
"grad_norm": 0.69140625,
|
| 1327 |
+
"learning_rate": 0.000767840654267247,
|
| 1328 |
+
"loss": 2.2041,
|
| 1329 |
+
"step": 88000
|
| 1330 |
+
},
|
| 1331 |
+
{
|
| 1332 |
+
"epoch": 11.673921646220816,
|
| 1333 |
+
"grad_norm": 0.59375,
|
| 1334 |
+
"learning_rate": 0.0007665215670755837,
|
| 1335 |
+
"loss": 2.2031,
|
| 1336 |
+
"step": 88500
|
| 1337 |
+
},
|
| 1338 |
+
{
|
| 1339 |
+
"epoch": 11.739876005803984,
|
| 1340 |
+
"grad_norm": 0.71875,
|
| 1341 |
+
"learning_rate": 0.0007652024798839203,
|
| 1342 |
+
"loss": 2.2059,
|
| 1343 |
+
"step": 89000
|
| 1344 |
+
},
|
| 1345 |
+
{
|
| 1346 |
+
"epoch": 11.805830365387152,
|
| 1347 |
+
"grad_norm": 0.65625,
|
| 1348 |
+
"learning_rate": 0.0007638833926922569,
|
| 1349 |
+
"loss": 2.21,
|
| 1350 |
+
"step": 89500
|
| 1351 |
+
},
|
| 1352 |
+
{
|
| 1353 |
+
"epoch": 11.87178472497032,
|
| 1354 |
+
"grad_norm": 0.6875,
|
| 1355 |
+
"learning_rate": 0.0007625643055005937,
|
| 1356 |
+
"loss": 2.2105,
|
| 1357 |
+
"step": 90000
|
| 1358 |
+
},
|
| 1359 |
+
{
|
| 1360 |
+
"epoch": 11.937739084553488,
|
| 1361 |
+
"grad_norm": 0.6015625,
|
| 1362 |
+
"learning_rate": 0.0007612452183089302,
|
| 1363 |
+
"loss": 2.2089,
|
| 1364 |
+
"step": 90500
|
| 1365 |
+
},
|
| 1366 |
+
{
|
| 1367 |
+
"epoch": 12.0,
|
| 1368 |
+
"eval_loss": 2.453855514526367,
|
| 1369 |
+
"eval_runtime": 3.8561,
|
| 1370 |
+
"eval_samples_per_second": 518.662,
|
| 1371 |
+
"eval_steps_per_second": 2.075,
|
| 1372 |
+
"step": 90972
|
| 1373 |
+
},
|
| 1374 |
+
{
|
| 1375 |
+
"epoch": 12.003693444136658,
|
| 1376 |
+
"grad_norm": 0.66796875,
|
| 1377 |
+
"learning_rate": 0.0007599261311172669,
|
| 1378 |
+
"loss": 2.2131,
|
| 1379 |
+
"step": 91000
|
| 1380 |
+
},
|
| 1381 |
+
{
|
| 1382 |
+
"epoch": 12.069647803719826,
|
| 1383 |
+
"grad_norm": 0.61328125,
|
| 1384 |
+
"learning_rate": 0.0007586070439256035,
|
| 1385 |
+
"loss": 2.1413,
|
| 1386 |
+
"step": 91500
|
| 1387 |
+
},
|
| 1388 |
+
{
|
| 1389 |
+
"epoch": 12.135602163302995,
|
| 1390 |
+
"grad_norm": 0.640625,
|
| 1391 |
+
"learning_rate": 0.0007572879567339402,
|
| 1392 |
+
"loss": 2.1529,
|
| 1393 |
+
"step": 92000
|
| 1394 |
+
},
|
| 1395 |
+
{
|
| 1396 |
+
"epoch": 12.201556522886163,
|
| 1397 |
+
"grad_norm": 0.609375,
|
| 1398 |
+
"learning_rate": 0.0007559688695422767,
|
| 1399 |
+
"loss": 2.1531,
|
| 1400 |
+
"step": 92500
|
| 1401 |
+
},
|
| 1402 |
+
{
|
| 1403 |
+
"epoch": 12.267510882469331,
|
| 1404 |
+
"grad_norm": 0.6640625,
|
| 1405 |
+
"learning_rate": 0.0007546497823506135,
|
| 1406 |
+
"loss": 2.1672,
|
| 1407 |
+
"step": 93000
|
| 1408 |
+
},
|
| 1409 |
+
{
|
| 1410 |
+
"epoch": 12.3334652420525,
|
| 1411 |
+
"grad_norm": 0.6484375,
|
| 1412 |
+
"learning_rate": 0.00075333069515895,
|
| 1413 |
+
"loss": 2.1653,
|
| 1414 |
+
"step": 93500
|
| 1415 |
+
},
|
| 1416 |
+
{
|
| 1417 |
+
"epoch": 12.399419601635667,
|
| 1418 |
+
"grad_norm": 0.62109375,
|
| 1419 |
+
"learning_rate": 0.0007520116079672867,
|
| 1420 |
+
"loss": 2.1711,
|
| 1421 |
+
"step": 94000
|
| 1422 |
+
},
|
| 1423 |
+
{
|
| 1424 |
+
"epoch": 12.465373961218837,
|
| 1425 |
+
"grad_norm": 0.59375,
|
| 1426 |
+
"learning_rate": 0.0007506925207756233,
|
| 1427 |
+
"loss": 2.1787,
|
| 1428 |
+
"step": 94500
|
| 1429 |
+
},
|
| 1430 |
+
{
|
| 1431 |
+
"epoch": 12.531328320802006,
|
| 1432 |
+
"grad_norm": 0.60546875,
|
| 1433 |
+
"learning_rate": 0.00074937343358396,
|
| 1434 |
+
"loss": 2.1737,
|
| 1435 |
+
"step": 95000
|
| 1436 |
+
},
|
| 1437 |
+
{
|
| 1438 |
+
"epoch": 12.597282680385174,
|
| 1439 |
+
"grad_norm": 0.65625,
|
| 1440 |
+
"learning_rate": 0.0007480543463922965,
|
| 1441 |
+
"loss": 2.1787,
|
| 1442 |
+
"step": 95500
|
| 1443 |
+
},
|
| 1444 |
+
{
|
| 1445 |
+
"epoch": 12.663237039968342,
|
| 1446 |
+
"grad_norm": 0.66015625,
|
| 1447 |
+
"learning_rate": 0.0007467352592006333,
|
| 1448 |
+
"loss": 2.1764,
|
| 1449 |
+
"step": 96000
|
| 1450 |
+
},
|
| 1451 |
+
{
|
| 1452 |
+
"epoch": 12.72919139955151,
|
| 1453 |
+
"grad_norm": 0.6484375,
|
| 1454 |
+
"learning_rate": 0.0007454161720089698,
|
| 1455 |
+
"loss": 2.1902,
|
| 1456 |
+
"step": 96500
|
| 1457 |
+
},
|
| 1458 |
+
{
|
| 1459 |
+
"epoch": 12.795145759134678,
|
| 1460 |
+
"grad_norm": 0.62109375,
|
| 1461 |
+
"learning_rate": 0.0007440970848173064,
|
| 1462 |
+
"loss": 2.1778,
|
| 1463 |
+
"step": 97000
|
| 1464 |
+
},
|
| 1465 |
+
{
|
| 1466 |
+
"epoch": 12.861100118717847,
|
| 1467 |
+
"grad_norm": 0.61328125,
|
| 1468 |
+
"learning_rate": 0.000742777997625643,
|
| 1469 |
+
"loss": 2.1791,
|
| 1470 |
+
"step": 97500
|
| 1471 |
+
},
|
| 1472 |
+
{
|
| 1473 |
+
"epoch": 12.927054478301017,
|
| 1474 |
+
"grad_norm": 0.875,
|
| 1475 |
+
"learning_rate": 0.0007414589104339797,
|
| 1476 |
+
"loss": 2.194,
|
| 1477 |
+
"step": 98000
|
| 1478 |
+
},
|
| 1479 |
+
{
|
| 1480 |
+
"epoch": 12.993008837884185,
|
| 1481 |
+
"grad_norm": 0.62890625,
|
| 1482 |
+
"learning_rate": 0.0007401398232423163,
|
| 1483 |
+
"loss": 2.1869,
|
| 1484 |
+
"step": 98500
|
| 1485 |
+
},
|
| 1486 |
+
{
|
| 1487 |
+
"epoch": 13.0,
|
| 1488 |
+
"eval_loss": 2.4518699645996094,
|
| 1489 |
+
"eval_runtime": 4.8206,
|
| 1490 |
+
"eval_samples_per_second": 414.883,
|
| 1491 |
+
"eval_steps_per_second": 1.66,
|
| 1492 |
+
"step": 98553
|
| 1493 |
+
},
|
| 1494 |
+
{
|
| 1495 |
+
"epoch": 13.058963197467353,
|
| 1496 |
+
"grad_norm": 0.65234375,
|
| 1497 |
+
"learning_rate": 0.0007388207360506529,
|
| 1498 |
+
"loss": 2.1188,
|
| 1499 |
+
"step": 99000
|
| 1500 |
+
},
|
| 1501 |
+
{
|
| 1502 |
+
"epoch": 13.124917557050521,
|
| 1503 |
+
"grad_norm": 0.640625,
|
| 1504 |
+
"learning_rate": 0.0007375016488589896,
|
| 1505 |
+
"loss": 2.1277,
|
| 1506 |
+
"step": 99500
|
| 1507 |
+
},
|
| 1508 |
+
{
|
| 1509 |
+
"epoch": 13.19087191663369,
|
| 1510 |
+
"grad_norm": 0.734375,
|
| 1511 |
+
"learning_rate": 0.0007361825616673262,
|
| 1512 |
+
"loss": 2.139,
|
| 1513 |
+
"step": 100000
|
| 1514 |
+
},
|
| 1515 |
+
{
|
| 1516 |
+
"epoch": 13.256826276216858,
|
| 1517 |
+
"grad_norm": 0.6796875,
|
| 1518 |
+
"learning_rate": 0.0007348634744756628,
|
| 1519 |
+
"loss": 2.1469,
|
| 1520 |
+
"step": 100500
|
| 1521 |
+
},
|
| 1522 |
+
{
|
| 1523 |
+
"epoch": 13.322780635800026,
|
| 1524 |
+
"grad_norm": 0.67578125,
|
| 1525 |
+
"learning_rate": 0.0007335443872839995,
|
| 1526 |
+
"loss": 2.1406,
|
| 1527 |
+
"step": 101000
|
| 1528 |
+
},
|
| 1529 |
+
{
|
| 1530 |
+
"epoch": 13.388734995383194,
|
| 1531 |
+
"grad_norm": 0.6640625,
|
| 1532 |
+
"learning_rate": 0.0007322253000923362,
|
| 1533 |
+
"loss": 2.1448,
|
| 1534 |
+
"step": 101500
|
| 1535 |
+
},
|
| 1536 |
+
{
|
| 1537 |
+
"epoch": 13.454689354966364,
|
| 1538 |
+
"grad_norm": 0.67578125,
|
| 1539 |
+
"learning_rate": 0.0007309062129006727,
|
| 1540 |
+
"loss": 2.1424,
|
| 1541 |
+
"step": 102000
|
| 1542 |
+
},
|
| 1543 |
+
{
|
| 1544 |
+
"epoch": 13.520643714549532,
|
| 1545 |
+
"grad_norm": 0.66796875,
|
| 1546 |
+
"learning_rate": 0.0007295871257090095,
|
| 1547 |
+
"loss": 2.1505,
|
| 1548 |
+
"step": 102500
|
| 1549 |
+
},
|
| 1550 |
+
{
|
| 1551 |
+
"epoch": 13.5865980741327,
|
| 1552 |
+
"grad_norm": 0.7578125,
|
| 1553 |
+
"learning_rate": 0.000728268038517346,
|
| 1554 |
+
"loss": 2.1533,
|
| 1555 |
+
"step": 103000
|
| 1556 |
+
},
|
| 1557 |
+
{
|
| 1558 |
+
"epoch": 13.652552433715869,
|
| 1559 |
+
"grad_norm": 0.671875,
|
| 1560 |
+
"learning_rate": 0.0007269489513256827,
|
| 1561 |
+
"loss": 2.1655,
|
| 1562 |
+
"step": 103500
|
| 1563 |
+
},
|
| 1564 |
+
{
|
| 1565 |
+
"epoch": 13.718506793299037,
|
| 1566 |
+
"grad_norm": 0.609375,
|
| 1567 |
+
"learning_rate": 0.0007256298641340193,
|
| 1568 |
+
"loss": 2.1599,
|
| 1569 |
+
"step": 104000
|
| 1570 |
+
},
|
| 1571 |
+
{
|
| 1572 |
+
"epoch": 13.784461152882205,
|
| 1573 |
+
"grad_norm": 0.59765625,
|
| 1574 |
+
"learning_rate": 0.000724310776942356,
|
| 1575 |
+
"loss": 2.1582,
|
| 1576 |
+
"step": 104500
|
| 1577 |
+
},
|
| 1578 |
+
{
|
| 1579 |
+
"epoch": 13.850415512465373,
|
| 1580 |
+
"grad_norm": 0.62890625,
|
| 1581 |
+
"learning_rate": 0.0007229916897506925,
|
| 1582 |
+
"loss": 2.1664,
|
| 1583 |
+
"step": 105000
|
| 1584 |
+
},
|
| 1585 |
+
{
|
| 1586 |
+
"epoch": 13.916369872048543,
|
| 1587 |
+
"grad_norm": 0.65625,
|
| 1588 |
+
"learning_rate": 0.0007216726025590291,
|
| 1589 |
+
"loss": 2.1638,
|
| 1590 |
+
"step": 105500
|
| 1591 |
+
},
|
| 1592 |
+
{
|
| 1593 |
+
"epoch": 13.982324231631711,
|
| 1594 |
+
"grad_norm": 0.7109375,
|
| 1595 |
+
"learning_rate": 0.0007203535153673658,
|
| 1596 |
+
"loss": 2.1643,
|
| 1597 |
+
"step": 106000
|
| 1598 |
+
},
|
| 1599 |
+
{
|
| 1600 |
+
"epoch": 14.0,
|
| 1601 |
+
"eval_loss": 2.442619562149048,
|
| 1602 |
+
"eval_runtime": 5.7628,
|
| 1603 |
+
"eval_samples_per_second": 347.051,
|
| 1604 |
+
"eval_steps_per_second": 1.388,
|
| 1605 |
+
"step": 106134
|
| 1606 |
+
},
|
| 1607 |
+
{
|
| 1608 |
+
"epoch": 14.04827859121488,
|
| 1609 |
+
"grad_norm": 0.68359375,
|
| 1610 |
+
"learning_rate": 0.0007190344281757024,
|
| 1611 |
+
"loss": 2.1108,
|
| 1612 |
+
"step": 106500
|
| 1613 |
+
},
|
| 1614 |
+
{
|
| 1615 |
+
"epoch": 14.114232950798048,
|
| 1616 |
+
"grad_norm": 0.67578125,
|
| 1617 |
+
"learning_rate": 0.000717715340984039,
|
| 1618 |
+
"loss": 2.1076,
|
| 1619 |
+
"step": 107000
|
| 1620 |
+
},
|
| 1621 |
+
{
|
| 1622 |
+
"epoch": 14.180187310381216,
|
| 1623 |
+
"grad_norm": 0.6484375,
|
| 1624 |
+
"learning_rate": 0.0007163962537923757,
|
| 1625 |
+
"loss": 2.1202,
|
| 1626 |
+
"step": 107500
|
| 1627 |
+
},
|
| 1628 |
+
{
|
| 1629 |
+
"epoch": 14.246141669964384,
|
| 1630 |
+
"grad_norm": 0.625,
|
| 1631 |
+
"learning_rate": 0.0007150771666007123,
|
| 1632 |
+
"loss": 2.1221,
|
| 1633 |
+
"step": 108000
|
| 1634 |
+
},
|
| 1635 |
+
{
|
| 1636 |
+
"epoch": 14.312096029547552,
|
| 1637 |
+
"grad_norm": 0.625,
|
| 1638 |
+
"learning_rate": 0.0007137580794090489,
|
| 1639 |
+
"loss": 2.1262,
|
| 1640 |
+
"step": 108500
|
| 1641 |
+
},
|
| 1642 |
+
{
|
| 1643 |
+
"epoch": 14.378050389130722,
|
| 1644 |
+
"grad_norm": 0.62109375,
|
| 1645 |
+
"learning_rate": 0.0007124389922173856,
|
| 1646 |
+
"loss": 2.1376,
|
| 1647 |
+
"step": 109000
|
| 1648 |
+
},
|
| 1649 |
+
{
|
| 1650 |
+
"epoch": 14.44400474871389,
|
| 1651 |
+
"grad_norm": 0.65234375,
|
| 1652 |
+
"learning_rate": 0.0007111199050257222,
|
| 1653 |
+
"loss": 2.1344,
|
| 1654 |
+
"step": 109500
|
| 1655 |
+
},
|
| 1656 |
+
{
|
| 1657 |
+
"epoch": 14.509959108297059,
|
| 1658 |
+
"grad_norm": 0.65625,
|
| 1659 |
+
"learning_rate": 0.0007098008178340588,
|
| 1660 |
+
"loss": 2.1379,
|
| 1661 |
+
"step": 110000
|
| 1662 |
+
},
|
| 1663 |
+
{
|
| 1664 |
+
"epoch": 14.575913467880227,
|
| 1665 |
+
"grad_norm": 0.671875,
|
| 1666 |
+
"learning_rate": 0.0007084817306423955,
|
| 1667 |
+
"loss": 2.1354,
|
| 1668 |
+
"step": 110500
|
| 1669 |
+
},
|
| 1670 |
+
{
|
| 1671 |
+
"epoch": 14.641867827463395,
|
| 1672 |
+
"grad_norm": 0.66796875,
|
| 1673 |
+
"learning_rate": 0.0007071626434507322,
|
| 1674 |
+
"loss": 2.1339,
|
| 1675 |
+
"step": 111000
|
| 1676 |
+
},
|
| 1677 |
+
{
|
| 1678 |
+
"epoch": 14.707822187046563,
|
| 1679 |
+
"grad_norm": 0.65625,
|
| 1680 |
+
"learning_rate": 0.0007058435562590687,
|
| 1681 |
+
"loss": 2.1297,
|
| 1682 |
+
"step": 111500
|
| 1683 |
+
},
|
| 1684 |
+
{
|
| 1685 |
+
"epoch": 14.773776546629731,
|
| 1686 |
+
"grad_norm": 0.69921875,
|
| 1687 |
+
"learning_rate": 0.0007045244690674055,
|
| 1688 |
+
"loss": 2.1355,
|
| 1689 |
+
"step": 112000
|
| 1690 |
+
},
|
| 1691 |
+
{
|
| 1692 |
+
"epoch": 14.839730906212901,
|
| 1693 |
+
"grad_norm": 0.6640625,
|
| 1694 |
+
"learning_rate": 0.000703205381875742,
|
| 1695 |
+
"loss": 2.1385,
|
| 1696 |
+
"step": 112500
|
| 1697 |
+
},
|
| 1698 |
+
{
|
| 1699 |
+
"epoch": 14.90568526579607,
|
| 1700 |
+
"grad_norm": 0.83203125,
|
| 1701 |
+
"learning_rate": 0.0007018862946840787,
|
| 1702 |
+
"loss": 2.1399,
|
| 1703 |
+
"step": 113000
|
| 1704 |
+
},
|
| 1705 |
+
{
|
| 1706 |
+
"epoch": 14.971639625379238,
|
| 1707 |
+
"grad_norm": 0.62890625,
|
| 1708 |
+
"learning_rate": 0.0007005672074924153,
|
| 1709 |
+
"loss": 2.149,
|
| 1710 |
+
"step": 113500
|
| 1711 |
+
},
|
| 1712 |
+
{
|
| 1713 |
+
"epoch": 15.0,
|
| 1714 |
+
"eval_loss": 2.43693208694458,
|
| 1715 |
+
"eval_runtime": 4.8208,
|
| 1716 |
+
"eval_samples_per_second": 414.869,
|
| 1717 |
+
"eval_steps_per_second": 1.659,
|
| 1718 |
+
"step": 113715
|
| 1719 |
+
},
|
| 1720 |
+
{
|
| 1721 |
+
"epoch": 15.037593984962406,
|
| 1722 |
+
"grad_norm": 0.6171875,
|
| 1723 |
+
"learning_rate": 0.0006992481203007519,
|
| 1724 |
+
"loss": 2.1078,
|
| 1725 |
+
"step": 114000
|
| 1726 |
+
},
|
| 1727 |
+
{
|
| 1728 |
+
"epoch": 15.103548344545574,
|
| 1729 |
+
"grad_norm": 0.6875,
|
| 1730 |
+
"learning_rate": 0.0006979290331090885,
|
| 1731 |
+
"loss": 2.0965,
|
| 1732 |
+
"step": 114500
|
| 1733 |
+
},
|
| 1734 |
+
{
|
| 1735 |
+
"epoch": 15.169502704128742,
|
| 1736 |
+
"grad_norm": 0.66796875,
|
| 1737 |
+
"learning_rate": 0.0006966099459174251,
|
| 1738 |
+
"loss": 2.1037,
|
| 1739 |
+
"step": 115000
|
| 1740 |
+
},
|
| 1741 |
+
{
|
| 1742 |
+
"epoch": 15.23545706371191,
|
| 1743 |
+
"grad_norm": 0.6640625,
|
| 1744 |
+
"learning_rate": 0.0006952908587257618,
|
| 1745 |
+
"loss": 2.1074,
|
| 1746 |
+
"step": 115500
|
| 1747 |
+
},
|
| 1748 |
+
{
|
| 1749 |
+
"epoch": 15.30141142329508,
|
| 1750 |
+
"grad_norm": 0.734375,
|
| 1751 |
+
"learning_rate": 0.0006939717715340984,
|
| 1752 |
+
"loss": 2.1103,
|
| 1753 |
+
"step": 116000
|
| 1754 |
+
},
|
| 1755 |
+
{
|
| 1756 |
+
"epoch": 15.367365782878249,
|
| 1757 |
+
"grad_norm": 0.69140625,
|
| 1758 |
+
"learning_rate": 0.000692652684342435,
|
| 1759 |
+
"loss": 2.1089,
|
| 1760 |
+
"step": 116500
|
| 1761 |
+
},
|
| 1762 |
+
{
|
| 1763 |
+
"epoch": 15.433320142461417,
|
| 1764 |
+
"grad_norm": 0.6328125,
|
| 1765 |
+
"learning_rate": 0.0006913335971507717,
|
| 1766 |
+
"loss": 2.1088,
|
| 1767 |
+
"step": 117000
|
| 1768 |
+
},
|
| 1769 |
+
{
|
| 1770 |
+
"epoch": 15.499274502044585,
|
| 1771 |
+
"grad_norm": 0.6640625,
|
| 1772 |
+
"learning_rate": 0.0006900145099591083,
|
| 1773 |
+
"loss": 2.1112,
|
| 1774 |
+
"step": 117500
|
| 1775 |
+
},
|
| 1776 |
+
{
|
| 1777 |
+
"epoch": 15.565228861627753,
|
| 1778 |
+
"grad_norm": 0.609375,
|
| 1779 |
+
"learning_rate": 0.0006886954227674449,
|
| 1780 |
+
"loss": 2.1153,
|
| 1781 |
+
"step": 118000
|
| 1782 |
+
},
|
| 1783 |
+
{
|
| 1784 |
+
"epoch": 15.631183221210922,
|
| 1785 |
+
"grad_norm": 0.6328125,
|
| 1786 |
+
"learning_rate": 0.0006873763355757816,
|
| 1787 |
+
"loss": 2.1321,
|
| 1788 |
+
"step": 118500
|
| 1789 |
+
},
|
| 1790 |
+
{
|
| 1791 |
+
"epoch": 15.69713758079409,
|
| 1792 |
+
"grad_norm": 0.93359375,
|
| 1793 |
+
"learning_rate": 0.0006860572483841182,
|
| 1794 |
+
"loss": 2.1233,
|
| 1795 |
+
"step": 119000
|
| 1796 |
+
},
|
| 1797 |
+
{
|
| 1798 |
+
"epoch": 15.763091940377258,
|
| 1799 |
+
"grad_norm": 0.68359375,
|
| 1800 |
+
"learning_rate": 0.0006847381611924548,
|
| 1801 |
+
"loss": 2.1361,
|
| 1802 |
+
"step": 119500
|
| 1803 |
+
},
|
| 1804 |
+
{
|
| 1805 |
+
"epoch": 15.829046299960428,
|
| 1806 |
+
"grad_norm": 0.76171875,
|
| 1807 |
+
"learning_rate": 0.0006834190740007915,
|
| 1808 |
+
"loss": 2.1296,
|
| 1809 |
+
"step": 120000
|
| 1810 |
+
},
|
| 1811 |
+
{
|
| 1812 |
+
"epoch": 15.895000659543596,
|
| 1813 |
+
"grad_norm": 0.62890625,
|
| 1814 |
+
"learning_rate": 0.0006820999868091281,
|
| 1815 |
+
"loss": 2.1236,
|
| 1816 |
+
"step": 120500
|
| 1817 |
+
},
|
| 1818 |
+
{
|
| 1819 |
+
"epoch": 15.960955019126764,
|
| 1820 |
+
"grad_norm": 0.6484375,
|
| 1821 |
+
"learning_rate": 0.0006807808996174647,
|
| 1822 |
+
"loss": 2.1354,
|
| 1823 |
+
"step": 121000
|
| 1824 |
+
},
|
| 1825 |
+
{
|
| 1826 |
+
"epoch": 16.0,
|
| 1827 |
+
"eval_loss": 2.431525945663452,
|
| 1828 |
+
"eval_runtime": 5.3462,
|
| 1829 |
+
"eval_samples_per_second": 374.095,
|
| 1830 |
+
"eval_steps_per_second": 1.496,
|
| 1831 |
+
"step": 121296
|
| 1832 |
+
},
|
| 1833 |
+
{
|
| 1834 |
+
"epoch": 16.026909378709934,
|
| 1835 |
+
"grad_norm": 0.69921875,
|
| 1836 |
+
"learning_rate": 0.0006794618124258015,
|
| 1837 |
+
"loss": 2.1031,
|
| 1838 |
+
"step": 121500
|
| 1839 |
+
},
|
| 1840 |
+
{
|
| 1841 |
+
"epoch": 16.0928637382931,
|
| 1842 |
+
"grad_norm": 0.66015625,
|
| 1843 |
+
"learning_rate": 0.000678142725234138,
|
| 1844 |
+
"loss": 2.0653,
|
| 1845 |
+
"step": 122000
|
| 1846 |
+
},
|
| 1847 |
+
{
|
| 1848 |
+
"epoch": 16.15881809787627,
|
| 1849 |
+
"grad_norm": 0.6484375,
|
| 1850 |
+
"learning_rate": 0.0006768236380424745,
|
| 1851 |
+
"loss": 2.0796,
|
| 1852 |
+
"step": 122500
|
| 1853 |
+
},
|
| 1854 |
+
{
|
| 1855 |
+
"epoch": 16.224772457459437,
|
| 1856 |
+
"grad_norm": 0.796875,
|
| 1857 |
+
"learning_rate": 0.0006755045508508113,
|
| 1858 |
+
"loss": 2.0916,
|
| 1859 |
+
"step": 123000
|
| 1860 |
+
},
|
| 1861 |
+
{
|
| 1862 |
+
"epoch": 16.290726817042607,
|
| 1863 |
+
"grad_norm": 0.6796875,
|
| 1864 |
+
"learning_rate": 0.0006741854636591479,
|
| 1865 |
+
"loss": 2.0888,
|
| 1866 |
+
"step": 123500
|
| 1867 |
+
},
|
| 1868 |
+
{
|
| 1869 |
+
"epoch": 16.356681176625774,
|
| 1870 |
+
"grad_norm": 0.62890625,
|
| 1871 |
+
"learning_rate": 0.0006728663764674845,
|
| 1872 |
+
"loss": 2.0981,
|
| 1873 |
+
"step": 124000
|
| 1874 |
+
},
|
| 1875 |
+
{
|
| 1876 |
+
"epoch": 16.422635536208944,
|
| 1877 |
+
"grad_norm": 0.65625,
|
| 1878 |
+
"learning_rate": 0.0006715472892758211,
|
| 1879 |
+
"loss": 2.1025,
|
| 1880 |
+
"step": 124500
|
| 1881 |
+
},
|
| 1882 |
+
{
|
| 1883 |
+
"epoch": 16.488589895792114,
|
| 1884 |
+
"grad_norm": 0.640625,
|
| 1885 |
+
"learning_rate": 0.0006702282020841578,
|
| 1886 |
+
"loss": 2.0996,
|
| 1887 |
+
"step": 125000
|
| 1888 |
+
},
|
| 1889 |
+
{
|
| 1890 |
+
"epoch": 16.55454425537528,
|
| 1891 |
+
"grad_norm": 0.6640625,
|
| 1892 |
+
"learning_rate": 0.0006689091148924944,
|
| 1893 |
+
"loss": 2.103,
|
| 1894 |
+
"step": 125500
|
| 1895 |
+
},
|
| 1896 |
+
{
|
| 1897 |
+
"epoch": 16.62049861495845,
|
| 1898 |
+
"grad_norm": 0.73828125,
|
| 1899 |
+
"learning_rate": 0.000667590027700831,
|
| 1900 |
+
"loss": 2.1105,
|
| 1901 |
+
"step": 126000
|
| 1902 |
+
},
|
| 1903 |
+
{
|
| 1904 |
+
"epoch": 16.686452974541616,
|
| 1905 |
+
"grad_norm": 0.6796875,
|
| 1906 |
+
"learning_rate": 0.0006662709405091677,
|
| 1907 |
+
"loss": 2.1138,
|
| 1908 |
+
"step": 126500
|
| 1909 |
+
},
|
| 1910 |
+
{
|
| 1911 |
+
"epoch": 16.752407334124786,
|
| 1912 |
+
"grad_norm": 0.65625,
|
| 1913 |
+
"learning_rate": 0.0006649518533175043,
|
| 1914 |
+
"loss": 2.1095,
|
| 1915 |
+
"step": 127000
|
| 1916 |
+
},
|
| 1917 |
+
{
|
| 1918 |
+
"epoch": 16.818361693707953,
|
| 1919 |
+
"grad_norm": 0.640625,
|
| 1920 |
+
"learning_rate": 0.0006636327661258409,
|
| 1921 |
+
"loss": 2.1114,
|
| 1922 |
+
"step": 127500
|
| 1923 |
+
},
|
| 1924 |
+
{
|
| 1925 |
+
"epoch": 16.884316053291123,
|
| 1926 |
+
"grad_norm": 0.7421875,
|
| 1927 |
+
"learning_rate": 0.0006623136789341776,
|
| 1928 |
+
"loss": 2.1012,
|
| 1929 |
+
"step": 128000
|
| 1930 |
+
},
|
| 1931 |
+
{
|
| 1932 |
+
"epoch": 16.950270412874293,
|
| 1933 |
+
"grad_norm": 0.69140625,
|
| 1934 |
+
"learning_rate": 0.0006609945917425142,
|
| 1935 |
+
"loss": 2.109,
|
| 1936 |
+
"step": 128500
|
| 1937 |
+
},
|
| 1938 |
+
{
|
| 1939 |
+
"epoch": 17.0,
|
| 1940 |
+
"eval_loss": 2.4235010147094727,
|
| 1941 |
+
"eval_runtime": 5.2236,
|
| 1942 |
+
"eval_samples_per_second": 382.874,
|
| 1943 |
+
"eval_steps_per_second": 1.531,
|
| 1944 |
+
"step": 128877
|
| 1945 |
+
},
|
| 1946 |
+
{
|
| 1947 |
+
"epoch": 17.01622477245746,
|
| 1948 |
+
"grad_norm": 0.6640625,
|
| 1949 |
+
"learning_rate": 0.0006596755045508508,
|
| 1950 |
+
"loss": 2.0946,
|
| 1951 |
+
"step": 129000
|
| 1952 |
+
},
|
| 1953 |
+
{
|
| 1954 |
+
"epoch": 17.08217913204063,
|
| 1955 |
+
"grad_norm": 0.67578125,
|
| 1956 |
+
"learning_rate": 0.0006583564173591875,
|
| 1957 |
+
"loss": 2.0571,
|
| 1958 |
+
"step": 129500
|
| 1959 |
+
},
|
| 1960 |
+
{
|
| 1961 |
+
"epoch": 17.148133491623796,
|
| 1962 |
+
"grad_norm": 0.7578125,
|
| 1963 |
+
"learning_rate": 0.0006570373301675241,
|
| 1964 |
+
"loss": 2.0727,
|
| 1965 |
+
"step": 130000
|
| 1966 |
+
},
|
| 1967 |
+
{
|
| 1968 |
+
"epoch": 17.214087851206965,
|
| 1969 |
+
"grad_norm": 0.6484375,
|
| 1970 |
+
"learning_rate": 0.0006557182429758607,
|
| 1971 |
+
"loss": 2.0699,
|
| 1972 |
+
"step": 130500
|
| 1973 |
+
},
|
| 1974 |
+
{
|
| 1975 |
+
"epoch": 17.280042210790132,
|
| 1976 |
+
"grad_norm": 0.6953125,
|
| 1977 |
+
"learning_rate": 0.0006543991557841973,
|
| 1978 |
+
"loss": 2.0763,
|
| 1979 |
+
"step": 131000
|
| 1980 |
+
},
|
| 1981 |
+
{
|
| 1982 |
+
"epoch": 17.345996570373302,
|
| 1983 |
+
"grad_norm": 0.7265625,
|
| 1984 |
+
"learning_rate": 0.000653080068592534,
|
| 1985 |
+
"loss": 2.0804,
|
| 1986 |
+
"step": 131500
|
| 1987 |
+
},
|
| 1988 |
+
{
|
| 1989 |
+
"epoch": 17.411950929956472,
|
| 1990 |
+
"grad_norm": 0.69140625,
|
| 1991 |
+
"learning_rate": 0.0006517609814008705,
|
| 1992 |
+
"loss": 2.0832,
|
| 1993 |
+
"step": 132000
|
| 1994 |
+
},
|
| 1995 |
+
{
|
| 1996 |
+
"epoch": 17.47790528953964,
|
| 1997 |
+
"grad_norm": 0.67578125,
|
| 1998 |
+
"learning_rate": 0.0006504418942092073,
|
| 1999 |
+
"loss": 2.0821,
|
| 2000 |
+
"step": 132500
|
| 2001 |
+
},
|
| 2002 |
+
{
|
| 2003 |
+
"epoch": 17.54385964912281,
|
| 2004 |
+
"grad_norm": 0.6875,
|
| 2005 |
+
"learning_rate": 0.0006491228070175439,
|
| 2006 |
+
"loss": 2.0821,
|
| 2007 |
+
"step": 133000
|
| 2008 |
+
},
|
| 2009 |
+
{
|
| 2010 |
+
"epoch": 17.609814008705975,
|
| 2011 |
+
"grad_norm": 0.7109375,
|
| 2012 |
+
"learning_rate": 0.0006478037198258805,
|
| 2013 |
+
"loss": 2.0886,
|
| 2014 |
+
"step": 133500
|
| 2015 |
+
},
|
| 2016 |
+
{
|
| 2017 |
+
"epoch": 17.675768368289145,
|
| 2018 |
+
"grad_norm": 0.66015625,
|
| 2019 |
+
"learning_rate": 0.000646484632634217,
|
| 2020 |
+
"loss": 2.0945,
|
| 2021 |
+
"step": 134000
|
| 2022 |
+
},
|
| 2023 |
+
{
|
| 2024 |
+
"epoch": 17.74172272787231,
|
| 2025 |
+
"grad_norm": 0.64453125,
|
| 2026 |
+
"learning_rate": 0.0006451655454425538,
|
| 2027 |
+
"loss": 2.0918,
|
| 2028 |
+
"step": 134500
|
| 2029 |
+
},
|
| 2030 |
+
{
|
| 2031 |
+
"epoch": 17.80767708745548,
|
| 2032 |
+
"grad_norm": 0.72265625,
|
| 2033 |
+
"learning_rate": 0.0006438464582508904,
|
| 2034 |
+
"loss": 2.099,
|
| 2035 |
+
"step": 135000
|
| 2036 |
+
},
|
| 2037 |
+
{
|
| 2038 |
+
"epoch": 17.873631447038647,
|
| 2039 |
+
"grad_norm": 0.68359375,
|
| 2040 |
+
"learning_rate": 0.000642527371059227,
|
| 2041 |
+
"loss": 2.097,
|
| 2042 |
+
"step": 135500
|
| 2043 |
+
},
|
| 2044 |
+
{
|
| 2045 |
+
"epoch": 17.939585806621817,
|
| 2046 |
+
"grad_norm": 0.68359375,
|
| 2047 |
+
"learning_rate": 0.0006412082838675637,
|
| 2048 |
+
"loss": 2.0928,
|
| 2049 |
+
"step": 136000
|
| 2050 |
+
},
|
| 2051 |
+
{
|
| 2052 |
+
"epoch": 18.0,
|
| 2053 |
+
"eval_loss": 2.4190404415130615,
|
| 2054 |
+
"eval_runtime": 4.6336,
|
| 2055 |
+
"eval_samples_per_second": 431.626,
|
| 2056 |
+
"eval_steps_per_second": 1.727,
|
| 2057 |
+
"step": 136458
|
| 2058 |
+
}
|
| 2059 |
+
],
|
| 2060 |
+
"logging_steps": 500,
|
| 2061 |
+
"max_steps": 379050,
|
| 2062 |
+
"num_input_tokens_seen": 0,
|
| 2063 |
+
"num_train_epochs": 50,
|
| 2064 |
+
"save_steps": 500,
|
| 2065 |
+
"stateful_callbacks": {
|
| 2066 |
+
"EarlyStoppingCallback": {
|
| 2067 |
+
"args": {
|
| 2068 |
+
"early_stopping_patience": 3,
|
| 2069 |
+
"early_stopping_threshold": 0.0
|
| 2070 |
+
},
|
| 2071 |
+
"attributes": {
|
| 2072 |
+
"early_stopping_patience_counter": 0
|
| 2073 |
+
}
|
| 2074 |
+
},
|
| 2075 |
+
"TrainerControl": {
|
| 2076 |
+
"args": {
|
| 2077 |
+
"should_epoch_stop": false,
|
| 2078 |
+
"should_evaluate": false,
|
| 2079 |
+
"should_log": false,
|
| 2080 |
+
"should_save": true,
|
| 2081 |
+
"should_training_stop": false
|
| 2082 |
+
},
|
| 2083 |
+
"attributes": {}
|
| 2084 |
+
}
|
| 2085 |
+
},
|
| 2086 |
+
"total_flos": 2.8932082563541893e+18,
|
| 2087 |
+
"train_batch_size": 256,
|
| 2088 |
+
"trial_name": null,
|
| 2089 |
+
"trial_params": null
|
| 2090 |
+
}
|
training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:eef93778f4df28bcb4b6b3504e3fda994c342be326fb8a4837a1a8cbf8b52b8a
|
| 3 |
+
size 5112
|