Auto upload zain 2026-08-12T22:04:31.568585
Browse files- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/config.json +36 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/model.safetensors +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/optimizer.pt +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/rng_state.pth +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/scheduler.pt +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/tokenizer.json +0 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/tokenizer_config.json +13 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/trainer_state.json +1258 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2400/training_args.bin +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/config.json +36 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/model.safetensors +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/optimizer.pt +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/rng_state.pth +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/scheduler.pt +3 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/tokenizer.json +0 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/tokenizer_config.json +13 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/trainer_state.json +1309 -0
- zain/Activation/out/glu-silu-100L_run/checkpoint-2500/training_args.bin +3 -0
- zain/Activation/out/glu-silu-100L_run/config.json +36 -0
- zain/Activation/out/glu-silu-100L_run/model.safetensors +3 -0
- zain/Activation/out/glu-silu-100L_run/tokenizer.json +0 -0
- zain/Activation/out/glu-silu-100L_run/tokenizer_config.json +13 -0
- zain/Activation/out/glu-silu-100L_run/training_args.bin +3 -0
- zain/Activation/out/glu-silu-100L_run/training_log.jsonl +14 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/config.json +36 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/model.safetensors +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/optimizer.pt +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/rng_state.pth +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/scheduler.pt +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/tokenizer.json +0 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/tokenizer_config.json +13 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/trainer_state.json +85 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/training_args.bin +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/config.json +36 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/model.safetensors +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/optimizer.pt +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/rng_state.pth +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/scheduler.pt +3 -0
- zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/tokenizer.json +0 -0
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/config.json
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "silu",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 100,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.16.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096,
|
| 35 |
+
"waleed_beta": 10.0
|
| 36 |
+
}
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4044bb875bf94c07051f2249e364d546e38e6ec2236445a4743feb0fab43eee5
|
| 3 |
+
size 33967272
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8eeb26579b747d9940ea28668f63001bae9950dd2289f7782ab0a03e676a8814
|
| 3 |
+
size 68504996
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:788fce37b66c4a897887d15a42b82c5e978ffd3ce1892b7282cbebe28b008c9d
|
| 3 |
+
size 14244
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cffe3454608d7821f042822654b62bd9020d78872ebdecde4f22f94320596378
|
| 3 |
+
size 1064
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/trainer_state.json
ADDED
|
@@ -0,0 +1,1258 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.16177957532861476,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 2400,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0013481631277384564,
|
| 14 |
+
"grad_norm": 2.078125,
|
| 15 |
+
"learning_rate": 2.66e-05,
|
| 16 |
+
"loss": 8.29962387084961,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.002696326255476913,
|
| 21 |
+
"grad_norm": 1.3515625,
|
| 22 |
+
"learning_rate": 5.46e-05,
|
| 23 |
+
"loss": 8.047650909423828,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.003370407819346141,
|
| 28 |
+
"eval_loss": 7.835535049438477,
|
| 29 |
+
"eval_runtime": 15.2165,
|
| 30 |
+
"eval_samples_per_second": 626.095,
|
| 31 |
+
"eval_steps_per_second": 0.657,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.004044489383215369,
|
| 36 |
+
"grad_norm": 1.2734375,
|
| 37 |
+
"learning_rate": 8.259999999999999e-05,
|
| 38 |
+
"loss": 7.8378746032714846,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.005392652510953826,
|
| 43 |
+
"grad_norm": 1.265625,
|
| 44 |
+
"learning_rate": 0.0001106,
|
| 45 |
+
"loss": 7.5728271484375,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.006740815638692282,
|
| 50 |
+
"grad_norm": 1.1953125,
|
| 51 |
+
"learning_rate": 0.0001386,
|
| 52 |
+
"loss": 7.2332916259765625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.006740815638692282,
|
| 57 |
+
"eval_loss": 7.039053440093994,
|
| 58 |
+
"eval_runtime": 15.2074,
|
| 59 |
+
"eval_samples_per_second": 626.471,
|
| 60 |
+
"eval_steps_per_second": 0.658,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.008088978766430738,
|
| 65 |
+
"grad_norm": 1.1328125,
|
| 66 |
+
"learning_rate": 0.00016659999999999998,
|
| 67 |
+
"loss": 6.8608039855957035,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.009437141894169195,
|
| 72 |
+
"grad_norm": 0.94921875,
|
| 73 |
+
"learning_rate": 0.00019460000000000001,
|
| 74 |
+
"loss": 6.517842864990234,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.010111223458038422,
|
| 79 |
+
"eval_loss": 6.229617595672607,
|
| 80 |
+
"eval_runtime": 15.1978,
|
| 81 |
+
"eval_samples_per_second": 626.867,
|
| 82 |
+
"eval_steps_per_second": 0.658,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.010785305021907651,
|
| 87 |
+
"grad_norm": 0.80078125,
|
| 88 |
+
"learning_rate": 0.0002226,
|
| 89 |
+
"loss": 6.2399169921875,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.012133468149646108,
|
| 94 |
+
"grad_norm": 0.6640625,
|
| 95 |
+
"learning_rate": 0.00025059999999999997,
|
| 96 |
+
"loss": 6.036548614501953,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.013481631277384564,
|
| 101 |
+
"grad_norm": 0.61328125,
|
| 102 |
+
"learning_rate": 0.0002786,
|
| 103 |
+
"loss": 5.857078552246094,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.013481631277384564,
|
| 108 |
+
"eval_loss": 5.765108585357666,
|
| 109 |
+
"eval_runtime": 15.5271,
|
| 110 |
+
"eval_samples_per_second": 613.574,
|
| 111 |
+
"eval_steps_per_second": 0.644,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.01482979440512302,
|
| 116 |
+
"grad_norm": 0.61328125,
|
| 117 |
+
"learning_rate": 0.00030659999999999997,
|
| 118 |
+
"loss": 5.680564117431641,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.016177957532861477,
|
| 123 |
+
"grad_norm": 2.0625,
|
| 124 |
+
"learning_rate": 0.0003346,
|
| 125 |
+
"loss": 5.472259902954102,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.016852039096730706,
|
| 130 |
+
"eval_loss": 5.240335464477539,
|
| 131 |
+
"eval_runtime": 15.2596,
|
| 132 |
+
"eval_samples_per_second": 624.33,
|
| 133 |
+
"eval_steps_per_second": 0.655,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.01752612066059993,
|
| 138 |
+
"grad_norm": 1.7421875,
|
| 139 |
+
"learning_rate": 0.00036260000000000003,
|
| 140 |
+
"loss": 5.24955940246582,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.01887428378833839,
|
| 145 |
+
"grad_norm": 0.62109375,
|
| 146 |
+
"learning_rate": 0.0003906,
|
| 147 |
+
"loss": 4.995759963989258,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.020222446916076844,
|
| 152 |
+
"grad_norm": 1.390625,
|
| 153 |
+
"learning_rate": 0.0004186,
|
| 154 |
+
"loss": 4.769353866577148,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.020222446916076844,
|
| 159 |
+
"eval_loss": 4.669629096984863,
|
| 160 |
+
"eval_runtime": 15.2595,
|
| 161 |
+
"eval_samples_per_second": 624.332,
|
| 162 |
+
"eval_steps_per_second": 0.655,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.021570610043815303,
|
| 167 |
+
"grad_norm": 0.73828125,
|
| 168 |
+
"learning_rate": 0.0004466,
|
| 169 |
+
"loss": 4.569867324829102,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.022918773171553757,
|
| 174 |
+
"grad_norm": 0.765625,
|
| 175 |
+
"learning_rate": 0.00047460000000000004,
|
| 176 |
+
"loss": 4.389446258544922,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.023592854735422986,
|
| 181 |
+
"eval_loss": 4.231207847595215,
|
| 182 |
+
"eval_runtime": 15.2143,
|
| 183 |
+
"eval_samples_per_second": 626.188,
|
| 184 |
+
"eval_steps_per_second": 0.657,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.024266936299292215,
|
| 189 |
+
"grad_norm": 1.2421875,
|
| 190 |
+
"learning_rate": 0.0005026,
|
| 191 |
+
"loss": 4.243848419189453,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.02561509942703067,
|
| 196 |
+
"grad_norm": 0.7109375,
|
| 197 |
+
"learning_rate": 0.0005306,
|
| 198 |
+
"loss": 4.1090648651123045,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.026963262554769128,
|
| 203 |
+
"grad_norm": 0.484375,
|
| 204 |
+
"learning_rate": 0.0005586,
|
| 205 |
+
"loss": 3.966172790527344,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.026963262554769128,
|
| 210 |
+
"eval_loss": 3.923773765563965,
|
| 211 |
+
"eval_runtime": 15.5308,
|
| 212 |
+
"eval_samples_per_second": 613.426,
|
| 213 |
+
"eval_steps_per_second": 0.644,
|
| 214 |
+
"step": 400
|
| 215 |
+
},
|
| 216 |
+
{
|
| 217 |
+
"epoch": 0.028311425682507583,
|
| 218 |
+
"grad_norm": 0.6640625,
|
| 219 |
+
"learning_rate": 0.0005866,
|
| 220 |
+
"loss": 3.8559597015380858,
|
| 221 |
+
"step": 420
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 0.02965958881024604,
|
| 225 |
+
"grad_norm": 0.515625,
|
| 226 |
+
"learning_rate": 0.0006146,
|
| 227 |
+
"loss": 3.7943866729736326,
|
| 228 |
+
"step": 440
|
| 229 |
+
},
|
| 230 |
+
{
|
| 231 |
+
"epoch": 0.030333670374115267,
|
| 232 |
+
"eval_loss": 3.6977553367614746,
|
| 233 |
+
"eval_runtime": 15.2001,
|
| 234 |
+
"eval_samples_per_second": 626.77,
|
| 235 |
+
"eval_steps_per_second": 0.658,
|
| 236 |
+
"step": 450
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.031007751937984496,
|
| 240 |
+
"grad_norm": 0.51953125,
|
| 241 |
+
"learning_rate": 0.0006426,
|
| 242 |
+
"loss": 3.686443328857422,
|
| 243 |
+
"step": 460
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"epoch": 0.032355915065722954,
|
| 247 |
+
"grad_norm": 0.62109375,
|
| 248 |
+
"learning_rate": 0.0006705999999999999,
|
| 249 |
+
"loss": 3.6123157501220704,
|
| 250 |
+
"step": 480
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.03370407819346141,
|
| 254 |
+
"grad_norm": 0.70703125,
|
| 255 |
+
"learning_rate": 0.0006986,
|
| 256 |
+
"loss": 3.542654800415039,
|
| 257 |
+
"step": 500
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"epoch": 0.03370407819346141,
|
| 261 |
+
"eval_loss": 3.513820171356201,
|
| 262 |
+
"eval_runtime": 15.2193,
|
| 263 |
+
"eval_samples_per_second": 625.983,
|
| 264 |
+
"eval_steps_per_second": 0.657,
|
| 265 |
+
"step": 500
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.03505224132119986,
|
| 269 |
+
"grad_norm": 0.48046875,
|
| 270 |
+
"learning_rate": 0.0007,
|
| 271 |
+
"loss": 3.472199249267578,
|
| 272 |
+
"step": 520
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.03640040444893832,
|
| 276 |
+
"grad_norm": 0.53515625,
|
| 277 |
+
"learning_rate": 0.0007,
|
| 278 |
+
"loss": 3.396291732788086,
|
| 279 |
+
"step": 540
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.03707448601280755,
|
| 283 |
+
"eval_loss": 3.3581483364105225,
|
| 284 |
+
"eval_runtime": 15.3044,
|
| 285 |
+
"eval_samples_per_second": 622.502,
|
| 286 |
+
"eval_steps_per_second": 0.653,
|
| 287 |
+
"step": 550
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"epoch": 0.03774856757667678,
|
| 291 |
+
"grad_norm": 0.47265625,
|
| 292 |
+
"learning_rate": 0.0007,
|
| 293 |
+
"loss": 3.347300720214844,
|
| 294 |
+
"step": 560
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.03909673070441524,
|
| 298 |
+
"grad_norm": 0.458984375,
|
| 299 |
+
"learning_rate": 0.0007,
|
| 300 |
+
"loss": 3.2885902404785154,
|
| 301 |
+
"step": 580
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"epoch": 0.04044489383215369,
|
| 305 |
+
"grad_norm": 0.4375,
|
| 306 |
+
"learning_rate": 0.0007,
|
| 307 |
+
"loss": 3.25723876953125,
|
| 308 |
+
"step": 600
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"epoch": 0.04044489383215369,
|
| 312 |
+
"eval_loss": 3.2229573726654053,
|
| 313 |
+
"eval_runtime": 15.2967,
|
| 314 |
+
"eval_samples_per_second": 622.812,
|
| 315 |
+
"eval_steps_per_second": 0.654,
|
| 316 |
+
"step": 600
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"epoch": 0.04179305695989215,
|
| 320 |
+
"grad_norm": 0.458984375,
|
| 321 |
+
"learning_rate": 0.0007,
|
| 322 |
+
"loss": 3.1902334213256838,
|
| 323 |
+
"step": 620
|
| 324 |
+
},
|
| 325 |
+
{
|
| 326 |
+
"epoch": 0.043141220087630605,
|
| 327 |
+
"grad_norm": 0.40625,
|
| 328 |
+
"learning_rate": 0.0007,
|
| 329 |
+
"loss": 3.167861557006836,
|
| 330 |
+
"step": 640
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.043815301651499834,
|
| 334 |
+
"eval_loss": 3.120636463165283,
|
| 335 |
+
"eval_runtime": 15.3751,
|
| 336 |
+
"eval_samples_per_second": 619.639,
|
| 337 |
+
"eval_steps_per_second": 0.65,
|
| 338 |
+
"step": 650
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.044489383215369056,
|
| 342 |
+
"grad_norm": 0.494140625,
|
| 343 |
+
"learning_rate": 0.0007,
|
| 344 |
+
"loss": 3.1089324951171875,
|
| 345 |
+
"step": 660
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.045837546343107514,
|
| 349 |
+
"grad_norm": 0.435546875,
|
| 350 |
+
"learning_rate": 0.0007,
|
| 351 |
+
"loss": 3.0828268051147463,
|
| 352 |
+
"step": 680
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.04718570947084597,
|
| 356 |
+
"grad_norm": 0.427734375,
|
| 357 |
+
"learning_rate": 0.0007,
|
| 358 |
+
"loss": 3.0396488189697264,
|
| 359 |
+
"step": 700
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.04718570947084597,
|
| 363 |
+
"eval_loss": 3.0276408195495605,
|
| 364 |
+
"eval_runtime": 15.2735,
|
| 365 |
+
"eval_samples_per_second": 623.761,
|
| 366 |
+
"eval_steps_per_second": 0.655,
|
| 367 |
+
"step": 700
|
| 368 |
+
},
|
| 369 |
+
{
|
| 370 |
+
"epoch": 0.04853387259858443,
|
| 371 |
+
"grad_norm": 0.4375,
|
| 372 |
+
"learning_rate": 0.0007,
|
| 373 |
+
"loss": 3.0034805297851563,
|
| 374 |
+
"step": 720
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 0.04988203572632288,
|
| 378 |
+
"grad_norm": 0.52734375,
|
| 379 |
+
"learning_rate": 0.0007,
|
| 380 |
+
"loss": 2.96556396484375,
|
| 381 |
+
"step": 740
|
| 382 |
+
},
|
| 383 |
+
{
|
| 384 |
+
"epoch": 0.05055611729019211,
|
| 385 |
+
"eval_loss": 2.950009346008301,
|
| 386 |
+
"eval_runtime": 15.243,
|
| 387 |
+
"eval_samples_per_second": 625.009,
|
| 388 |
+
"eval_steps_per_second": 0.656,
|
| 389 |
+
"step": 750
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.05123019885406134,
|
| 393 |
+
"grad_norm": 0.451171875,
|
| 394 |
+
"learning_rate": 0.0007,
|
| 395 |
+
"loss": 2.950779151916504,
|
| 396 |
+
"step": 760
|
| 397 |
+
},
|
| 398 |
+
{
|
| 399 |
+
"epoch": 0.0525783619817998,
|
| 400 |
+
"grad_norm": 0.400390625,
|
| 401 |
+
"learning_rate": 0.0007,
|
| 402 |
+
"loss": 2.956203079223633,
|
| 403 |
+
"step": 780
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 0.053926525109538256,
|
| 407 |
+
"grad_norm": 0.40625,
|
| 408 |
+
"learning_rate": 0.0007,
|
| 409 |
+
"loss": 2.9017578125,
|
| 410 |
+
"step": 800
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.053926525109538256,
|
| 414 |
+
"eval_loss": 2.8844006061553955,
|
| 415 |
+
"eval_runtime": 15.2903,
|
| 416 |
+
"eval_samples_per_second": 623.073,
|
| 417 |
+
"eval_steps_per_second": 0.654,
|
| 418 |
+
"step": 800
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.05527468823727671,
|
| 422 |
+
"grad_norm": 0.41015625,
|
| 423 |
+
"learning_rate": 0.0007,
|
| 424 |
+
"loss": 2.887515640258789,
|
| 425 |
+
"step": 820
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.056622851365015166,
|
| 429 |
+
"grad_norm": 0.41796875,
|
| 430 |
+
"learning_rate": 0.0007,
|
| 431 |
+
"loss": 2.8556928634643555,
|
| 432 |
+
"step": 840
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.057296932928884395,
|
| 436 |
+
"eval_loss": 2.821610450744629,
|
| 437 |
+
"eval_runtime": 15.2253,
|
| 438 |
+
"eval_samples_per_second": 625.733,
|
| 439 |
+
"eval_steps_per_second": 0.657,
|
| 440 |
+
"step": 850
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.057971014492753624,
|
| 444 |
+
"grad_norm": 0.423828125,
|
| 445 |
+
"learning_rate": 0.0007,
|
| 446 |
+
"loss": 2.8130855560302734,
|
| 447 |
+
"step": 860
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.05931917762049208,
|
| 451 |
+
"grad_norm": 0.38671875,
|
| 452 |
+
"learning_rate": 0.0007,
|
| 453 |
+
"loss": 2.790329170227051,
|
| 454 |
+
"step": 880
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.06066734074823053,
|
| 458 |
+
"grad_norm": 0.427734375,
|
| 459 |
+
"learning_rate": 0.0007,
|
| 460 |
+
"loss": 2.7760162353515625,
|
| 461 |
+
"step": 900
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.06066734074823053,
|
| 465 |
+
"eval_loss": 2.7661373615264893,
|
| 466 |
+
"eval_runtime": 15.2736,
|
| 467 |
+
"eval_samples_per_second": 623.755,
|
| 468 |
+
"eval_steps_per_second": 0.655,
|
| 469 |
+
"step": 900
|
| 470 |
+
},
|
| 471 |
+
{
|
| 472 |
+
"epoch": 0.06201550387596899,
|
| 473 |
+
"grad_norm": 0.435546875,
|
| 474 |
+
"learning_rate": 0.0007,
|
| 475 |
+
"loss": 2.7462047576904296,
|
| 476 |
+
"step": 920
|
| 477 |
+
},
|
| 478 |
+
{
|
| 479 |
+
"epoch": 0.06336366700370745,
|
| 480 |
+
"grad_norm": 0.4140625,
|
| 481 |
+
"learning_rate": 0.0007,
|
| 482 |
+
"loss": 2.732274627685547,
|
| 483 |
+
"step": 940
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.06403774856757667,
|
| 487 |
+
"eval_loss": 2.713942289352417,
|
| 488 |
+
"eval_runtime": 15.6062,
|
| 489 |
+
"eval_samples_per_second": 610.463,
|
| 490 |
+
"eval_steps_per_second": 0.641,
|
| 491 |
+
"step": 950
|
| 492 |
+
},
|
| 493 |
+
{
|
| 494 |
+
"epoch": 0.06471183013144591,
|
| 495 |
+
"grad_norm": 0.419921875,
|
| 496 |
+
"learning_rate": 0.0007,
|
| 497 |
+
"loss": 2.705318069458008,
|
| 498 |
+
"step": 960
|
| 499 |
+
},
|
| 500 |
+
{
|
| 501 |
+
"epoch": 0.06605999325918437,
|
| 502 |
+
"grad_norm": 0.4296875,
|
| 503 |
+
"learning_rate": 0.0007,
|
| 504 |
+
"loss": 2.6892818450927733,
|
| 505 |
+
"step": 980
|
| 506 |
+
},
|
| 507 |
+
{
|
| 508 |
+
"epoch": 0.06740815638692282,
|
| 509 |
+
"grad_norm": 0.4453125,
|
| 510 |
+
"learning_rate": 0.0007,
|
| 511 |
+
"loss": 2.687115287780762,
|
| 512 |
+
"step": 1000
|
| 513 |
+
},
|
| 514 |
+
{
|
| 515 |
+
"epoch": 0.06740815638692282,
|
| 516 |
+
"eval_loss": 2.668647289276123,
|
| 517 |
+
"eval_runtime": 15.2597,
|
| 518 |
+
"eval_samples_per_second": 624.326,
|
| 519 |
+
"eval_steps_per_second": 0.655,
|
| 520 |
+
"step": 1000
|
| 521 |
+
},
|
| 522 |
+
{
|
| 523 |
+
"epoch": 0.06875631951466127,
|
| 524 |
+
"grad_norm": 0.388671875,
|
| 525 |
+
"learning_rate": 0.0007,
|
| 526 |
+
"loss": 2.6533071517944338,
|
| 527 |
+
"step": 1020
|
| 528 |
+
},
|
| 529 |
+
{
|
| 530 |
+
"epoch": 0.07010448264239973,
|
| 531 |
+
"grad_norm": 0.392578125,
|
| 532 |
+
"learning_rate": 0.0007,
|
| 533 |
+
"loss": 2.6390922546386717,
|
| 534 |
+
"step": 1040
|
| 535 |
+
},
|
| 536 |
+
{
|
| 537 |
+
"epoch": 0.07077856420626896,
|
| 538 |
+
"eval_loss": 2.6291816234588623,
|
| 539 |
+
"eval_runtime": 15.2457,
|
| 540 |
+
"eval_samples_per_second": 624.896,
|
| 541 |
+
"eval_steps_per_second": 0.656,
|
| 542 |
+
"step": 1050
|
| 543 |
+
},
|
| 544 |
+
{
|
| 545 |
+
"epoch": 0.07145264577013818,
|
| 546 |
+
"grad_norm": 0.4375,
|
| 547 |
+
"learning_rate": 0.0007,
|
| 548 |
+
"loss": 2.624461555480957,
|
| 549 |
+
"step": 1060
|
| 550 |
+
},
|
| 551 |
+
{
|
| 552 |
+
"epoch": 0.07280080889787664,
|
| 553 |
+
"grad_norm": 0.396484375,
|
| 554 |
+
"learning_rate": 0.0007,
|
| 555 |
+
"loss": 2.6044427871704103,
|
| 556 |
+
"step": 1080
|
| 557 |
+
},
|
| 558 |
+
{
|
| 559 |
+
"epoch": 0.0741489720256151,
|
| 560 |
+
"grad_norm": 0.423828125,
|
| 561 |
+
"learning_rate": 0.0007,
|
| 562 |
+
"loss": 2.597646713256836,
|
| 563 |
+
"step": 1100
|
| 564 |
+
},
|
| 565 |
+
{
|
| 566 |
+
"epoch": 0.0741489720256151,
|
| 567 |
+
"eval_loss": 2.589696168899536,
|
| 568 |
+
"eval_runtime": 15.3967,
|
| 569 |
+
"eval_samples_per_second": 618.768,
|
| 570 |
+
"eval_steps_per_second": 0.649,
|
| 571 |
+
"step": 1100
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 0.07549713515335356,
|
| 575 |
+
"grad_norm": 0.369140625,
|
| 576 |
+
"learning_rate": 0.0007,
|
| 577 |
+
"loss": 2.561947250366211,
|
| 578 |
+
"step": 1120
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 0.07684529828109202,
|
| 582 |
+
"grad_norm": 0.390625,
|
| 583 |
+
"learning_rate": 0.0007,
|
| 584 |
+
"loss": 2.54862117767334,
|
| 585 |
+
"step": 1140
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 0.07751937984496124,
|
| 589 |
+
"eval_loss": 2.559185743331909,
|
| 590 |
+
"eval_runtime": 15.2877,
|
| 591 |
+
"eval_samples_per_second": 623.181,
|
| 592 |
+
"eval_steps_per_second": 0.654,
|
| 593 |
+
"step": 1150
|
| 594 |
+
},
|
| 595 |
+
{
|
| 596 |
+
"epoch": 0.07819346140883048,
|
| 597 |
+
"grad_norm": 0.37890625,
|
| 598 |
+
"learning_rate": 0.0007,
|
| 599 |
+
"loss": 2.5469717025756835,
|
| 600 |
+
"step": 1160
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"epoch": 0.07954162453656892,
|
| 604 |
+
"grad_norm": 0.400390625,
|
| 605 |
+
"learning_rate": 0.0007,
|
| 606 |
+
"loss": 2.540451240539551,
|
| 607 |
+
"step": 1180
|
| 608 |
+
},
|
| 609 |
+
{
|
| 610 |
+
"epoch": 0.08088978766430738,
|
| 611 |
+
"grad_norm": 0.40625,
|
| 612 |
+
"learning_rate": 0.0007,
|
| 613 |
+
"loss": 2.53134708404541,
|
| 614 |
+
"step": 1200
|
| 615 |
+
},
|
| 616 |
+
{
|
| 617 |
+
"epoch": 0.08088978766430738,
|
| 618 |
+
"eval_loss": 2.521665096282959,
|
| 619 |
+
"eval_runtime": 15.2822,
|
| 620 |
+
"eval_samples_per_second": 623.407,
|
| 621 |
+
"eval_steps_per_second": 0.654,
|
| 622 |
+
"step": 1200
|
| 623 |
+
},
|
| 624 |
+
{
|
| 625 |
+
"epoch": 0.08223795079204584,
|
| 626 |
+
"grad_norm": 0.37109375,
|
| 627 |
+
"learning_rate": 0.0007,
|
| 628 |
+
"loss": 2.513539123535156,
|
| 629 |
+
"step": 1220
|
| 630 |
+
},
|
| 631 |
+
{
|
| 632 |
+
"epoch": 0.0835861139197843,
|
| 633 |
+
"grad_norm": 0.36328125,
|
| 634 |
+
"learning_rate": 0.0007,
|
| 635 |
+
"loss": 2.5009738922119142,
|
| 636 |
+
"step": 1240
|
| 637 |
+
},
|
| 638 |
+
{
|
| 639 |
+
"epoch": 0.08426019548365352,
|
| 640 |
+
"eval_loss": 2.489459753036499,
|
| 641 |
+
"eval_runtime": 15.2493,
|
| 642 |
+
"eval_samples_per_second": 624.749,
|
| 643 |
+
"eval_steps_per_second": 0.656,
|
| 644 |
+
"step": 1250
|
| 645 |
+
},
|
| 646 |
+
{
|
| 647 |
+
"epoch": 0.08493427704752275,
|
| 648 |
+
"grad_norm": 0.388671875,
|
| 649 |
+
"learning_rate": 0.0007,
|
| 650 |
+
"loss": 2.4894077301025392,
|
| 651 |
+
"step": 1260
|
| 652 |
+
},
|
| 653 |
+
{
|
| 654 |
+
"epoch": 0.08628244017526121,
|
| 655 |
+
"grad_norm": 0.369140625,
|
| 656 |
+
"learning_rate": 0.0007,
|
| 657 |
+
"loss": 2.4697439193725588,
|
| 658 |
+
"step": 1280
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"epoch": 0.08763060330299967,
|
| 662 |
+
"grad_norm": 0.423828125,
|
| 663 |
+
"learning_rate": 0.0007,
|
| 664 |
+
"loss": 2.461294174194336,
|
| 665 |
+
"step": 1300
|
| 666 |
+
},
|
| 667 |
+
{
|
| 668 |
+
"epoch": 0.08763060330299967,
|
| 669 |
+
"eval_loss": 2.4627816677093506,
|
| 670 |
+
"eval_runtime": 15.421,
|
| 671 |
+
"eval_samples_per_second": 617.795,
|
| 672 |
+
"eval_steps_per_second": 0.648,
|
| 673 |
+
"step": 1300
|
| 674 |
+
},
|
| 675 |
+
{
|
| 676 |
+
"epoch": 0.08897876643073811,
|
| 677 |
+
"grad_norm": 0.36328125,
|
| 678 |
+
"learning_rate": 0.0007,
|
| 679 |
+
"loss": 2.452680206298828,
|
| 680 |
+
"step": 1320
|
| 681 |
+
},
|
| 682 |
+
{
|
| 683 |
+
"epoch": 0.09032692955847657,
|
| 684 |
+
"grad_norm": 0.404296875,
|
| 685 |
+
"learning_rate": 0.0007,
|
| 686 |
+
"loss": 2.4396684646606444,
|
| 687 |
+
"step": 1340
|
| 688 |
+
},
|
| 689 |
+
{
|
| 690 |
+
"epoch": 0.0910010111223458,
|
| 691 |
+
"eval_loss": 2.437220335006714,
|
| 692 |
+
"eval_runtime": 15.3026,
|
| 693 |
+
"eval_samples_per_second": 622.576,
|
| 694 |
+
"eval_steps_per_second": 0.653,
|
| 695 |
+
"step": 1350
|
| 696 |
+
},
|
| 697 |
+
{
|
| 698 |
+
"epoch": 0.09167509268621503,
|
| 699 |
+
"grad_norm": 0.412109375,
|
| 700 |
+
"learning_rate": 0.0007,
|
| 701 |
+
"loss": 2.4397760391235352,
|
| 702 |
+
"step": 1360
|
| 703 |
+
},
|
| 704 |
+
{
|
| 705 |
+
"epoch": 0.09302325581395349,
|
| 706 |
+
"grad_norm": 0.353515625,
|
| 707 |
+
"learning_rate": 0.0007,
|
| 708 |
+
"loss": 2.4158536911010744,
|
| 709 |
+
"step": 1380
|
| 710 |
+
},
|
| 711 |
+
{
|
| 712 |
+
"epoch": 0.09437141894169195,
|
| 713 |
+
"grad_norm": 0.345703125,
|
| 714 |
+
"learning_rate": 0.0007,
|
| 715 |
+
"loss": 2.406942367553711,
|
| 716 |
+
"step": 1400
|
| 717 |
+
},
|
| 718 |
+
{
|
| 719 |
+
"epoch": 0.09437141894169195,
|
| 720 |
+
"eval_loss": 2.4096601009368896,
|
| 721 |
+
"eval_runtime": 15.357,
|
| 722 |
+
"eval_samples_per_second": 620.368,
|
| 723 |
+
"eval_steps_per_second": 0.651,
|
| 724 |
+
"step": 1400
|
| 725 |
+
},
|
| 726 |
+
{
|
| 727 |
+
"epoch": 0.0957195820694304,
|
| 728 |
+
"grad_norm": 0.36328125,
|
| 729 |
+
"learning_rate": 0.0007,
|
| 730 |
+
"loss": 2.4161144256591798,
|
| 731 |
+
"step": 1420
|
| 732 |
+
},
|
| 733 |
+
{
|
| 734 |
+
"epoch": 0.09706774519716886,
|
| 735 |
+
"grad_norm": 0.36328125,
|
| 736 |
+
"learning_rate": 0.0007,
|
| 737 |
+
"loss": 2.392984390258789,
|
| 738 |
+
"step": 1440
|
| 739 |
+
},
|
| 740 |
+
{
|
| 741 |
+
"epoch": 0.09774182676103808,
|
| 742 |
+
"eval_loss": 2.3937113285064697,
|
| 743 |
+
"eval_runtime": 15.3555,
|
| 744 |
+
"eval_samples_per_second": 620.43,
|
| 745 |
+
"eval_steps_per_second": 0.651,
|
| 746 |
+
"step": 1450
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 0.09841590832490732,
|
| 750 |
+
"grad_norm": 0.384765625,
|
| 751 |
+
"learning_rate": 0.0007,
|
| 752 |
+
"loss": 2.3820331573486326,
|
| 753 |
+
"step": 1460
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 0.09976407145264576,
|
| 757 |
+
"grad_norm": 0.33984375,
|
| 758 |
+
"learning_rate": 0.0007,
|
| 759 |
+
"loss": 2.3687986373901366,
|
| 760 |
+
"step": 1480
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 0.10111223458038422,
|
| 764 |
+
"grad_norm": 0.38671875,
|
| 765 |
+
"learning_rate": 0.0007,
|
| 766 |
+
"loss": 2.355032730102539,
|
| 767 |
+
"step": 1500
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 0.10111223458038422,
|
| 771 |
+
"eval_loss": 2.3681657314300537,
|
| 772 |
+
"eval_runtime": 15.3336,
|
| 773 |
+
"eval_samples_per_second": 621.317,
|
| 774 |
+
"eval_steps_per_second": 0.652,
|
| 775 |
+
"step": 1500
|
| 776 |
+
},
|
| 777 |
+
{
|
| 778 |
+
"epoch": 0.10246039770812268,
|
| 779 |
+
"grad_norm": 0.392578125,
|
| 780 |
+
"learning_rate": 0.0007,
|
| 781 |
+
"loss": 2.3572179794311525,
|
| 782 |
+
"step": 1520
|
| 783 |
+
},
|
| 784 |
+
{
|
| 785 |
+
"epoch": 0.10380856083586114,
|
| 786 |
+
"grad_norm": 0.35546875,
|
| 787 |
+
"learning_rate": 0.0007,
|
| 788 |
+
"loss": 2.345840835571289,
|
| 789 |
+
"step": 1540
|
| 790 |
+
},
|
| 791 |
+
{
|
| 792 |
+
"epoch": 0.10448264239973036,
|
| 793 |
+
"eval_loss": 2.3475193977355957,
|
| 794 |
+
"eval_runtime": 15.3453,
|
| 795 |
+
"eval_samples_per_second": 620.842,
|
| 796 |
+
"eval_steps_per_second": 0.652,
|
| 797 |
+
"step": 1550
|
| 798 |
+
},
|
| 799 |
+
{
|
| 800 |
+
"epoch": 0.1051567239635996,
|
| 801 |
+
"grad_norm": 0.388671875,
|
| 802 |
+
"learning_rate": 0.0007,
|
| 803 |
+
"loss": 2.335231971740723,
|
| 804 |
+
"step": 1560
|
| 805 |
+
},
|
| 806 |
+
{
|
| 807 |
+
"epoch": 0.10650488709133805,
|
| 808 |
+
"grad_norm": 0.359375,
|
| 809 |
+
"learning_rate": 0.0007,
|
| 810 |
+
"loss": 2.3543283462524416,
|
| 811 |
+
"step": 1580
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 0.10785305021907651,
|
| 815 |
+
"grad_norm": 0.39453125,
|
| 816 |
+
"learning_rate": 0.0007,
|
| 817 |
+
"loss": 2.3237049102783205,
|
| 818 |
+
"step": 1600
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 0.10785305021907651,
|
| 822 |
+
"eval_loss": 2.3377676010131836,
|
| 823 |
+
"eval_runtime": 15.2858,
|
| 824 |
+
"eval_samples_per_second": 623.259,
|
| 825 |
+
"eval_steps_per_second": 0.654,
|
| 826 |
+
"step": 1600
|
| 827 |
+
},
|
| 828 |
+
{
|
| 829 |
+
"epoch": 0.10920121334681497,
|
| 830 |
+
"grad_norm": 0.357421875,
|
| 831 |
+
"learning_rate": 0.0007,
|
| 832 |
+
"loss": 2.3196277618408203,
|
| 833 |
+
"step": 1620
|
| 834 |
+
},
|
| 835 |
+
{
|
| 836 |
+
"epoch": 0.11054937647455342,
|
| 837 |
+
"grad_norm": 0.341796875,
|
| 838 |
+
"learning_rate": 0.0007,
|
| 839 |
+
"loss": 2.3308162689208984,
|
| 840 |
+
"step": 1640
|
| 841 |
+
},
|
| 842 |
+
{
|
| 843 |
+
"epoch": 0.11122345803842265,
|
| 844 |
+
"eval_loss": 2.313627004623413,
|
| 845 |
+
"eval_runtime": 15.276,
|
| 846 |
+
"eval_samples_per_second": 623.659,
|
| 847 |
+
"eval_steps_per_second": 0.655,
|
| 848 |
+
"step": 1650
|
| 849 |
+
},
|
| 850 |
+
{
|
| 851 |
+
"epoch": 0.11189753960229187,
|
| 852 |
+
"grad_norm": 0.353515625,
|
| 853 |
+
"learning_rate": 0.0007,
|
| 854 |
+
"loss": 2.3104904174804686,
|
| 855 |
+
"step": 1660
|
| 856 |
+
},
|
| 857 |
+
{
|
| 858 |
+
"epoch": 0.11324570273003033,
|
| 859 |
+
"grad_norm": 0.353515625,
|
| 860 |
+
"learning_rate": 0.0007,
|
| 861 |
+
"loss": 2.2955816268920897,
|
| 862 |
+
"step": 1680
|
| 863 |
+
},
|
| 864 |
+
{
|
| 865 |
+
"epoch": 0.11459386585776879,
|
| 866 |
+
"grad_norm": 0.345703125,
|
| 867 |
+
"learning_rate": 0.0007,
|
| 868 |
+
"loss": 2.3044401168823243,
|
| 869 |
+
"step": 1700
|
| 870 |
+
},
|
| 871 |
+
{
|
| 872 |
+
"epoch": 0.11459386585776879,
|
| 873 |
+
"eval_loss": 2.2976956367492676,
|
| 874 |
+
"eval_runtime": 15.3404,
|
| 875 |
+
"eval_samples_per_second": 621.039,
|
| 876 |
+
"eval_steps_per_second": 0.652,
|
| 877 |
+
"step": 1700
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 0.11594202898550725,
|
| 881 |
+
"grad_norm": 0.373046875,
|
| 882 |
+
"learning_rate": 0.0007,
|
| 883 |
+
"loss": 2.280081939697266,
|
| 884 |
+
"step": 1720
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 0.1172901921132457,
|
| 888 |
+
"grad_norm": 0.373046875,
|
| 889 |
+
"learning_rate": 0.0007,
|
| 890 |
+
"loss": 2.269576644897461,
|
| 891 |
+
"step": 1740
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 0.11796427367711493,
|
| 895 |
+
"eval_loss": 2.28181791305542,
|
| 896 |
+
"eval_runtime": 15.2137,
|
| 897 |
+
"eval_samples_per_second": 626.211,
|
| 898 |
+
"eval_steps_per_second": 0.657,
|
| 899 |
+
"step": 1750
|
| 900 |
+
},
|
| 901 |
+
{
|
| 902 |
+
"epoch": 0.11863835524098416,
|
| 903 |
+
"grad_norm": 0.359375,
|
| 904 |
+
"learning_rate": 0.0007,
|
| 905 |
+
"loss": 2.273081398010254,
|
| 906 |
+
"step": 1760
|
| 907 |
+
},
|
| 908 |
+
{
|
| 909 |
+
"epoch": 0.11998651836872262,
|
| 910 |
+
"grad_norm": 0.345703125,
|
| 911 |
+
"learning_rate": 0.0007,
|
| 912 |
+
"loss": 2.2877025604248047,
|
| 913 |
+
"step": 1780
|
| 914 |
+
},
|
| 915 |
+
{
|
| 916 |
+
"epoch": 0.12133468149646107,
|
| 917 |
+
"grad_norm": 0.3671875,
|
| 918 |
+
"learning_rate": 0.0007,
|
| 919 |
+
"loss": 2.2828989028930664,
|
| 920 |
+
"step": 1800
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 0.12133468149646107,
|
| 924 |
+
"eval_loss": 2.2671334743499756,
|
| 925 |
+
"eval_runtime": 15.1615,
|
| 926 |
+
"eval_samples_per_second": 628.369,
|
| 927 |
+
"eval_steps_per_second": 0.66,
|
| 928 |
+
"step": 1800
|
| 929 |
+
},
|
| 930 |
+
{
|
| 931 |
+
"epoch": 0.12268284462419952,
|
| 932 |
+
"grad_norm": 0.388671875,
|
| 933 |
+
"learning_rate": 0.0007,
|
| 934 |
+
"loss": 2.2743486404418944,
|
| 935 |
+
"step": 1820
|
| 936 |
+
},
|
| 937 |
+
{
|
| 938 |
+
"epoch": 0.12403100775193798,
|
| 939 |
+
"grad_norm": 0.3671875,
|
| 940 |
+
"learning_rate": 0.0007,
|
| 941 |
+
"loss": 2.2429489135742187,
|
| 942 |
+
"step": 1840
|
| 943 |
+
},
|
| 944 |
+
{
|
| 945 |
+
"epoch": 0.12470508931580722,
|
| 946 |
+
"eval_loss": 2.252441644668579,
|
| 947 |
+
"eval_runtime": 15.3255,
|
| 948 |
+
"eval_samples_per_second": 621.643,
|
| 949 |
+
"eval_steps_per_second": 0.653,
|
| 950 |
+
"step": 1850
|
| 951 |
+
},
|
| 952 |
+
{
|
| 953 |
+
"epoch": 0.12537917087967643,
|
| 954 |
+
"grad_norm": 0.33984375,
|
| 955 |
+
"learning_rate": 0.0007,
|
| 956 |
+
"loss": 2.242429733276367,
|
| 957 |
+
"step": 1860
|
| 958 |
+
},
|
| 959 |
+
{
|
| 960 |
+
"epoch": 0.1267273340074149,
|
| 961 |
+
"grad_norm": 0.353515625,
|
| 962 |
+
"learning_rate": 0.0007,
|
| 963 |
+
"loss": 2.2348007202148437,
|
| 964 |
+
"step": 1880
|
| 965 |
+
},
|
| 966 |
+
{
|
| 967 |
+
"epoch": 0.12807549713515334,
|
| 968 |
+
"grad_norm": 0.330078125,
|
| 969 |
+
"learning_rate": 0.0007,
|
| 970 |
+
"loss": 2.241002655029297,
|
| 971 |
+
"step": 1900
|
| 972 |
+
},
|
| 973 |
+
{
|
| 974 |
+
"epoch": 0.12807549713515334,
|
| 975 |
+
"eval_loss": 2.2387285232543945,
|
| 976 |
+
"eval_runtime": 15.2942,
|
| 977 |
+
"eval_samples_per_second": 622.915,
|
| 978 |
+
"eval_steps_per_second": 0.654,
|
| 979 |
+
"step": 1900
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 0.12942366026289182,
|
| 983 |
+
"grad_norm": 0.34765625,
|
| 984 |
+
"learning_rate": 0.0007,
|
| 985 |
+
"loss": 2.2172693252563476,
|
| 986 |
+
"step": 1920
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 0.13077182339063026,
|
| 990 |
+
"grad_norm": 0.357421875,
|
| 991 |
+
"learning_rate": 0.0007,
|
| 992 |
+
"loss": 2.2283090591430663,
|
| 993 |
+
"step": 1940
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 0.13144590495449948,
|
| 997 |
+
"eval_loss": 2.225919246673584,
|
| 998 |
+
"eval_runtime": 15.3259,
|
| 999 |
+
"eval_samples_per_second": 621.627,
|
| 1000 |
+
"eval_steps_per_second": 0.652,
|
| 1001 |
+
"step": 1950
|
| 1002 |
+
},
|
| 1003 |
+
{
|
| 1004 |
+
"epoch": 0.13211998651836873,
|
| 1005 |
+
"grad_norm": 0.333984375,
|
| 1006 |
+
"learning_rate": 0.0007,
|
| 1007 |
+
"loss": 2.2027830123901366,
|
| 1008 |
+
"step": 1960
|
| 1009 |
+
},
|
| 1010 |
+
{
|
| 1011 |
+
"epoch": 0.13346814964610718,
|
| 1012 |
+
"grad_norm": 0.359375,
|
| 1013 |
+
"learning_rate": 0.0007,
|
| 1014 |
+
"loss": 2.224252128601074,
|
| 1015 |
+
"step": 1980
|
| 1016 |
+
},
|
| 1017 |
+
{
|
| 1018 |
+
"epoch": 0.13481631277384565,
|
| 1019 |
+
"grad_norm": 0.35546875,
|
| 1020 |
+
"learning_rate": 0.0007,
|
| 1021 |
+
"loss": 2.1990388870239257,
|
| 1022 |
+
"step": 2000
|
| 1023 |
+
},
|
| 1024 |
+
{
|
| 1025 |
+
"epoch": 0.13481631277384565,
|
| 1026 |
+
"eval_loss": 2.215721368789673,
|
| 1027 |
+
"eval_runtime": 15.3022,
|
| 1028 |
+
"eval_samples_per_second": 622.59,
|
| 1029 |
+
"eval_steps_per_second": 0.654,
|
| 1030 |
+
"step": 2000
|
| 1031 |
+
},
|
| 1032 |
+
{
|
| 1033 |
+
"epoch": 0.1361644759015841,
|
| 1034 |
+
"grad_norm": 0.345703125,
|
| 1035 |
+
"learning_rate": 0.0007,
|
| 1036 |
+
"loss": 2.207795524597168,
|
| 1037 |
+
"step": 2020
|
| 1038 |
+
},
|
| 1039 |
+
{
|
| 1040 |
+
"epoch": 0.13751263902932254,
|
| 1041 |
+
"grad_norm": 0.396484375,
|
| 1042 |
+
"learning_rate": 0.0007,
|
| 1043 |
+
"loss": 2.1882051467895507,
|
| 1044 |
+
"step": 2040
|
| 1045 |
+
},
|
| 1046 |
+
{
|
| 1047 |
+
"epoch": 0.1381867205931918,
|
| 1048 |
+
"eval_loss": 2.2054405212402344,
|
| 1049 |
+
"eval_runtime": 15.4183,
|
| 1050 |
+
"eval_samples_per_second": 617.904,
|
| 1051 |
+
"eval_steps_per_second": 0.649,
|
| 1052 |
+
"step": 2050
|
| 1053 |
+
},
|
| 1054 |
+
{
|
| 1055 |
+
"epoch": 0.138860802157061,
|
| 1056 |
+
"grad_norm": 0.365234375,
|
| 1057 |
+
"learning_rate": 0.0007,
|
| 1058 |
+
"loss": 2.206527900695801,
|
| 1059 |
+
"step": 2060
|
| 1060 |
+
},
|
| 1061 |
+
{
|
| 1062 |
+
"epoch": 0.14020896528479945,
|
| 1063 |
+
"grad_norm": 0.361328125,
|
| 1064 |
+
"learning_rate": 0.0007,
|
| 1065 |
+
"loss": 2.1837987899780273,
|
| 1066 |
+
"step": 2080
|
| 1067 |
+
},
|
| 1068 |
+
{
|
| 1069 |
+
"epoch": 0.14155712841253792,
|
| 1070 |
+
"grad_norm": 0.33203125,
|
| 1071 |
+
"learning_rate": 0.0007,
|
| 1072 |
+
"loss": 2.1744739532470705,
|
| 1073 |
+
"step": 2100
|
| 1074 |
+
},
|
| 1075 |
+
{
|
| 1076 |
+
"epoch": 0.14155712841253792,
|
| 1077 |
+
"eval_loss": 2.1948788166046143,
|
| 1078 |
+
"eval_runtime": 15.1797,
|
| 1079 |
+
"eval_samples_per_second": 627.615,
|
| 1080 |
+
"eval_steps_per_second": 0.659,
|
| 1081 |
+
"step": 2100
|
| 1082 |
+
},
|
| 1083 |
+
{
|
| 1084 |
+
"epoch": 0.14290529154027637,
|
| 1085 |
+
"grad_norm": 0.361328125,
|
| 1086 |
+
"learning_rate": 0.0007,
|
| 1087 |
+
"loss": 2.182137298583984,
|
| 1088 |
+
"step": 2120
|
| 1089 |
+
},
|
| 1090 |
+
{
|
| 1091 |
+
"epoch": 0.14425345466801484,
|
| 1092 |
+
"grad_norm": 0.337890625,
|
| 1093 |
+
"learning_rate": 0.0007,
|
| 1094 |
+
"loss": 2.1851844787597656,
|
| 1095 |
+
"step": 2140
|
| 1096 |
+
},
|
| 1097 |
+
{
|
| 1098 |
+
"epoch": 0.14492753623188406,
|
| 1099 |
+
"eval_loss": 2.1849257946014404,
|
| 1100 |
+
"eval_runtime": 15.2467,
|
| 1101 |
+
"eval_samples_per_second": 624.858,
|
| 1102 |
+
"eval_steps_per_second": 0.656,
|
| 1103 |
+
"step": 2150
|
| 1104 |
+
},
|
| 1105 |
+
{
|
| 1106 |
+
"epoch": 0.14560161779575329,
|
| 1107 |
+
"grad_norm": 0.3515625,
|
| 1108 |
+
"learning_rate": 0.0007,
|
| 1109 |
+
"loss": 2.1681228637695313,
|
| 1110 |
+
"step": 2160
|
| 1111 |
+
},
|
| 1112 |
+
{
|
| 1113 |
+
"epoch": 0.14694978092349173,
|
| 1114 |
+
"grad_norm": 0.34375,
|
| 1115 |
+
"learning_rate": 0.0007,
|
| 1116 |
+
"loss": 2.1777414321899413,
|
| 1117 |
+
"step": 2180
|
| 1118 |
+
},
|
| 1119 |
+
{
|
| 1120 |
+
"epoch": 0.1482979440512302,
|
| 1121 |
+
"grad_norm": 0.33203125,
|
| 1122 |
+
"learning_rate": 0.0007,
|
| 1123 |
+
"loss": 2.1520442962646484,
|
| 1124 |
+
"step": 2200
|
| 1125 |
+
},
|
| 1126 |
+
{
|
| 1127 |
+
"epoch": 0.1482979440512302,
|
| 1128 |
+
"eval_loss": 2.1730399131774902,
|
| 1129 |
+
"eval_runtime": 15.3329,
|
| 1130 |
+
"eval_samples_per_second": 621.344,
|
| 1131 |
+
"eval_steps_per_second": 0.652,
|
| 1132 |
+
"step": 2200
|
| 1133 |
+
},
|
| 1134 |
+
{
|
| 1135 |
+
"epoch": 0.14964610717896865,
|
| 1136 |
+
"grad_norm": 0.333984375,
|
| 1137 |
+
"learning_rate": 0.0007,
|
| 1138 |
+
"loss": 2.159272384643555,
|
| 1139 |
+
"step": 2220
|
| 1140 |
+
},
|
| 1141 |
+
{
|
| 1142 |
+
"epoch": 0.15099427030670712,
|
| 1143 |
+
"grad_norm": 0.357421875,
|
| 1144 |
+
"learning_rate": 0.0007,
|
| 1145 |
+
"loss": 2.1691991806030275,
|
| 1146 |
+
"step": 2240
|
| 1147 |
+
},
|
| 1148 |
+
{
|
| 1149 |
+
"epoch": 0.15166835187057634,
|
| 1150 |
+
"eval_loss": 2.1636791229248047,
|
| 1151 |
+
"eval_runtime": 15.4687,
|
| 1152 |
+
"eval_samples_per_second": 615.888,
|
| 1153 |
+
"eval_steps_per_second": 0.646,
|
| 1154 |
+
"step": 2250
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 0.15234243343444556,
|
| 1158 |
+
"grad_norm": 0.36328125,
|
| 1159 |
+
"learning_rate": 0.0007,
|
| 1160 |
+
"loss": 2.147579383850098,
|
| 1161 |
+
"step": 2260
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 0.15369059656218403,
|
| 1165 |
+
"grad_norm": 0.34765625,
|
| 1166 |
+
"learning_rate": 0.0007,
|
| 1167 |
+
"loss": 2.1597997665405275,
|
| 1168 |
+
"step": 2280
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 0.15503875968992248,
|
| 1172 |
+
"grad_norm": 0.333984375,
|
| 1173 |
+
"learning_rate": 0.0007,
|
| 1174 |
+
"loss": 2.1413923263549806,
|
| 1175 |
+
"step": 2300
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 0.15503875968992248,
|
| 1179 |
+
"eval_loss": 2.1545023918151855,
|
| 1180 |
+
"eval_runtime": 15.4261,
|
| 1181 |
+
"eval_samples_per_second": 617.588,
|
| 1182 |
+
"eval_steps_per_second": 0.648,
|
| 1183 |
+
"step": 2300
|
| 1184 |
+
},
|
| 1185 |
+
{
|
| 1186 |
+
"epoch": 0.15638692281766095,
|
| 1187 |
+
"grad_norm": 0.33203125,
|
| 1188 |
+
"learning_rate": 0.0007,
|
| 1189 |
+
"loss": 2.1491439819335936,
|
| 1190 |
+
"step": 2320
|
| 1191 |
+
},
|
| 1192 |
+
{
|
| 1193 |
+
"epoch": 0.1577350859453994,
|
| 1194 |
+
"grad_norm": 0.33203125,
|
| 1195 |
+
"learning_rate": 0.0007,
|
| 1196 |
+
"loss": 2.137864112854004,
|
| 1197 |
+
"step": 2340
|
| 1198 |
+
},
|
| 1199 |
+
{
|
| 1200 |
+
"epoch": 0.15840916750926862,
|
| 1201 |
+
"eval_loss": 2.145824432373047,
|
| 1202 |
+
"eval_runtime": 15.654,
|
| 1203 |
+
"eval_samples_per_second": 608.599,
|
| 1204 |
+
"eval_steps_per_second": 0.639,
|
| 1205 |
+
"step": 2350
|
| 1206 |
+
},
|
| 1207 |
+
{
|
| 1208 |
+
"epoch": 0.15908324907313784,
|
| 1209 |
+
"grad_norm": 0.326171875,
|
| 1210 |
+
"learning_rate": 0.0007,
|
| 1211 |
+
"loss": 2.1423198699951174,
|
| 1212 |
+
"step": 2360
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 0.1604314122008763,
|
| 1216 |
+
"grad_norm": 0.33203125,
|
| 1217 |
+
"learning_rate": 0.0007,
|
| 1218 |
+
"loss": 2.143879699707031,
|
| 1219 |
+
"step": 2380
|
| 1220 |
+
},
|
| 1221 |
+
{
|
| 1222 |
+
"epoch": 0.16177957532861476,
|
| 1223 |
+
"grad_norm": 0.3671875,
|
| 1224 |
+
"learning_rate": 0.0007,
|
| 1225 |
+
"loss": 2.1223339080810546,
|
| 1226 |
+
"step": 2400
|
| 1227 |
+
},
|
| 1228 |
+
{
|
| 1229 |
+
"epoch": 0.16177957532861476,
|
| 1230 |
+
"eval_loss": 2.142425537109375,
|
| 1231 |
+
"eval_runtime": 15.4359,
|
| 1232 |
+
"eval_samples_per_second": 617.196,
|
| 1233 |
+
"eval_steps_per_second": 0.648,
|
| 1234 |
+
"step": 2400
|
| 1235 |
+
}
|
| 1236 |
+
],
|
| 1237 |
+
"logging_steps": 20,
|
| 1238 |
+
"max_steps": 2500,
|
| 1239 |
+
"num_input_tokens_seen": 0,
|
| 1240 |
+
"num_train_epochs": 1,
|
| 1241 |
+
"save_steps": 100,
|
| 1242 |
+
"stateful_callbacks": {
|
| 1243 |
+
"TrainerControl": {
|
| 1244 |
+
"args": {
|
| 1245 |
+
"should_epoch_stop": false,
|
| 1246 |
+
"should_evaluate": false,
|
| 1247 |
+
"should_log": false,
|
| 1248 |
+
"should_save": true,
|
| 1249 |
+
"should_training_stop": false
|
| 1250 |
+
},
|
| 1251 |
+
"attributes": {}
|
| 1252 |
+
}
|
| 1253 |
+
},
|
| 1254 |
+
"total_flos": 7743081126297600.0,
|
| 1255 |
+
"train_batch_size": 64,
|
| 1256 |
+
"trial_name": null,
|
| 1257 |
+
"trial_params": null
|
| 1258 |
+
}
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2400/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fd90512e56f813612dc668676d6f8cab72da110a42f23c61cafebf91064625a1
|
| 3 |
+
size 4920
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/config.json
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "silu",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 100,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.16.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096,
|
| 35 |
+
"waleed_beta": 10.0
|
| 36 |
+
}
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f045fa85b5f282452577f935fd9e86fdb9b94b3442e1820631a0c81d9f1f95c6
|
| 3 |
+
size 33967272
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d30ccf9af11823d2d85f7e89da0ed14291a06b4d77d92784611145a5de328539
|
| 3 |
+
size 68504996
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8dccfcc3afceab4d2d2b25fbb67392f946fa9f1c9efbf11612cbcfa90ce2d337
|
| 3 |
+
size 14244
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fae3f7a9f86d9663c06f6e22221bc9e6879cd9691e2b1bba04ee6412e2572f45
|
| 3 |
+
size 1064
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/trainer_state.json
ADDED
|
@@ -0,0 +1,1309 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.16852039096730703,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 2500,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0013481631277384564,
|
| 14 |
+
"grad_norm": 2.078125,
|
| 15 |
+
"learning_rate": 2.66e-05,
|
| 16 |
+
"loss": 8.29962387084961,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.002696326255476913,
|
| 21 |
+
"grad_norm": 1.3515625,
|
| 22 |
+
"learning_rate": 5.46e-05,
|
| 23 |
+
"loss": 8.047650909423828,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.003370407819346141,
|
| 28 |
+
"eval_loss": 7.835535049438477,
|
| 29 |
+
"eval_runtime": 15.2165,
|
| 30 |
+
"eval_samples_per_second": 626.095,
|
| 31 |
+
"eval_steps_per_second": 0.657,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.004044489383215369,
|
| 36 |
+
"grad_norm": 1.2734375,
|
| 37 |
+
"learning_rate": 8.259999999999999e-05,
|
| 38 |
+
"loss": 7.8378746032714846,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.005392652510953826,
|
| 43 |
+
"grad_norm": 1.265625,
|
| 44 |
+
"learning_rate": 0.0001106,
|
| 45 |
+
"loss": 7.5728271484375,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.006740815638692282,
|
| 50 |
+
"grad_norm": 1.1953125,
|
| 51 |
+
"learning_rate": 0.0001386,
|
| 52 |
+
"loss": 7.2332916259765625,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.006740815638692282,
|
| 57 |
+
"eval_loss": 7.039053440093994,
|
| 58 |
+
"eval_runtime": 15.2074,
|
| 59 |
+
"eval_samples_per_second": 626.471,
|
| 60 |
+
"eval_steps_per_second": 0.658,
|
| 61 |
+
"step": 100
|
| 62 |
+
},
|
| 63 |
+
{
|
| 64 |
+
"epoch": 0.008088978766430738,
|
| 65 |
+
"grad_norm": 1.1328125,
|
| 66 |
+
"learning_rate": 0.00016659999999999998,
|
| 67 |
+
"loss": 6.8608039855957035,
|
| 68 |
+
"step": 120
|
| 69 |
+
},
|
| 70 |
+
{
|
| 71 |
+
"epoch": 0.009437141894169195,
|
| 72 |
+
"grad_norm": 0.94921875,
|
| 73 |
+
"learning_rate": 0.00019460000000000001,
|
| 74 |
+
"loss": 6.517842864990234,
|
| 75 |
+
"step": 140
|
| 76 |
+
},
|
| 77 |
+
{
|
| 78 |
+
"epoch": 0.010111223458038422,
|
| 79 |
+
"eval_loss": 6.229617595672607,
|
| 80 |
+
"eval_runtime": 15.1978,
|
| 81 |
+
"eval_samples_per_second": 626.867,
|
| 82 |
+
"eval_steps_per_second": 0.658,
|
| 83 |
+
"step": 150
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"epoch": 0.010785305021907651,
|
| 87 |
+
"grad_norm": 0.80078125,
|
| 88 |
+
"learning_rate": 0.0002226,
|
| 89 |
+
"loss": 6.2399169921875,
|
| 90 |
+
"step": 160
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"epoch": 0.012133468149646108,
|
| 94 |
+
"grad_norm": 0.6640625,
|
| 95 |
+
"learning_rate": 0.00025059999999999997,
|
| 96 |
+
"loss": 6.036548614501953,
|
| 97 |
+
"step": 180
|
| 98 |
+
},
|
| 99 |
+
{
|
| 100 |
+
"epoch": 0.013481631277384564,
|
| 101 |
+
"grad_norm": 0.61328125,
|
| 102 |
+
"learning_rate": 0.0002786,
|
| 103 |
+
"loss": 5.857078552246094,
|
| 104 |
+
"step": 200
|
| 105 |
+
},
|
| 106 |
+
{
|
| 107 |
+
"epoch": 0.013481631277384564,
|
| 108 |
+
"eval_loss": 5.765108585357666,
|
| 109 |
+
"eval_runtime": 15.5271,
|
| 110 |
+
"eval_samples_per_second": 613.574,
|
| 111 |
+
"eval_steps_per_second": 0.644,
|
| 112 |
+
"step": 200
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"epoch": 0.01482979440512302,
|
| 116 |
+
"grad_norm": 0.61328125,
|
| 117 |
+
"learning_rate": 0.00030659999999999997,
|
| 118 |
+
"loss": 5.680564117431641,
|
| 119 |
+
"step": 220
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"epoch": 0.016177957532861477,
|
| 123 |
+
"grad_norm": 2.0625,
|
| 124 |
+
"learning_rate": 0.0003346,
|
| 125 |
+
"loss": 5.472259902954102,
|
| 126 |
+
"step": 240
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"epoch": 0.016852039096730706,
|
| 130 |
+
"eval_loss": 5.240335464477539,
|
| 131 |
+
"eval_runtime": 15.2596,
|
| 132 |
+
"eval_samples_per_second": 624.33,
|
| 133 |
+
"eval_steps_per_second": 0.655,
|
| 134 |
+
"step": 250
|
| 135 |
+
},
|
| 136 |
+
{
|
| 137 |
+
"epoch": 0.01752612066059993,
|
| 138 |
+
"grad_norm": 1.7421875,
|
| 139 |
+
"learning_rate": 0.00036260000000000003,
|
| 140 |
+
"loss": 5.24955940246582,
|
| 141 |
+
"step": 260
|
| 142 |
+
},
|
| 143 |
+
{
|
| 144 |
+
"epoch": 0.01887428378833839,
|
| 145 |
+
"grad_norm": 0.62109375,
|
| 146 |
+
"learning_rate": 0.0003906,
|
| 147 |
+
"loss": 4.995759963989258,
|
| 148 |
+
"step": 280
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"epoch": 0.020222446916076844,
|
| 152 |
+
"grad_norm": 1.390625,
|
| 153 |
+
"learning_rate": 0.0004186,
|
| 154 |
+
"loss": 4.769353866577148,
|
| 155 |
+
"step": 300
|
| 156 |
+
},
|
| 157 |
+
{
|
| 158 |
+
"epoch": 0.020222446916076844,
|
| 159 |
+
"eval_loss": 4.669629096984863,
|
| 160 |
+
"eval_runtime": 15.2595,
|
| 161 |
+
"eval_samples_per_second": 624.332,
|
| 162 |
+
"eval_steps_per_second": 0.655,
|
| 163 |
+
"step": 300
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 0.021570610043815303,
|
| 167 |
+
"grad_norm": 0.73828125,
|
| 168 |
+
"learning_rate": 0.0004466,
|
| 169 |
+
"loss": 4.569867324829102,
|
| 170 |
+
"step": 320
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 0.022918773171553757,
|
| 174 |
+
"grad_norm": 0.765625,
|
| 175 |
+
"learning_rate": 0.00047460000000000004,
|
| 176 |
+
"loss": 4.389446258544922,
|
| 177 |
+
"step": 340
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.023592854735422986,
|
| 181 |
+
"eval_loss": 4.231207847595215,
|
| 182 |
+
"eval_runtime": 15.2143,
|
| 183 |
+
"eval_samples_per_second": 626.188,
|
| 184 |
+
"eval_steps_per_second": 0.657,
|
| 185 |
+
"step": 350
|
| 186 |
+
},
|
| 187 |
+
{
|
| 188 |
+
"epoch": 0.024266936299292215,
|
| 189 |
+
"grad_norm": 1.2421875,
|
| 190 |
+
"learning_rate": 0.0005026,
|
| 191 |
+
"loss": 4.243848419189453,
|
| 192 |
+
"step": 360
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.02561509942703067,
|
| 196 |
+
"grad_norm": 0.7109375,
|
| 197 |
+
"learning_rate": 0.0005306,
|
| 198 |
+
"loss": 4.1090648651123045,
|
| 199 |
+
"step": 380
|
| 200 |
+
},
|
| 201 |
+
{
|
| 202 |
+
"epoch": 0.026963262554769128,
|
| 203 |
+
"grad_norm": 0.484375,
|
| 204 |
+
"learning_rate": 0.0005586,
|
| 205 |
+
"loss": 3.966172790527344,
|
| 206 |
+
"step": 400
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 0.026963262554769128,
|
| 210 |
+
"eval_loss": 3.923773765563965,
|
| 211 |
+
"eval_runtime": 15.5308,
|
| 212 |
+
"eval_samples_per_second": 613.426,
|
| 213 |
+
"eval_steps_per_second": 0.644,
|
| 214 |
+
"step": 400
|
| 215 |
+
},
|
| 216 |
+
{
|
| 217 |
+
"epoch": 0.028311425682507583,
|
| 218 |
+
"grad_norm": 0.6640625,
|
| 219 |
+
"learning_rate": 0.0005866,
|
| 220 |
+
"loss": 3.8559597015380858,
|
| 221 |
+
"step": 420
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 0.02965958881024604,
|
| 225 |
+
"grad_norm": 0.515625,
|
| 226 |
+
"learning_rate": 0.0006146,
|
| 227 |
+
"loss": 3.7943866729736326,
|
| 228 |
+
"step": 440
|
| 229 |
+
},
|
| 230 |
+
{
|
| 231 |
+
"epoch": 0.030333670374115267,
|
| 232 |
+
"eval_loss": 3.6977553367614746,
|
| 233 |
+
"eval_runtime": 15.2001,
|
| 234 |
+
"eval_samples_per_second": 626.77,
|
| 235 |
+
"eval_steps_per_second": 0.658,
|
| 236 |
+
"step": 450
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.031007751937984496,
|
| 240 |
+
"grad_norm": 0.51953125,
|
| 241 |
+
"learning_rate": 0.0006426,
|
| 242 |
+
"loss": 3.686443328857422,
|
| 243 |
+
"step": 460
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"epoch": 0.032355915065722954,
|
| 247 |
+
"grad_norm": 0.62109375,
|
| 248 |
+
"learning_rate": 0.0006705999999999999,
|
| 249 |
+
"loss": 3.6123157501220704,
|
| 250 |
+
"step": 480
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.03370407819346141,
|
| 254 |
+
"grad_norm": 0.70703125,
|
| 255 |
+
"learning_rate": 0.0006986,
|
| 256 |
+
"loss": 3.542654800415039,
|
| 257 |
+
"step": 500
|
| 258 |
+
},
|
| 259 |
+
{
|
| 260 |
+
"epoch": 0.03370407819346141,
|
| 261 |
+
"eval_loss": 3.513820171356201,
|
| 262 |
+
"eval_runtime": 15.2193,
|
| 263 |
+
"eval_samples_per_second": 625.983,
|
| 264 |
+
"eval_steps_per_second": 0.657,
|
| 265 |
+
"step": 500
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.03505224132119986,
|
| 269 |
+
"grad_norm": 0.48046875,
|
| 270 |
+
"learning_rate": 0.0007,
|
| 271 |
+
"loss": 3.472199249267578,
|
| 272 |
+
"step": 520
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.03640040444893832,
|
| 276 |
+
"grad_norm": 0.53515625,
|
| 277 |
+
"learning_rate": 0.0007,
|
| 278 |
+
"loss": 3.396291732788086,
|
| 279 |
+
"step": 540
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.03707448601280755,
|
| 283 |
+
"eval_loss": 3.3581483364105225,
|
| 284 |
+
"eval_runtime": 15.3044,
|
| 285 |
+
"eval_samples_per_second": 622.502,
|
| 286 |
+
"eval_steps_per_second": 0.653,
|
| 287 |
+
"step": 550
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"epoch": 0.03774856757667678,
|
| 291 |
+
"grad_norm": 0.47265625,
|
| 292 |
+
"learning_rate": 0.0007,
|
| 293 |
+
"loss": 3.347300720214844,
|
| 294 |
+
"step": 560
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.03909673070441524,
|
| 298 |
+
"grad_norm": 0.458984375,
|
| 299 |
+
"learning_rate": 0.0007,
|
| 300 |
+
"loss": 3.2885902404785154,
|
| 301 |
+
"step": 580
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"epoch": 0.04044489383215369,
|
| 305 |
+
"grad_norm": 0.4375,
|
| 306 |
+
"learning_rate": 0.0007,
|
| 307 |
+
"loss": 3.25723876953125,
|
| 308 |
+
"step": 600
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"epoch": 0.04044489383215369,
|
| 312 |
+
"eval_loss": 3.2229573726654053,
|
| 313 |
+
"eval_runtime": 15.2967,
|
| 314 |
+
"eval_samples_per_second": 622.812,
|
| 315 |
+
"eval_steps_per_second": 0.654,
|
| 316 |
+
"step": 600
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"epoch": 0.04179305695989215,
|
| 320 |
+
"grad_norm": 0.458984375,
|
| 321 |
+
"learning_rate": 0.0007,
|
| 322 |
+
"loss": 3.1902334213256838,
|
| 323 |
+
"step": 620
|
| 324 |
+
},
|
| 325 |
+
{
|
| 326 |
+
"epoch": 0.043141220087630605,
|
| 327 |
+
"grad_norm": 0.40625,
|
| 328 |
+
"learning_rate": 0.0007,
|
| 329 |
+
"loss": 3.167861557006836,
|
| 330 |
+
"step": 640
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"epoch": 0.043815301651499834,
|
| 334 |
+
"eval_loss": 3.120636463165283,
|
| 335 |
+
"eval_runtime": 15.3751,
|
| 336 |
+
"eval_samples_per_second": 619.639,
|
| 337 |
+
"eval_steps_per_second": 0.65,
|
| 338 |
+
"step": 650
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 0.044489383215369056,
|
| 342 |
+
"grad_norm": 0.494140625,
|
| 343 |
+
"learning_rate": 0.0007,
|
| 344 |
+
"loss": 3.1089324951171875,
|
| 345 |
+
"step": 660
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 0.045837546343107514,
|
| 349 |
+
"grad_norm": 0.435546875,
|
| 350 |
+
"learning_rate": 0.0007,
|
| 351 |
+
"loss": 3.0828268051147463,
|
| 352 |
+
"step": 680
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.04718570947084597,
|
| 356 |
+
"grad_norm": 0.427734375,
|
| 357 |
+
"learning_rate": 0.0007,
|
| 358 |
+
"loss": 3.0396488189697264,
|
| 359 |
+
"step": 700
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 0.04718570947084597,
|
| 363 |
+
"eval_loss": 3.0276408195495605,
|
| 364 |
+
"eval_runtime": 15.2735,
|
| 365 |
+
"eval_samples_per_second": 623.761,
|
| 366 |
+
"eval_steps_per_second": 0.655,
|
| 367 |
+
"step": 700
|
| 368 |
+
},
|
| 369 |
+
{
|
| 370 |
+
"epoch": 0.04853387259858443,
|
| 371 |
+
"grad_norm": 0.4375,
|
| 372 |
+
"learning_rate": 0.0007,
|
| 373 |
+
"loss": 3.0034805297851563,
|
| 374 |
+
"step": 720
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 0.04988203572632288,
|
| 378 |
+
"grad_norm": 0.52734375,
|
| 379 |
+
"learning_rate": 0.0007,
|
| 380 |
+
"loss": 2.96556396484375,
|
| 381 |
+
"step": 740
|
| 382 |
+
},
|
| 383 |
+
{
|
| 384 |
+
"epoch": 0.05055611729019211,
|
| 385 |
+
"eval_loss": 2.950009346008301,
|
| 386 |
+
"eval_runtime": 15.243,
|
| 387 |
+
"eval_samples_per_second": 625.009,
|
| 388 |
+
"eval_steps_per_second": 0.656,
|
| 389 |
+
"step": 750
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.05123019885406134,
|
| 393 |
+
"grad_norm": 0.451171875,
|
| 394 |
+
"learning_rate": 0.0007,
|
| 395 |
+
"loss": 2.950779151916504,
|
| 396 |
+
"step": 760
|
| 397 |
+
},
|
| 398 |
+
{
|
| 399 |
+
"epoch": 0.0525783619817998,
|
| 400 |
+
"grad_norm": 0.400390625,
|
| 401 |
+
"learning_rate": 0.0007,
|
| 402 |
+
"loss": 2.956203079223633,
|
| 403 |
+
"step": 780
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 0.053926525109538256,
|
| 407 |
+
"grad_norm": 0.40625,
|
| 408 |
+
"learning_rate": 0.0007,
|
| 409 |
+
"loss": 2.9017578125,
|
| 410 |
+
"step": 800
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 0.053926525109538256,
|
| 414 |
+
"eval_loss": 2.8844006061553955,
|
| 415 |
+
"eval_runtime": 15.2903,
|
| 416 |
+
"eval_samples_per_second": 623.073,
|
| 417 |
+
"eval_steps_per_second": 0.654,
|
| 418 |
+
"step": 800
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.05527468823727671,
|
| 422 |
+
"grad_norm": 0.41015625,
|
| 423 |
+
"learning_rate": 0.0007,
|
| 424 |
+
"loss": 2.887515640258789,
|
| 425 |
+
"step": 820
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.056622851365015166,
|
| 429 |
+
"grad_norm": 0.41796875,
|
| 430 |
+
"learning_rate": 0.0007,
|
| 431 |
+
"loss": 2.8556928634643555,
|
| 432 |
+
"step": 840
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.057296932928884395,
|
| 436 |
+
"eval_loss": 2.821610450744629,
|
| 437 |
+
"eval_runtime": 15.2253,
|
| 438 |
+
"eval_samples_per_second": 625.733,
|
| 439 |
+
"eval_steps_per_second": 0.657,
|
| 440 |
+
"step": 850
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.057971014492753624,
|
| 444 |
+
"grad_norm": 0.423828125,
|
| 445 |
+
"learning_rate": 0.0007,
|
| 446 |
+
"loss": 2.8130855560302734,
|
| 447 |
+
"step": 860
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.05931917762049208,
|
| 451 |
+
"grad_norm": 0.38671875,
|
| 452 |
+
"learning_rate": 0.0007,
|
| 453 |
+
"loss": 2.790329170227051,
|
| 454 |
+
"step": 880
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.06066734074823053,
|
| 458 |
+
"grad_norm": 0.427734375,
|
| 459 |
+
"learning_rate": 0.0007,
|
| 460 |
+
"loss": 2.7760162353515625,
|
| 461 |
+
"step": 900
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.06066734074823053,
|
| 465 |
+
"eval_loss": 2.7661373615264893,
|
| 466 |
+
"eval_runtime": 15.2736,
|
| 467 |
+
"eval_samples_per_second": 623.755,
|
| 468 |
+
"eval_steps_per_second": 0.655,
|
| 469 |
+
"step": 900
|
| 470 |
+
},
|
| 471 |
+
{
|
| 472 |
+
"epoch": 0.06201550387596899,
|
| 473 |
+
"grad_norm": 0.435546875,
|
| 474 |
+
"learning_rate": 0.0007,
|
| 475 |
+
"loss": 2.7462047576904296,
|
| 476 |
+
"step": 920
|
| 477 |
+
},
|
| 478 |
+
{
|
| 479 |
+
"epoch": 0.06336366700370745,
|
| 480 |
+
"grad_norm": 0.4140625,
|
| 481 |
+
"learning_rate": 0.0007,
|
| 482 |
+
"loss": 2.732274627685547,
|
| 483 |
+
"step": 940
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.06403774856757667,
|
| 487 |
+
"eval_loss": 2.713942289352417,
|
| 488 |
+
"eval_runtime": 15.6062,
|
| 489 |
+
"eval_samples_per_second": 610.463,
|
| 490 |
+
"eval_steps_per_second": 0.641,
|
| 491 |
+
"step": 950
|
| 492 |
+
},
|
| 493 |
+
{
|
| 494 |
+
"epoch": 0.06471183013144591,
|
| 495 |
+
"grad_norm": 0.419921875,
|
| 496 |
+
"learning_rate": 0.0007,
|
| 497 |
+
"loss": 2.705318069458008,
|
| 498 |
+
"step": 960
|
| 499 |
+
},
|
| 500 |
+
{
|
| 501 |
+
"epoch": 0.06605999325918437,
|
| 502 |
+
"grad_norm": 0.4296875,
|
| 503 |
+
"learning_rate": 0.0007,
|
| 504 |
+
"loss": 2.6892818450927733,
|
| 505 |
+
"step": 980
|
| 506 |
+
},
|
| 507 |
+
{
|
| 508 |
+
"epoch": 0.06740815638692282,
|
| 509 |
+
"grad_norm": 0.4453125,
|
| 510 |
+
"learning_rate": 0.0007,
|
| 511 |
+
"loss": 2.687115287780762,
|
| 512 |
+
"step": 1000
|
| 513 |
+
},
|
| 514 |
+
{
|
| 515 |
+
"epoch": 0.06740815638692282,
|
| 516 |
+
"eval_loss": 2.668647289276123,
|
| 517 |
+
"eval_runtime": 15.2597,
|
| 518 |
+
"eval_samples_per_second": 624.326,
|
| 519 |
+
"eval_steps_per_second": 0.655,
|
| 520 |
+
"step": 1000
|
| 521 |
+
},
|
| 522 |
+
{
|
| 523 |
+
"epoch": 0.06875631951466127,
|
| 524 |
+
"grad_norm": 0.388671875,
|
| 525 |
+
"learning_rate": 0.0007,
|
| 526 |
+
"loss": 2.6533071517944338,
|
| 527 |
+
"step": 1020
|
| 528 |
+
},
|
| 529 |
+
{
|
| 530 |
+
"epoch": 0.07010448264239973,
|
| 531 |
+
"grad_norm": 0.392578125,
|
| 532 |
+
"learning_rate": 0.0007,
|
| 533 |
+
"loss": 2.6390922546386717,
|
| 534 |
+
"step": 1040
|
| 535 |
+
},
|
| 536 |
+
{
|
| 537 |
+
"epoch": 0.07077856420626896,
|
| 538 |
+
"eval_loss": 2.6291816234588623,
|
| 539 |
+
"eval_runtime": 15.2457,
|
| 540 |
+
"eval_samples_per_second": 624.896,
|
| 541 |
+
"eval_steps_per_second": 0.656,
|
| 542 |
+
"step": 1050
|
| 543 |
+
},
|
| 544 |
+
{
|
| 545 |
+
"epoch": 0.07145264577013818,
|
| 546 |
+
"grad_norm": 0.4375,
|
| 547 |
+
"learning_rate": 0.0007,
|
| 548 |
+
"loss": 2.624461555480957,
|
| 549 |
+
"step": 1060
|
| 550 |
+
},
|
| 551 |
+
{
|
| 552 |
+
"epoch": 0.07280080889787664,
|
| 553 |
+
"grad_norm": 0.396484375,
|
| 554 |
+
"learning_rate": 0.0007,
|
| 555 |
+
"loss": 2.6044427871704103,
|
| 556 |
+
"step": 1080
|
| 557 |
+
},
|
| 558 |
+
{
|
| 559 |
+
"epoch": 0.0741489720256151,
|
| 560 |
+
"grad_norm": 0.423828125,
|
| 561 |
+
"learning_rate": 0.0007,
|
| 562 |
+
"loss": 2.597646713256836,
|
| 563 |
+
"step": 1100
|
| 564 |
+
},
|
| 565 |
+
{
|
| 566 |
+
"epoch": 0.0741489720256151,
|
| 567 |
+
"eval_loss": 2.589696168899536,
|
| 568 |
+
"eval_runtime": 15.3967,
|
| 569 |
+
"eval_samples_per_second": 618.768,
|
| 570 |
+
"eval_steps_per_second": 0.649,
|
| 571 |
+
"step": 1100
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 0.07549713515335356,
|
| 575 |
+
"grad_norm": 0.369140625,
|
| 576 |
+
"learning_rate": 0.0007,
|
| 577 |
+
"loss": 2.561947250366211,
|
| 578 |
+
"step": 1120
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 0.07684529828109202,
|
| 582 |
+
"grad_norm": 0.390625,
|
| 583 |
+
"learning_rate": 0.0007,
|
| 584 |
+
"loss": 2.54862117767334,
|
| 585 |
+
"step": 1140
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 0.07751937984496124,
|
| 589 |
+
"eval_loss": 2.559185743331909,
|
| 590 |
+
"eval_runtime": 15.2877,
|
| 591 |
+
"eval_samples_per_second": 623.181,
|
| 592 |
+
"eval_steps_per_second": 0.654,
|
| 593 |
+
"step": 1150
|
| 594 |
+
},
|
| 595 |
+
{
|
| 596 |
+
"epoch": 0.07819346140883048,
|
| 597 |
+
"grad_norm": 0.37890625,
|
| 598 |
+
"learning_rate": 0.0007,
|
| 599 |
+
"loss": 2.5469717025756835,
|
| 600 |
+
"step": 1160
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"epoch": 0.07954162453656892,
|
| 604 |
+
"grad_norm": 0.400390625,
|
| 605 |
+
"learning_rate": 0.0007,
|
| 606 |
+
"loss": 2.540451240539551,
|
| 607 |
+
"step": 1180
|
| 608 |
+
},
|
| 609 |
+
{
|
| 610 |
+
"epoch": 0.08088978766430738,
|
| 611 |
+
"grad_norm": 0.40625,
|
| 612 |
+
"learning_rate": 0.0007,
|
| 613 |
+
"loss": 2.53134708404541,
|
| 614 |
+
"step": 1200
|
| 615 |
+
},
|
| 616 |
+
{
|
| 617 |
+
"epoch": 0.08088978766430738,
|
| 618 |
+
"eval_loss": 2.521665096282959,
|
| 619 |
+
"eval_runtime": 15.2822,
|
| 620 |
+
"eval_samples_per_second": 623.407,
|
| 621 |
+
"eval_steps_per_second": 0.654,
|
| 622 |
+
"step": 1200
|
| 623 |
+
},
|
| 624 |
+
{
|
| 625 |
+
"epoch": 0.08223795079204584,
|
| 626 |
+
"grad_norm": 0.37109375,
|
| 627 |
+
"learning_rate": 0.0007,
|
| 628 |
+
"loss": 2.513539123535156,
|
| 629 |
+
"step": 1220
|
| 630 |
+
},
|
| 631 |
+
{
|
| 632 |
+
"epoch": 0.0835861139197843,
|
| 633 |
+
"grad_norm": 0.36328125,
|
| 634 |
+
"learning_rate": 0.0007,
|
| 635 |
+
"loss": 2.5009738922119142,
|
| 636 |
+
"step": 1240
|
| 637 |
+
},
|
| 638 |
+
{
|
| 639 |
+
"epoch": 0.08426019548365352,
|
| 640 |
+
"eval_loss": 2.489459753036499,
|
| 641 |
+
"eval_runtime": 15.2493,
|
| 642 |
+
"eval_samples_per_second": 624.749,
|
| 643 |
+
"eval_steps_per_second": 0.656,
|
| 644 |
+
"step": 1250
|
| 645 |
+
},
|
| 646 |
+
{
|
| 647 |
+
"epoch": 0.08493427704752275,
|
| 648 |
+
"grad_norm": 0.388671875,
|
| 649 |
+
"learning_rate": 0.0007,
|
| 650 |
+
"loss": 2.4894077301025392,
|
| 651 |
+
"step": 1260
|
| 652 |
+
},
|
| 653 |
+
{
|
| 654 |
+
"epoch": 0.08628244017526121,
|
| 655 |
+
"grad_norm": 0.369140625,
|
| 656 |
+
"learning_rate": 0.0007,
|
| 657 |
+
"loss": 2.4697439193725588,
|
| 658 |
+
"step": 1280
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"epoch": 0.08763060330299967,
|
| 662 |
+
"grad_norm": 0.423828125,
|
| 663 |
+
"learning_rate": 0.0007,
|
| 664 |
+
"loss": 2.461294174194336,
|
| 665 |
+
"step": 1300
|
| 666 |
+
},
|
| 667 |
+
{
|
| 668 |
+
"epoch": 0.08763060330299967,
|
| 669 |
+
"eval_loss": 2.4627816677093506,
|
| 670 |
+
"eval_runtime": 15.421,
|
| 671 |
+
"eval_samples_per_second": 617.795,
|
| 672 |
+
"eval_steps_per_second": 0.648,
|
| 673 |
+
"step": 1300
|
| 674 |
+
},
|
| 675 |
+
{
|
| 676 |
+
"epoch": 0.08897876643073811,
|
| 677 |
+
"grad_norm": 0.36328125,
|
| 678 |
+
"learning_rate": 0.0007,
|
| 679 |
+
"loss": 2.452680206298828,
|
| 680 |
+
"step": 1320
|
| 681 |
+
},
|
| 682 |
+
{
|
| 683 |
+
"epoch": 0.09032692955847657,
|
| 684 |
+
"grad_norm": 0.404296875,
|
| 685 |
+
"learning_rate": 0.0007,
|
| 686 |
+
"loss": 2.4396684646606444,
|
| 687 |
+
"step": 1340
|
| 688 |
+
},
|
| 689 |
+
{
|
| 690 |
+
"epoch": 0.0910010111223458,
|
| 691 |
+
"eval_loss": 2.437220335006714,
|
| 692 |
+
"eval_runtime": 15.3026,
|
| 693 |
+
"eval_samples_per_second": 622.576,
|
| 694 |
+
"eval_steps_per_second": 0.653,
|
| 695 |
+
"step": 1350
|
| 696 |
+
},
|
| 697 |
+
{
|
| 698 |
+
"epoch": 0.09167509268621503,
|
| 699 |
+
"grad_norm": 0.412109375,
|
| 700 |
+
"learning_rate": 0.0007,
|
| 701 |
+
"loss": 2.4397760391235352,
|
| 702 |
+
"step": 1360
|
| 703 |
+
},
|
| 704 |
+
{
|
| 705 |
+
"epoch": 0.09302325581395349,
|
| 706 |
+
"grad_norm": 0.353515625,
|
| 707 |
+
"learning_rate": 0.0007,
|
| 708 |
+
"loss": 2.4158536911010744,
|
| 709 |
+
"step": 1380
|
| 710 |
+
},
|
| 711 |
+
{
|
| 712 |
+
"epoch": 0.09437141894169195,
|
| 713 |
+
"grad_norm": 0.345703125,
|
| 714 |
+
"learning_rate": 0.0007,
|
| 715 |
+
"loss": 2.406942367553711,
|
| 716 |
+
"step": 1400
|
| 717 |
+
},
|
| 718 |
+
{
|
| 719 |
+
"epoch": 0.09437141894169195,
|
| 720 |
+
"eval_loss": 2.4096601009368896,
|
| 721 |
+
"eval_runtime": 15.357,
|
| 722 |
+
"eval_samples_per_second": 620.368,
|
| 723 |
+
"eval_steps_per_second": 0.651,
|
| 724 |
+
"step": 1400
|
| 725 |
+
},
|
| 726 |
+
{
|
| 727 |
+
"epoch": 0.0957195820694304,
|
| 728 |
+
"grad_norm": 0.36328125,
|
| 729 |
+
"learning_rate": 0.0007,
|
| 730 |
+
"loss": 2.4161144256591798,
|
| 731 |
+
"step": 1420
|
| 732 |
+
},
|
| 733 |
+
{
|
| 734 |
+
"epoch": 0.09706774519716886,
|
| 735 |
+
"grad_norm": 0.36328125,
|
| 736 |
+
"learning_rate": 0.0007,
|
| 737 |
+
"loss": 2.392984390258789,
|
| 738 |
+
"step": 1440
|
| 739 |
+
},
|
| 740 |
+
{
|
| 741 |
+
"epoch": 0.09774182676103808,
|
| 742 |
+
"eval_loss": 2.3937113285064697,
|
| 743 |
+
"eval_runtime": 15.3555,
|
| 744 |
+
"eval_samples_per_second": 620.43,
|
| 745 |
+
"eval_steps_per_second": 0.651,
|
| 746 |
+
"step": 1450
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 0.09841590832490732,
|
| 750 |
+
"grad_norm": 0.384765625,
|
| 751 |
+
"learning_rate": 0.0007,
|
| 752 |
+
"loss": 2.3820331573486326,
|
| 753 |
+
"step": 1460
|
| 754 |
+
},
|
| 755 |
+
{
|
| 756 |
+
"epoch": 0.09976407145264576,
|
| 757 |
+
"grad_norm": 0.33984375,
|
| 758 |
+
"learning_rate": 0.0007,
|
| 759 |
+
"loss": 2.3687986373901366,
|
| 760 |
+
"step": 1480
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"epoch": 0.10111223458038422,
|
| 764 |
+
"grad_norm": 0.38671875,
|
| 765 |
+
"learning_rate": 0.0007,
|
| 766 |
+
"loss": 2.355032730102539,
|
| 767 |
+
"step": 1500
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 0.10111223458038422,
|
| 771 |
+
"eval_loss": 2.3681657314300537,
|
| 772 |
+
"eval_runtime": 15.3336,
|
| 773 |
+
"eval_samples_per_second": 621.317,
|
| 774 |
+
"eval_steps_per_second": 0.652,
|
| 775 |
+
"step": 1500
|
| 776 |
+
},
|
| 777 |
+
{
|
| 778 |
+
"epoch": 0.10246039770812268,
|
| 779 |
+
"grad_norm": 0.392578125,
|
| 780 |
+
"learning_rate": 0.0007,
|
| 781 |
+
"loss": 2.3572179794311525,
|
| 782 |
+
"step": 1520
|
| 783 |
+
},
|
| 784 |
+
{
|
| 785 |
+
"epoch": 0.10380856083586114,
|
| 786 |
+
"grad_norm": 0.35546875,
|
| 787 |
+
"learning_rate": 0.0007,
|
| 788 |
+
"loss": 2.345840835571289,
|
| 789 |
+
"step": 1540
|
| 790 |
+
},
|
| 791 |
+
{
|
| 792 |
+
"epoch": 0.10448264239973036,
|
| 793 |
+
"eval_loss": 2.3475193977355957,
|
| 794 |
+
"eval_runtime": 15.3453,
|
| 795 |
+
"eval_samples_per_second": 620.842,
|
| 796 |
+
"eval_steps_per_second": 0.652,
|
| 797 |
+
"step": 1550
|
| 798 |
+
},
|
| 799 |
+
{
|
| 800 |
+
"epoch": 0.1051567239635996,
|
| 801 |
+
"grad_norm": 0.388671875,
|
| 802 |
+
"learning_rate": 0.0007,
|
| 803 |
+
"loss": 2.335231971740723,
|
| 804 |
+
"step": 1560
|
| 805 |
+
},
|
| 806 |
+
{
|
| 807 |
+
"epoch": 0.10650488709133805,
|
| 808 |
+
"grad_norm": 0.359375,
|
| 809 |
+
"learning_rate": 0.0007,
|
| 810 |
+
"loss": 2.3543283462524416,
|
| 811 |
+
"step": 1580
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 0.10785305021907651,
|
| 815 |
+
"grad_norm": 0.39453125,
|
| 816 |
+
"learning_rate": 0.0007,
|
| 817 |
+
"loss": 2.3237049102783205,
|
| 818 |
+
"step": 1600
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 0.10785305021907651,
|
| 822 |
+
"eval_loss": 2.3377676010131836,
|
| 823 |
+
"eval_runtime": 15.2858,
|
| 824 |
+
"eval_samples_per_second": 623.259,
|
| 825 |
+
"eval_steps_per_second": 0.654,
|
| 826 |
+
"step": 1600
|
| 827 |
+
},
|
| 828 |
+
{
|
| 829 |
+
"epoch": 0.10920121334681497,
|
| 830 |
+
"grad_norm": 0.357421875,
|
| 831 |
+
"learning_rate": 0.0007,
|
| 832 |
+
"loss": 2.3196277618408203,
|
| 833 |
+
"step": 1620
|
| 834 |
+
},
|
| 835 |
+
{
|
| 836 |
+
"epoch": 0.11054937647455342,
|
| 837 |
+
"grad_norm": 0.341796875,
|
| 838 |
+
"learning_rate": 0.0007,
|
| 839 |
+
"loss": 2.3308162689208984,
|
| 840 |
+
"step": 1640
|
| 841 |
+
},
|
| 842 |
+
{
|
| 843 |
+
"epoch": 0.11122345803842265,
|
| 844 |
+
"eval_loss": 2.313627004623413,
|
| 845 |
+
"eval_runtime": 15.276,
|
| 846 |
+
"eval_samples_per_second": 623.659,
|
| 847 |
+
"eval_steps_per_second": 0.655,
|
| 848 |
+
"step": 1650
|
| 849 |
+
},
|
| 850 |
+
{
|
| 851 |
+
"epoch": 0.11189753960229187,
|
| 852 |
+
"grad_norm": 0.353515625,
|
| 853 |
+
"learning_rate": 0.0007,
|
| 854 |
+
"loss": 2.3104904174804686,
|
| 855 |
+
"step": 1660
|
| 856 |
+
},
|
| 857 |
+
{
|
| 858 |
+
"epoch": 0.11324570273003033,
|
| 859 |
+
"grad_norm": 0.353515625,
|
| 860 |
+
"learning_rate": 0.0007,
|
| 861 |
+
"loss": 2.2955816268920897,
|
| 862 |
+
"step": 1680
|
| 863 |
+
},
|
| 864 |
+
{
|
| 865 |
+
"epoch": 0.11459386585776879,
|
| 866 |
+
"grad_norm": 0.345703125,
|
| 867 |
+
"learning_rate": 0.0007,
|
| 868 |
+
"loss": 2.3044401168823243,
|
| 869 |
+
"step": 1700
|
| 870 |
+
},
|
| 871 |
+
{
|
| 872 |
+
"epoch": 0.11459386585776879,
|
| 873 |
+
"eval_loss": 2.2976956367492676,
|
| 874 |
+
"eval_runtime": 15.3404,
|
| 875 |
+
"eval_samples_per_second": 621.039,
|
| 876 |
+
"eval_steps_per_second": 0.652,
|
| 877 |
+
"step": 1700
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 0.11594202898550725,
|
| 881 |
+
"grad_norm": 0.373046875,
|
| 882 |
+
"learning_rate": 0.0007,
|
| 883 |
+
"loss": 2.280081939697266,
|
| 884 |
+
"step": 1720
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 0.1172901921132457,
|
| 888 |
+
"grad_norm": 0.373046875,
|
| 889 |
+
"learning_rate": 0.0007,
|
| 890 |
+
"loss": 2.269576644897461,
|
| 891 |
+
"step": 1740
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 0.11796427367711493,
|
| 895 |
+
"eval_loss": 2.28181791305542,
|
| 896 |
+
"eval_runtime": 15.2137,
|
| 897 |
+
"eval_samples_per_second": 626.211,
|
| 898 |
+
"eval_steps_per_second": 0.657,
|
| 899 |
+
"step": 1750
|
| 900 |
+
},
|
| 901 |
+
{
|
| 902 |
+
"epoch": 0.11863835524098416,
|
| 903 |
+
"grad_norm": 0.359375,
|
| 904 |
+
"learning_rate": 0.0007,
|
| 905 |
+
"loss": 2.273081398010254,
|
| 906 |
+
"step": 1760
|
| 907 |
+
},
|
| 908 |
+
{
|
| 909 |
+
"epoch": 0.11998651836872262,
|
| 910 |
+
"grad_norm": 0.345703125,
|
| 911 |
+
"learning_rate": 0.0007,
|
| 912 |
+
"loss": 2.2877025604248047,
|
| 913 |
+
"step": 1780
|
| 914 |
+
},
|
| 915 |
+
{
|
| 916 |
+
"epoch": 0.12133468149646107,
|
| 917 |
+
"grad_norm": 0.3671875,
|
| 918 |
+
"learning_rate": 0.0007,
|
| 919 |
+
"loss": 2.2828989028930664,
|
| 920 |
+
"step": 1800
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 0.12133468149646107,
|
| 924 |
+
"eval_loss": 2.2671334743499756,
|
| 925 |
+
"eval_runtime": 15.1615,
|
| 926 |
+
"eval_samples_per_second": 628.369,
|
| 927 |
+
"eval_steps_per_second": 0.66,
|
| 928 |
+
"step": 1800
|
| 929 |
+
},
|
| 930 |
+
{
|
| 931 |
+
"epoch": 0.12268284462419952,
|
| 932 |
+
"grad_norm": 0.388671875,
|
| 933 |
+
"learning_rate": 0.0007,
|
| 934 |
+
"loss": 2.2743486404418944,
|
| 935 |
+
"step": 1820
|
| 936 |
+
},
|
| 937 |
+
{
|
| 938 |
+
"epoch": 0.12403100775193798,
|
| 939 |
+
"grad_norm": 0.3671875,
|
| 940 |
+
"learning_rate": 0.0007,
|
| 941 |
+
"loss": 2.2429489135742187,
|
| 942 |
+
"step": 1840
|
| 943 |
+
},
|
| 944 |
+
{
|
| 945 |
+
"epoch": 0.12470508931580722,
|
| 946 |
+
"eval_loss": 2.252441644668579,
|
| 947 |
+
"eval_runtime": 15.3255,
|
| 948 |
+
"eval_samples_per_second": 621.643,
|
| 949 |
+
"eval_steps_per_second": 0.653,
|
| 950 |
+
"step": 1850
|
| 951 |
+
},
|
| 952 |
+
{
|
| 953 |
+
"epoch": 0.12537917087967643,
|
| 954 |
+
"grad_norm": 0.33984375,
|
| 955 |
+
"learning_rate": 0.0007,
|
| 956 |
+
"loss": 2.242429733276367,
|
| 957 |
+
"step": 1860
|
| 958 |
+
},
|
| 959 |
+
{
|
| 960 |
+
"epoch": 0.1267273340074149,
|
| 961 |
+
"grad_norm": 0.353515625,
|
| 962 |
+
"learning_rate": 0.0007,
|
| 963 |
+
"loss": 2.2348007202148437,
|
| 964 |
+
"step": 1880
|
| 965 |
+
},
|
| 966 |
+
{
|
| 967 |
+
"epoch": 0.12807549713515334,
|
| 968 |
+
"grad_norm": 0.330078125,
|
| 969 |
+
"learning_rate": 0.0007,
|
| 970 |
+
"loss": 2.241002655029297,
|
| 971 |
+
"step": 1900
|
| 972 |
+
},
|
| 973 |
+
{
|
| 974 |
+
"epoch": 0.12807549713515334,
|
| 975 |
+
"eval_loss": 2.2387285232543945,
|
| 976 |
+
"eval_runtime": 15.2942,
|
| 977 |
+
"eval_samples_per_second": 622.915,
|
| 978 |
+
"eval_steps_per_second": 0.654,
|
| 979 |
+
"step": 1900
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 0.12942366026289182,
|
| 983 |
+
"grad_norm": 0.34765625,
|
| 984 |
+
"learning_rate": 0.0007,
|
| 985 |
+
"loss": 2.2172693252563476,
|
| 986 |
+
"step": 1920
|
| 987 |
+
},
|
| 988 |
+
{
|
| 989 |
+
"epoch": 0.13077182339063026,
|
| 990 |
+
"grad_norm": 0.357421875,
|
| 991 |
+
"learning_rate": 0.0007,
|
| 992 |
+
"loss": 2.2283090591430663,
|
| 993 |
+
"step": 1940
|
| 994 |
+
},
|
| 995 |
+
{
|
| 996 |
+
"epoch": 0.13144590495449948,
|
| 997 |
+
"eval_loss": 2.225919246673584,
|
| 998 |
+
"eval_runtime": 15.3259,
|
| 999 |
+
"eval_samples_per_second": 621.627,
|
| 1000 |
+
"eval_steps_per_second": 0.652,
|
| 1001 |
+
"step": 1950
|
| 1002 |
+
},
|
| 1003 |
+
{
|
| 1004 |
+
"epoch": 0.13211998651836873,
|
| 1005 |
+
"grad_norm": 0.333984375,
|
| 1006 |
+
"learning_rate": 0.0007,
|
| 1007 |
+
"loss": 2.2027830123901366,
|
| 1008 |
+
"step": 1960
|
| 1009 |
+
},
|
| 1010 |
+
{
|
| 1011 |
+
"epoch": 0.13346814964610718,
|
| 1012 |
+
"grad_norm": 0.359375,
|
| 1013 |
+
"learning_rate": 0.0007,
|
| 1014 |
+
"loss": 2.224252128601074,
|
| 1015 |
+
"step": 1980
|
| 1016 |
+
},
|
| 1017 |
+
{
|
| 1018 |
+
"epoch": 0.13481631277384565,
|
| 1019 |
+
"grad_norm": 0.35546875,
|
| 1020 |
+
"learning_rate": 0.0007,
|
| 1021 |
+
"loss": 2.1990388870239257,
|
| 1022 |
+
"step": 2000
|
| 1023 |
+
},
|
| 1024 |
+
{
|
| 1025 |
+
"epoch": 0.13481631277384565,
|
| 1026 |
+
"eval_loss": 2.215721368789673,
|
| 1027 |
+
"eval_runtime": 15.3022,
|
| 1028 |
+
"eval_samples_per_second": 622.59,
|
| 1029 |
+
"eval_steps_per_second": 0.654,
|
| 1030 |
+
"step": 2000
|
| 1031 |
+
},
|
| 1032 |
+
{
|
| 1033 |
+
"epoch": 0.1361644759015841,
|
| 1034 |
+
"grad_norm": 0.345703125,
|
| 1035 |
+
"learning_rate": 0.0007,
|
| 1036 |
+
"loss": 2.207795524597168,
|
| 1037 |
+
"step": 2020
|
| 1038 |
+
},
|
| 1039 |
+
{
|
| 1040 |
+
"epoch": 0.13751263902932254,
|
| 1041 |
+
"grad_norm": 0.396484375,
|
| 1042 |
+
"learning_rate": 0.0007,
|
| 1043 |
+
"loss": 2.1882051467895507,
|
| 1044 |
+
"step": 2040
|
| 1045 |
+
},
|
| 1046 |
+
{
|
| 1047 |
+
"epoch": 0.1381867205931918,
|
| 1048 |
+
"eval_loss": 2.2054405212402344,
|
| 1049 |
+
"eval_runtime": 15.4183,
|
| 1050 |
+
"eval_samples_per_second": 617.904,
|
| 1051 |
+
"eval_steps_per_second": 0.649,
|
| 1052 |
+
"step": 2050
|
| 1053 |
+
},
|
| 1054 |
+
{
|
| 1055 |
+
"epoch": 0.138860802157061,
|
| 1056 |
+
"grad_norm": 0.365234375,
|
| 1057 |
+
"learning_rate": 0.0007,
|
| 1058 |
+
"loss": 2.206527900695801,
|
| 1059 |
+
"step": 2060
|
| 1060 |
+
},
|
| 1061 |
+
{
|
| 1062 |
+
"epoch": 0.14020896528479945,
|
| 1063 |
+
"grad_norm": 0.361328125,
|
| 1064 |
+
"learning_rate": 0.0007,
|
| 1065 |
+
"loss": 2.1837987899780273,
|
| 1066 |
+
"step": 2080
|
| 1067 |
+
},
|
| 1068 |
+
{
|
| 1069 |
+
"epoch": 0.14155712841253792,
|
| 1070 |
+
"grad_norm": 0.33203125,
|
| 1071 |
+
"learning_rate": 0.0007,
|
| 1072 |
+
"loss": 2.1744739532470705,
|
| 1073 |
+
"step": 2100
|
| 1074 |
+
},
|
| 1075 |
+
{
|
| 1076 |
+
"epoch": 0.14155712841253792,
|
| 1077 |
+
"eval_loss": 2.1948788166046143,
|
| 1078 |
+
"eval_runtime": 15.1797,
|
| 1079 |
+
"eval_samples_per_second": 627.615,
|
| 1080 |
+
"eval_steps_per_second": 0.659,
|
| 1081 |
+
"step": 2100
|
| 1082 |
+
},
|
| 1083 |
+
{
|
| 1084 |
+
"epoch": 0.14290529154027637,
|
| 1085 |
+
"grad_norm": 0.361328125,
|
| 1086 |
+
"learning_rate": 0.0007,
|
| 1087 |
+
"loss": 2.182137298583984,
|
| 1088 |
+
"step": 2120
|
| 1089 |
+
},
|
| 1090 |
+
{
|
| 1091 |
+
"epoch": 0.14425345466801484,
|
| 1092 |
+
"grad_norm": 0.337890625,
|
| 1093 |
+
"learning_rate": 0.0007,
|
| 1094 |
+
"loss": 2.1851844787597656,
|
| 1095 |
+
"step": 2140
|
| 1096 |
+
},
|
| 1097 |
+
{
|
| 1098 |
+
"epoch": 0.14492753623188406,
|
| 1099 |
+
"eval_loss": 2.1849257946014404,
|
| 1100 |
+
"eval_runtime": 15.2467,
|
| 1101 |
+
"eval_samples_per_second": 624.858,
|
| 1102 |
+
"eval_steps_per_second": 0.656,
|
| 1103 |
+
"step": 2150
|
| 1104 |
+
},
|
| 1105 |
+
{
|
| 1106 |
+
"epoch": 0.14560161779575329,
|
| 1107 |
+
"grad_norm": 0.3515625,
|
| 1108 |
+
"learning_rate": 0.0007,
|
| 1109 |
+
"loss": 2.1681228637695313,
|
| 1110 |
+
"step": 2160
|
| 1111 |
+
},
|
| 1112 |
+
{
|
| 1113 |
+
"epoch": 0.14694978092349173,
|
| 1114 |
+
"grad_norm": 0.34375,
|
| 1115 |
+
"learning_rate": 0.0007,
|
| 1116 |
+
"loss": 2.1777414321899413,
|
| 1117 |
+
"step": 2180
|
| 1118 |
+
},
|
| 1119 |
+
{
|
| 1120 |
+
"epoch": 0.1482979440512302,
|
| 1121 |
+
"grad_norm": 0.33203125,
|
| 1122 |
+
"learning_rate": 0.0007,
|
| 1123 |
+
"loss": 2.1520442962646484,
|
| 1124 |
+
"step": 2200
|
| 1125 |
+
},
|
| 1126 |
+
{
|
| 1127 |
+
"epoch": 0.1482979440512302,
|
| 1128 |
+
"eval_loss": 2.1730399131774902,
|
| 1129 |
+
"eval_runtime": 15.3329,
|
| 1130 |
+
"eval_samples_per_second": 621.344,
|
| 1131 |
+
"eval_steps_per_second": 0.652,
|
| 1132 |
+
"step": 2200
|
| 1133 |
+
},
|
| 1134 |
+
{
|
| 1135 |
+
"epoch": 0.14964610717896865,
|
| 1136 |
+
"grad_norm": 0.333984375,
|
| 1137 |
+
"learning_rate": 0.0007,
|
| 1138 |
+
"loss": 2.159272384643555,
|
| 1139 |
+
"step": 2220
|
| 1140 |
+
},
|
| 1141 |
+
{
|
| 1142 |
+
"epoch": 0.15099427030670712,
|
| 1143 |
+
"grad_norm": 0.357421875,
|
| 1144 |
+
"learning_rate": 0.0007,
|
| 1145 |
+
"loss": 2.1691991806030275,
|
| 1146 |
+
"step": 2240
|
| 1147 |
+
},
|
| 1148 |
+
{
|
| 1149 |
+
"epoch": 0.15166835187057634,
|
| 1150 |
+
"eval_loss": 2.1636791229248047,
|
| 1151 |
+
"eval_runtime": 15.4687,
|
| 1152 |
+
"eval_samples_per_second": 615.888,
|
| 1153 |
+
"eval_steps_per_second": 0.646,
|
| 1154 |
+
"step": 2250
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 0.15234243343444556,
|
| 1158 |
+
"grad_norm": 0.36328125,
|
| 1159 |
+
"learning_rate": 0.0007,
|
| 1160 |
+
"loss": 2.147579383850098,
|
| 1161 |
+
"step": 2260
|
| 1162 |
+
},
|
| 1163 |
+
{
|
| 1164 |
+
"epoch": 0.15369059656218403,
|
| 1165 |
+
"grad_norm": 0.34765625,
|
| 1166 |
+
"learning_rate": 0.0007,
|
| 1167 |
+
"loss": 2.1597997665405275,
|
| 1168 |
+
"step": 2280
|
| 1169 |
+
},
|
| 1170 |
+
{
|
| 1171 |
+
"epoch": 0.15503875968992248,
|
| 1172 |
+
"grad_norm": 0.333984375,
|
| 1173 |
+
"learning_rate": 0.0007,
|
| 1174 |
+
"loss": 2.1413923263549806,
|
| 1175 |
+
"step": 2300
|
| 1176 |
+
},
|
| 1177 |
+
{
|
| 1178 |
+
"epoch": 0.15503875968992248,
|
| 1179 |
+
"eval_loss": 2.1545023918151855,
|
| 1180 |
+
"eval_runtime": 15.4261,
|
| 1181 |
+
"eval_samples_per_second": 617.588,
|
| 1182 |
+
"eval_steps_per_second": 0.648,
|
| 1183 |
+
"step": 2300
|
| 1184 |
+
},
|
| 1185 |
+
{
|
| 1186 |
+
"epoch": 0.15638692281766095,
|
| 1187 |
+
"grad_norm": 0.33203125,
|
| 1188 |
+
"learning_rate": 0.0007,
|
| 1189 |
+
"loss": 2.1491439819335936,
|
| 1190 |
+
"step": 2320
|
| 1191 |
+
},
|
| 1192 |
+
{
|
| 1193 |
+
"epoch": 0.1577350859453994,
|
| 1194 |
+
"grad_norm": 0.33203125,
|
| 1195 |
+
"learning_rate": 0.0007,
|
| 1196 |
+
"loss": 2.137864112854004,
|
| 1197 |
+
"step": 2340
|
| 1198 |
+
},
|
| 1199 |
+
{
|
| 1200 |
+
"epoch": 0.15840916750926862,
|
| 1201 |
+
"eval_loss": 2.145824432373047,
|
| 1202 |
+
"eval_runtime": 15.654,
|
| 1203 |
+
"eval_samples_per_second": 608.599,
|
| 1204 |
+
"eval_steps_per_second": 0.639,
|
| 1205 |
+
"step": 2350
|
| 1206 |
+
},
|
| 1207 |
+
{
|
| 1208 |
+
"epoch": 0.15908324907313784,
|
| 1209 |
+
"grad_norm": 0.326171875,
|
| 1210 |
+
"learning_rate": 0.0007,
|
| 1211 |
+
"loss": 2.1423198699951174,
|
| 1212 |
+
"step": 2360
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 0.1604314122008763,
|
| 1216 |
+
"grad_norm": 0.33203125,
|
| 1217 |
+
"learning_rate": 0.0007,
|
| 1218 |
+
"loss": 2.143879699707031,
|
| 1219 |
+
"step": 2380
|
| 1220 |
+
},
|
| 1221 |
+
{
|
| 1222 |
+
"epoch": 0.16177957532861476,
|
| 1223 |
+
"grad_norm": 0.3671875,
|
| 1224 |
+
"learning_rate": 0.0007,
|
| 1225 |
+
"loss": 2.1223339080810546,
|
| 1226 |
+
"step": 2400
|
| 1227 |
+
},
|
| 1228 |
+
{
|
| 1229 |
+
"epoch": 0.16177957532861476,
|
| 1230 |
+
"eval_loss": 2.142425537109375,
|
| 1231 |
+
"eval_runtime": 15.4359,
|
| 1232 |
+
"eval_samples_per_second": 617.196,
|
| 1233 |
+
"eval_steps_per_second": 0.648,
|
| 1234 |
+
"step": 2400
|
| 1235 |
+
},
|
| 1236 |
+
{
|
| 1237 |
+
"epoch": 0.16312773845635323,
|
| 1238 |
+
"grad_norm": 0.341796875,
|
| 1239 |
+
"learning_rate": 0.0007,
|
| 1240 |
+
"loss": 2.1280586242675783,
|
| 1241 |
+
"step": 2420
|
| 1242 |
+
},
|
| 1243 |
+
{
|
| 1244 |
+
"epoch": 0.16447590158409167,
|
| 1245 |
+
"grad_norm": 0.33984375,
|
| 1246 |
+
"learning_rate": 0.0007,
|
| 1247 |
+
"loss": 2.1163003921508787,
|
| 1248 |
+
"step": 2440
|
| 1249 |
+
},
|
| 1250 |
+
{
|
| 1251 |
+
"epoch": 0.1651499831479609,
|
| 1252 |
+
"eval_loss": 2.130929708480835,
|
| 1253 |
+
"eval_runtime": 15.3525,
|
| 1254 |
+
"eval_samples_per_second": 620.55,
|
| 1255 |
+
"eval_steps_per_second": 0.651,
|
| 1256 |
+
"step": 2450
|
| 1257 |
+
},
|
| 1258 |
+
{
|
| 1259 |
+
"epoch": 0.16582406471183014,
|
| 1260 |
+
"grad_norm": 0.326171875,
|
| 1261 |
+
"learning_rate": 0.0007,
|
| 1262 |
+
"loss": 2.1273950576782226,
|
| 1263 |
+
"step": 2460
|
| 1264 |
+
},
|
| 1265 |
+
{
|
| 1266 |
+
"epoch": 0.1671722278395686,
|
| 1267 |
+
"grad_norm": 0.34375,
|
| 1268 |
+
"learning_rate": 0.0007,
|
| 1269 |
+
"loss": 2.129487419128418,
|
| 1270 |
+
"step": 2480
|
| 1271 |
+
},
|
| 1272 |
+
{
|
| 1273 |
+
"epoch": 0.16852039096730703,
|
| 1274 |
+
"grad_norm": 0.328125,
|
| 1275 |
+
"learning_rate": 0.0007,
|
| 1276 |
+
"loss": 2.109725570678711,
|
| 1277 |
+
"step": 2500
|
| 1278 |
+
},
|
| 1279 |
+
{
|
| 1280 |
+
"epoch": 0.16852039096730703,
|
| 1281 |
+
"eval_loss": 2.1222457885742188,
|
| 1282 |
+
"eval_runtime": 15.3726,
|
| 1283 |
+
"eval_samples_per_second": 619.739,
|
| 1284 |
+
"eval_steps_per_second": 0.651,
|
| 1285 |
+
"step": 2500
|
| 1286 |
+
}
|
| 1287 |
+
],
|
| 1288 |
+
"logging_steps": 20,
|
| 1289 |
+
"max_steps": 2500,
|
| 1290 |
+
"num_input_tokens_seen": 0,
|
| 1291 |
+
"num_train_epochs": 1,
|
| 1292 |
+
"save_steps": 100,
|
| 1293 |
+
"stateful_callbacks": {
|
| 1294 |
+
"TrainerControl": {
|
| 1295 |
+
"args": {
|
| 1296 |
+
"should_epoch_stop": false,
|
| 1297 |
+
"should_evaluate": false,
|
| 1298 |
+
"should_log": false,
|
| 1299 |
+
"should_save": true,
|
| 1300 |
+
"should_training_stop": true
|
| 1301 |
+
},
|
| 1302 |
+
"attributes": {}
|
| 1303 |
+
}
|
| 1304 |
+
},
|
| 1305 |
+
"total_flos": 8065709506560000.0,
|
| 1306 |
+
"train_batch_size": 64,
|
| 1307 |
+
"trial_name": null,
|
| 1308 |
+
"trial_params": null
|
| 1309 |
+
}
|
zain/Activation/out/glu-silu-100L_run/checkpoint-2500/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fd90512e56f813612dc668676d6f8cab72da110a42f23c61cafebf91064625a1
|
| 3 |
+
size 4920
|
zain/Activation/out/glu-silu-100L_run/config.json
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "silu",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 100,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.16.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096,
|
| 35 |
+
"waleed_beta": 10.0
|
| 36 |
+
}
|
zain/Activation/out/glu-silu-100L_run/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f045fa85b5f282452577f935fd9e86fdb9b94b3442e1820631a0c81d9f1f95c6
|
| 3 |
+
size 33967272
|
zain/Activation/out/glu-silu-100L_run/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/glu-silu-100L_run/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
zain/Activation/out/glu-silu-100L_run/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fd90512e56f813612dc668676d6f8cab72da110a42f23c61cafebf91064625a1
|
| 3 |
+
size 4920
|
zain/Activation/out/glu-silu-100L_run/training_log.jsonl
CHANGED
|
@@ -163,3 +163,17 @@
|
|
| 163 |
{"step": 2300, "epoch": 0.15503875968992248, "timestamp": 1786571954.7473707, "eval_loss": 2.1545023918151855, "eval_runtime": 15.4261, "eval_samples_per_second": 617.588, "eval_steps_per_second": 0.648, "train/total_time_seconds": 544.8871062174439, "train/time_per_step_avg": 0.23878501038998365, "train/epoch_time_elapsed": 1369.4057754948735, "train/estimated_remaining_minutes": 0.7896914582861506}
|
| 164 |
{"step": 2320, "epoch": 0.15638692281766095, "timestamp": 1786571960.8122804, "loss": 2.1491439819335936, "grad_norm": 0.33203125, "learning_rate": 0.0007, "train/total_time_seconds": 549.6645726785064, "train/time_per_step_avg": 0.2389586379006505, "train/epoch_time_elapsed": 1375.4706870652735, "train/estimated_remaining_minutes": 0.7107731543256548}
|
| 165 |
{"step": 2340, "epoch": 0.1577350859453994, "timestamp": 1786571966.612273, "loss": 2.137864112854004, "grad_norm": 0.33203125, "learning_rate": 0.0007, "train/total_time_seconds": 554.4445280842483, "train/time_per_step_avg": 0.23901270151138307, "train/epoch_time_elapsed": 1381.2706793919206, "train/estimated_remaining_minutes": 0.6318456160504254}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 163 |
{"step": 2300, "epoch": 0.15503875968992248, "timestamp": 1786571954.7473707, "eval_loss": 2.1545023918151855, "eval_runtime": 15.4261, "eval_samples_per_second": 617.588, "eval_steps_per_second": 0.648, "train/total_time_seconds": 544.8871062174439, "train/time_per_step_avg": 0.23878501038998365, "train/epoch_time_elapsed": 1369.4057754948735, "train/estimated_remaining_minutes": 0.7896914582861506}
|
| 164 |
{"step": 2320, "epoch": 0.15638692281766095, "timestamp": 1786571960.8122804, "loss": 2.1491439819335936, "grad_norm": 0.33203125, "learning_rate": 0.0007, "train/total_time_seconds": 549.6645726785064, "train/time_per_step_avg": 0.2389586379006505, "train/epoch_time_elapsed": 1375.4706870652735, "train/estimated_remaining_minutes": 0.7107731543256548}
|
| 165 |
{"step": 2340, "epoch": 0.1577350859453994, "timestamp": 1786571966.612273, "loss": 2.137864112854004, "grad_norm": 0.33203125, "learning_rate": 0.0007, "train/total_time_seconds": 554.4445280842483, "train/time_per_step_avg": 0.23901270151138307, "train/epoch_time_elapsed": 1381.2706793919206, "train/estimated_remaining_minutes": 0.6318456160504254}
|
| 166 |
+
{"step": 2350, "epoch": 0.15840916750926862, "timestamp": 1786571985.1526456, "eval_loss": 2.145824432373047, "eval_runtime": 15.654, "eval_samples_per_second": 608.599, "eval_steps_per_second": 0.639, "train/total_time_seconds": 556.8283236846328, "train/time_per_step_avg": 0.23877111468464135, "train/epoch_time_elapsed": 1399.8110508657992, "train/estimated_remaining_minutes": 0.5923705571113114}
|
| 167 |
+
{"step": 2360, "epoch": 0.15908324907313784, "timestamp": 1786571988.0664167, "loss": 2.1423198699951174, "grad_norm": 0.326171875, "learning_rate": 0.0007, "train/total_time_seconds": 559.2268124930561, "train/time_per_step_avg": 0.2389866068586707, "train/epoch_time_elapsed": 1402.7248232401907, "train/estimated_remaining_minutes": 0.5529078654592362}
|
| 168 |
+
{"step": 2380, "epoch": 0.1604314122008763, "timestamp": 1786571993.8099024, "loss": 2.143879699707031, "grad_norm": 0.33203125, "learning_rate": 0.0007, "train/total_time_seconds": 563.9822333268821, "train/time_per_step_avg": 0.23892352018505336, "train/epoch_time_elapsed": 1408.468309007585, "train/estimated_remaining_minutes": 0.47393464985452277}
|
| 169 |
+
{"step": 2400, "epoch": 0.16177957532861476, "timestamp": 1786571999.6064997, "loss": 2.1223339080810546, "grad_norm": 0.3671875, "learning_rate": 0.0007, "train/total_time_seconds": 568.7588178738952, "train/time_per_step_avg": 0.23871711656451225, "train/epoch_time_elapsed": 1414.2649062387645, "train/estimated_remaining_minutes": 0.3949714013013161}
|
| 170 |
+
{"step": 2400, "epoch": 0.16177957532861476, "timestamp": 1786572015.044009, "eval_loss": 2.142425537109375, "eval_runtime": 15.4359, "eval_samples_per_second": 617.196, "eval_steps_per_second": 0.648, "train/total_time_seconds": 568.7588178738952, "train/time_per_step_avg": 0.23871711656451225, "train/epoch_time_elapsed": 1429.702413905412, "train/estimated_remaining_minutes": 0.3949714013013161}
|
| 171 |
+
{"step": 2420, "epoch": 0.16312773845635323, "timestamp": 1786572021.136019, "loss": 2.1280586242675783, "grad_norm": 0.341796875, "learning_rate": 0.0007, "train/total_time_seconds": 573.538488149643, "train/time_per_step_avg": 0.2387391547113657, "train/epoch_time_elapsed": 1435.794425956905, "train/estimated_remaining_minutes": 0.31599916702459663}
|
| 172 |
+
{"step": 2440, "epoch": 0.16447590158409167, "timestamp": 1786572026.9621944, "loss": 2.1163003921508787, "grad_norm": 0.33984375, "learning_rate": 0.0007, "train/total_time_seconds": 578.3542434461415, "train/time_per_step_avg": 0.23909715361893177, "train/epoch_time_elapsed": 1441.6206013932824, "train/estimated_remaining_minutes": 0.23703042764186127}
|
| 173 |
+
{"step": 2450, "epoch": 0.1651499831479609, "timestamp": 1786572045.1965752, "eval_loss": 2.130929708480835, "eval_runtime": 15.3525, "eval_samples_per_second": 620.55, "eval_steps_per_second": 0.651, "train/total_time_seconds": 580.7406451590359, "train/time_per_step_avg": 0.23912321474403142, "train/epoch_time_elapsed": 1459.8549805395305, "train/estimated_remaining_minutes": 0.19753083168674693}
|
| 174 |
+
{"step": 2460, "epoch": 0.16582406471183014, "timestamp": 1786572048.1221642, "loss": 2.1273950576782226, "grad_norm": 0.326171875, "learning_rate": 0.0007, "train/total_time_seconds": 583.1407674327493, "train/time_per_step_avg": 0.23913954939693213, "train/epoch_time_elapsed": 1462.780570179224, "train/estimated_remaining_minutes": 0.15803272830155807}
|
| 175 |
+
{"step": 2480, "epoch": 0.1671722278395686, "timestamp": 1786572053.8920004, "loss": 2.129487419128418, "grad_norm": 0.34375, "learning_rate": 0.0007, "train/total_time_seconds": 587.9142602346838, "train/time_per_step_avg": 0.2393202690780163, "train/epoch_time_elapsed": 1468.5504074655473, "train/estimated_remaining_minutes": 0.07902073390251126}
|
| 176 |
+
{"step": 2500, "epoch": 0.16852039096730703, "timestamp": 1786572059.68274, "loss": 2.109725570678711, "grad_norm": 0.328125, "learning_rate": 0.0007, "train/total_time_seconds": 592.685107909143, "train/time_per_step_avg": 0.23926290035247802, "train/epoch_time_elapsed": 1474.3411462754011, "train/estimated_remaining_minutes": 0.0}
|
| 177 |
+
{"step": 2500, "epoch": 0.16852039096730703, "timestamp": 1786572075.0569973, "eval_loss": 2.1222457885742188, "eval_runtime": 15.3726, "eval_samples_per_second": 619.739, "eval_steps_per_second": 0.651, "train/total_time_seconds": 592.685107909143, "train/time_per_step_avg": 0.23926290035247802, "train/epoch_time_elapsed": 1489.7154013849795, "train/estimated_remaining_minutes": 0.0}
|
| 178 |
+
{"step": 2500, "epoch": 0.16852039096730703, "timestamp": 1786572075.3533838, "train_runtime": 1491.0291, "train_samples_per_second": 107.308, "train_steps_per_second": 1.677, "total_flos": 8065709506560000.0, "train_loss": 3.075948629760742, "train/total_time_seconds": 592.685107909143, "train/time_per_step_avg": 0.23926290035247802, "train/epoch_time_elapsed": 1490.0117889344692, "train/estimated_remaining_minutes": 0.0}
|
| 179 |
+
{"step": 2500, "epoch": 0.16852039096730703, "timestamp": 1786572090.6925077, "eval_loss": 2.1222457885742188, "eval_runtime": 15.3355, "eval_samples_per_second": 621.239, "eval_steps_per_second": 0.652, "train/total_time_seconds": 592.685107909143, "train/time_per_step_avg": 0.23926290035247802, "train/epoch_time_elapsed": 1505.3509125448763, "train/estimated_remaining_minutes": 0.0}
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/config.json
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "waleed10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 100,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.16.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096,
|
| 35 |
+
"waleed_beta": 10.0
|
| 36 |
+
}
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e849d2aa9701ac2658fd762e3709646c628759e1c8baa34a1eeb6c00a70e9776
|
| 3 |
+
size 33967272
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e4d18f60303922e49bfed23c274dab4db12c630ef3a23a75a6dfecbe2ac9bb88
|
| 3 |
+
size 68504996
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3cf9097d4513154245c48236b6ec5137b7ee2a21c9f58f2cba798ea275c6026f
|
| 3 |
+
size 14244
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3001486ba00eb51d89ef078fe70e7e37535cd4e2311d7dd589266d045c9fb892
|
| 3 |
+
size 1064
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/tokenizer_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": false,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<|endoftext|>",
|
| 5 |
+
"eos_token": "<|endoftext|>",
|
| 6 |
+
"errors": "replace",
|
| 7 |
+
"is_local": false,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1024,
|
| 10 |
+
"pad_token": "<|endoftext|>",
|
| 11 |
+
"tokenizer_class": "GPT2Tokenizer",
|
| 12 |
+
"unk_token": "<|endoftext|>"
|
| 13 |
+
}
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/trainer_state.json
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.006740815638692282,
|
| 6 |
+
"eval_steps": 50,
|
| 7 |
+
"global_step": 100,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0013481631277384564,
|
| 14 |
+
"grad_norm": 2.125,
|
| 15 |
+
"learning_rate": 2.66e-05,
|
| 16 |
+
"loss": 8.29942626953125,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.002696326255476913,
|
| 21 |
+
"grad_norm": 1.3046875,
|
| 22 |
+
"learning_rate": 5.46e-05,
|
| 23 |
+
"loss": 8.036360168457032,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.003370407819346141,
|
| 28 |
+
"eval_loss": 7.832070827484131,
|
| 29 |
+
"eval_runtime": 15.4822,
|
| 30 |
+
"eval_samples_per_second": 615.352,
|
| 31 |
+
"eval_steps_per_second": 0.646,
|
| 32 |
+
"step": 50
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 0.004044489383215369,
|
| 36 |
+
"grad_norm": 1.2734375,
|
| 37 |
+
"learning_rate": 8.259999999999999e-05,
|
| 38 |
+
"loss": 7.834564208984375,
|
| 39 |
+
"step": 60
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.005392652510953826,
|
| 43 |
+
"grad_norm": 1.265625,
|
| 44 |
+
"learning_rate": 0.0001106,
|
| 45 |
+
"loss": 7.5718849182128904,
|
| 46 |
+
"step": 80
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"epoch": 0.006740815638692282,
|
| 50 |
+
"grad_norm": 1.1953125,
|
| 51 |
+
"learning_rate": 0.0001386,
|
| 52 |
+
"loss": 7.2372383117675785,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.006740815638692282,
|
| 57 |
+
"eval_loss": 7.0443572998046875,
|
| 58 |
+
"eval_runtime": 15.5098,
|
| 59 |
+
"eval_samples_per_second": 614.257,
|
| 60 |
+
"eval_steps_per_second": 0.645,
|
| 61 |
+
"step": 100
|
| 62 |
+
}
|
| 63 |
+
],
|
| 64 |
+
"logging_steps": 20,
|
| 65 |
+
"max_steps": 2500,
|
| 66 |
+
"num_input_tokens_seen": 0,
|
| 67 |
+
"num_train_epochs": 1,
|
| 68 |
+
"save_steps": 100,
|
| 69 |
+
"stateful_callbacks": {
|
| 70 |
+
"TrainerControl": {
|
| 71 |
+
"args": {
|
| 72 |
+
"should_epoch_stop": false,
|
| 73 |
+
"should_evaluate": false,
|
| 74 |
+
"should_log": false,
|
| 75 |
+
"should_save": true,
|
| 76 |
+
"should_training_stop": false
|
| 77 |
+
},
|
| 78 |
+
"attributes": {}
|
| 79 |
+
}
|
| 80 |
+
},
|
| 81 |
+
"total_flos": 322628380262400.0,
|
| 82 |
+
"train_batch_size": 64,
|
| 83 |
+
"trial_name": null,
|
| 84 |
+
"trial_params": null
|
| 85 |
+
}
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-100/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:95042bce3268a58e39bf6898ae4dc3d7f41ae0258221813fdf6720b6f7599e44
|
| 3 |
+
size 4920
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/config.json
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "waleed10",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"TinyLlamaForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 1,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eos_token_id": 2,
|
| 11 |
+
"head_dim": 32,
|
| 12 |
+
"hidden_act": "silu",
|
| 13 |
+
"hidden_size": 128,
|
| 14 |
+
"initializer_range": 0.02,
|
| 15 |
+
"intermediate_size": 256,
|
| 16 |
+
"max_position_embeddings": 512,
|
| 17 |
+
"mlp_bias": false,
|
| 18 |
+
"mlp_type": "glu",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 100,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"pretraining_tp": 1,
|
| 25 |
+
"rms_norm_eps": 1e-06,
|
| 26 |
+
"rope_parameters": {
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"rope_type": "default"
|
| 29 |
+
},
|
| 30 |
+
"tie_word_embeddings": true,
|
| 31 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 32 |
+
"transformers_version": "5.16.0.dev0",
|
| 33 |
+
"use_cache": false,
|
| 34 |
+
"vocab_size": 4096,
|
| 35 |
+
"waleed_beta": 10.0
|
| 36 |
+
}
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0d1fa8cb579eecffad489fd4cec7e2a3d9d3fe35881a7cc2399dc4ac53186207
|
| 3 |
+
size 33967272
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c59b0e8734f31cdf307508221b8cd33e3928c5a4fc4e7ec93316f115d367fdc4
|
| 3 |
+
size 68504996
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ecefbb3f17bb76b6655eb0157c98b5287c17fa4b4c72a6b9068b0823ce9fd18d
|
| 3 |
+
size 14244
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:28668eba09299968c820a8f82f4368de9f79bd22806c6be52cebe6b8700b6584
|
| 3 |
+
size 1064
|
zain/Activation/out/glu-waleed10-100L_run/checkpoint-200/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|