Training in progress, step 2000, checkpoint
Browse files- checkpoint-2000/config.json +40 -0
- checkpoint-2000/generation_config.json +12 -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 +1433 -0
- checkpoint-2000/training_args.bin +3 -0
checkpoint-2000/config.json
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "meta-llama/Llama-3.2-1B-Instruct",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 128000,
|
| 9 |
+
"eos_token_id": [
|
| 10 |
+
128001,
|
| 11 |
+
128008,
|
| 12 |
+
128009
|
| 13 |
+
],
|
| 14 |
+
"head_dim": 64,
|
| 15 |
+
"hidden_act": "silu",
|
| 16 |
+
"hidden_size": 2048,
|
| 17 |
+
"initializer_range": 0.02,
|
| 18 |
+
"intermediate_size": 8192,
|
| 19 |
+
"max_position_embeddings": 131072,
|
| 20 |
+
"mlp_bias": false,
|
| 21 |
+
"model_type": "llama",
|
| 22 |
+
"num_attention_heads": 32,
|
| 23 |
+
"num_hidden_layers": 16,
|
| 24 |
+
"num_key_value_heads": 8,
|
| 25 |
+
"pretraining_tp": 1,
|
| 26 |
+
"rms_norm_eps": 1e-05,
|
| 27 |
+
"rope_scaling": {
|
| 28 |
+
"factor": 32.0,
|
| 29 |
+
"high_freq_factor": 4.0,
|
| 30 |
+
"low_freq_factor": 1.0,
|
| 31 |
+
"original_max_position_embeddings": 8192,
|
| 32 |
+
"rope_type": "llama3"
|
| 33 |
+
},
|
| 34 |
+
"rope_theta": 500000.0,
|
| 35 |
+
"tie_word_embeddings": true,
|
| 36 |
+
"torch_dtype": "bfloat16",
|
| 37 |
+
"transformers_version": "4.46.3",
|
| 38 |
+
"use_cache": true,
|
| 39 |
+
"vocab_size": 128256
|
| 40 |
+
}
|
checkpoint-2000/generation_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 128000,
|
| 3 |
+
"do_sample": true,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
128001,
|
| 6 |
+
128008,
|
| 7 |
+
128009
|
| 8 |
+
],
|
| 9 |
+
"temperature": 0.6,
|
| 10 |
+
"top_p": 0.9,
|
| 11 |
+
"transformers_version": "4.46.3"
|
| 12 |
+
}
|
checkpoint-2000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:76b57509f3c9954cbdf3aac7fd9748b4b96ec7ed2919a186fda5500a5d86e95f
|
| 3 |
+
size 2471645608
|
checkpoint-2000/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5e2ee0c13c0b517a1a56f6200d8284c90ad273d5723b839afabae3cc3b5749ca
|
| 3 |
+
size 3724602
|
checkpoint-2000/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0a748e6d108c40cd56a220b92e90bf3609e0fdb4f0a58a51298be5058b3c2817
|
| 3 |
+
size 14244
|
checkpoint-2000/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cd38d316dd9f41fef58099ade0ad1204cda80689b93c579586f4bde2645634fe
|
| 3 |
+
size 1064
|
checkpoint-2000/trainer_state.json
ADDED
|
@@ -0,0 +1,1433 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 10.468933943754088,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 2000,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.052321778940483975,
|
| 13 |
+
"grad_norm": 50.25,
|
| 14 |
+
"learning_rate": 8.000000000000001e-07,
|
| 15 |
+
"loss": 3.2827,
|
| 16 |
+
"step": 10
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.10464355788096795,
|
| 20 |
+
"grad_norm": 282.0,
|
| 21 |
+
"learning_rate": 1.6000000000000001e-06,
|
| 22 |
+
"loss": 3.2733,
|
| 23 |
+
"step": 20
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.15696533682145192,
|
| 27 |
+
"grad_norm": 390.0,
|
| 28 |
+
"learning_rate": 2.4000000000000003e-06,
|
| 29 |
+
"loss": 3.1571,
|
| 30 |
+
"step": 30
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.2092871157619359,
|
| 34 |
+
"grad_norm": 322.0,
|
| 35 |
+
"learning_rate": 3.2000000000000003e-06,
|
| 36 |
+
"loss": 3.393,
|
| 37 |
+
"step": 40
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.2616088947024199,
|
| 41 |
+
"grad_norm": 700.0,
|
| 42 |
+
"learning_rate": 4.000000000000001e-06,
|
| 43 |
+
"loss": 3.4911,
|
| 44 |
+
"step": 50
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.31393067364290383,
|
| 48 |
+
"grad_norm": 1704.0,
|
| 49 |
+
"learning_rate": 4.800000000000001e-06,
|
| 50 |
+
"loss": 3.9171,
|
| 51 |
+
"step": 60
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.36625245258338784,
|
| 55 |
+
"grad_norm": 214.0,
|
| 56 |
+
"learning_rate": 5.600000000000001e-06,
|
| 57 |
+
"loss": 4.6157,
|
| 58 |
+
"step": 70
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.4185742315238718,
|
| 62 |
+
"grad_norm": 1608.0,
|
| 63 |
+
"learning_rate": 6.4000000000000006e-06,
|
| 64 |
+
"loss": 4.779,
|
| 65 |
+
"step": 80
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.4708960104643558,
|
| 69 |
+
"grad_norm": 162.0,
|
| 70 |
+
"learning_rate": 7.2000000000000005e-06,
|
| 71 |
+
"loss": 4.2508,
|
| 72 |
+
"step": 90
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.5232177894048398,
|
| 76 |
+
"grad_norm": 149.0,
|
| 77 |
+
"learning_rate": 8.000000000000001e-06,
|
| 78 |
+
"loss": 4.6611,
|
| 79 |
+
"step": 100
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.5755395683453237,
|
| 83 |
+
"grad_norm": 214.0,
|
| 84 |
+
"learning_rate": 8.8e-06,
|
| 85 |
+
"loss": 3.9426,
|
| 86 |
+
"step": 110
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 0.6278613472858077,
|
| 90 |
+
"grad_norm": 408.0,
|
| 91 |
+
"learning_rate": 9.600000000000001e-06,
|
| 92 |
+
"loss": 3.7952,
|
| 93 |
+
"step": 120
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 0.6801831262262917,
|
| 97 |
+
"grad_norm": 71.0,
|
| 98 |
+
"learning_rate": 1.04e-05,
|
| 99 |
+
"loss": 3.647,
|
| 100 |
+
"step": 130
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.7325049051667757,
|
| 104 |
+
"grad_norm": 59.0,
|
| 105 |
+
"learning_rate": 1.1200000000000001e-05,
|
| 106 |
+
"loss": 3.6275,
|
| 107 |
+
"step": 140
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 0.7848266841072596,
|
| 111 |
+
"grad_norm": 45.25,
|
| 112 |
+
"learning_rate": 1.2e-05,
|
| 113 |
+
"loss": 3.4353,
|
| 114 |
+
"step": 150
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.8371484630477436,
|
| 118 |
+
"grad_norm": 47.0,
|
| 119 |
+
"learning_rate": 1.2800000000000001e-05,
|
| 120 |
+
"loss": 3.69,
|
| 121 |
+
"step": 160
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 0.8894702419882276,
|
| 125 |
+
"grad_norm": 33.25,
|
| 126 |
+
"learning_rate": 1.3600000000000002e-05,
|
| 127 |
+
"loss": 3.6108,
|
| 128 |
+
"step": 170
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 0.9417920209287116,
|
| 132 |
+
"grad_norm": 134.0,
|
| 133 |
+
"learning_rate": 1.4400000000000001e-05,
|
| 134 |
+
"loss": 3.7036,
|
| 135 |
+
"step": 180
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.9941137998691956,
|
| 139 |
+
"grad_norm": 45.25,
|
| 140 |
+
"learning_rate": 1.5200000000000002e-05,
|
| 141 |
+
"loss": 3.4621,
|
| 142 |
+
"step": 190
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 1.0464355788096795,
|
| 146 |
+
"grad_norm": 48.25,
|
| 147 |
+
"learning_rate": 1.6000000000000003e-05,
|
| 148 |
+
"loss": 3.4607,
|
| 149 |
+
"step": 200
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 1.0987573577501635,
|
| 153 |
+
"grad_norm": 42.0,
|
| 154 |
+
"learning_rate": 1.6800000000000002e-05,
|
| 155 |
+
"loss": 18.2652,
|
| 156 |
+
"step": 210
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 1.1510791366906474,
|
| 160 |
+
"grad_norm": 41.0,
|
| 161 |
+
"learning_rate": 1.76e-05,
|
| 162 |
+
"loss": 3.6676,
|
| 163 |
+
"step": 220
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 1.2034009156311314,
|
| 167 |
+
"grad_norm": 25.375,
|
| 168 |
+
"learning_rate": 1.8400000000000003e-05,
|
| 169 |
+
"loss": 3.3636,
|
| 170 |
+
"step": 230
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 1.2557226945716153,
|
| 174 |
+
"grad_norm": 35.25,
|
| 175 |
+
"learning_rate": 1.9200000000000003e-05,
|
| 176 |
+
"loss": 3.1784,
|
| 177 |
+
"step": 240
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 1.3080444735120995,
|
| 181 |
+
"grad_norm": 42.0,
|
| 182 |
+
"learning_rate": 2e-05,
|
| 183 |
+
"loss": 3.7673,
|
| 184 |
+
"step": 250
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 1.3603662524525835,
|
| 188 |
+
"grad_norm": 32.75,
|
| 189 |
+
"learning_rate": 1.9999025240093045e-05,
|
| 190 |
+
"loss": 3.7099,
|
| 191 |
+
"step": 260
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 1.4126880313930674,
|
| 195 |
+
"grad_norm": 27.625,
|
| 196 |
+
"learning_rate": 1.9996101150403543e-05,
|
| 197 |
+
"loss": 3.3748,
|
| 198 |
+
"step": 270
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 1.4650098103335514,
|
| 202 |
+
"grad_norm": 24.375,
|
| 203 |
+
"learning_rate": 1.9991228300988586e-05,
|
| 204 |
+
"loss": 3.1567,
|
| 205 |
+
"step": 280
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 1.5173315892740353,
|
| 209 |
+
"grad_norm": 26.625,
|
| 210 |
+
"learning_rate": 1.9984407641819812e-05,
|
| 211 |
+
"loss": 3.1535,
|
| 212 |
+
"step": 290
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 1.5696533682145193,
|
| 216 |
+
"grad_norm": 20.125,
|
| 217 |
+
"learning_rate": 1.9975640502598243e-05,
|
| 218 |
+
"loss": 3.0614,
|
| 219 |
+
"step": 300
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 1.6219751471550032,
|
| 223 |
+
"grad_norm": 23.125,
|
| 224 |
+
"learning_rate": 1.9964928592495046e-05,
|
| 225 |
+
"loss": 3.071,
|
| 226 |
+
"step": 310
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 1.6742969260954872,
|
| 230 |
+
"grad_norm": 32.25,
|
| 231 |
+
"learning_rate": 1.9952273999818312e-05,
|
| 232 |
+
"loss": 3.1193,
|
| 233 |
+
"step": 320
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 1.7266187050359711,
|
| 237 |
+
"grad_norm": 19.375,
|
| 238 |
+
"learning_rate": 1.9937679191605964e-05,
|
| 239 |
+
"loss": 3.0596,
|
| 240 |
+
"step": 330
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 1.7789404839764553,
|
| 244 |
+
"grad_norm": 36.75,
|
| 245 |
+
"learning_rate": 1.9921147013144782e-05,
|
| 246 |
+
"loss": 2.9772,
|
| 247 |
+
"step": 340
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 1.831262262916939,
|
| 251 |
+
"grad_norm": 27.25,
|
| 252 |
+
"learning_rate": 1.9902680687415704e-05,
|
| 253 |
+
"loss": 3.0119,
|
| 254 |
+
"step": 350
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 1.8835840418574232,
|
| 258 |
+
"grad_norm": 26.375,
|
| 259 |
+
"learning_rate": 1.988228381446553e-05,
|
| 260 |
+
"loss": 2.8278,
|
| 261 |
+
"step": 360
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 1.9359058207979072,
|
| 265 |
+
"grad_norm": 27.75,
|
| 266 |
+
"learning_rate": 1.985996037070505e-05,
|
| 267 |
+
"loss": 2.6952,
|
| 268 |
+
"step": 370
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 1.9882275997383911,
|
| 272 |
+
"grad_norm": 27.5,
|
| 273 |
+
"learning_rate": 1.983571470813386e-05,
|
| 274 |
+
"loss": 2.5663,
|
| 275 |
+
"step": 380
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 2.040549378678875,
|
| 279 |
+
"grad_norm": 27.375,
|
| 280 |
+
"learning_rate": 1.9809551553491918e-05,
|
| 281 |
+
"loss": 2.2019,
|
| 282 |
+
"step": 390
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 2.092871157619359,
|
| 286 |
+
"grad_norm": 15.75,
|
| 287 |
+
"learning_rate": 1.9781476007338058e-05,
|
| 288 |
+
"loss": 2.2746,
|
| 289 |
+
"step": 400
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 2.145192936559843,
|
| 293 |
+
"grad_norm": 38.0,
|
| 294 |
+
"learning_rate": 1.9751493543055634e-05,
|
| 295 |
+
"loss": 2.4245,
|
| 296 |
+
"step": 410
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 2.197514715500327,
|
| 300 |
+
"grad_norm": 30.125,
|
| 301 |
+
"learning_rate": 1.9719610005785466e-05,
|
| 302 |
+
"loss": 2.1032,
|
| 303 |
+
"step": 420
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 2.249836494440811,
|
| 307 |
+
"grad_norm": 18.5,
|
| 308 |
+
"learning_rate": 1.9685831611286312e-05,
|
| 309 |
+
"loss": 2.3059,
|
| 310 |
+
"step": 430
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 2.302158273381295,
|
| 314 |
+
"grad_norm": 26.5,
|
| 315 |
+
"learning_rate": 1.9650164944723116e-05,
|
| 316 |
+
"loss": 2.025,
|
| 317 |
+
"step": 440
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 2.354480052321779,
|
| 321 |
+
"grad_norm": 18.625,
|
| 322 |
+
"learning_rate": 1.961261695938319e-05,
|
| 323 |
+
"loss": 2.0708,
|
| 324 |
+
"step": 450
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 2.4068018312622628,
|
| 328 |
+
"grad_norm": 28.25,
|
| 329 |
+
"learning_rate": 1.9573194975320672e-05,
|
| 330 |
+
"loss": 1.926,
|
| 331 |
+
"step": 460
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 2.459123610202747,
|
| 335 |
+
"grad_norm": 24.75,
|
| 336 |
+
"learning_rate": 1.9531906677929472e-05,
|
| 337 |
+
"loss": 1.9348,
|
| 338 |
+
"step": 470
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 2.5114453891432307,
|
| 342 |
+
"grad_norm": 48.5,
|
| 343 |
+
"learning_rate": 1.9488760116444966e-05,
|
| 344 |
+
"loss": 1.8206,
|
| 345 |
+
"step": 480
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 2.563767168083715,
|
| 349 |
+
"grad_norm": 19.875,
|
| 350 |
+
"learning_rate": 1.944376370237481e-05,
|
| 351 |
+
"loss": 1.8137,
|
| 352 |
+
"step": 490
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 2.616088947024199,
|
| 356 |
+
"grad_norm": 25.125,
|
| 357 |
+
"learning_rate": 1.9396926207859085e-05,
|
| 358 |
+
"loss": 1.8683,
|
| 359 |
+
"step": 500
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 2.6684107259646828,
|
| 363 |
+
"grad_norm": 21.5,
|
| 364 |
+
"learning_rate": 1.9348256763960146e-05,
|
| 365 |
+
"loss": 1.6886,
|
| 366 |
+
"step": 510
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 2.720732504905167,
|
| 370 |
+
"grad_norm": 27.625,
|
| 371 |
+
"learning_rate": 1.9297764858882516e-05,
|
| 372 |
+
"loss": 1.6966,
|
| 373 |
+
"step": 520
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 2.7730542838456507,
|
| 377 |
+
"grad_norm": 16.375,
|
| 378 |
+
"learning_rate": 1.9245460336123136e-05,
|
| 379 |
+
"loss": 1.5617,
|
| 380 |
+
"step": 530
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"epoch": 2.825376062786135,
|
| 384 |
+
"grad_norm": 27.5,
|
| 385 |
+
"learning_rate": 1.9191353392552346e-05,
|
| 386 |
+
"loss": 1.4437,
|
| 387 |
+
"step": 540
|
| 388 |
+
},
|
| 389 |
+
{
|
| 390 |
+
"epoch": 2.8776978417266186,
|
| 391 |
+
"grad_norm": 7.59375,
|
| 392 |
+
"learning_rate": 1.913545457642601e-05,
|
| 393 |
+
"loss": 1.3535,
|
| 394 |
+
"step": 550
|
| 395 |
+
},
|
| 396 |
+
{
|
| 397 |
+
"epoch": 2.9300196206671028,
|
| 398 |
+
"grad_norm": 23.625,
|
| 399 |
+
"learning_rate": 1.907777478532909e-05,
|
| 400 |
+
"loss": 1.3569,
|
| 401 |
+
"step": 560
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 2.982341399607587,
|
| 405 |
+
"grad_norm": 21.75,
|
| 406 |
+
"learning_rate": 1.901832526405114e-05,
|
| 407 |
+
"loss": 1.5882,
|
| 408 |
+
"step": 570
|
| 409 |
+
},
|
| 410 |
+
{
|
| 411 |
+
"epoch": 3.0346631785480707,
|
| 412 |
+
"grad_norm": 24.125,
|
| 413 |
+
"learning_rate": 1.895711760239413e-05,
|
| 414 |
+
"loss": 1.4584,
|
| 415 |
+
"step": 580
|
| 416 |
+
},
|
| 417 |
+
{
|
| 418 |
+
"epoch": 3.0869849574885544,
|
| 419 |
+
"grad_norm": 24.375,
|
| 420 |
+
"learning_rate": 1.889416373291298e-05,
|
| 421 |
+
"loss": 1.266,
|
| 422 |
+
"step": 590
|
| 423 |
+
},
|
| 424 |
+
{
|
| 425 |
+
"epoch": 3.1393067364290386,
|
| 426 |
+
"grad_norm": 9.3125,
|
| 427 |
+
"learning_rate": 1.8829475928589272e-05,
|
| 428 |
+
"loss": 1.1857,
|
| 429 |
+
"step": 600
|
| 430 |
+
},
|
| 431 |
+
{
|
| 432 |
+
"epoch": 3.1916285153695227,
|
| 433 |
+
"grad_norm": 12.875,
|
| 434 |
+
"learning_rate": 1.8763066800438638e-05,
|
| 435 |
+
"loss": 1.1123,
|
| 436 |
+
"step": 610
|
| 437 |
+
},
|
| 438 |
+
{
|
| 439 |
+
"epoch": 3.2439502943100065,
|
| 440 |
+
"grad_norm": 19.5,
|
| 441 |
+
"learning_rate": 1.869494929505219e-05,
|
| 442 |
+
"loss": 1.1477,
|
| 443 |
+
"step": 620
|
| 444 |
+
},
|
| 445 |
+
{
|
| 446 |
+
"epoch": 3.2962720732504907,
|
| 447 |
+
"grad_norm": 14.625,
|
| 448 |
+
"learning_rate": 1.8625136692072577e-05,
|
| 449 |
+
"loss": 1.1091,
|
| 450 |
+
"step": 630
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"epoch": 3.3485938521909744,
|
| 454 |
+
"grad_norm": 5.90625,
|
| 455 |
+
"learning_rate": 1.855364260160507e-05,
|
| 456 |
+
"loss": 1.1665,
|
| 457 |
+
"step": 640
|
| 458 |
+
},
|
| 459 |
+
{
|
| 460 |
+
"epoch": 3.4009156311314586,
|
| 461 |
+
"grad_norm": 12.6875,
|
| 462 |
+
"learning_rate": 1.848048096156426e-05,
|
| 463 |
+
"loss": 1.0396,
|
| 464 |
+
"step": 650
|
| 465 |
+
},
|
| 466 |
+
{
|
| 467 |
+
"epoch": 3.4532374100719423,
|
| 468 |
+
"grad_norm": 6.65625,
|
| 469 |
+
"learning_rate": 1.8405666034956842e-05,
|
| 470 |
+
"loss": 1.0918,
|
| 471 |
+
"step": 660
|
| 472 |
+
},
|
| 473 |
+
{
|
| 474 |
+
"epoch": 3.5055591890124265,
|
| 475 |
+
"grad_norm": 18.625,
|
| 476 |
+
"learning_rate": 1.8329212407100996e-05,
|
| 477 |
+
"loss": 1.1203,
|
| 478 |
+
"step": 670
|
| 479 |
+
},
|
| 480 |
+
{
|
| 481 |
+
"epoch": 3.5578809679529106,
|
| 482 |
+
"grad_norm": 11.875,
|
| 483 |
+
"learning_rate": 1.8251134982782952e-05,
|
| 484 |
+
"loss": 1.1369,
|
| 485 |
+
"step": 680
|
| 486 |
+
},
|
| 487 |
+
{
|
| 488 |
+
"epoch": 3.6102027468933944,
|
| 489 |
+
"grad_norm": 17.125,
|
| 490 |
+
"learning_rate": 1.8171448983351284e-05,
|
| 491 |
+
"loss": 1.1366,
|
| 492 |
+
"step": 690
|
| 493 |
+
},
|
| 494 |
+
{
|
| 495 |
+
"epoch": 3.662524525833878,
|
| 496 |
+
"grad_norm": 47.25,
|
| 497 |
+
"learning_rate": 1.8090169943749477e-05,
|
| 498 |
+
"loss": 1.3546,
|
| 499 |
+
"step": 700
|
| 500 |
+
},
|
| 501 |
+
{
|
| 502 |
+
"epoch": 3.7148463047743623,
|
| 503 |
+
"grad_norm": 10.75,
|
| 504 |
+
"learning_rate": 1.8007313709487334e-05,
|
| 505 |
+
"loss": 1.2935,
|
| 506 |
+
"step": 710
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 3.7671680837148465,
|
| 510 |
+
"grad_norm": 11.125,
|
| 511 |
+
"learning_rate": 1.792289643355191e-05,
|
| 512 |
+
"loss": 1.024,
|
| 513 |
+
"step": 720
|
| 514 |
+
},
|
| 515 |
+
{
|
| 516 |
+
"epoch": 3.81948986265533,
|
| 517 |
+
"grad_norm": 11.75,
|
| 518 |
+
"learning_rate": 1.78369345732584e-05,
|
| 519 |
+
"loss": 1.0973,
|
| 520 |
+
"step": 730
|
| 521 |
+
},
|
| 522 |
+
{
|
| 523 |
+
"epoch": 3.8718116415958144,
|
| 524 |
+
"grad_norm": 27.0,
|
| 525 |
+
"learning_rate": 1.7749444887041797e-05,
|
| 526 |
+
"loss": 1.0991,
|
| 527 |
+
"step": 740
|
| 528 |
+
},
|
| 529 |
+
{
|
| 530 |
+
"epoch": 3.924133420536298,
|
| 531 |
+
"grad_norm": 14.5,
|
| 532 |
+
"learning_rate": 1.766044443118978e-05,
|
| 533 |
+
"loss": 1.0702,
|
| 534 |
+
"step": 750
|
| 535 |
+
},
|
| 536 |
+
{
|
| 537 |
+
"epoch": 3.9764551994767823,
|
| 538 |
+
"grad_norm": 9.0,
|
| 539 |
+
"learning_rate": 1.7569950556517566e-05,
|
| 540 |
+
"loss": 1.0792,
|
| 541 |
+
"step": 760
|
| 542 |
+
},
|
| 543 |
+
{
|
| 544 |
+
"epoch": 4.028776978417266,
|
| 545 |
+
"grad_norm": 10.6875,
|
| 546 |
+
"learning_rate": 1.747798090498532e-05,
|
| 547 |
+
"loss": 0.9713,
|
| 548 |
+
"step": 770
|
| 549 |
+
},
|
| 550 |
+
{
|
| 551 |
+
"epoch": 4.08109875735775,
|
| 552 |
+
"grad_norm": 11.125,
|
| 553 |
+
"learning_rate": 1.7384553406258842e-05,
|
| 554 |
+
"loss": 0.9195,
|
| 555 |
+
"step": 780
|
| 556 |
+
},
|
| 557 |
+
{
|
| 558 |
+
"epoch": 4.133420536298234,
|
| 559 |
+
"grad_norm": 9.4375,
|
| 560 |
+
"learning_rate": 1.7289686274214116e-05,
|
| 561 |
+
"loss": 0.8511,
|
| 562 |
+
"step": 790
|
| 563 |
+
},
|
| 564 |
+
{
|
| 565 |
+
"epoch": 4.185742315238718,
|
| 566 |
+
"grad_norm": 11.5,
|
| 567 |
+
"learning_rate": 1.7193398003386514e-05,
|
| 568 |
+
"loss": 0.8347,
|
| 569 |
+
"step": 800
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 4.238064094179202,
|
| 573 |
+
"grad_norm": 7.71875,
|
| 574 |
+
"learning_rate": 1.709570736536521e-05,
|
| 575 |
+
"loss": 0.861,
|
| 576 |
+
"step": 810
|
| 577 |
+
},
|
| 578 |
+
{
|
| 579 |
+
"epoch": 4.290385873119686,
|
| 580 |
+
"grad_norm": 8.5,
|
| 581 |
+
"learning_rate": 1.6996633405133656e-05,
|
| 582 |
+
"loss": 0.8834,
|
| 583 |
+
"step": 820
|
| 584 |
+
},
|
| 585 |
+
{
|
| 586 |
+
"epoch": 4.34270765206017,
|
| 587 |
+
"grad_norm": 9.9375,
|
| 588 |
+
"learning_rate": 1.68961954373567e-05,
|
| 589 |
+
"loss": 0.8895,
|
| 590 |
+
"step": 830
|
| 591 |
+
},
|
| 592 |
+
{
|
| 593 |
+
"epoch": 4.395029431000654,
|
| 594 |
+
"grad_norm": 8.875,
|
| 595 |
+
"learning_rate": 1.6794413042615168e-05,
|
| 596 |
+
"loss": 0.8292,
|
| 597 |
+
"step": 840
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 4.447351209941138,
|
| 601 |
+
"grad_norm": 9.125,
|
| 602 |
+
"learning_rate": 1.6691306063588583e-05,
|
| 603 |
+
"loss": 0.9389,
|
| 604 |
+
"step": 850
|
| 605 |
+
},
|
| 606 |
+
{
|
| 607 |
+
"epoch": 4.499672988881622,
|
| 608 |
+
"grad_norm": 5.875,
|
| 609 |
+
"learning_rate": 1.6586894601186804e-05,
|
| 610 |
+
"loss": 0.8355,
|
| 611 |
+
"step": 860
|
| 612 |
+
},
|
| 613 |
+
{
|
| 614 |
+
"epoch": 4.551994767822106,
|
| 615 |
+
"grad_norm": 8.4375,
|
| 616 |
+
"learning_rate": 1.6481199010631312e-05,
|
| 617 |
+
"loss": 0.8835,
|
| 618 |
+
"step": 870
|
| 619 |
+
},
|
| 620 |
+
{
|
| 621 |
+
"epoch": 4.60431654676259,
|
| 622 |
+
"grad_norm": 14.0625,
|
| 623 |
+
"learning_rate": 1.63742398974869e-05,
|
| 624 |
+
"loss": 0.8394,
|
| 625 |
+
"step": 880
|
| 626 |
+
},
|
| 627 |
+
{
|
| 628 |
+
"epoch": 4.656638325703074,
|
| 629 |
+
"grad_norm": 6.3125,
|
| 630 |
+
"learning_rate": 1.6266038113644605e-05,
|
| 631 |
+
"loss": 0.8661,
|
| 632 |
+
"step": 890
|
| 633 |
+
},
|
| 634 |
+
{
|
| 635 |
+
"epoch": 4.708960104643558,
|
| 636 |
+
"grad_norm": 5.4375,
|
| 637 |
+
"learning_rate": 1.6156614753256583e-05,
|
| 638 |
+
"loss": 0.828,
|
| 639 |
+
"step": 900
|
| 640 |
+
},
|
| 641 |
+
{
|
| 642 |
+
"epoch": 4.761281883584042,
|
| 643 |
+
"grad_norm": 6.5,
|
| 644 |
+
"learning_rate": 1.6045991148623752e-05,
|
| 645 |
+
"loss": 0.8028,
|
| 646 |
+
"step": 910
|
| 647 |
+
},
|
| 648 |
+
{
|
| 649 |
+
"epoch": 4.8136036625245255,
|
| 650 |
+
"grad_norm": 6.4375,
|
| 651 |
+
"learning_rate": 1.5934188866037017e-05,
|
| 652 |
+
"loss": 0.816,
|
| 653 |
+
"step": 920
|
| 654 |
+
},
|
| 655 |
+
{
|
| 656 |
+
"epoch": 4.86592544146501,
|
| 657 |
+
"grad_norm": 11.9375,
|
| 658 |
+
"learning_rate": 1.5821229701572897e-05,
|
| 659 |
+
"loss": 0.8221,
|
| 660 |
+
"step": 930
|
| 661 |
+
},
|
| 662 |
+
{
|
| 663 |
+
"epoch": 4.918247220405494,
|
| 664 |
+
"grad_norm": 8.8125,
|
| 665 |
+
"learning_rate": 1.570713567684432e-05,
|
| 666 |
+
"loss": 0.8955,
|
| 667 |
+
"step": 940
|
| 668 |
+
},
|
| 669 |
+
{
|
| 670 |
+
"epoch": 4.970568999345978,
|
| 671 |
+
"grad_norm": 5.4375,
|
| 672 |
+
"learning_rate": 1.5591929034707468e-05,
|
| 673 |
+
"loss": 0.8606,
|
| 674 |
+
"step": 950
|
| 675 |
+
},
|
| 676 |
+
{
|
| 677 |
+
"epoch": 5.022890778286461,
|
| 678 |
+
"grad_norm": 13.5,
|
| 679 |
+
"learning_rate": 1.5475632234925505e-05,
|
| 680 |
+
"loss": 0.7756,
|
| 681 |
+
"step": 960
|
| 682 |
+
},
|
| 683 |
+
{
|
| 684 |
+
"epoch": 5.075212557226946,
|
| 685 |
+
"grad_norm": 9.5625,
|
| 686 |
+
"learning_rate": 1.5358267949789968e-05,
|
| 687 |
+
"loss": 0.7097,
|
| 688 |
+
"step": 970
|
| 689 |
+
},
|
| 690 |
+
{
|
| 691 |
+
"epoch": 5.12753433616743,
|
| 692 |
+
"grad_norm": 7.5,
|
| 693 |
+
"learning_rate": 1.5239859059700794e-05,
|
| 694 |
+
"loss": 0.6738,
|
| 695 |
+
"step": 980
|
| 696 |
+
},
|
| 697 |
+
{
|
| 698 |
+
"epoch": 5.179856115107913,
|
| 699 |
+
"grad_norm": 10.9375,
|
| 700 |
+
"learning_rate": 1.5120428648705716e-05,
|
| 701 |
+
"loss": 0.6686,
|
| 702 |
+
"step": 990
|
| 703 |
+
},
|
| 704 |
+
{
|
| 705 |
+
"epoch": 5.232177894048398,
|
| 706 |
+
"grad_norm": 9.25,
|
| 707 |
+
"learning_rate": 1.5000000000000002e-05,
|
| 708 |
+
"loss": 0.6961,
|
| 709 |
+
"step": 1000
|
| 710 |
+
},
|
| 711 |
+
{
|
| 712 |
+
"epoch": 5.284499672988882,
|
| 713 |
+
"grad_norm": 8.75,
|
| 714 |
+
"learning_rate": 1.4878596591387329e-05,
|
| 715 |
+
"loss": 0.707,
|
| 716 |
+
"step": 1010
|
| 717 |
+
},
|
| 718 |
+
{
|
| 719 |
+
"epoch": 5.3368214519293655,
|
| 720 |
+
"grad_norm": 12.25,
|
| 721 |
+
"learning_rate": 1.4756242090702756e-05,
|
| 722 |
+
"loss": 0.7172,
|
| 723 |
+
"step": 1020
|
| 724 |
+
},
|
| 725 |
+
{
|
| 726 |
+
"epoch": 5.389143230869849,
|
| 727 |
+
"grad_norm": 6.59375,
|
| 728 |
+
"learning_rate": 1.463296035119862e-05,
|
| 729 |
+
"loss": 0.6868,
|
| 730 |
+
"step": 1030
|
| 731 |
+
},
|
| 732 |
+
{
|
| 733 |
+
"epoch": 5.441465009810334,
|
| 734 |
+
"grad_norm": 5.3125,
|
| 735 |
+
"learning_rate": 1.4508775406894308e-05,
|
| 736 |
+
"loss": 0.6678,
|
| 737 |
+
"step": 1040
|
| 738 |
+
},
|
| 739 |
+
{
|
| 740 |
+
"epoch": 5.493786788750818,
|
| 741 |
+
"grad_norm": 5.84375,
|
| 742 |
+
"learning_rate": 1.4383711467890776e-05,
|
| 743 |
+
"loss": 0.6353,
|
| 744 |
+
"step": 1050
|
| 745 |
+
},
|
| 746 |
+
{
|
| 747 |
+
"epoch": 5.546108567691301,
|
| 748 |
+
"grad_norm": 6.6875,
|
| 749 |
+
"learning_rate": 1.4257792915650728e-05,
|
| 750 |
+
"loss": 0.702,
|
| 751 |
+
"step": 1060
|
| 752 |
+
},
|
| 753 |
+
{
|
| 754 |
+
"epoch": 5.598430346631785,
|
| 755 |
+
"grad_norm": 5.09375,
|
| 756 |
+
"learning_rate": 1.413104429824542e-05,
|
| 757 |
+
"loss": 0.7138,
|
| 758 |
+
"step": 1070
|
| 759 |
+
},
|
| 760 |
+
{
|
| 761 |
+
"epoch": 5.65075212557227,
|
| 762 |
+
"grad_norm": 10.0625,
|
| 763 |
+
"learning_rate": 1.4003490325568953e-05,
|
| 764 |
+
"loss": 0.6754,
|
| 765 |
+
"step": 1080
|
| 766 |
+
},
|
| 767 |
+
{
|
| 768 |
+
"epoch": 5.703073904512753,
|
| 769 |
+
"grad_norm": 10.625,
|
| 770 |
+
"learning_rate": 1.3875155864521031e-05,
|
| 771 |
+
"loss": 0.693,
|
| 772 |
+
"step": 1090
|
| 773 |
+
},
|
| 774 |
+
{
|
| 775 |
+
"epoch": 5.755395683453237,
|
| 776 |
+
"grad_norm": 5.9375,
|
| 777 |
+
"learning_rate": 1.3746065934159123e-05,
|
| 778 |
+
"loss": 0.6458,
|
| 779 |
+
"step": 1100
|
| 780 |
+
},
|
| 781 |
+
{
|
| 782 |
+
"epoch": 5.807717462393722,
|
| 783 |
+
"grad_norm": 8.3125,
|
| 784 |
+
"learning_rate": 1.3616245700820922e-05,
|
| 785 |
+
"loss": 0.673,
|
| 786 |
+
"step": 1110
|
| 787 |
+
},
|
| 788 |
+
{
|
| 789 |
+
"epoch": 5.8600392413342055,
|
| 790 |
+
"grad_norm": 7.71875,
|
| 791 |
+
"learning_rate": 1.3485720473218153e-05,
|
| 792 |
+
"loss": 0.7141,
|
| 793 |
+
"step": 1120
|
| 794 |
+
},
|
| 795 |
+
{
|
| 796 |
+
"epoch": 5.912361020274689,
|
| 797 |
+
"grad_norm": 11.4375,
|
| 798 |
+
"learning_rate": 1.3354515697502552e-05,
|
| 799 |
+
"loss": 0.664,
|
| 800 |
+
"step": 1130
|
| 801 |
+
},
|
| 802 |
+
{
|
| 803 |
+
"epoch": 5.964682799215173,
|
| 804 |
+
"grad_norm": 7.28125,
|
| 805 |
+
"learning_rate": 1.3222656952305113e-05,
|
| 806 |
+
"loss": 0.6634,
|
| 807 |
+
"step": 1140
|
| 808 |
+
},
|
| 809 |
+
{
|
| 810 |
+
"epoch": 6.017004578155658,
|
| 811 |
+
"grad_norm": 7.75,
|
| 812 |
+
"learning_rate": 1.3090169943749475e-05,
|
| 813 |
+
"loss": 0.5929,
|
| 814 |
+
"step": 1150
|
| 815 |
+
},
|
| 816 |
+
{
|
| 817 |
+
"epoch": 6.069326357096141,
|
| 818 |
+
"grad_norm": 10.625,
|
| 819 |
+
"learning_rate": 1.2957080500440469e-05,
|
| 820 |
+
"loss": 0.557,
|
| 821 |
+
"step": 1160
|
| 822 |
+
},
|
| 823 |
+
{
|
| 824 |
+
"epoch": 6.121648136036625,
|
| 825 |
+
"grad_norm": 4.90625,
|
| 826 |
+
"learning_rate": 1.2823414568428767e-05,
|
| 827 |
+
"loss": 0.5868,
|
| 828 |
+
"step": 1170
|
| 829 |
+
},
|
| 830 |
+
{
|
| 831 |
+
"epoch": 6.173969914977109,
|
| 832 |
+
"grad_norm": 7.40625,
|
| 833 |
+
"learning_rate": 1.2689198206152657e-05,
|
| 834 |
+
"loss": 0.5627,
|
| 835 |
+
"step": 1180
|
| 836 |
+
},
|
| 837 |
+
{
|
| 838 |
+
"epoch": 6.226291693917593,
|
| 839 |
+
"grad_norm": 7.53125,
|
| 840 |
+
"learning_rate": 1.2554457579357906e-05,
|
| 841 |
+
"loss": 0.5412,
|
| 842 |
+
"step": 1190
|
| 843 |
+
},
|
| 844 |
+
{
|
| 845 |
+
"epoch": 6.278613472858077,
|
| 846 |
+
"grad_norm": 6.03125,
|
| 847 |
+
"learning_rate": 1.2419218955996677e-05,
|
| 848 |
+
"loss": 0.5498,
|
| 849 |
+
"step": 1200
|
| 850 |
+
},
|
| 851 |
+
{
|
| 852 |
+
"epoch": 6.330935251798561,
|
| 853 |
+
"grad_norm": 7.5625,
|
| 854 |
+
"learning_rate": 1.2283508701106559e-05,
|
| 855 |
+
"loss": 0.589,
|
| 856 |
+
"step": 1210
|
| 857 |
+
},
|
| 858 |
+
{
|
| 859 |
+
"epoch": 6.3832570307390455,
|
| 860 |
+
"grad_norm": 6.90625,
|
| 861 |
+
"learning_rate": 1.2147353271670634e-05,
|
| 862 |
+
"loss": 0.5565,
|
| 863 |
+
"step": 1220
|
| 864 |
+
},
|
| 865 |
+
{
|
| 866 |
+
"epoch": 6.435578809679529,
|
| 867 |
+
"grad_norm": 8.0625,
|
| 868 |
+
"learning_rate": 1.2010779211459649e-05,
|
| 869 |
+
"loss": 0.564,
|
| 870 |
+
"step": 1230
|
| 871 |
+
},
|
| 872 |
+
{
|
| 873 |
+
"epoch": 6.487900588620013,
|
| 874 |
+
"grad_norm": 4.90625,
|
| 875 |
+
"learning_rate": 1.187381314585725e-05,
|
| 876 |
+
"loss": 0.5612,
|
| 877 |
+
"step": 1240
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 6.540222367560497,
|
| 881 |
+
"grad_norm": 5.34375,
|
| 882 |
+
"learning_rate": 1.1736481776669307e-05,
|
| 883 |
+
"loss": 0.5532,
|
| 884 |
+
"step": 1250
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 6.592544146500981,
|
| 888 |
+
"grad_norm": 6.0625,
|
| 889 |
+
"learning_rate": 1.159881187691835e-05,
|
| 890 |
+
"loss": 0.5796,
|
| 891 |
+
"step": 1260
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 6.644865925441465,
|
| 895 |
+
"grad_norm": 7.125,
|
| 896 |
+
"learning_rate": 1.1460830285624119e-05,
|
| 897 |
+
"loss": 0.5547,
|
| 898 |
+
"step": 1270
|
| 899 |
+
},
|
| 900 |
+
{
|
| 901 |
+
"epoch": 6.697187704381949,
|
| 902 |
+
"grad_norm": 6.75,
|
| 903 |
+
"learning_rate": 1.1322563902571227e-05,
|
| 904 |
+
"loss": 0.5652,
|
| 905 |
+
"step": 1280
|
| 906 |
+
},
|
| 907 |
+
{
|
| 908 |
+
"epoch": 6.7495094833224325,
|
| 909 |
+
"grad_norm": 7.59375,
|
| 910 |
+
"learning_rate": 1.1184039683065014e-05,
|
| 911 |
+
"loss": 0.5366,
|
| 912 |
+
"step": 1290
|
| 913 |
+
},
|
| 914 |
+
{
|
| 915 |
+
"epoch": 6.801831262262917,
|
| 916 |
+
"grad_norm": 4.6875,
|
| 917 |
+
"learning_rate": 1.1045284632676535e-05,
|
| 918 |
+
"loss": 0.5645,
|
| 919 |
+
"step": 1300
|
| 920 |
+
},
|
| 921 |
+
{
|
| 922 |
+
"epoch": 6.854153041203401,
|
| 923 |
+
"grad_norm": 4.84375,
|
| 924 |
+
"learning_rate": 1.0906325801977804e-05,
|
| 925 |
+
"loss": 0.5297,
|
| 926 |
+
"step": 1310
|
| 927 |
+
},
|
| 928 |
+
{
|
| 929 |
+
"epoch": 6.906474820143885,
|
| 930 |
+
"grad_norm": 5.9375,
|
| 931 |
+
"learning_rate": 1.0767190281268187e-05,
|
| 932 |
+
"loss": 0.5563,
|
| 933 |
+
"step": 1320
|
| 934 |
+
},
|
| 935 |
+
{
|
| 936 |
+
"epoch": 6.958796599084369,
|
| 937 |
+
"grad_norm": 7.78125,
|
| 938 |
+
"learning_rate": 1.0627905195293135e-05,
|
| 939 |
+
"loss": 0.5587,
|
| 940 |
+
"step": 1330
|
| 941 |
+
},
|
| 942 |
+
{
|
| 943 |
+
"epoch": 7.011118378024853,
|
| 944 |
+
"grad_norm": 5.34375,
|
| 945 |
+
"learning_rate": 1.0488497697956134e-05,
|
| 946 |
+
"loss": 0.5201,
|
| 947 |
+
"step": 1340
|
| 948 |
+
},
|
| 949 |
+
{
|
| 950 |
+
"epoch": 7.063440156965337,
|
| 951 |
+
"grad_norm": 8.0625,
|
| 952 |
+
"learning_rate": 1.0348994967025012e-05,
|
| 953 |
+
"loss": 0.4576,
|
| 954 |
+
"step": 1350
|
| 955 |
+
},
|
| 956 |
+
{
|
| 957 |
+
"epoch": 7.11576193590582,
|
| 958 |
+
"grad_norm": 5.09375,
|
| 959 |
+
"learning_rate": 1.0209424198833571e-05,
|
| 960 |
+
"loss": 0.5009,
|
| 961 |
+
"step": 1360
|
| 962 |
+
},
|
| 963 |
+
{
|
| 964 |
+
"epoch": 7.168083714846305,
|
| 965 |
+
"grad_norm": 4.0625,
|
| 966 |
+
"learning_rate": 1.0069812602979617e-05,
|
| 967 |
+
"loss": 0.4734,
|
| 968 |
+
"step": 1370
|
| 969 |
+
},
|
| 970 |
+
{
|
| 971 |
+
"epoch": 7.220405493786789,
|
| 972 |
+
"grad_norm": 3.609375,
|
| 973 |
+
"learning_rate": 9.930187397020385e-06,
|
| 974 |
+
"loss": 0.4506,
|
| 975 |
+
"step": 1380
|
| 976 |
+
},
|
| 977 |
+
{
|
| 978 |
+
"epoch": 7.2727272727272725,
|
| 979 |
+
"grad_norm": 8.0625,
|
| 980 |
+
"learning_rate": 9.790575801166432e-06,
|
| 981 |
+
"loss": 0.5113,
|
| 982 |
+
"step": 1390
|
| 983 |
+
},
|
| 984 |
+
{
|
| 985 |
+
"epoch": 7.325049051667757,
|
| 986 |
+
"grad_norm": 5.8125,
|
| 987 |
+
"learning_rate": 9.651005032974994e-06,
|
| 988 |
+
"loss": 0.4691,
|
| 989 |
+
"step": 1400
|
| 990 |
+
},
|
| 991 |
+
{
|
| 992 |
+
"epoch": 7.377370830608241,
|
| 993 |
+
"grad_norm": 5.03125,
|
| 994 |
+
"learning_rate": 9.511502302043867e-06,
|
| 995 |
+
"loss": 0.4878,
|
| 996 |
+
"step": 1410
|
| 997 |
+
},
|
| 998 |
+
{
|
| 999 |
+
"epoch": 7.429692609548725,
|
| 1000 |
+
"grad_norm": 6.1875,
|
| 1001 |
+
"learning_rate": 9.372094804706867e-06,
|
| 1002 |
+
"loss": 0.4673,
|
| 1003 |
+
"step": 1420
|
| 1004 |
+
},
|
| 1005 |
+
{
|
| 1006 |
+
"epoch": 7.482014388489208,
|
| 1007 |
+
"grad_norm": 4.34375,
|
| 1008 |
+
"learning_rate": 9.232809718731815e-06,
|
| 1009 |
+
"loss": 0.4654,
|
| 1010 |
+
"step": 1430
|
| 1011 |
+
},
|
| 1012 |
+
{
|
| 1013 |
+
"epoch": 7.534336167429693,
|
| 1014 |
+
"grad_norm": 6.0625,
|
| 1015 |
+
"learning_rate": 9.093674198022201e-06,
|
| 1016 |
+
"loss": 0.4706,
|
| 1017 |
+
"step": 1440
|
| 1018 |
+
},
|
| 1019 |
+
{
|
| 1020 |
+
"epoch": 7.586657946370177,
|
| 1021 |
+
"grad_norm": 4.125,
|
| 1022 |
+
"learning_rate": 8.954715367323468e-06,
|
| 1023 |
+
"loss": 0.4622,
|
| 1024 |
+
"step": 1450
|
| 1025 |
+
},
|
| 1026 |
+
{
|
| 1027 |
+
"epoch": 7.63897972531066,
|
| 1028 |
+
"grad_norm": 4.78125,
|
| 1029 |
+
"learning_rate": 8.815960316934991e-06,
|
| 1030 |
+
"loss": 0.4614,
|
| 1031 |
+
"step": 1460
|
| 1032 |
+
},
|
| 1033 |
+
{
|
| 1034 |
+
"epoch": 7.691301504251145,
|
| 1035 |
+
"grad_norm": 3.8125,
|
| 1036 |
+
"learning_rate": 8.677436097428775e-06,
|
| 1037 |
+
"loss": 0.435,
|
| 1038 |
+
"step": 1470
|
| 1039 |
+
},
|
| 1040 |
+
{
|
| 1041 |
+
"epoch": 7.743623283191629,
|
| 1042 |
+
"grad_norm": 5.625,
|
| 1043 |
+
"learning_rate": 8.539169714375885e-06,
|
| 1044 |
+
"loss": 0.4581,
|
| 1045 |
+
"step": 1480
|
| 1046 |
+
},
|
| 1047 |
+
{
|
| 1048 |
+
"epoch": 7.7959450621321125,
|
| 1049 |
+
"grad_norm": 4.59375,
|
| 1050 |
+
"learning_rate": 8.401188123081653e-06,
|
| 1051 |
+
"loss": 0.4656,
|
| 1052 |
+
"step": 1490
|
| 1053 |
+
},
|
| 1054 |
+
{
|
| 1055 |
+
"epoch": 7.848266841072596,
|
| 1056 |
+
"grad_norm": 5.375,
|
| 1057 |
+
"learning_rate": 8.263518223330698e-06,
|
| 1058 |
+
"loss": 0.4525,
|
| 1059 |
+
"step": 1500
|
| 1060 |
+
},
|
| 1061 |
+
{
|
| 1062 |
+
"epoch": 7.905166775670373,
|
| 1063 |
+
"grad_norm": 7.3125,
|
| 1064 |
+
"learning_rate": 8.126186854142752e-06,
|
| 1065 |
+
"loss": 0.4519,
|
| 1066 |
+
"step": 1510
|
| 1067 |
+
},
|
| 1068 |
+
{
|
| 1069 |
+
"epoch": 7.9574885546108565,
|
| 1070 |
+
"grad_norm": 5.5,
|
| 1071 |
+
"learning_rate": 7.989220788540356e-06,
|
| 1072 |
+
"loss": 0.4917,
|
| 1073 |
+
"step": 1520
|
| 1074 |
+
},
|
| 1075 |
+
{
|
| 1076 |
+
"epoch": 8.009810333551341,
|
| 1077 |
+
"grad_norm": 3.640625,
|
| 1078 |
+
"learning_rate": 7.852646728329368e-06,
|
| 1079 |
+
"loss": 0.4613,
|
| 1080 |
+
"step": 1530
|
| 1081 |
+
},
|
| 1082 |
+
{
|
| 1083 |
+
"epoch": 8.062132112491824,
|
| 1084 |
+
"grad_norm": 3.90625,
|
| 1085 |
+
"learning_rate": 7.716491298893443e-06,
|
| 1086 |
+
"loss": 0.3999,
|
| 1087 |
+
"step": 1540
|
| 1088 |
+
},
|
| 1089 |
+
{
|
| 1090 |
+
"epoch": 8.114453891432309,
|
| 1091 |
+
"grad_norm": 3.53125,
|
| 1092 |
+
"learning_rate": 7.580781044003324e-06,
|
| 1093 |
+
"loss": 0.3909,
|
| 1094 |
+
"step": 1550
|
| 1095 |
+
},
|
| 1096 |
+
{
|
| 1097 |
+
"epoch": 8.166775670372793,
|
| 1098 |
+
"grad_norm": 5.84375,
|
| 1099 |
+
"learning_rate": 7.445542420642097e-06,
|
| 1100 |
+
"loss": 0.4046,
|
| 1101 |
+
"step": 1560
|
| 1102 |
+
},
|
| 1103 |
+
{
|
| 1104 |
+
"epoch": 8.219097449313276,
|
| 1105 |
+
"grad_norm": 3.5,
|
| 1106 |
+
"learning_rate": 7.310801793847344e-06,
|
| 1107 |
+
"loss": 0.3953,
|
| 1108 |
+
"step": 1570
|
| 1109 |
+
},
|
| 1110 |
+
{
|
| 1111 |
+
"epoch": 8.27141922825376,
|
| 1112 |
+
"grad_norm": 3.953125,
|
| 1113 |
+
"learning_rate": 7.176585431571235e-06,
|
| 1114 |
+
"loss": 0.4144,
|
| 1115 |
+
"step": 1580
|
| 1116 |
+
},
|
| 1117 |
+
{
|
| 1118 |
+
"epoch": 8.323741007194245,
|
| 1119 |
+
"grad_norm": 3.59375,
|
| 1120 |
+
"learning_rate": 7.042919499559538e-06,
|
| 1121 |
+
"loss": 0.4091,
|
| 1122 |
+
"step": 1590
|
| 1123 |
+
},
|
| 1124 |
+
{
|
| 1125 |
+
"epoch": 8.376062786134728,
|
| 1126 |
+
"grad_norm": 3.71875,
|
| 1127 |
+
"learning_rate": 6.909830056250527e-06,
|
| 1128 |
+
"loss": 0.4056,
|
| 1129 |
+
"step": 1600
|
| 1130 |
+
},
|
| 1131 |
+
{
|
| 1132 |
+
"epoch": 8.428384565075213,
|
| 1133 |
+
"grad_norm": 4.4375,
|
| 1134 |
+
"learning_rate": 6.777343047694891e-06,
|
| 1135 |
+
"loss": 0.4064,
|
| 1136 |
+
"step": 1610
|
| 1137 |
+
},
|
| 1138 |
+
{
|
| 1139 |
+
"epoch": 8.480706344015697,
|
| 1140 |
+
"grad_norm": 3.578125,
|
| 1141 |
+
"learning_rate": 6.645484302497452e-06,
|
| 1142 |
+
"loss": 0.4054,
|
| 1143 |
+
"step": 1620
|
| 1144 |
+
},
|
| 1145 |
+
{
|
| 1146 |
+
"epoch": 8.53302812295618,
|
| 1147 |
+
"grad_norm": 4.15625,
|
| 1148 |
+
"learning_rate": 6.5142795267818505e-06,
|
| 1149 |
+
"loss": 0.4314,
|
| 1150 |
+
"step": 1630
|
| 1151 |
+
},
|
| 1152 |
+
{
|
| 1153 |
+
"epoch": 8.585349901896665,
|
| 1154 |
+
"grad_norm": 3.96875,
|
| 1155 |
+
"learning_rate": 6.383754299179079e-06,
|
| 1156 |
+
"loss": 0.4023,
|
| 1157 |
+
"step": 1640
|
| 1158 |
+
},
|
| 1159 |
+
{
|
| 1160 |
+
"epoch": 8.637671680837148,
|
| 1161 |
+
"grad_norm": 3.5,
|
| 1162 |
+
"learning_rate": 6.25393406584088e-06,
|
| 1163 |
+
"loss": 0.4027,
|
| 1164 |
+
"step": 1650
|
| 1165 |
+
},
|
| 1166 |
+
{
|
| 1167 |
+
"epoch": 8.689993459777632,
|
| 1168 |
+
"grad_norm": 4.3125,
|
| 1169 |
+
"learning_rate": 6.124844135478971e-06,
|
| 1170 |
+
"loss": 0.4086,
|
| 1171 |
+
"step": 1660
|
| 1172 |
+
},
|
| 1173 |
+
{
|
| 1174 |
+
"epoch": 8.742315238718117,
|
| 1175 |
+
"grad_norm": 3.34375,
|
| 1176 |
+
"learning_rate": 5.996509674431053e-06,
|
| 1177 |
+
"loss": 0.3924,
|
| 1178 |
+
"step": 1670
|
| 1179 |
+
},
|
| 1180 |
+
{
|
| 1181 |
+
"epoch": 8.7946370176586,
|
| 1182 |
+
"grad_norm": 5.78125,
|
| 1183 |
+
"learning_rate": 5.868955701754584e-06,
|
| 1184 |
+
"loss": 0.3933,
|
| 1185 |
+
"step": 1680
|
| 1186 |
+
},
|
| 1187 |
+
{
|
| 1188 |
+
"epoch": 8.846958796599084,
|
| 1189 |
+
"grad_norm": 3.640625,
|
| 1190 |
+
"learning_rate": 5.742207084349274e-06,
|
| 1191 |
+
"loss": 0.4018,
|
| 1192 |
+
"step": 1690
|
| 1193 |
+
},
|
| 1194 |
+
{
|
| 1195 |
+
"epoch": 8.899280575539569,
|
| 1196 |
+
"grad_norm": 4.65625,
|
| 1197 |
+
"learning_rate": 5.616288532109225e-06,
|
| 1198 |
+
"loss": 0.4096,
|
| 1199 |
+
"step": 1700
|
| 1200 |
+
},
|
| 1201 |
+
{
|
| 1202 |
+
"epoch": 8.951602354480052,
|
| 1203 |
+
"grad_norm": 4.3125,
|
| 1204 |
+
"learning_rate": 5.491224593105695e-06,
|
| 1205 |
+
"loss": 0.4171,
|
| 1206 |
+
"step": 1710
|
| 1207 |
+
},
|
| 1208 |
+
{
|
| 1209 |
+
"epoch": 9.003924133420536,
|
| 1210 |
+
"grad_norm": 2.703125,
|
| 1211 |
+
"learning_rate": 5.367039648801386e-06,
|
| 1212 |
+
"loss": 0.4009,
|
| 1213 |
+
"step": 1720
|
| 1214 |
+
},
|
| 1215 |
+
{
|
| 1216 |
+
"epoch": 9.056245912361021,
|
| 1217 |
+
"grad_norm": 3.34375,
|
| 1218 |
+
"learning_rate": 5.243757909297247e-06,
|
| 1219 |
+
"loss": 0.3646,
|
| 1220 |
+
"step": 1730
|
| 1221 |
+
},
|
| 1222 |
+
{
|
| 1223 |
+
"epoch": 9.108567691301504,
|
| 1224 |
+
"grad_norm": 3.640625,
|
| 1225 |
+
"learning_rate": 5.121403408612672e-06,
|
| 1226 |
+
"loss": 0.3731,
|
| 1227 |
+
"step": 1740
|
| 1228 |
+
},
|
| 1229 |
+
{
|
| 1230 |
+
"epoch": 9.160889470241989,
|
| 1231 |
+
"grad_norm": 3.28125,
|
| 1232 |
+
"learning_rate": 5.000000000000003e-06,
|
| 1233 |
+
"loss": 0.3763,
|
| 1234 |
+
"step": 1750
|
| 1235 |
+
},
|
| 1236 |
+
{
|
| 1237 |
+
"epoch": 9.213211249182471,
|
| 1238 |
+
"grad_norm": 3.59375,
|
| 1239 |
+
"learning_rate": 4.879571351294287e-06,
|
| 1240 |
+
"loss": 0.3662,
|
| 1241 |
+
"step": 1760
|
| 1242 |
+
},
|
| 1243 |
+
{
|
| 1244 |
+
"epoch": 9.265533028122956,
|
| 1245 |
+
"grad_norm": 3.671875,
|
| 1246 |
+
"learning_rate": 4.76014094029921e-06,
|
| 1247 |
+
"loss": 0.37,
|
| 1248 |
+
"step": 1770
|
| 1249 |
+
},
|
| 1250 |
+
{
|
| 1251 |
+
"epoch": 9.31785480706344,
|
| 1252 |
+
"grad_norm": 3.09375,
|
| 1253 |
+
"learning_rate": 4.641732050210032e-06,
|
| 1254 |
+
"loss": 0.3498,
|
| 1255 |
+
"step": 1780
|
| 1256 |
+
},
|
| 1257 |
+
{
|
| 1258 |
+
"epoch": 9.370176586003923,
|
| 1259 |
+
"grad_norm": 3.59375,
|
| 1260 |
+
"learning_rate": 4.524367765074499e-06,
|
| 1261 |
+
"loss": 0.3578,
|
| 1262 |
+
"step": 1790
|
| 1263 |
+
},
|
| 1264 |
+
{
|
| 1265 |
+
"epoch": 9.422498364944408,
|
| 1266 |
+
"grad_norm": 3.390625,
|
| 1267 |
+
"learning_rate": 4.408070965292534e-06,
|
| 1268 |
+
"loss": 0.3482,
|
| 1269 |
+
"step": 1800
|
| 1270 |
+
},
|
| 1271 |
+
{
|
| 1272 |
+
"epoch": 9.474820143884893,
|
| 1273 |
+
"grad_norm": 2.984375,
|
| 1274 |
+
"learning_rate": 4.292864323155684e-06,
|
| 1275 |
+
"loss": 0.3684,
|
| 1276 |
+
"step": 1810
|
| 1277 |
+
},
|
| 1278 |
+
{
|
| 1279 |
+
"epoch": 9.527141922825376,
|
| 1280 |
+
"grad_norm": 2.890625,
|
| 1281 |
+
"learning_rate": 4.178770298427107e-06,
|
| 1282 |
+
"loss": 0.3806,
|
| 1283 |
+
"step": 1820
|
| 1284 |
+
},
|
| 1285 |
+
{
|
| 1286 |
+
"epoch": 9.57946370176586,
|
| 1287 |
+
"grad_norm": 3.765625,
|
| 1288 |
+
"learning_rate": 4.065811133962987e-06,
|
| 1289 |
+
"loss": 0.3718,
|
| 1290 |
+
"step": 1830
|
| 1291 |
+
},
|
| 1292 |
+
{
|
| 1293 |
+
"epoch": 9.631785480706345,
|
| 1294 |
+
"grad_norm": 2.921875,
|
| 1295 |
+
"learning_rate": 3.954008851376252e-06,
|
| 1296 |
+
"loss": 0.3469,
|
| 1297 |
+
"step": 1840
|
| 1298 |
+
},
|
| 1299 |
+
{
|
| 1300 |
+
"epoch": 9.684107259646828,
|
| 1301 |
+
"grad_norm": 3.53125,
|
| 1302 |
+
"learning_rate": 3.8433852467434175e-06,
|
| 1303 |
+
"loss": 0.3776,
|
| 1304 |
+
"step": 1850
|
| 1305 |
+
},
|
| 1306 |
+
{
|
| 1307 |
+
"epoch": 9.736429038587312,
|
| 1308 |
+
"grad_norm": 3.78125,
|
| 1309 |
+
"learning_rate": 3.7339618863553983e-06,
|
| 1310 |
+
"loss": 0.3682,
|
| 1311 |
+
"step": 1860
|
| 1312 |
+
},
|
| 1313 |
+
{
|
| 1314 |
+
"epoch": 9.788750817527795,
|
| 1315 |
+
"grad_norm": 3.375,
|
| 1316 |
+
"learning_rate": 3.625760102513103e-06,
|
| 1317 |
+
"loss": 0.3637,
|
| 1318 |
+
"step": 1870
|
| 1319 |
+
},
|
| 1320 |
+
{
|
| 1321 |
+
"epoch": 9.84107259646828,
|
| 1322 |
+
"grad_norm": 3.6875,
|
| 1323 |
+
"learning_rate": 3.5188009893686916e-06,
|
| 1324 |
+
"loss": 0.367,
|
| 1325 |
+
"step": 1880
|
| 1326 |
+
},
|
| 1327 |
+
{
|
| 1328 |
+
"epoch": 9.893394375408764,
|
| 1329 |
+
"grad_norm": 3.625,
|
| 1330 |
+
"learning_rate": 3.4131053988131947e-06,
|
| 1331 |
+
"loss": 0.3591,
|
| 1332 |
+
"step": 1890
|
| 1333 |
+
},
|
| 1334 |
+
{
|
| 1335 |
+
"epoch": 9.945716154349247,
|
| 1336 |
+
"grad_norm": 3.375,
|
| 1337 |
+
"learning_rate": 3.308693936411421e-06,
|
| 1338 |
+
"loss": 0.3601,
|
| 1339 |
+
"step": 1900
|
| 1340 |
+
},
|
| 1341 |
+
{
|
| 1342 |
+
"epoch": 9.998037933289732,
|
| 1343 |
+
"grad_norm": 3.59375,
|
| 1344 |
+
"learning_rate": 3.2055869573848374e-06,
|
| 1345 |
+
"loss": 0.344,
|
| 1346 |
+
"step": 1910
|
| 1347 |
+
},
|
| 1348 |
+
{
|
| 1349 |
+
"epoch": 10.050359712230216,
|
| 1350 |
+
"grad_norm": 3.1875,
|
| 1351 |
+
"learning_rate": 3.103804562643302e-06,
|
| 1352 |
+
"loss": 0.3447,
|
| 1353 |
+
"step": 1920
|
| 1354 |
+
},
|
| 1355 |
+
{
|
| 1356 |
+
"epoch": 10.1026814911707,
|
| 1357 |
+
"grad_norm": 2.921875,
|
| 1358 |
+
"learning_rate": 3.003366594866345e-06,
|
| 1359 |
+
"loss": 0.3423,
|
| 1360 |
+
"step": 1930
|
| 1361 |
+
},
|
| 1362 |
+
{
|
| 1363 |
+
"epoch": 10.155003270111184,
|
| 1364 |
+
"grad_norm": 3.484375,
|
| 1365 |
+
"learning_rate": 2.9042926346347932e-06,
|
| 1366 |
+
"loss": 0.3488,
|
| 1367 |
+
"step": 1940
|
| 1368 |
+
},
|
| 1369 |
+
{
|
| 1370 |
+
"epoch": 10.207325049051668,
|
| 1371 |
+
"grad_norm": 3.390625,
|
| 1372 |
+
"learning_rate": 2.8066019966134907e-06,
|
| 1373 |
+
"loss": 0.3577,
|
| 1374 |
+
"step": 1950
|
| 1375 |
+
},
|
| 1376 |
+
{
|
| 1377 |
+
"epoch": 10.259646827992151,
|
| 1378 |
+
"grad_norm": 3.109375,
|
| 1379 |
+
"learning_rate": 2.7103137257858867e-06,
|
| 1380 |
+
"loss": 0.3234,
|
| 1381 |
+
"step": 1960
|
| 1382 |
+
},
|
| 1383 |
+
{
|
| 1384 |
+
"epoch": 10.311968606932636,
|
| 1385 |
+
"grad_norm": 2.9375,
|
| 1386 |
+
"learning_rate": 2.615446593741161e-06,
|
| 1387 |
+
"loss": 0.3506,
|
| 1388 |
+
"step": 1970
|
| 1389 |
+
},
|
| 1390 |
+
{
|
| 1391 |
+
"epoch": 10.364290385873119,
|
| 1392 |
+
"grad_norm": 3.078125,
|
| 1393 |
+
"learning_rate": 2.522019095014683e-06,
|
| 1394 |
+
"loss": 0.3399,
|
| 1395 |
+
"step": 1980
|
| 1396 |
+
},
|
| 1397 |
+
{
|
| 1398 |
+
"epoch": 10.416612164813603,
|
| 1399 |
+
"grad_norm": 2.984375,
|
| 1400 |
+
"learning_rate": 2.4300494434824373e-06,
|
| 1401 |
+
"loss": 0.3303,
|
| 1402 |
+
"step": 1990
|
| 1403 |
+
},
|
| 1404 |
+
{
|
| 1405 |
+
"epoch": 10.468933943754088,
|
| 1406 |
+
"grad_norm": 2.71875,
|
| 1407 |
+
"learning_rate": 2.339555568810221e-06,
|
| 1408 |
+
"loss": 0.3284,
|
| 1409 |
+
"step": 2000
|
| 1410 |
+
}
|
| 1411 |
+
],
|
| 1412 |
+
"logging_steps": 10,
|
| 1413 |
+
"max_steps": 2500,
|
| 1414 |
+
"num_input_tokens_seen": 0,
|
| 1415 |
+
"num_train_epochs": 14,
|
| 1416 |
+
"save_steps": 500,
|
| 1417 |
+
"stateful_callbacks": {
|
| 1418 |
+
"TrainerControl": {
|
| 1419 |
+
"args": {
|
| 1420 |
+
"should_epoch_stop": false,
|
| 1421 |
+
"should_evaluate": false,
|
| 1422 |
+
"should_log": false,
|
| 1423 |
+
"should_save": true,
|
| 1424 |
+
"should_training_stop": false
|
| 1425 |
+
},
|
| 1426 |
+
"attributes": {}
|
| 1427 |
+
}
|
| 1428 |
+
},
|
| 1429 |
+
"total_flos": 7.650740827638989e+17,
|
| 1430 |
+
"train_batch_size": 4,
|
| 1431 |
+
"trial_name": null,
|
| 1432 |
+
"trial_params": null
|
| 1433 |
+
}
|
checkpoint-2000/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4eb4ade74cb9775d95b0be7d21771ef0cc6f6f2f38b8820e306b00ddc2cb2097
|
| 3 |
+
size 5368
|