Training in progress, step 2000, checkpoint
Browse files- checkpoint-2000/config.json +26 -0
- checkpoint-2000/generation_config.json +6 -0
- checkpoint-2000/model.safetensors +3 -0
- checkpoint-2000/optimizer.pt +3 -0
- checkpoint-2000/rng_state.pth +3 -0
- checkpoint-2000/scheduler.pt +3 -0
- checkpoint-2000/trainer_state.json +1434 -0
- checkpoint-2000/training_args.bin +3 -0
checkpoint-2000/config.json
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"Olmo2ForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"eos_token_id": 100257,
|
| 8 |
+
"hidden_act": "silu",
|
| 9 |
+
"hidden_size": 2048,
|
| 10 |
+
"initializer_range": 0.02,
|
| 11 |
+
"intermediate_size": 8192,
|
| 12 |
+
"max_position_embeddings": 4096,
|
| 13 |
+
"model_type": "olmo2",
|
| 14 |
+
"num_attention_heads": 16,
|
| 15 |
+
"num_hidden_layers": 16,
|
| 16 |
+
"num_key_value_heads": 16,
|
| 17 |
+
"pad_token_id": 100277,
|
| 18 |
+
"rms_norm_eps": 1e-06,
|
| 19 |
+
"rope_scaling": null,
|
| 20 |
+
"rope_theta": 500000,
|
| 21 |
+
"tie_word_embeddings": false,
|
| 22 |
+
"torch_dtype": "bfloat16",
|
| 23 |
+
"transformers_version": "4.51.3",
|
| 24 |
+
"use_cache": false,
|
| 25 |
+
"vocab_size": 100352
|
| 26 |
+
}
|
checkpoint-2000/generation_config.json
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"eos_token_id": 100257,
|
| 4 |
+
"pad_token_id": 100277,
|
| 5 |
+
"transformers_version": "4.51.3"
|
| 6 |
+
}
|
checkpoint-2000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:df08cd3f0c9f85558458cd58b3a8156ff85c34a2dedb427d33111ed85c4776bf
|
| 3 |
+
size 2969854224
|
checkpoint-2000/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f641257cac95875e371c1b39d873a40524bd134d2de639fe68c12f399de2f6db
|
| 3 |
+
size 4500154
|
checkpoint-2000/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a94923ae51ff9eb76ea399515bf553d510c05f5577b00f9d38151d90de6da9b8
|
| 3 |
+
size 15006
|
checkpoint-2000/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9b77e739d3a466a8c8b45ef8be095e17ac60ac81a252719e575140b2e39245b7
|
| 3 |
+
size 1064
|
checkpoint-2000/trainer_state.json
ADDED
|
@@ -0,0 +1,1434 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.5,
|
| 6 |
+
"eval_steps": 500,
|
| 7 |
+
"global_step": 2000,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0025,
|
| 14 |
+
"grad_norm": 156.0,
|
| 15 |
+
"learning_rate": 4.5000000000000003e-07,
|
| 16 |
+
"loss": 15.8812,
|
| 17 |
+
"step": 10
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.005,
|
| 21 |
+
"grad_norm": 106.5,
|
| 22 |
+
"learning_rate": 9.500000000000001e-07,
|
| 23 |
+
"loss": 15.2812,
|
| 24 |
+
"step": 20
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.0075,
|
| 28 |
+
"grad_norm": 181.0,
|
| 29 |
+
"learning_rate": 1.45e-06,
|
| 30 |
+
"loss": 15.3875,
|
| 31 |
+
"step": 30
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"epoch": 0.01,
|
| 35 |
+
"grad_norm": 174.0,
|
| 36 |
+
"learning_rate": 1.9500000000000004e-06,
|
| 37 |
+
"loss": 15.4062,
|
| 38 |
+
"step": 40
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.0125,
|
| 42 |
+
"grad_norm": 218.0,
|
| 43 |
+
"learning_rate": 2.4500000000000003e-06,
|
| 44 |
+
"loss": 16.0312,
|
| 45 |
+
"step": 50
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.015,
|
| 49 |
+
"grad_norm": 178.0,
|
| 50 |
+
"learning_rate": 2.95e-06,
|
| 51 |
+
"loss": 15.35,
|
| 52 |
+
"step": 60
|
| 53 |
+
},
|
| 54 |
+
{
|
| 55 |
+
"epoch": 0.0175,
|
| 56 |
+
"grad_norm": 171.0,
|
| 57 |
+
"learning_rate": 3.45e-06,
|
| 58 |
+
"loss": 15.2188,
|
| 59 |
+
"step": 70
|
| 60 |
+
},
|
| 61 |
+
{
|
| 62 |
+
"epoch": 0.02,
|
| 63 |
+
"grad_norm": 220.0,
|
| 64 |
+
"learning_rate": 3.95e-06,
|
| 65 |
+
"loss": 15.4125,
|
| 66 |
+
"step": 80
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 0.0225,
|
| 70 |
+
"grad_norm": 138.0,
|
| 71 |
+
"learning_rate": 4.450000000000001e-06,
|
| 72 |
+
"loss": 15.0563,
|
| 73 |
+
"step": 90
|
| 74 |
+
},
|
| 75 |
+
{
|
| 76 |
+
"epoch": 0.025,
|
| 77 |
+
"grad_norm": 125.0,
|
| 78 |
+
"learning_rate": 4.95e-06,
|
| 79 |
+
"loss": 14.8938,
|
| 80 |
+
"step": 100
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"epoch": 0.0275,
|
| 84 |
+
"grad_norm": 126.5,
|
| 85 |
+
"learning_rate": 5.450000000000001e-06,
|
| 86 |
+
"loss": 14.475,
|
| 87 |
+
"step": 110
|
| 88 |
+
},
|
| 89 |
+
{
|
| 90 |
+
"epoch": 0.03,
|
| 91 |
+
"grad_norm": 134.0,
|
| 92 |
+
"learning_rate": 5.950000000000001e-06,
|
| 93 |
+
"loss": 14.0312,
|
| 94 |
+
"step": 120
|
| 95 |
+
},
|
| 96 |
+
{
|
| 97 |
+
"epoch": 0.0325,
|
| 98 |
+
"grad_norm": 127.5,
|
| 99 |
+
"learning_rate": 6.450000000000001e-06,
|
| 100 |
+
"loss": 13.7312,
|
| 101 |
+
"step": 130
|
| 102 |
+
},
|
| 103 |
+
{
|
| 104 |
+
"epoch": 0.035,
|
| 105 |
+
"grad_norm": 138.0,
|
| 106 |
+
"learning_rate": 6.95e-06,
|
| 107 |
+
"loss": 13.5125,
|
| 108 |
+
"step": 140
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"epoch": 0.0375,
|
| 112 |
+
"grad_norm": 174.0,
|
| 113 |
+
"learning_rate": 7.450000000000001e-06,
|
| 114 |
+
"loss": 12.6125,
|
| 115 |
+
"step": 150
|
| 116 |
+
},
|
| 117 |
+
{
|
| 118 |
+
"epoch": 0.04,
|
| 119 |
+
"grad_norm": 205.0,
|
| 120 |
+
"learning_rate": 7.950000000000002e-06,
|
| 121 |
+
"loss": 11.8187,
|
| 122 |
+
"step": 160
|
| 123 |
+
},
|
| 124 |
+
{
|
| 125 |
+
"epoch": 0.0425,
|
| 126 |
+
"grad_norm": 175.0,
|
| 127 |
+
"learning_rate": 8.45e-06,
|
| 128 |
+
"loss": 11.0125,
|
| 129 |
+
"step": 170
|
| 130 |
+
},
|
| 131 |
+
{
|
| 132 |
+
"epoch": 0.045,
|
| 133 |
+
"grad_norm": 180.0,
|
| 134 |
+
"learning_rate": 8.95e-06,
|
| 135 |
+
"loss": 9.9812,
|
| 136 |
+
"step": 180
|
| 137 |
+
},
|
| 138 |
+
{
|
| 139 |
+
"epoch": 0.0475,
|
| 140 |
+
"grad_norm": 106.0,
|
| 141 |
+
"learning_rate": 9.450000000000001e-06,
|
| 142 |
+
"loss": 9.2812,
|
| 143 |
+
"step": 190
|
| 144 |
+
},
|
| 145 |
+
{
|
| 146 |
+
"epoch": 0.05,
|
| 147 |
+
"grad_norm": 106.5,
|
| 148 |
+
"learning_rate": 9.950000000000001e-06,
|
| 149 |
+
"loss": 8.3938,
|
| 150 |
+
"step": 200
|
| 151 |
+
},
|
| 152 |
+
{
|
| 153 |
+
"epoch": 0.0525,
|
| 154 |
+
"grad_norm": 96.0,
|
| 155 |
+
"learning_rate": 1.045e-05,
|
| 156 |
+
"loss": 7.7094,
|
| 157 |
+
"step": 210
|
| 158 |
+
},
|
| 159 |
+
{
|
| 160 |
+
"epoch": 0.055,
|
| 161 |
+
"grad_norm": 100.5,
|
| 162 |
+
"learning_rate": 1.095e-05,
|
| 163 |
+
"loss": 7.1562,
|
| 164 |
+
"step": 220
|
| 165 |
+
},
|
| 166 |
+
{
|
| 167 |
+
"epoch": 0.0575,
|
| 168 |
+
"grad_norm": 108.5,
|
| 169 |
+
"learning_rate": 1.145e-05,
|
| 170 |
+
"loss": 6.3125,
|
| 171 |
+
"step": 230
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"epoch": 0.06,
|
| 175 |
+
"grad_norm": 98.5,
|
| 176 |
+
"learning_rate": 1.195e-05,
|
| 177 |
+
"loss": 5.5938,
|
| 178 |
+
"step": 240
|
| 179 |
+
},
|
| 180 |
+
{
|
| 181 |
+
"epoch": 0.0625,
|
| 182 |
+
"grad_norm": 97.5,
|
| 183 |
+
"learning_rate": 1.2450000000000003e-05,
|
| 184 |
+
"loss": 5.0406,
|
| 185 |
+
"step": 250
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.065,
|
| 189 |
+
"grad_norm": 92.0,
|
| 190 |
+
"learning_rate": 1.295e-05,
|
| 191 |
+
"loss": 4.6641,
|
| 192 |
+
"step": 260
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.0675,
|
| 196 |
+
"grad_norm": 87.5,
|
| 197 |
+
"learning_rate": 1.3450000000000002e-05,
|
| 198 |
+
"loss": 3.9469,
|
| 199 |
+
"step": 270
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.07,
|
| 203 |
+
"grad_norm": 80.0,
|
| 204 |
+
"learning_rate": 1.3950000000000002e-05,
|
| 205 |
+
"loss": 3.2578,
|
| 206 |
+
"step": 280
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.0725,
|
| 210 |
+
"grad_norm": 71.5,
|
| 211 |
+
"learning_rate": 1.4450000000000002e-05,
|
| 212 |
+
"loss": 2.475,
|
| 213 |
+
"step": 290
|
| 214 |
+
},
|
| 215 |
+
{
|
| 216 |
+
"epoch": 0.075,
|
| 217 |
+
"grad_norm": 60.25,
|
| 218 |
+
"learning_rate": 1.4950000000000003e-05,
|
| 219 |
+
"loss": 2.1828,
|
| 220 |
+
"step": 300
|
| 221 |
+
},
|
| 222 |
+
{
|
| 223 |
+
"epoch": 0.0775,
|
| 224 |
+
"grad_norm": 40.0,
|
| 225 |
+
"learning_rate": 1.545e-05,
|
| 226 |
+
"loss": 1.7547,
|
| 227 |
+
"step": 310
|
| 228 |
+
},
|
| 229 |
+
{
|
| 230 |
+
"epoch": 0.08,
|
| 231 |
+
"grad_norm": 32.75,
|
| 232 |
+
"learning_rate": 1.595e-05,
|
| 233 |
+
"loss": 1.5266,
|
| 234 |
+
"step": 320
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"epoch": 0.0825,
|
| 238 |
+
"grad_norm": 14.9375,
|
| 239 |
+
"learning_rate": 1.645e-05,
|
| 240 |
+
"loss": 1.1621,
|
| 241 |
+
"step": 330
|
| 242 |
+
},
|
| 243 |
+
{
|
| 244 |
+
"epoch": 0.085,
|
| 245 |
+
"grad_norm": 12.625,
|
| 246 |
+
"learning_rate": 1.6950000000000002e-05,
|
| 247 |
+
"loss": 1.1262,
|
| 248 |
+
"step": 340
|
| 249 |
+
},
|
| 250 |
+
{
|
| 251 |
+
"epoch": 0.0875,
|
| 252 |
+
"grad_norm": 6.59375,
|
| 253 |
+
"learning_rate": 1.7450000000000004e-05,
|
| 254 |
+
"loss": 1.3047,
|
| 255 |
+
"step": 350
|
| 256 |
+
},
|
| 257 |
+
{
|
| 258 |
+
"epoch": 0.09,
|
| 259 |
+
"grad_norm": 5.0,
|
| 260 |
+
"learning_rate": 1.795e-05,
|
| 261 |
+
"loss": 1.1195,
|
| 262 |
+
"step": 360
|
| 263 |
+
},
|
| 264 |
+
{
|
| 265 |
+
"epoch": 0.0925,
|
| 266 |
+
"grad_norm": 4.78125,
|
| 267 |
+
"learning_rate": 1.845e-05,
|
| 268 |
+
"loss": 0.9902,
|
| 269 |
+
"step": 370
|
| 270 |
+
},
|
| 271 |
+
{
|
| 272 |
+
"epoch": 0.095,
|
| 273 |
+
"grad_norm": 4.9375,
|
| 274 |
+
"learning_rate": 1.8950000000000003e-05,
|
| 275 |
+
"loss": 0.8932,
|
| 276 |
+
"step": 380
|
| 277 |
+
},
|
| 278 |
+
{
|
| 279 |
+
"epoch": 0.0975,
|
| 280 |
+
"grad_norm": 3.859375,
|
| 281 |
+
"learning_rate": 1.9450000000000002e-05,
|
| 282 |
+
"loss": 0.9434,
|
| 283 |
+
"step": 390
|
| 284 |
+
},
|
| 285 |
+
{
|
| 286 |
+
"epoch": 0.1,
|
| 287 |
+
"grad_norm": 5.875,
|
| 288 |
+
"learning_rate": 1.9950000000000004e-05,
|
| 289 |
+
"loss": 0.859,
|
| 290 |
+
"step": 400
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"epoch": 0.1025,
|
| 294 |
+
"grad_norm": 6.71875,
|
| 295 |
+
"learning_rate": 1.99996915764479e-05,
|
| 296 |
+
"loss": 0.9338,
|
| 297 |
+
"step": 410
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"epoch": 0.105,
|
| 301 |
+
"grad_norm": 2.96875,
|
| 302 |
+
"learning_rate": 1.9998625445384374e-05,
|
| 303 |
+
"loss": 0.8354,
|
| 304 |
+
"step": 420
|
| 305 |
+
},
|
| 306 |
+
{
|
| 307 |
+
"epoch": 0.1075,
|
| 308 |
+
"grad_norm": 3.03125,
|
| 309 |
+
"learning_rate": 1.9996797880281935e-05,
|
| 310 |
+
"loss": 0.9184,
|
| 311 |
+
"step": 430
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"epoch": 0.11,
|
| 315 |
+
"grad_norm": 3.53125,
|
| 316 |
+
"learning_rate": 1.999420902031673e-05,
|
| 317 |
+
"loss": 0.9055,
|
| 318 |
+
"step": 440
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 0.1125,
|
| 322 |
+
"grad_norm": 2.640625,
|
| 323 |
+
"learning_rate": 1.9990859062640478e-05,
|
| 324 |
+
"loss": 0.8426,
|
| 325 |
+
"step": 450
|
| 326 |
+
},
|
| 327 |
+
{
|
| 328 |
+
"epoch": 0.115,
|
| 329 |
+
"grad_norm": 3.453125,
|
| 330 |
+
"learning_rate": 1.998674826236542e-05,
|
| 331 |
+
"loss": 0.8184,
|
| 332 |
+
"step": 460
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"epoch": 0.1175,
|
| 336 |
+
"grad_norm": 3.390625,
|
| 337 |
+
"learning_rate": 1.9981876932544918e-05,
|
| 338 |
+
"loss": 0.7809,
|
| 339 |
+
"step": 470
|
| 340 |
+
},
|
| 341 |
+
{
|
| 342 |
+
"epoch": 0.12,
|
| 343 |
+
"grad_norm": 2.1875,
|
| 344 |
+
"learning_rate": 1.997624544414959e-05,
|
| 345 |
+
"loss": 0.733,
|
| 346 |
+
"step": 480
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.1225,
|
| 350 |
+
"grad_norm": 3.515625,
|
| 351 |
+
"learning_rate": 1.9969854226039088e-05,
|
| 352 |
+
"loss": 0.6943,
|
| 353 |
+
"step": 490
|
| 354 |
+
},
|
| 355 |
+
{
|
| 356 |
+
"epoch": 0.125,
|
| 357 |
+
"grad_norm": 6.15625,
|
| 358 |
+
"learning_rate": 1.9962703764929413e-05,
|
| 359 |
+
"loss": 0.7104,
|
| 360 |
+
"step": 500
|
| 361 |
+
},
|
| 362 |
+
{
|
| 363 |
+
"epoch": 0.1275,
|
| 364 |
+
"grad_norm": 2.671875,
|
| 365 |
+
"learning_rate": 1.995479460535586e-05,
|
| 366 |
+
"loss": 0.7924,
|
| 367 |
+
"step": 510
|
| 368 |
+
},
|
| 369 |
+
{
|
| 370 |
+
"epoch": 0.13,
|
| 371 |
+
"grad_norm": 7.125,
|
| 372 |
+
"learning_rate": 1.9946127349631564e-05,
|
| 373 |
+
"loss": 0.8195,
|
| 374 |
+
"step": 520
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 0.1325,
|
| 378 |
+
"grad_norm": 3.65625,
|
| 379 |
+
"learning_rate": 1.9936702657801586e-05,
|
| 380 |
+
"loss": 0.6439,
|
| 381 |
+
"step": 530
|
| 382 |
+
},
|
| 383 |
+
{
|
| 384 |
+
"epoch": 0.135,
|
| 385 |
+
"grad_norm": 6.875,
|
| 386 |
+
"learning_rate": 1.992652124759271e-05,
|
| 387 |
+
"loss": 0.8467,
|
| 388 |
+
"step": 540
|
| 389 |
+
},
|
| 390 |
+
{
|
| 391 |
+
"epoch": 0.1375,
|
| 392 |
+
"grad_norm": 2.6875,
|
| 393 |
+
"learning_rate": 1.9915583894358744e-05,
|
| 394 |
+
"loss": 0.8398,
|
| 395 |
+
"step": 550
|
| 396 |
+
},
|
| 397 |
+
{
|
| 398 |
+
"epoch": 0.14,
|
| 399 |
+
"grad_norm": 3.390625,
|
| 400 |
+
"learning_rate": 1.9903891431021477e-05,
|
| 401 |
+
"loss": 0.7451,
|
| 402 |
+
"step": 560
|
| 403 |
+
},
|
| 404 |
+
{
|
| 405 |
+
"epoch": 0.1425,
|
| 406 |
+
"grad_norm": 2.625,
|
| 407 |
+
"learning_rate": 1.989144474800726e-05,
|
| 408 |
+
"loss": 0.5564,
|
| 409 |
+
"step": 570
|
| 410 |
+
},
|
| 411 |
+
{
|
| 412 |
+
"epoch": 0.145,
|
| 413 |
+
"grad_norm": 2.359375,
|
| 414 |
+
"learning_rate": 1.9878244793179198e-05,
|
| 415 |
+
"loss": 0.6477,
|
| 416 |
+
"step": 580
|
| 417 |
+
},
|
| 418 |
+
{
|
| 419 |
+
"epoch": 0.1475,
|
| 420 |
+
"grad_norm": 6.6875,
|
| 421 |
+
"learning_rate": 1.9864292571764956e-05,
|
| 422 |
+
"loss": 0.6457,
|
| 423 |
+
"step": 590
|
| 424 |
+
},
|
| 425 |
+
{
|
| 426 |
+
"epoch": 0.15,
|
| 427 |
+
"grad_norm": 4.03125,
|
| 428 |
+
"learning_rate": 1.9849589146280212e-05,
|
| 429 |
+
"loss": 0.648,
|
| 430 |
+
"step": 600
|
| 431 |
+
},
|
| 432 |
+
{
|
| 433 |
+
"epoch": 0.1525,
|
| 434 |
+
"grad_norm": 3.265625,
|
| 435 |
+
"learning_rate": 1.9834135636447745e-05,
|
| 436 |
+
"loss": 0.5932,
|
| 437 |
+
"step": 610
|
| 438 |
+
},
|
| 439 |
+
{
|
| 440 |
+
"epoch": 0.155,
|
| 441 |
+
"grad_norm": 2.546875,
|
| 442 |
+
"learning_rate": 1.981793321911216e-05,
|
| 443 |
+
"loss": 0.6838,
|
| 444 |
+
"step": 620
|
| 445 |
+
},
|
| 446 |
+
{
|
| 447 |
+
"epoch": 0.1575,
|
| 448 |
+
"grad_norm": 7.21875,
|
| 449 |
+
"learning_rate": 1.9800983128150263e-05,
|
| 450 |
+
"loss": 0.7277,
|
| 451 |
+
"step": 630
|
| 452 |
+
},
|
| 453 |
+
{
|
| 454 |
+
"epoch": 0.16,
|
| 455 |
+
"grad_norm": 2.25,
|
| 456 |
+
"learning_rate": 1.978328665437711e-05,
|
| 457 |
+
"loss": 0.5895,
|
| 458 |
+
"step": 640
|
| 459 |
+
},
|
| 460 |
+
{
|
| 461 |
+
"epoch": 0.1625,
|
| 462 |
+
"grad_norm": 2.3125,
|
| 463 |
+
"learning_rate": 1.9764845145447687e-05,
|
| 464 |
+
"loss": 0.5982,
|
| 465 |
+
"step": 650
|
| 466 |
+
},
|
| 467 |
+
{
|
| 468 |
+
"epoch": 0.165,
|
| 469 |
+
"grad_norm": 2.5,
|
| 470 |
+
"learning_rate": 1.974566000575431e-05,
|
| 471 |
+
"loss": 0.5367,
|
| 472 |
+
"step": 660
|
| 473 |
+
},
|
| 474 |
+
{
|
| 475 |
+
"epoch": 0.1675,
|
| 476 |
+
"grad_norm": 2.25,
|
| 477 |
+
"learning_rate": 1.9725732696319634e-05,
|
| 478 |
+
"loss": 0.6245,
|
| 479 |
+
"step": 670
|
| 480 |
+
},
|
| 481 |
+
{
|
| 482 |
+
"epoch": 0.17,
|
| 483 |
+
"grad_norm": 1.59375,
|
| 484 |
+
"learning_rate": 1.9705064734685425e-05,
|
| 485 |
+
"loss": 0.5006,
|
| 486 |
+
"step": 680
|
| 487 |
+
},
|
| 488 |
+
{
|
| 489 |
+
"epoch": 0.1725,
|
| 490 |
+
"grad_norm": 0.9609375,
|
| 491 |
+
"learning_rate": 1.9683657694796988e-05,
|
| 492 |
+
"loss": 0.5548,
|
| 493 |
+
"step": 690
|
| 494 |
+
},
|
| 495 |
+
{
|
| 496 |
+
"epoch": 0.175,
|
| 497 |
+
"grad_norm": 3.78125,
|
| 498 |
+
"learning_rate": 1.9661513206883288e-05,
|
| 499 |
+
"loss": 0.5451,
|
| 500 |
+
"step": 700
|
| 501 |
+
},
|
| 502 |
+
{
|
| 503 |
+
"epoch": 0.1775,
|
| 504 |
+
"grad_norm": 2.6875,
|
| 505 |
+
"learning_rate": 1.963863295733281e-05,
|
| 506 |
+
"loss": 0.518,
|
| 507 |
+
"step": 710
|
| 508 |
+
},
|
| 509 |
+
{
|
| 510 |
+
"epoch": 0.18,
|
| 511 |
+
"grad_norm": 2.1875,
|
| 512 |
+
"learning_rate": 1.961501868856515e-05,
|
| 513 |
+
"loss": 0.509,
|
| 514 |
+
"step": 720
|
| 515 |
+
},
|
| 516 |
+
{
|
| 517 |
+
"epoch": 0.1825,
|
| 518 |
+
"grad_norm": 2.40625,
|
| 519 |
+
"learning_rate": 1.9590672198898297e-05,
|
| 520 |
+
"loss": 0.5361,
|
| 521 |
+
"step": 730
|
| 522 |
+
},
|
| 523 |
+
{
|
| 524 |
+
"epoch": 0.185,
|
| 525 |
+
"grad_norm": 2.234375,
|
| 526 |
+
"learning_rate": 1.95655953424117e-05,
|
| 527 |
+
"loss": 0.4965,
|
| 528 |
+
"step": 740
|
| 529 |
+
},
|
| 530 |
+
{
|
| 531 |
+
"epoch": 0.1875,
|
| 532 |
+
"grad_norm": 4.53125,
|
| 533 |
+
"learning_rate": 1.953979002880507e-05,
|
| 534 |
+
"loss": 0.5505,
|
| 535 |
+
"step": 750
|
| 536 |
+
},
|
| 537 |
+
{
|
| 538 |
+
"epoch": 0.19,
|
| 539 |
+
"grad_norm": 2.09375,
|
| 540 |
+
"learning_rate": 1.951325822325295e-05,
|
| 541 |
+
"loss": 0.4783,
|
| 542 |
+
"step": 760
|
| 543 |
+
},
|
| 544 |
+
{
|
| 545 |
+
"epoch": 0.1925,
|
| 546 |
+
"grad_norm": 1.0234375,
|
| 547 |
+
"learning_rate": 1.9486001946255046e-05,
|
| 548 |
+
"loss": 0.452,
|
| 549 |
+
"step": 770
|
| 550 |
+
},
|
| 551 |
+
{
|
| 552 |
+
"epoch": 0.195,
|
| 553 |
+
"grad_norm": 1.765625,
|
| 554 |
+
"learning_rate": 1.945802327348239e-05,
|
| 555 |
+
"loss": 0.4703,
|
| 556 |
+
"step": 780
|
| 557 |
+
},
|
| 558 |
+
{
|
| 559 |
+
"epoch": 0.1975,
|
| 560 |
+
"grad_norm": 1.9375,
|
| 561 |
+
"learning_rate": 1.9429324335619234e-05,
|
| 562 |
+
"loss": 0.4996,
|
| 563 |
+
"step": 790
|
| 564 |
+
},
|
| 565 |
+
{
|
| 566 |
+
"epoch": 0.2,
|
| 567 |
+
"grad_norm": 0.875,
|
| 568 |
+
"learning_rate": 1.93999073182008e-05,
|
| 569 |
+
"loss": 0.4553,
|
| 570 |
+
"step": 800
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 0.2025,
|
| 574 |
+
"grad_norm": 1.2890625,
|
| 575 |
+
"learning_rate": 1.936977446144687e-05,
|
| 576 |
+
"loss": 0.4842,
|
| 577 |
+
"step": 810
|
| 578 |
+
},
|
| 579 |
+
{
|
| 580 |
+
"epoch": 0.205,
|
| 581 |
+
"grad_norm": 1.03125,
|
| 582 |
+
"learning_rate": 1.9338928060091145e-05,
|
| 583 |
+
"loss": 0.4132,
|
| 584 |
+
"step": 820
|
| 585 |
+
},
|
| 586 |
+
{
|
| 587 |
+
"epoch": 0.2075,
|
| 588 |
+
"grad_norm": 1.375,
|
| 589 |
+
"learning_rate": 1.9307370463206512e-05,
|
| 590 |
+
"loss": 0.459,
|
| 591 |
+
"step": 830
|
| 592 |
+
},
|
| 593 |
+
{
|
| 594 |
+
"epoch": 0.21,
|
| 595 |
+
"grad_norm": 1.9609375,
|
| 596 |
+
"learning_rate": 1.9275104074026152e-05,
|
| 597 |
+
"loss": 0.4533,
|
| 598 |
+
"step": 840
|
| 599 |
+
},
|
| 600 |
+
{
|
| 601 |
+
"epoch": 0.2125,
|
| 602 |
+
"grad_norm": 1.5859375,
|
| 603 |
+
"learning_rate": 1.9242131349760536e-05,
|
| 604 |
+
"loss": 0.4134,
|
| 605 |
+
"step": 850
|
| 606 |
+
},
|
| 607 |
+
{
|
| 608 |
+
"epoch": 0.215,
|
| 609 |
+
"grad_norm": 1.25,
|
| 610 |
+
"learning_rate": 1.9208454801410267e-05,
|
| 611 |
+
"loss": 0.3762,
|
| 612 |
+
"step": 860
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.2175,
|
| 616 |
+
"grad_norm": 1.7578125,
|
| 617 |
+
"learning_rate": 1.9174076993574883e-05,
|
| 618 |
+
"loss": 0.4325,
|
| 619 |
+
"step": 870
|
| 620 |
+
},
|
| 621 |
+
{
|
| 622 |
+
"epoch": 0.22,
|
| 623 |
+
"grad_norm": 1.46875,
|
| 624 |
+
"learning_rate": 1.913900054425756e-05,
|
| 625 |
+
"loss": 0.4203,
|
| 626 |
+
"step": 880
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.2225,
|
| 630 |
+
"grad_norm": 1.046875,
|
| 631 |
+
"learning_rate": 1.9103228124665713e-05,
|
| 632 |
+
"loss": 0.3394,
|
| 633 |
+
"step": 890
|
| 634 |
+
},
|
| 635 |
+
{
|
| 636 |
+
"epoch": 0.225,
|
| 637 |
+
"grad_norm": 1.2109375,
|
| 638 |
+
"learning_rate": 1.906676245900759e-05,
|
| 639 |
+
"loss": 0.417,
|
| 640 |
+
"step": 900
|
| 641 |
+
},
|
| 642 |
+
{
|
| 643 |
+
"epoch": 0.2275,
|
| 644 |
+
"grad_norm": 1.53125,
|
| 645 |
+
"learning_rate": 1.9029606324284814e-05,
|
| 646 |
+
"loss": 0.3961,
|
| 647 |
+
"step": 910
|
| 648 |
+
},
|
| 649 |
+
{
|
| 650 |
+
"epoch": 0.23,
|
| 651 |
+
"grad_norm": 2.078125,
|
| 652 |
+
"learning_rate": 1.8991762550080905e-05,
|
| 653 |
+
"loss": 0.4698,
|
| 654 |
+
"step": 920
|
| 655 |
+
},
|
| 656 |
+
{
|
| 657 |
+
"epoch": 0.2325,
|
| 658 |
+
"grad_norm": 1.8828125,
|
| 659 |
+
"learning_rate": 1.895323401834578e-05,
|
| 660 |
+
"loss": 0.3908,
|
| 661 |
+
"step": 930
|
| 662 |
+
},
|
| 663 |
+
{
|
| 664 |
+
"epoch": 0.235,
|
| 665 |
+
"grad_norm": 1.0234375,
|
| 666 |
+
"learning_rate": 1.8914023663176306e-05,
|
| 667 |
+
"loss": 0.449,
|
| 668 |
+
"step": 940
|
| 669 |
+
},
|
| 670 |
+
{
|
| 671 |
+
"epoch": 0.2375,
|
| 672 |
+
"grad_norm": 1.609375,
|
| 673 |
+
"learning_rate": 1.8874134470592836e-05,
|
| 674 |
+
"loss": 0.4523,
|
| 675 |
+
"step": 950
|
| 676 |
+
},
|
| 677 |
+
{
|
| 678 |
+
"epoch": 0.24,
|
| 679 |
+
"grad_norm": 1.078125,
|
| 680 |
+
"learning_rate": 1.8833569478311818e-05,
|
| 681 |
+
"loss": 0.3848,
|
| 682 |
+
"step": 960
|
| 683 |
+
},
|
| 684 |
+
{
|
| 685 |
+
"epoch": 0.2425,
|
| 686 |
+
"grad_norm": 1.2421875,
|
| 687 |
+
"learning_rate": 1.879233177551447e-05,
|
| 688 |
+
"loss": 0.412,
|
| 689 |
+
"step": 970
|
| 690 |
+
},
|
| 691 |
+
{
|
| 692 |
+
"epoch": 0.245,
|
| 693 |
+
"grad_norm": 1.3125,
|
| 694 |
+
"learning_rate": 1.8750424502611528e-05,
|
| 695 |
+
"loss": 0.4363,
|
| 696 |
+
"step": 980
|
| 697 |
+
},
|
| 698 |
+
{
|
| 699 |
+
"epoch": 0.2475,
|
| 700 |
+
"grad_norm": 1.5546875,
|
| 701 |
+
"learning_rate": 1.870785085100406e-05,
|
| 702 |
+
"loss": 0.4064,
|
| 703 |
+
"step": 990
|
| 704 |
+
},
|
| 705 |
+
{
|
| 706 |
+
"epoch": 0.25,
|
| 707 |
+
"grad_norm": 1.0859375,
|
| 708 |
+
"learning_rate": 1.8664614062840472e-05,
|
| 709 |
+
"loss": 0.4113,
|
| 710 |
+
"step": 1000
|
| 711 |
+
},
|
| 712 |
+
{
|
| 713 |
+
"epoch": 0.2525,
|
| 714 |
+
"grad_norm": 45.5,
|
| 715 |
+
"learning_rate": 1.8620717430769586e-05,
|
| 716 |
+
"loss": 0.5559,
|
| 717 |
+
"step": 1010
|
| 718 |
+
},
|
| 719 |
+
{
|
| 720 |
+
"epoch": 0.255,
|
| 721 |
+
"grad_norm": 1.03125,
|
| 722 |
+
"learning_rate": 1.8576164297689877e-05,
|
| 723 |
+
"loss": 0.3405,
|
| 724 |
+
"step": 1020
|
| 725 |
+
},
|
| 726 |
+
{
|
| 727 |
+
"epoch": 0.2575,
|
| 728 |
+
"grad_norm": 1.375,
|
| 729 |
+
"learning_rate": 1.8530958056494934e-05,
|
| 730 |
+
"loss": 0.3953,
|
| 731 |
+
"step": 1030
|
| 732 |
+
},
|
| 733 |
+
{
|
| 734 |
+
"epoch": 0.26,
|
| 735 |
+
"grad_norm": 1.5,
|
| 736 |
+
"learning_rate": 1.8485102149815038e-05,
|
| 737 |
+
"loss": 0.4022,
|
| 738 |
+
"step": 1040
|
| 739 |
+
},
|
| 740 |
+
{
|
| 741 |
+
"epoch": 0.2625,
|
| 742 |
+
"grad_norm": 1.078125,
|
| 743 |
+
"learning_rate": 1.8438600069755027e-05,
|
| 744 |
+
"loss": 0.4464,
|
| 745 |
+
"step": 1050
|
| 746 |
+
},
|
| 747 |
+
{
|
| 748 |
+
"epoch": 0.265,
|
| 749 |
+
"grad_norm": 1.1953125,
|
| 750 |
+
"learning_rate": 1.8391455357628334e-05,
|
| 751 |
+
"loss": 0.4125,
|
| 752 |
+
"step": 1060
|
| 753 |
+
},
|
| 754 |
+
{
|
| 755 |
+
"epoch": 0.2675,
|
| 756 |
+
"grad_norm": 1.28125,
|
| 757 |
+
"learning_rate": 1.834367160368732e-05,
|
| 758 |
+
"loss": 0.4108,
|
| 759 |
+
"step": 1070
|
| 760 |
+
},
|
| 761 |
+
{
|
| 762 |
+
"epoch": 0.27,
|
| 763 |
+
"grad_norm": 1.046875,
|
| 764 |
+
"learning_rate": 1.8295252446849842e-05,
|
| 765 |
+
"loss": 0.4479,
|
| 766 |
+
"step": 1080
|
| 767 |
+
},
|
| 768 |
+
{
|
| 769 |
+
"epoch": 0.2725,
|
| 770 |
+
"grad_norm": 1.2578125,
|
| 771 |
+
"learning_rate": 1.8246201574422164e-05,
|
| 772 |
+
"loss": 0.4491,
|
| 773 |
+
"step": 1090
|
| 774 |
+
},
|
| 775 |
+
{
|
| 776 |
+
"epoch": 0.275,
|
| 777 |
+
"grad_norm": 1.40625,
|
| 778 |
+
"learning_rate": 1.8196522721818128e-05,
|
| 779 |
+
"loss": 0.4101,
|
| 780 |
+
"step": 1100
|
| 781 |
+
},
|
| 782 |
+
{
|
| 783 |
+
"epoch": 0.2775,
|
| 784 |
+
"grad_norm": 0.88671875,
|
| 785 |
+
"learning_rate": 1.8146219672274693e-05,
|
| 786 |
+
"loss": 0.3836,
|
| 787 |
+
"step": 1110
|
| 788 |
+
},
|
| 789 |
+
{
|
| 790 |
+
"epoch": 0.28,
|
| 791 |
+
"grad_norm": 0.84765625,
|
| 792 |
+
"learning_rate": 1.8095296256563845e-05,
|
| 793 |
+
"loss": 0.4341,
|
| 794 |
+
"step": 1120
|
| 795 |
+
},
|
| 796 |
+
{
|
| 797 |
+
"epoch": 0.2825,
|
| 798 |
+
"grad_norm": 1.328125,
|
| 799 |
+
"learning_rate": 1.8043756352700844e-05,
|
| 800 |
+
"loss": 0.3781,
|
| 801 |
+
"step": 1130
|
| 802 |
+
},
|
| 803 |
+
{
|
| 804 |
+
"epoch": 0.285,
|
| 805 |
+
"grad_norm": 1.0390625,
|
| 806 |
+
"learning_rate": 1.799160388564892e-05,
|
| 807 |
+
"loss": 0.3866,
|
| 808 |
+
"step": 1140
|
| 809 |
+
},
|
| 810 |
+
{
|
| 811 |
+
"epoch": 0.2875,
|
| 812 |
+
"grad_norm": 1.1328125,
|
| 813 |
+
"learning_rate": 1.7938842827020347e-05,
|
| 814 |
+
"loss": 0.4173,
|
| 815 |
+
"step": 1150
|
| 816 |
+
},
|
| 817 |
+
{
|
| 818 |
+
"epoch": 0.29,
|
| 819 |
+
"grad_norm": 1.171875,
|
| 820 |
+
"learning_rate": 1.788547719477402e-05,
|
| 821 |
+
"loss": 0.3896,
|
| 822 |
+
"step": 1160
|
| 823 |
+
},
|
| 824 |
+
{
|
| 825 |
+
"epoch": 0.2925,
|
| 826 |
+
"grad_norm": 1.0390625,
|
| 827 |
+
"learning_rate": 1.7831511052909442e-05,
|
| 828 |
+
"loss": 0.393,
|
| 829 |
+
"step": 1170
|
| 830 |
+
},
|
| 831 |
+
{
|
| 832 |
+
"epoch": 0.295,
|
| 833 |
+
"grad_norm": 0.90234375,
|
| 834 |
+
"learning_rate": 1.777694851115726e-05,
|
| 835 |
+
"loss": 0.3941,
|
| 836 |
+
"step": 1180
|
| 837 |
+
},
|
| 838 |
+
{
|
| 839 |
+
"epoch": 0.2975,
|
| 840 |
+
"grad_norm": 1.2578125,
|
| 841 |
+
"learning_rate": 1.772179372466627e-05,
|
| 842 |
+
"loss": 0.3998,
|
| 843 |
+
"step": 1190
|
| 844 |
+
},
|
| 845 |
+
{
|
| 846 |
+
"epoch": 0.3,
|
| 847 |
+
"grad_norm": 1.0390625,
|
| 848 |
+
"learning_rate": 1.7666050893687007e-05,
|
| 849 |
+
"loss": 0.4119,
|
| 850 |
+
"step": 1200
|
| 851 |
+
},
|
| 852 |
+
{
|
| 853 |
+
"epoch": 0.3025,
|
| 854 |
+
"grad_norm": 1.3515625,
|
| 855 |
+
"learning_rate": 1.760972426325187e-05,
|
| 856 |
+
"loss": 0.3405,
|
| 857 |
+
"step": 1210
|
| 858 |
+
},
|
| 859 |
+
{
|
| 860 |
+
"epoch": 0.305,
|
| 861 |
+
"grad_norm": 1.546875,
|
| 862 |
+
"learning_rate": 1.7552818122851838e-05,
|
| 863 |
+
"loss": 0.5383,
|
| 864 |
+
"step": 1220
|
| 865 |
+
},
|
| 866 |
+
{
|
| 867 |
+
"epoch": 0.3075,
|
| 868 |
+
"grad_norm": 1.6796875,
|
| 869 |
+
"learning_rate": 1.7495336806109828e-05,
|
| 870 |
+
"loss": 0.4734,
|
| 871 |
+
"step": 1230
|
| 872 |
+
},
|
| 873 |
+
{
|
| 874 |
+
"epoch": 0.31,
|
| 875 |
+
"grad_norm": 1.28125,
|
| 876 |
+
"learning_rate": 1.7437284690450656e-05,
|
| 877 |
+
"loss": 0.4313,
|
| 878 |
+
"step": 1240
|
| 879 |
+
},
|
| 880 |
+
{
|
| 881 |
+
"epoch": 0.3125,
|
| 882 |
+
"grad_norm": 1.3359375,
|
| 883 |
+
"learning_rate": 1.7378666196767685e-05,
|
| 884 |
+
"loss": 0.432,
|
| 885 |
+
"step": 1250
|
| 886 |
+
},
|
| 887 |
+
{
|
| 888 |
+
"epoch": 0.315,
|
| 889 |
+
"grad_norm": 1.515625,
|
| 890 |
+
"learning_rate": 1.7319485789086164e-05,
|
| 891 |
+
"loss": 0.401,
|
| 892 |
+
"step": 1260
|
| 893 |
+
},
|
| 894 |
+
{
|
| 895 |
+
"epoch": 0.3175,
|
| 896 |
+
"grad_norm": 1.1796875,
|
| 897 |
+
"learning_rate": 1.7259747974223264e-05,
|
| 898 |
+
"loss": 0.3599,
|
| 899 |
+
"step": 1270
|
| 900 |
+
},
|
| 901 |
+
{
|
| 902 |
+
"epoch": 0.32,
|
| 903 |
+
"grad_norm": 1.109375,
|
| 904 |
+
"learning_rate": 1.719945730144487e-05,
|
| 905 |
+
"loss": 0.3218,
|
| 906 |
+
"step": 1280
|
| 907 |
+
},
|
| 908 |
+
{
|
| 909 |
+
"epoch": 0.3225,
|
| 910 |
+
"grad_norm": 0.86328125,
|
| 911 |
+
"learning_rate": 1.7138618362119136e-05,
|
| 912 |
+
"loss": 0.3248,
|
| 913 |
+
"step": 1290
|
| 914 |
+
},
|
| 915 |
+
{
|
| 916 |
+
"epoch": 0.325,
|
| 917 |
+
"grad_norm": 2.59375,
|
| 918 |
+
"learning_rate": 1.7077235789366845e-05,
|
| 919 |
+
"loss": 0.5289,
|
| 920 |
+
"step": 1300
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 0.3275,
|
| 924 |
+
"grad_norm": 1.15625,
|
| 925 |
+
"learning_rate": 1.701531425770856e-05,
|
| 926 |
+
"loss": 0.4097,
|
| 927 |
+
"step": 1310
|
| 928 |
+
},
|
| 929 |
+
{
|
| 930 |
+
"epoch": 0.33,
|
| 931 |
+
"grad_norm": 0.84765625,
|
| 932 |
+
"learning_rate": 1.6952858482708657e-05,
|
| 933 |
+
"loss": 0.3437,
|
| 934 |
+
"step": 1320
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 0.3325,
|
| 938 |
+
"grad_norm": 0.9921875,
|
| 939 |
+
"learning_rate": 1.6889873220616208e-05,
|
| 940 |
+
"loss": 0.3796,
|
| 941 |
+
"step": 1330
|
| 942 |
+
},
|
| 943 |
+
{
|
| 944 |
+
"epoch": 0.335,
|
| 945 |
+
"grad_norm": 1.2421875,
|
| 946 |
+
"learning_rate": 1.682636326800278e-05,
|
| 947 |
+
"loss": 0.3579,
|
| 948 |
+
"step": 1340
|
| 949 |
+
},
|
| 950 |
+
{
|
| 951 |
+
"epoch": 0.3375,
|
| 952 |
+
"grad_norm": 1.6796875,
|
| 953 |
+
"learning_rate": 1.6762333461397157e-05,
|
| 954 |
+
"loss": 0.4375,
|
| 955 |
+
"step": 1350
|
| 956 |
+
},
|
| 957 |
+
{
|
| 958 |
+
"epoch": 0.34,
|
| 959 |
+
"grad_norm": 1.09375,
|
| 960 |
+
"learning_rate": 1.6697788676917007e-05,
|
| 961 |
+
"loss": 0.4392,
|
| 962 |
+
"step": 1360
|
| 963 |
+
},
|
| 964 |
+
{
|
| 965 |
+
"epoch": 0.3425,
|
| 966 |
+
"grad_norm": 0.8359375,
|
| 967 |
+
"learning_rate": 1.6632733829897567e-05,
|
| 968 |
+
"loss": 0.3189,
|
| 969 |
+
"step": 1370
|
| 970 |
+
},
|
| 971 |
+
{
|
| 972 |
+
"epoch": 0.345,
|
| 973 |
+
"grad_norm": 1.296875,
|
| 974 |
+
"learning_rate": 1.656717387451731e-05,
|
| 975 |
+
"loss": 0.3032,
|
| 976 |
+
"step": 1380
|
| 977 |
+
},
|
| 978 |
+
{
|
| 979 |
+
"epoch": 0.3475,
|
| 980 |
+
"grad_norm": 1.390625,
|
| 981 |
+
"learning_rate": 1.650111380342066e-05,
|
| 982 |
+
"loss": 0.4189,
|
| 983 |
+
"step": 1390
|
| 984 |
+
},
|
| 985 |
+
{
|
| 986 |
+
"epoch": 0.35,
|
| 987 |
+
"grad_norm": 1.375,
|
| 988 |
+
"learning_rate": 1.643455864733779e-05,
|
| 989 |
+
"loss": 0.3173,
|
| 990 |
+
"step": 1400
|
| 991 |
+
},
|
| 992 |
+
{
|
| 993 |
+
"epoch": 0.3525,
|
| 994 |
+
"grad_norm": 1.75,
|
| 995 |
+
"learning_rate": 1.636751347470152e-05,
|
| 996 |
+
"loss": 0.3714,
|
| 997 |
+
"step": 1410
|
| 998 |
+
},
|
| 999 |
+
{
|
| 1000 |
+
"epoch": 0.355,
|
| 1001 |
+
"grad_norm": 2.203125,
|
| 1002 |
+
"learning_rate": 1.6299983391261323e-05,
|
| 1003 |
+
"loss": 0.3614,
|
| 1004 |
+
"step": 1420
|
| 1005 |
+
},
|
| 1006 |
+
{
|
| 1007 |
+
"epoch": 0.3575,
|
| 1008 |
+
"grad_norm": 1.2734375,
|
| 1009 |
+
"learning_rate": 1.6231973539694506e-05,
|
| 1010 |
+
"loss": 0.4073,
|
| 1011 |
+
"step": 1430
|
| 1012 |
+
},
|
| 1013 |
+
{
|
| 1014 |
+
"epoch": 0.36,
|
| 1015 |
+
"grad_norm": 1.375,
|
| 1016 |
+
"learning_rate": 1.616348909921457e-05,
|
| 1017 |
+
"loss": 0.3504,
|
| 1018 |
+
"step": 1440
|
| 1019 |
+
},
|
| 1020 |
+
{
|
| 1021 |
+
"epoch": 0.3625,
|
| 1022 |
+
"grad_norm": 1.28125,
|
| 1023 |
+
"learning_rate": 1.6094535285176812e-05,
|
| 1024 |
+
"loss": 0.3663,
|
| 1025 |
+
"step": 1450
|
| 1026 |
+
},
|
| 1027 |
+
{
|
| 1028 |
+
"epoch": 0.365,
|
| 1029 |
+
"grad_norm": 0.8046875,
|
| 1030 |
+
"learning_rate": 1.6025117348681133e-05,
|
| 1031 |
+
"loss": 0.3766,
|
| 1032 |
+
"step": 1460
|
| 1033 |
+
},
|
| 1034 |
+
{
|
| 1035 |
+
"epoch": 0.3675,
|
| 1036 |
+
"grad_norm": 0.94140625,
|
| 1037 |
+
"learning_rate": 1.5955240576172165e-05,
|
| 1038 |
+
"loss": 0.3743,
|
| 1039 |
+
"step": 1470
|
| 1040 |
+
},
|
| 1041 |
+
{
|
| 1042 |
+
"epoch": 0.37,
|
| 1043 |
+
"grad_norm": 0.91796875,
|
| 1044 |
+
"learning_rate": 1.588491028903667e-05,
|
| 1045 |
+
"loss": 0.3538,
|
| 1046 |
+
"step": 1480
|
| 1047 |
+
},
|
| 1048 |
+
{
|
| 1049 |
+
"epoch": 0.3725,
|
| 1050 |
+
"grad_norm": 0.82421875,
|
| 1051 |
+
"learning_rate": 1.5814131843198307e-05,
|
| 1052 |
+
"loss": 0.3508,
|
| 1053 |
+
"step": 1490
|
| 1054 |
+
},
|
| 1055 |
+
{
|
| 1056 |
+
"epoch": 0.375,
|
| 1057 |
+
"grad_norm": 1.578125,
|
| 1058 |
+
"learning_rate": 1.5742910628709757e-05,
|
| 1059 |
+
"loss": 0.4406,
|
| 1060 |
+
"step": 1500
|
| 1061 |
+
},
|
| 1062 |
+
{
|
| 1063 |
+
"epoch": 0.3775,
|
| 1064 |
+
"grad_norm": 0.90234375,
|
| 1065 |
+
"learning_rate": 1.567125206934225e-05,
|
| 1066 |
+
"loss": 0.3217,
|
| 1067 |
+
"step": 1510
|
| 1068 |
+
},
|
| 1069 |
+
{
|
| 1070 |
+
"epoch": 0.38,
|
| 1071 |
+
"grad_norm": 1.0078125,
|
| 1072 |
+
"learning_rate": 1.5599161622172515e-05,
|
| 1073 |
+
"loss": 0.3889,
|
| 1074 |
+
"step": 1520
|
| 1075 |
+
},
|
| 1076 |
+
{
|
| 1077 |
+
"epoch": 0.3825,
|
| 1078 |
+
"grad_norm": 1.4453125,
|
| 1079 |
+
"learning_rate": 1.552664477716722e-05,
|
| 1080 |
+
"loss": 0.3753,
|
| 1081 |
+
"step": 1530
|
| 1082 |
+
},
|
| 1083 |
+
{
|
| 1084 |
+
"epoch": 0.385,
|
| 1085 |
+
"grad_norm": 1.0625,
|
| 1086 |
+
"learning_rate": 1.5453707056764865e-05,
|
| 1087 |
+
"loss": 0.3065,
|
| 1088 |
+
"step": 1540
|
| 1089 |
+
},
|
| 1090 |
+
{
|
| 1091 |
+
"epoch": 0.3875,
|
| 1092 |
+
"grad_norm": 1.3359375,
|
| 1093 |
+
"learning_rate": 1.538035401545525e-05,
|
| 1094 |
+
"loss": 0.362,
|
| 1095 |
+
"step": 1550
|
| 1096 |
+
},
|
| 1097 |
+
{
|
| 1098 |
+
"epoch": 0.39,
|
| 1099 |
+
"grad_norm": 1.4609375,
|
| 1100 |
+
"learning_rate": 1.5306591239356477e-05,
|
| 1101 |
+
"loss": 0.3934,
|
| 1102 |
+
"step": 1560
|
| 1103 |
+
},
|
| 1104 |
+
{
|
| 1105 |
+
"epoch": 0.3925,
|
| 1106 |
+
"grad_norm": 1.4609375,
|
| 1107 |
+
"learning_rate": 1.5232424345789517e-05,
|
| 1108 |
+
"loss": 0.4311,
|
| 1109 |
+
"step": 1570
|
| 1110 |
+
},
|
| 1111 |
+
{
|
| 1112 |
+
"epoch": 0.395,
|
| 1113 |
+
"grad_norm": 1.109375,
|
| 1114 |
+
"learning_rate": 1.5157858982850475e-05,
|
| 1115 |
+
"loss": 0.3334,
|
| 1116 |
+
"step": 1580
|
| 1117 |
+
},
|
| 1118 |
+
{
|
| 1119 |
+
"epoch": 0.3975,
|
| 1120 |
+
"grad_norm": 1.0546875,
|
| 1121 |
+
"learning_rate": 1.5082900828980425e-05,
|
| 1122 |
+
"loss": 0.4175,
|
| 1123 |
+
"step": 1590
|
| 1124 |
+
},
|
| 1125 |
+
{
|
| 1126 |
+
"epoch": 0.4,
|
| 1127 |
+
"grad_norm": 0.6640625,
|
| 1128 |
+
"learning_rate": 1.5007555592532996e-05,
|
| 1129 |
+
"loss": 0.3453,
|
| 1130 |
+
"step": 1600
|
| 1131 |
+
},
|
| 1132 |
+
{
|
| 1133 |
+
"epoch": 0.4025,
|
| 1134 |
+
"grad_norm": 0.9140625,
|
| 1135 |
+
"learning_rate": 1.4931829011339658e-05,
|
| 1136 |
+
"loss": 0.3693,
|
| 1137 |
+
"step": 1610
|
| 1138 |
+
},
|
| 1139 |
+
{
|
| 1140 |
+
"epoch": 0.405,
|
| 1141 |
+
"grad_norm": 0.88671875,
|
| 1142 |
+
"learning_rate": 1.4855726852272754e-05,
|
| 1143 |
+
"loss": 0.3871,
|
| 1144 |
+
"step": 1620
|
| 1145 |
+
},
|
| 1146 |
+
{
|
| 1147 |
+
"epoch": 0.4075,
|
| 1148 |
+
"grad_norm": 0.796875,
|
| 1149 |
+
"learning_rate": 1.4779254910806336e-05,
|
| 1150 |
+
"loss": 0.3512,
|
| 1151 |
+
"step": 1630
|
| 1152 |
+
},
|
| 1153 |
+
{
|
| 1154 |
+
"epoch": 0.41,
|
| 1155 |
+
"grad_norm": 1.3984375,
|
| 1156 |
+
"learning_rate": 1.4702419010574824e-05,
|
| 1157 |
+
"loss": 0.411,
|
| 1158 |
+
"step": 1640
|
| 1159 |
+
},
|
| 1160 |
+
{
|
| 1161 |
+
"epoch": 0.4125,
|
| 1162 |
+
"grad_norm": 1.3828125,
|
| 1163 |
+
"learning_rate": 1.4625225002929501e-05,
|
| 1164 |
+
"loss": 0.383,
|
| 1165 |
+
"step": 1650
|
| 1166 |
+
},
|
| 1167 |
+
{
|
| 1168 |
+
"epoch": 0.415,
|
| 1169 |
+
"grad_norm": 1.3203125,
|
| 1170 |
+
"learning_rate": 1.4547678766492919e-05,
|
| 1171 |
+
"loss": 0.3493,
|
| 1172 |
+
"step": 1660
|
| 1173 |
+
},
|
| 1174 |
+
{
|
| 1175 |
+
"epoch": 0.4175,
|
| 1176 |
+
"grad_norm": 1.359375,
|
| 1177 |
+
"learning_rate": 1.4469786206711213e-05,
|
| 1178 |
+
"loss": 0.3665,
|
| 1179 |
+
"step": 1670
|
| 1180 |
+
},
|
| 1181 |
+
{
|
| 1182 |
+
"epoch": 0.42,
|
| 1183 |
+
"grad_norm": 1.8359375,
|
| 1184 |
+
"learning_rate": 1.4391553255404385e-05,
|
| 1185 |
+
"loss": 0.327,
|
| 1186 |
+
"step": 1680
|
| 1187 |
+
},
|
| 1188 |
+
{
|
| 1189 |
+
"epoch": 0.4225,
|
| 1190 |
+
"grad_norm": 0.953125,
|
| 1191 |
+
"learning_rate": 1.4312985870314568e-05,
|
| 1192 |
+
"loss": 0.3758,
|
| 1193 |
+
"step": 1690
|
| 1194 |
+
},
|
| 1195 |
+
{
|
| 1196 |
+
"epoch": 0.425,
|
| 1197 |
+
"grad_norm": 1.2578125,
|
| 1198 |
+
"learning_rate": 1.4234090034652324e-05,
|
| 1199 |
+
"loss": 0.337,
|
| 1200 |
+
"step": 1700
|
| 1201 |
+
},
|
| 1202 |
+
{
|
| 1203 |
+
"epoch": 0.4275,
|
| 1204 |
+
"grad_norm": 0.875,
|
| 1205 |
+
"learning_rate": 1.4154871756640996e-05,
|
| 1206 |
+
"loss": 0.2903,
|
| 1207 |
+
"step": 1710
|
| 1208 |
+
},
|
| 1209 |
+
{
|
| 1210 |
+
"epoch": 0.43,
|
| 1211 |
+
"grad_norm": 0.9921875,
|
| 1212 |
+
"learning_rate": 1.4075337069059159e-05,
|
| 1213 |
+
"loss": 0.4283,
|
| 1214 |
+
"step": 1720
|
| 1215 |
+
},
|
| 1216 |
+
{
|
| 1217 |
+
"epoch": 0.4325,
|
| 1218 |
+
"grad_norm": 1.0546875,
|
| 1219 |
+
"learning_rate": 1.3995492028781202e-05,
|
| 1220 |
+
"loss": 0.4288,
|
| 1221 |
+
"step": 1730
|
| 1222 |
+
},
|
| 1223 |
+
{
|
| 1224 |
+
"epoch": 0.435,
|
| 1225 |
+
"grad_norm": 1.0625,
|
| 1226 |
+
"learning_rate": 1.3915342716316075e-05,
|
| 1227 |
+
"loss": 0.4184,
|
| 1228 |
+
"step": 1740
|
| 1229 |
+
},
|
| 1230 |
+
{
|
| 1231 |
+
"epoch": 0.4375,
|
| 1232 |
+
"grad_norm": 1.1796875,
|
| 1233 |
+
"learning_rate": 1.3834895235344242e-05,
|
| 1234 |
+
"loss": 0.3089,
|
| 1235 |
+
"step": 1750
|
| 1236 |
+
},
|
| 1237 |
+
{
|
| 1238 |
+
"epoch": 0.44,
|
| 1239 |
+
"grad_norm": 0.8671875,
|
| 1240 |
+
"learning_rate": 1.3754155712252832e-05,
|
| 1241 |
+
"loss": 0.3169,
|
| 1242 |
+
"step": 1760
|
| 1243 |
+
},
|
| 1244 |
+
{
|
| 1245 |
+
"epoch": 0.4425,
|
| 1246 |
+
"grad_norm": 1.1640625,
|
| 1247 |
+
"learning_rate": 1.367313029566913e-05,
|
| 1248 |
+
"loss": 0.3522,
|
| 1249 |
+
"step": 1770
|
| 1250 |
+
},
|
| 1251 |
+
{
|
| 1252 |
+
"epoch": 0.445,
|
| 1253 |
+
"grad_norm": 1.109375,
|
| 1254 |
+
"learning_rate": 1.3591825155992311e-05,
|
| 1255 |
+
"loss": 0.3641,
|
| 1256 |
+
"step": 1780
|
| 1257 |
+
},
|
| 1258 |
+
{
|
| 1259 |
+
"epoch": 0.4475,
|
| 1260 |
+
"grad_norm": 0.84375,
|
| 1261 |
+
"learning_rate": 1.3510246484923547e-05,
|
| 1262 |
+
"loss": 0.3394,
|
| 1263 |
+
"step": 1790
|
| 1264 |
+
},
|
| 1265 |
+
{
|
| 1266 |
+
"epoch": 0.45,
|
| 1267 |
+
"grad_norm": 0.9765625,
|
| 1268 |
+
"learning_rate": 1.3428400494994484e-05,
|
| 1269 |
+
"loss": 0.3107,
|
| 1270 |
+
"step": 1800
|
| 1271 |
+
},
|
| 1272 |
+
{
|
| 1273 |
+
"epoch": 0.4525,
|
| 1274 |
+
"grad_norm": 1.03125,
|
| 1275 |
+
"learning_rate": 1.3346293419094134e-05,
|
| 1276 |
+
"loss": 0.3344,
|
| 1277 |
+
"step": 1810
|
| 1278 |
+
},
|
| 1279 |
+
{
|
| 1280 |
+
"epoch": 0.455,
|
| 1281 |
+
"grad_norm": 1.3046875,
|
| 1282 |
+
"learning_rate": 1.326393150999422e-05,
|
| 1283 |
+
"loss": 0.3364,
|
| 1284 |
+
"step": 1820
|
| 1285 |
+
},
|
| 1286 |
+
{
|
| 1287 |
+
"epoch": 0.4575,
|
| 1288 |
+
"grad_norm": 0.9921875,
|
| 1289 |
+
"learning_rate": 1.3181321039872995e-05,
|
| 1290 |
+
"loss": 0.3396,
|
| 1291 |
+
"step": 1830
|
| 1292 |
+
},
|
| 1293 |
+
{
|
| 1294 |
+
"epoch": 0.46,
|
| 1295 |
+
"grad_norm": 0.90625,
|
| 1296 |
+
"learning_rate": 1.3098468299837599e-05,
|
| 1297 |
+
"loss": 0.3707,
|
| 1298 |
+
"step": 1840
|
| 1299 |
+
},
|
| 1300 |
+
{
|
| 1301 |
+
"epoch": 0.4625,
|
| 1302 |
+
"grad_norm": 1.1953125,
|
| 1303 |
+
"learning_rate": 1.3015379599444958e-05,
|
| 1304 |
+
"loss": 0.3758,
|
| 1305 |
+
"step": 1850
|
| 1306 |
+
},
|
| 1307 |
+
{
|
| 1308 |
+
"epoch": 0.465,
|
| 1309 |
+
"grad_norm": 1.3359375,
|
| 1310 |
+
"learning_rate": 1.2932061266221304e-05,
|
| 1311 |
+
"loss": 0.3853,
|
| 1312 |
+
"step": 1860
|
| 1313 |
+
},
|
| 1314 |
+
{
|
| 1315 |
+
"epoch": 0.4675,
|
| 1316 |
+
"grad_norm": 0.96484375,
|
| 1317 |
+
"learning_rate": 1.2848519645180296e-05,
|
| 1318 |
+
"loss": 0.3396,
|
| 1319 |
+
"step": 1870
|
| 1320 |
+
},
|
| 1321 |
+
{
|
| 1322 |
+
"epoch": 0.47,
|
| 1323 |
+
"grad_norm": 1.453125,
|
| 1324 |
+
"learning_rate": 1.276476109833981e-05,
|
| 1325 |
+
"loss": 0.3241,
|
| 1326 |
+
"step": 1880
|
| 1327 |
+
},
|
| 1328 |
+
{
|
| 1329 |
+
"epoch": 0.4725,
|
| 1330 |
+
"grad_norm": 1.1015625,
|
| 1331 |
+
"learning_rate": 1.268079200423748e-05,
|
| 1332 |
+
"loss": 0.3541,
|
| 1333 |
+
"step": 1890
|
| 1334 |
+
},
|
| 1335 |
+
{
|
| 1336 |
+
"epoch": 0.475,
|
| 1337 |
+
"grad_norm": 1.015625,
|
| 1338 |
+
"learning_rate": 1.2596618757444918e-05,
|
| 1339 |
+
"loss": 0.372,
|
| 1340 |
+
"step": 1900
|
| 1341 |
+
},
|
| 1342 |
+
{
|
| 1343 |
+
"epoch": 0.4775,
|
| 1344 |
+
"grad_norm": 1.46875,
|
| 1345 |
+
"learning_rate": 1.2512247768080756e-05,
|
| 1346 |
+
"loss": 0.4219,
|
| 1347 |
+
"step": 1910
|
| 1348 |
+
},
|
| 1349 |
+
{
|
| 1350 |
+
"epoch": 0.48,
|
| 1351 |
+
"grad_norm": 1.453125,
|
| 1352 |
+
"learning_rate": 1.2427685461322497e-05,
|
| 1353 |
+
"loss": 0.3663,
|
| 1354 |
+
"step": 1920
|
| 1355 |
+
},
|
| 1356 |
+
{
|
| 1357 |
+
"epoch": 0.4825,
|
| 1358 |
+
"grad_norm": 1.015625,
|
| 1359 |
+
"learning_rate": 1.2342938276917187e-05,
|
| 1360 |
+
"loss": 0.3584,
|
| 1361 |
+
"step": 1930
|
| 1362 |
+
},
|
| 1363 |
+
{
|
| 1364 |
+
"epoch": 0.485,
|
| 1365 |
+
"grad_norm": 1.2109375,
|
| 1366 |
+
"learning_rate": 1.2258012668691039e-05,
|
| 1367 |
+
"loss": 0.3991,
|
| 1368 |
+
"step": 1940
|
| 1369 |
+
},
|
| 1370 |
+
{
|
| 1371 |
+
"epoch": 0.4875,
|
| 1372 |
+
"grad_norm": 1.765625,
|
| 1373 |
+
"learning_rate": 1.2172915104057919e-05,
|
| 1374 |
+
"loss": 0.4113,
|
| 1375 |
+
"step": 1950
|
| 1376 |
+
},
|
| 1377 |
+
{
|
| 1378 |
+
"epoch": 0.49,
|
| 1379 |
+
"grad_norm": 1.953125,
|
| 1380 |
+
"learning_rate": 1.2087652063526837e-05,
|
| 1381 |
+
"loss": 0.3711,
|
| 1382 |
+
"step": 1960
|
| 1383 |
+
},
|
| 1384 |
+
{
|
| 1385 |
+
"epoch": 0.4925,
|
| 1386 |
+
"grad_norm": 0.6953125,
|
| 1387 |
+
"learning_rate": 1.2002230040208446e-05,
|
| 1388 |
+
"loss": 0.3359,
|
| 1389 |
+
"step": 1970
|
| 1390 |
+
},
|
| 1391 |
+
{
|
| 1392 |
+
"epoch": 0.495,
|
| 1393 |
+
"grad_norm": 1.0859375,
|
| 1394 |
+
"learning_rate": 1.1916655539320547e-05,
|
| 1395 |
+
"loss": 0.3334,
|
| 1396 |
+
"step": 1980
|
| 1397 |
+
},
|
| 1398 |
+
{
|
| 1399 |
+
"epoch": 0.4975,
|
| 1400 |
+
"grad_norm": 0.93359375,
|
| 1401 |
+
"learning_rate": 1.1830935077692696e-05,
|
| 1402 |
+
"loss": 0.3273,
|
| 1403 |
+
"step": 1990
|
| 1404 |
+
},
|
| 1405 |
+
{
|
| 1406 |
+
"epoch": 0.5,
|
| 1407 |
+
"grad_norm": 0.796875,
|
| 1408 |
+
"learning_rate": 1.174507518326992e-05,
|
| 1409 |
+
"loss": 0.3102,
|
| 1410 |
+
"step": 2000
|
| 1411 |
+
}
|
| 1412 |
+
],
|
| 1413 |
+
"logging_steps": 10,
|
| 1414 |
+
"max_steps": 4000,
|
| 1415 |
+
"num_input_tokens_seen": 0,
|
| 1416 |
+
"num_train_epochs": 9223372036854775807,
|
| 1417 |
+
"save_steps": 2000,
|
| 1418 |
+
"stateful_callbacks": {
|
| 1419 |
+
"TrainerControl": {
|
| 1420 |
+
"args": {
|
| 1421 |
+
"should_epoch_stop": false,
|
| 1422 |
+
"should_evaluate": false,
|
| 1423 |
+
"should_log": false,
|
| 1424 |
+
"should_save": true,
|
| 1425 |
+
"should_training_stop": false
|
| 1426 |
+
},
|
| 1427 |
+
"attributes": {}
|
| 1428 |
+
}
|
| 1429 |
+
},
|
| 1430 |
+
"total_flos": 3.773091725443072e+17,
|
| 1431 |
+
"train_batch_size": 12,
|
| 1432 |
+
"trial_name": null,
|
| 1433 |
+
"trial_params": null
|
| 1434 |
+
}
|
checkpoint-2000/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a800295fc140e3564e52e1657b1b68041f26583e3976b0286ca43212b311f516
|
| 3 |
+
size 5368
|