Auto upload zain 2026-08-20T23:29:28.088837 (part 2)
Browse filesThis view is limited to 50 files because it contains too many changes. See raw diff
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/model.safetensors +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/optimizer.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/rng_state.pth +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/scheduler.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/tokenizer.json +0 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/tokenizer_config.json +13 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/trainer_state.json +894 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/training_args.bin +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/config.json +37 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/model.safetensors +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/optimizer.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/rng_state.pth +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/scheduler.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/tokenizer.json +0 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/tokenizer_config.json +13 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/trainer_state.json +937 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/training_args.bin +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/config.json +37 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/model.safetensors +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/optimizer.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/rng_state.pth +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/scheduler.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/tokenizer.json +0 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/tokenizer_config.json +13 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/trainer_state.json +980 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/training_args.bin +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/config.json +37 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/model.safetensors +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/optimizer.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/rng_state.pth +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/scheduler.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/tokenizer.json +0 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/tokenizer_config.json +13 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/trainer_state.json +1023 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/training_args.bin +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/config.json +37 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/model.safetensors +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/optimizer.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/rng_state.pth +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/scheduler.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/tokenizer.json +0 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/tokenizer_config.json +13 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/trainer_state.json +1066 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/training_args.bin +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/config.json +37 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/model.safetensors +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/optimizer.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/rng_state.pth +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/scheduler.pt +3 -0
- zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/tokenizer.json +0 -0
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ebdc3d1268f9aca9cdfcf3b91c4b79ddf61f9eff3403c03bf3235306049bb9e9
|
| 3 |
+
size 4010544
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0f23d734b75fe871a7dda809ea71ae896af619d940f35f6487be7e4c3b3cbada
|
| 3 |
+
size 8068282
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:645b71843375baf4cc10bc75a2f0f04e91b8e3f8e8929f518a87022898b3bc20
|
| 3 |
+
size 14244
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:56cab430bf2883ab08ecdb0930091aaa86b7be034aaf50147f79e0060e5cbc6c
|
| 3 |
+
size 1064
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/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/mlp-squared_relu-9L_run/checkpoint-2000/trainer_state.json
ADDED
|
@@ -0,0 +1,894 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.06740815638692282,
|
| 6 |
+
"eval_steps": 100,
|
| 7 |
+
"global_step": 2000,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0006740815638692282,
|
| 14 |
+
"grad_norm": 2.71875,
|
| 15 |
+
"learning_rate": 1.14e-05,
|
| 16 |
+
"loss": 8.314944458007812,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.0013481631277384564,
|
| 21 |
+
"grad_norm": 1.8515625,
|
| 22 |
+
"learning_rate": 2.34e-05,
|
| 23 |
+
"loss": 8.251383972167968,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.0020222446916076846,
|
| 28 |
+
"grad_norm": 1.3203125,
|
| 29 |
+
"learning_rate": 3.539999999999999e-05,
|
| 30 |
+
"loss": 8.104505920410157,
|
| 31 |
+
"step": 60
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"epoch": 0.002696326255476913,
|
| 35 |
+
"grad_norm": 1.296875,
|
| 36 |
+
"learning_rate": 4.7399999999999993e-05,
|
| 37 |
+
"loss": 7.968361663818359,
|
| 38 |
+
"step": 80
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.003370407819346141,
|
| 42 |
+
"grad_norm": 1.2578125,
|
| 43 |
+
"learning_rate": 5.94e-05,
|
| 44 |
+
"loss": 7.804827880859375,
|
| 45 |
+
"step": 100
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.003370407819346141,
|
| 49 |
+
"eval_loss": 7.703802585601807,
|
| 50 |
+
"eval_runtime": 8.0325,
|
| 51 |
+
"eval_samples_per_second": 1186.059,
|
| 52 |
+
"eval_steps_per_second": 0.871,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.004044489383215369,
|
| 57 |
+
"grad_norm": 1.2734375,
|
| 58 |
+
"learning_rate": 7.139999999999999e-05,
|
| 59 |
+
"loss": 7.600141906738282,
|
| 60 |
+
"step": 120
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"epoch": 0.0047185709470845974,
|
| 64 |
+
"grad_norm": 1.265625,
|
| 65 |
+
"learning_rate": 8.34e-05,
|
| 66 |
+
"loss": 7.368203735351562,
|
| 67 |
+
"step": 140
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"epoch": 0.005392652510953826,
|
| 71 |
+
"grad_norm": 1.265625,
|
| 72 |
+
"learning_rate": 9.539999999999999e-05,
|
| 73 |
+
"loss": 7.133045196533203,
|
| 74 |
+
"step": 160
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.006066734074823054,
|
| 78 |
+
"grad_norm": 1.203125,
|
| 79 |
+
"learning_rate": 0.00010739999999999998,
|
| 80 |
+
"loss": 6.885277557373047,
|
| 81 |
+
"step": 180
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"epoch": 0.006740815638692282,
|
| 85 |
+
"grad_norm": 1.15625,
|
| 86 |
+
"learning_rate": 0.0001194,
|
| 87 |
+
"loss": 6.629558563232422,
|
| 88 |
+
"step": 200
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"epoch": 0.006740815638692282,
|
| 92 |
+
"eval_loss": 6.4997687339782715,
|
| 93 |
+
"eval_runtime": 7.9479,
|
| 94 |
+
"eval_samples_per_second": 1198.683,
|
| 95 |
+
"eval_steps_per_second": 0.881,
|
| 96 |
+
"step": 200
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"epoch": 0.00741489720256151,
|
| 100 |
+
"grad_norm": 1.109375,
|
| 101 |
+
"learning_rate": 0.0001314,
|
| 102 |
+
"loss": 6.391195297241211,
|
| 103 |
+
"step": 220
|
| 104 |
+
},
|
| 105 |
+
{
|
| 106 |
+
"epoch": 0.008088978766430738,
|
| 107 |
+
"grad_norm": 1.21875,
|
| 108 |
+
"learning_rate": 0.0001434,
|
| 109 |
+
"loss": 6.139227676391601,
|
| 110 |
+
"step": 240
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"epoch": 0.008763060330299966,
|
| 114 |
+
"grad_norm": 1.0625,
|
| 115 |
+
"learning_rate": 0.00015539999999999998,
|
| 116 |
+
"loss": 5.917041778564453,
|
| 117 |
+
"step": 260
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"epoch": 0.009437141894169195,
|
| 121 |
+
"grad_norm": 1.1015625,
|
| 122 |
+
"learning_rate": 0.0001674,
|
| 123 |
+
"loss": 5.702054595947265,
|
| 124 |
+
"step": 280
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"epoch": 0.010111223458038422,
|
| 128 |
+
"grad_norm": 0.921875,
|
| 129 |
+
"learning_rate": 0.00017939999999999997,
|
| 130 |
+
"loss": 5.500263214111328,
|
| 131 |
+
"step": 300
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"epoch": 0.010111223458038422,
|
| 135 |
+
"eval_loss": 5.393505573272705,
|
| 136 |
+
"eval_runtime": 8.0449,
|
| 137 |
+
"eval_samples_per_second": 1184.223,
|
| 138 |
+
"eval_steps_per_second": 0.87,
|
| 139 |
+
"step": 300
|
| 140 |
+
},
|
| 141 |
+
{
|
| 142 |
+
"epoch": 0.010785305021907651,
|
| 143 |
+
"grad_norm": 2.5,
|
| 144 |
+
"learning_rate": 0.0001914,
|
| 145 |
+
"loss": 5.319306182861328,
|
| 146 |
+
"step": 320
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.011459386585776879,
|
| 150 |
+
"grad_norm": 1.078125,
|
| 151 |
+
"learning_rate": 0.00020339999999999998,
|
| 152 |
+
"loss": 5.202053070068359,
|
| 153 |
+
"step": 340
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"epoch": 0.012133468149646108,
|
| 157 |
+
"grad_norm": 1.3203125,
|
| 158 |
+
"learning_rate": 0.00021539999999999998,
|
| 159 |
+
"loss": 5.020883941650391,
|
| 160 |
+
"step": 360
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 0.012807549713515335,
|
| 164 |
+
"grad_norm": 1.1171875,
|
| 165 |
+
"learning_rate": 0.00022739999999999997,
|
| 166 |
+
"loss": 4.873059844970703,
|
| 167 |
+
"step": 380
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"epoch": 0.013481631277384564,
|
| 171 |
+
"grad_norm": 1.234375,
|
| 172 |
+
"learning_rate": 0.0002394,
|
| 173 |
+
"loss": 4.741780853271484,
|
| 174 |
+
"step": 400
|
| 175 |
+
},
|
| 176 |
+
{
|
| 177 |
+
"epoch": 0.013481631277384564,
|
| 178 |
+
"eval_loss": 4.676144123077393,
|
| 179 |
+
"eval_runtime": 8.0457,
|
| 180 |
+
"eval_samples_per_second": 1184.113,
|
| 181 |
+
"eval_steps_per_second": 0.87,
|
| 182 |
+
"step": 400
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"epoch": 0.014155712841253791,
|
| 186 |
+
"grad_norm": 1.3046875,
|
| 187 |
+
"learning_rate": 0.0002514,
|
| 188 |
+
"loss": 4.606548690795899,
|
| 189 |
+
"step": 420
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 0.01482979440512302,
|
| 193 |
+
"grad_norm": 1.2578125,
|
| 194 |
+
"learning_rate": 0.00026339999999999995,
|
| 195 |
+
"loss": 4.515376281738281,
|
| 196 |
+
"step": 440
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"epoch": 0.015503875968992248,
|
| 200 |
+
"grad_norm": 1.0625,
|
| 201 |
+
"learning_rate": 0.00027539999999999997,
|
| 202 |
+
"loss": 4.416887664794922,
|
| 203 |
+
"step": 460
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"epoch": 0.016177957532861477,
|
| 207 |
+
"grad_norm": 1.3671875,
|
| 208 |
+
"learning_rate": 0.00028739999999999994,
|
| 209 |
+
"loss": 4.3046520233154295,
|
| 210 |
+
"step": 480
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.016852039096730706,
|
| 214 |
+
"grad_norm": 1.2109375,
|
| 215 |
+
"learning_rate": 0.00029939999999999996,
|
| 216 |
+
"loss": 4.238706970214844,
|
| 217 |
+
"step": 500
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"epoch": 0.016852039096730706,
|
| 221 |
+
"eval_loss": 4.196270942687988,
|
| 222 |
+
"eval_runtime": 7.9301,
|
| 223 |
+
"eval_samples_per_second": 1201.377,
|
| 224 |
+
"eval_steps_per_second": 0.883,
|
| 225 |
+
"step": 500
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"epoch": 0.01752612066059993,
|
| 229 |
+
"grad_norm": 0.97265625,
|
| 230 |
+
"learning_rate": 0.0003,
|
| 231 |
+
"loss": 4.1756237030029295,
|
| 232 |
+
"step": 520
|
| 233 |
+
},
|
| 234 |
+
{
|
| 235 |
+
"epoch": 0.01820020222446916,
|
| 236 |
+
"grad_norm": 0.9140625,
|
| 237 |
+
"learning_rate": 0.0003,
|
| 238 |
+
"loss": 4.0915069580078125,
|
| 239 |
+
"step": 540
|
| 240 |
+
},
|
| 241 |
+
{
|
| 242 |
+
"epoch": 0.01887428378833839,
|
| 243 |
+
"grad_norm": 1.2890625,
|
| 244 |
+
"learning_rate": 0.0003,
|
| 245 |
+
"loss": 4.031630325317383,
|
| 246 |
+
"step": 560
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"epoch": 0.01954836535220762,
|
| 250 |
+
"grad_norm": 1.125,
|
| 251 |
+
"learning_rate": 0.0003,
|
| 252 |
+
"loss": 3.9763336181640625,
|
| 253 |
+
"step": 580
|
| 254 |
+
},
|
| 255 |
+
{
|
| 256 |
+
"epoch": 0.020222446916076844,
|
| 257 |
+
"grad_norm": 1.21875,
|
| 258 |
+
"learning_rate": 0.0003,
|
| 259 |
+
"loss": 3.9347896575927734,
|
| 260 |
+
"step": 600
|
| 261 |
+
},
|
| 262 |
+
{
|
| 263 |
+
"epoch": 0.020222446916076844,
|
| 264 |
+
"eval_loss": 3.9157378673553467,
|
| 265 |
+
"eval_runtime": 8.0606,
|
| 266 |
+
"eval_samples_per_second": 1181.917,
|
| 267 |
+
"eval_steps_per_second": 0.868,
|
| 268 |
+
"step": 600
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.020896528479946073,
|
| 272 |
+
"grad_norm": 1.1640625,
|
| 273 |
+
"learning_rate": 0.0003,
|
| 274 |
+
"loss": 3.878151702880859,
|
| 275 |
+
"step": 620
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.021570610043815303,
|
| 279 |
+
"grad_norm": 1.484375,
|
| 280 |
+
"learning_rate": 0.0003,
|
| 281 |
+
"loss": 3.855076217651367,
|
| 282 |
+
"step": 640
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.022244691607684528,
|
| 286 |
+
"grad_norm": 1.3359375,
|
| 287 |
+
"learning_rate": 0.0003,
|
| 288 |
+
"loss": 3.8232196807861327,
|
| 289 |
+
"step": 660
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.022918773171553757,
|
| 293 |
+
"grad_norm": 1.1640625,
|
| 294 |
+
"learning_rate": 0.0003,
|
| 295 |
+
"loss": 3.782532501220703,
|
| 296 |
+
"step": 680
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.023592854735422986,
|
| 300 |
+
"grad_norm": 1.1328125,
|
| 301 |
+
"learning_rate": 0.0003,
|
| 302 |
+
"loss": 3.7758609771728517,
|
| 303 |
+
"step": 700
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.023592854735422986,
|
| 307 |
+
"eval_loss": 3.7596347332000732,
|
| 308 |
+
"eval_runtime": 7.9743,
|
| 309 |
+
"eval_samples_per_second": 1194.709,
|
| 310 |
+
"eval_steps_per_second": 0.878,
|
| 311 |
+
"step": 700
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"epoch": 0.024266936299292215,
|
| 315 |
+
"grad_norm": 1.25,
|
| 316 |
+
"learning_rate": 0.0003,
|
| 317 |
+
"loss": 3.7433414459228516,
|
| 318 |
+
"step": 720
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 0.02494101786316144,
|
| 322 |
+
"grad_norm": 1.3125,
|
| 323 |
+
"learning_rate": 0.0003,
|
| 324 |
+
"loss": 3.732330322265625,
|
| 325 |
+
"step": 740
|
| 326 |
+
},
|
| 327 |
+
{
|
| 328 |
+
"epoch": 0.02561509942703067,
|
| 329 |
+
"grad_norm": 1.265625,
|
| 330 |
+
"learning_rate": 0.0003,
|
| 331 |
+
"loss": 3.6843395233154297,
|
| 332 |
+
"step": 760
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"epoch": 0.0262891809908999,
|
| 336 |
+
"grad_norm": 1.1640625,
|
| 337 |
+
"learning_rate": 0.0003,
|
| 338 |
+
"loss": 3.6463153839111326,
|
| 339 |
+
"step": 780
|
| 340 |
+
},
|
| 341 |
+
{
|
| 342 |
+
"epoch": 0.026963262554769128,
|
| 343 |
+
"grad_norm": 1.1171875,
|
| 344 |
+
"learning_rate": 0.0003,
|
| 345 |
+
"loss": 3.6544403076171874,
|
| 346 |
+
"step": 800
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.026963262554769128,
|
| 350 |
+
"eval_loss": 3.652294635772705,
|
| 351 |
+
"eval_runtime": 7.9753,
|
| 352 |
+
"eval_samples_per_second": 1194.559,
|
| 353 |
+
"eval_steps_per_second": 0.878,
|
| 354 |
+
"step": 800
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.027637344118638354,
|
| 358 |
+
"grad_norm": 1.1171875,
|
| 359 |
+
"learning_rate": 0.0003,
|
| 360 |
+
"loss": 3.6159641265869142,
|
| 361 |
+
"step": 820
|
| 362 |
+
},
|
| 363 |
+
{
|
| 364 |
+
"epoch": 0.028311425682507583,
|
| 365 |
+
"grad_norm": 1.078125,
|
| 366 |
+
"learning_rate": 0.0003,
|
| 367 |
+
"loss": 3.6099781036376952,
|
| 368 |
+
"step": 840
|
| 369 |
+
},
|
| 370 |
+
{
|
| 371 |
+
"epoch": 0.028985507246376812,
|
| 372 |
+
"grad_norm": 1.2734375,
|
| 373 |
+
"learning_rate": 0.0003,
|
| 374 |
+
"loss": 3.6344444274902346,
|
| 375 |
+
"step": 860
|
| 376 |
+
},
|
| 377 |
+
{
|
| 378 |
+
"epoch": 0.02965958881024604,
|
| 379 |
+
"grad_norm": 1.140625,
|
| 380 |
+
"learning_rate": 0.0003,
|
| 381 |
+
"loss": 3.5855178833007812,
|
| 382 |
+
"step": 880
|
| 383 |
+
},
|
| 384 |
+
{
|
| 385 |
+
"epoch": 0.030333670374115267,
|
| 386 |
+
"grad_norm": 1.109375,
|
| 387 |
+
"learning_rate": 0.0003,
|
| 388 |
+
"loss": 3.5488441467285154,
|
| 389 |
+
"step": 900
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.030333670374115267,
|
| 393 |
+
"eval_loss": 3.5648436546325684,
|
| 394 |
+
"eval_runtime": 8.321,
|
| 395 |
+
"eval_samples_per_second": 1144.931,
|
| 396 |
+
"eval_steps_per_second": 0.841,
|
| 397 |
+
"step": 900
|
| 398 |
+
},
|
| 399 |
+
{
|
| 400 |
+
"epoch": 0.031007751937984496,
|
| 401 |
+
"grad_norm": 1.1015625,
|
| 402 |
+
"learning_rate": 0.0003,
|
| 403 |
+
"loss": 3.5646598815917967,
|
| 404 |
+
"step": 920
|
| 405 |
+
},
|
| 406 |
+
{
|
| 407 |
+
"epoch": 0.031681833501853725,
|
| 408 |
+
"grad_norm": 1.25,
|
| 409 |
+
"learning_rate": 0.0003,
|
| 410 |
+
"loss": 3.5480445861816405,
|
| 411 |
+
"step": 940
|
| 412 |
+
},
|
| 413 |
+
{
|
| 414 |
+
"epoch": 0.032355915065722954,
|
| 415 |
+
"grad_norm": 1.140625,
|
| 416 |
+
"learning_rate": 0.0003,
|
| 417 |
+
"loss": 3.5140933990478516,
|
| 418 |
+
"step": 960
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.03302999662959218,
|
| 422 |
+
"grad_norm": 1.140625,
|
| 423 |
+
"learning_rate": 0.0003,
|
| 424 |
+
"loss": 3.501856231689453,
|
| 425 |
+
"step": 980
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.03370407819346141,
|
| 429 |
+
"grad_norm": 1.359375,
|
| 430 |
+
"learning_rate": 0.0003,
|
| 431 |
+
"loss": 3.500592803955078,
|
| 432 |
+
"step": 1000
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.03370407819346141,
|
| 436 |
+
"eval_loss": 3.4981162548065186,
|
| 437 |
+
"eval_runtime": 8.1013,
|
| 438 |
+
"eval_samples_per_second": 1175.987,
|
| 439 |
+
"eval_steps_per_second": 0.864,
|
| 440 |
+
"step": 1000
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.034378159757330634,
|
| 444 |
+
"grad_norm": 1.1328125,
|
| 445 |
+
"learning_rate": 0.0003,
|
| 446 |
+
"loss": 3.4914066314697267,
|
| 447 |
+
"step": 1020
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.03505224132119986,
|
| 451 |
+
"grad_norm": 1.1171875,
|
| 452 |
+
"learning_rate": 0.0003,
|
| 453 |
+
"loss": 3.474192810058594,
|
| 454 |
+
"step": 1040
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.03572632288506909,
|
| 458 |
+
"grad_norm": 1.28125,
|
| 459 |
+
"learning_rate": 0.0003,
|
| 460 |
+
"loss": 3.4552947998046877,
|
| 461 |
+
"step": 1060
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.03640040444893832,
|
| 465 |
+
"grad_norm": 1.203125,
|
| 466 |
+
"learning_rate": 0.0003,
|
| 467 |
+
"loss": 3.4411731719970704,
|
| 468 |
+
"step": 1080
|
| 469 |
+
},
|
| 470 |
+
{
|
| 471 |
+
"epoch": 0.03707448601280755,
|
| 472 |
+
"grad_norm": 1.2734375,
|
| 473 |
+
"learning_rate": 0.0003,
|
| 474 |
+
"loss": 3.435680389404297,
|
| 475 |
+
"step": 1100
|
| 476 |
+
},
|
| 477 |
+
{
|
| 478 |
+
"epoch": 0.03707448601280755,
|
| 479 |
+
"eval_loss": 3.445117712020874,
|
| 480 |
+
"eval_runtime": 8.1641,
|
| 481 |
+
"eval_samples_per_second": 1166.938,
|
| 482 |
+
"eval_steps_per_second": 0.857,
|
| 483 |
+
"step": 1100
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.03774856757667678,
|
| 487 |
+
"grad_norm": 1.03125,
|
| 488 |
+
"learning_rate": 0.0003,
|
| 489 |
+
"loss": 3.4368762969970703,
|
| 490 |
+
"step": 1120
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.03842264914054601,
|
| 494 |
+
"grad_norm": 1.2421875,
|
| 495 |
+
"learning_rate": 0.0003,
|
| 496 |
+
"loss": 3.410645294189453,
|
| 497 |
+
"step": 1140
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"epoch": 0.03909673070441524,
|
| 501 |
+
"grad_norm": 1.0703125,
|
| 502 |
+
"learning_rate": 0.0003,
|
| 503 |
+
"loss": 3.4069515228271485,
|
| 504 |
+
"step": 1160
|
| 505 |
+
},
|
| 506 |
+
{
|
| 507 |
+
"epoch": 0.03977081226828446,
|
| 508 |
+
"grad_norm": 1.03125,
|
| 509 |
+
"learning_rate": 0.0003,
|
| 510 |
+
"loss": 3.4227649688720705,
|
| 511 |
+
"step": 1180
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"epoch": 0.04044489383215369,
|
| 515 |
+
"grad_norm": 1.203125,
|
| 516 |
+
"learning_rate": 0.0003,
|
| 517 |
+
"loss": 3.3914558410644533,
|
| 518 |
+
"step": 1200
|
| 519 |
+
},
|
| 520 |
+
{
|
| 521 |
+
"epoch": 0.04044489383215369,
|
| 522 |
+
"eval_loss": 3.3945467472076416,
|
| 523 |
+
"eval_runtime": 8.0433,
|
| 524 |
+
"eval_samples_per_second": 1184.469,
|
| 525 |
+
"eval_steps_per_second": 0.87,
|
| 526 |
+
"step": 1200
|
| 527 |
+
},
|
| 528 |
+
{
|
| 529 |
+
"epoch": 0.04111897539602292,
|
| 530 |
+
"grad_norm": 1.0546875,
|
| 531 |
+
"learning_rate": 0.0003,
|
| 532 |
+
"loss": 3.3614086151123046,
|
| 533 |
+
"step": 1220
|
| 534 |
+
},
|
| 535 |
+
{
|
| 536 |
+
"epoch": 0.04179305695989215,
|
| 537 |
+
"grad_norm": 1.09375,
|
| 538 |
+
"learning_rate": 0.0003,
|
| 539 |
+
"loss": 3.3790237426757814,
|
| 540 |
+
"step": 1240
|
| 541 |
+
},
|
| 542 |
+
{
|
| 543 |
+
"epoch": 0.042467138523761376,
|
| 544 |
+
"grad_norm": 1.3515625,
|
| 545 |
+
"learning_rate": 0.0003,
|
| 546 |
+
"loss": 3.38794059753418,
|
| 547 |
+
"step": 1260
|
| 548 |
+
},
|
| 549 |
+
{
|
| 550 |
+
"epoch": 0.043141220087630605,
|
| 551 |
+
"grad_norm": 1.28125,
|
| 552 |
+
"learning_rate": 0.0003,
|
| 553 |
+
"loss": 3.3654109954833986,
|
| 554 |
+
"step": 1280
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.043815301651499834,
|
| 558 |
+
"grad_norm": 1.1796875,
|
| 559 |
+
"learning_rate": 0.0003,
|
| 560 |
+
"loss": 3.3572288513183595,
|
| 561 |
+
"step": 1300
|
| 562 |
+
},
|
| 563 |
+
{
|
| 564 |
+
"epoch": 0.043815301651499834,
|
| 565 |
+
"eval_loss": 3.3501007556915283,
|
| 566 |
+
"eval_runtime": 8.0621,
|
| 567 |
+
"eval_samples_per_second": 1181.702,
|
| 568 |
+
"eval_steps_per_second": 0.868,
|
| 569 |
+
"step": 1300
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 0.044489383215369056,
|
| 573 |
+
"grad_norm": 1.2734375,
|
| 574 |
+
"learning_rate": 0.0003,
|
| 575 |
+
"loss": 3.321734619140625,
|
| 576 |
+
"step": 1320
|
| 577 |
+
},
|
| 578 |
+
{
|
| 579 |
+
"epoch": 0.045163464779238285,
|
| 580 |
+
"grad_norm": 1.2890625,
|
| 581 |
+
"learning_rate": 0.0003,
|
| 582 |
+
"loss": 3.3493595123291016,
|
| 583 |
+
"step": 1340
|
| 584 |
+
},
|
| 585 |
+
{
|
| 586 |
+
"epoch": 0.045837546343107514,
|
| 587 |
+
"grad_norm": 1.1875,
|
| 588 |
+
"learning_rate": 0.0003,
|
| 589 |
+
"loss": 3.3272212982177733,
|
| 590 |
+
"step": 1360
|
| 591 |
+
},
|
| 592 |
+
{
|
| 593 |
+
"epoch": 0.046511627906976744,
|
| 594 |
+
"grad_norm": 1.1328125,
|
| 595 |
+
"learning_rate": 0.0003,
|
| 596 |
+
"loss": 3.3202045440673826,
|
| 597 |
+
"step": 1380
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 0.04718570947084597,
|
| 601 |
+
"grad_norm": 1.1171875,
|
| 602 |
+
"learning_rate": 0.0003,
|
| 603 |
+
"loss": 3.2998809814453125,
|
| 604 |
+
"step": 1400
|
| 605 |
+
},
|
| 606 |
+
{
|
| 607 |
+
"epoch": 0.04718570947084597,
|
| 608 |
+
"eval_loss": 3.311476707458496,
|
| 609 |
+
"eval_runtime": 8.1361,
|
| 610 |
+
"eval_samples_per_second": 1170.957,
|
| 611 |
+
"eval_steps_per_second": 0.86,
|
| 612 |
+
"step": 1400
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.0478597910347152,
|
| 616 |
+
"grad_norm": 1.140625,
|
| 617 |
+
"learning_rate": 0.0003,
|
| 618 |
+
"loss": 3.3057735443115233,
|
| 619 |
+
"step": 1420
|
| 620 |
+
},
|
| 621 |
+
{
|
| 622 |
+
"epoch": 0.04853387259858443,
|
| 623 |
+
"grad_norm": 1.109375,
|
| 624 |
+
"learning_rate": 0.0003,
|
| 625 |
+
"loss": 3.2907379150390623,
|
| 626 |
+
"step": 1440
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.04920795416245366,
|
| 630 |
+
"grad_norm": 1.078125,
|
| 631 |
+
"learning_rate": 0.0003,
|
| 632 |
+
"loss": 3.264795684814453,
|
| 633 |
+
"step": 1460
|
| 634 |
+
},
|
| 635 |
+
{
|
| 636 |
+
"epoch": 0.04988203572632288,
|
| 637 |
+
"grad_norm": 1.3671875,
|
| 638 |
+
"learning_rate": 0.0003,
|
| 639 |
+
"loss": 3.2918243408203125,
|
| 640 |
+
"step": 1480
|
| 641 |
+
},
|
| 642 |
+
{
|
| 643 |
+
"epoch": 0.05055611729019211,
|
| 644 |
+
"grad_norm": 1.2421875,
|
| 645 |
+
"learning_rate": 0.0003,
|
| 646 |
+
"loss": 3.2844181060791016,
|
| 647 |
+
"step": 1500
|
| 648 |
+
},
|
| 649 |
+
{
|
| 650 |
+
"epoch": 0.05055611729019211,
|
| 651 |
+
"eval_loss": 3.277984142303467,
|
| 652 |
+
"eval_runtime": 8.0313,
|
| 653 |
+
"eval_samples_per_second": 1186.24,
|
| 654 |
+
"eval_steps_per_second": 0.872,
|
| 655 |
+
"step": 1500
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 0.05123019885406134,
|
| 659 |
+
"grad_norm": 1.265625,
|
| 660 |
+
"learning_rate": 0.0003,
|
| 661 |
+
"loss": 3.2729434967041016,
|
| 662 |
+
"step": 1520
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 0.05190428041793057,
|
| 666 |
+
"grad_norm": 1.1015625,
|
| 667 |
+
"learning_rate": 0.0003,
|
| 668 |
+
"loss": 3.279766845703125,
|
| 669 |
+
"step": 1540
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 0.0525783619817998,
|
| 673 |
+
"grad_norm": 1.25,
|
| 674 |
+
"learning_rate": 0.0003,
|
| 675 |
+
"loss": 3.281978988647461,
|
| 676 |
+
"step": 1560
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 0.05325244354566903,
|
| 680 |
+
"grad_norm": 1.125,
|
| 681 |
+
"learning_rate": 0.0003,
|
| 682 |
+
"loss": 3.2608470916748047,
|
| 683 |
+
"step": 1580
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 0.053926525109538256,
|
| 687 |
+
"grad_norm": 1.2265625,
|
| 688 |
+
"learning_rate": 0.0003,
|
| 689 |
+
"loss": 3.2503360748291015,
|
| 690 |
+
"step": 1600
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.053926525109538256,
|
| 694 |
+
"eval_loss": 3.2511110305786133,
|
| 695 |
+
"eval_runtime": 8.097,
|
| 696 |
+
"eval_samples_per_second": 1176.602,
|
| 697 |
+
"eval_steps_per_second": 0.865,
|
| 698 |
+
"step": 1600
|
| 699 |
+
},
|
| 700 |
+
{
|
| 701 |
+
"epoch": 0.054600606673407485,
|
| 702 |
+
"grad_norm": 1.234375,
|
| 703 |
+
"learning_rate": 0.0003,
|
| 704 |
+
"loss": 3.2713451385498047,
|
| 705 |
+
"step": 1620
|
| 706 |
+
},
|
| 707 |
+
{
|
| 708 |
+
"epoch": 0.05527468823727671,
|
| 709 |
+
"grad_norm": 1.015625,
|
| 710 |
+
"learning_rate": 0.0003,
|
| 711 |
+
"loss": 3.2468338012695312,
|
| 712 |
+
"step": 1640
|
| 713 |
+
},
|
| 714 |
+
{
|
| 715 |
+
"epoch": 0.05594876980114594,
|
| 716 |
+
"grad_norm": 1.15625,
|
| 717 |
+
"learning_rate": 0.0003,
|
| 718 |
+
"loss": 3.255104827880859,
|
| 719 |
+
"step": 1660
|
| 720 |
+
},
|
| 721 |
+
{
|
| 722 |
+
"epoch": 0.056622851365015166,
|
| 723 |
+
"grad_norm": 1.1875,
|
| 724 |
+
"learning_rate": 0.0003,
|
| 725 |
+
"loss": 3.2280048370361327,
|
| 726 |
+
"step": 1680
|
| 727 |
+
},
|
| 728 |
+
{
|
| 729 |
+
"epoch": 0.057296932928884395,
|
| 730 |
+
"grad_norm": 1.265625,
|
| 731 |
+
"learning_rate": 0.0003,
|
| 732 |
+
"loss": 3.216978073120117,
|
| 733 |
+
"step": 1700
|
| 734 |
+
},
|
| 735 |
+
{
|
| 736 |
+
"epoch": 0.057296932928884395,
|
| 737 |
+
"eval_loss": 3.2238714694976807,
|
| 738 |
+
"eval_runtime": 8.059,
|
| 739 |
+
"eval_samples_per_second": 1182.16,
|
| 740 |
+
"eval_steps_per_second": 0.869,
|
| 741 |
+
"step": 1700
|
| 742 |
+
},
|
| 743 |
+
{
|
| 744 |
+
"epoch": 0.057971014492753624,
|
| 745 |
+
"grad_norm": 1.125,
|
| 746 |
+
"learning_rate": 0.0003,
|
| 747 |
+
"loss": 3.213521194458008,
|
| 748 |
+
"step": 1720
|
| 749 |
+
},
|
| 750 |
+
{
|
| 751 |
+
"epoch": 0.05864509605662285,
|
| 752 |
+
"grad_norm": 1.234375,
|
| 753 |
+
"learning_rate": 0.0003,
|
| 754 |
+
"loss": 3.2087985992431642,
|
| 755 |
+
"step": 1740
|
| 756 |
+
},
|
| 757 |
+
{
|
| 758 |
+
"epoch": 0.05931917762049208,
|
| 759 |
+
"grad_norm": 1.2109375,
|
| 760 |
+
"learning_rate": 0.0003,
|
| 761 |
+
"loss": 3.198755645751953,
|
| 762 |
+
"step": 1760
|
| 763 |
+
},
|
| 764 |
+
{
|
| 765 |
+
"epoch": 0.05999325918436131,
|
| 766 |
+
"grad_norm": 1.4453125,
|
| 767 |
+
"learning_rate": 0.0003,
|
| 768 |
+
"loss": 3.2067222595214844,
|
| 769 |
+
"step": 1780
|
| 770 |
+
},
|
| 771 |
+
{
|
| 772 |
+
"epoch": 0.06066734074823053,
|
| 773 |
+
"grad_norm": 1.140625,
|
| 774 |
+
"learning_rate": 0.0003,
|
| 775 |
+
"loss": 3.195578956604004,
|
| 776 |
+
"step": 1800
|
| 777 |
+
},
|
| 778 |
+
{
|
| 779 |
+
"epoch": 0.06066734074823053,
|
| 780 |
+
"eval_loss": 3.1985833644866943,
|
| 781 |
+
"eval_runtime": 8.2334,
|
| 782 |
+
"eval_samples_per_second": 1157.109,
|
| 783 |
+
"eval_steps_per_second": 0.85,
|
| 784 |
+
"step": 1800
|
| 785 |
+
},
|
| 786 |
+
{
|
| 787 |
+
"epoch": 0.06134142231209976,
|
| 788 |
+
"grad_norm": 1.1640625,
|
| 789 |
+
"learning_rate": 0.0003,
|
| 790 |
+
"loss": 3.186159133911133,
|
| 791 |
+
"step": 1820
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"epoch": 0.06201550387596899,
|
| 795 |
+
"grad_norm": 1.21875,
|
| 796 |
+
"learning_rate": 0.0003,
|
| 797 |
+
"loss": 3.1850618362426757,
|
| 798 |
+
"step": 1840
|
| 799 |
+
},
|
| 800 |
+
{
|
| 801 |
+
"epoch": 0.06268958543983821,
|
| 802 |
+
"grad_norm": 1.3515625,
|
| 803 |
+
"learning_rate": 0.0003,
|
| 804 |
+
"loss": 3.1967243194580077,
|
| 805 |
+
"step": 1860
|
| 806 |
+
},
|
| 807 |
+
{
|
| 808 |
+
"epoch": 0.06336366700370745,
|
| 809 |
+
"grad_norm": 1.3125,
|
| 810 |
+
"learning_rate": 0.0003,
|
| 811 |
+
"loss": 3.1668914794921874,
|
| 812 |
+
"step": 1880
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 0.06403774856757667,
|
| 816 |
+
"grad_norm": 1.1484375,
|
| 817 |
+
"learning_rate": 0.0003,
|
| 818 |
+
"loss": 3.1733448028564455,
|
| 819 |
+
"step": 1900
|
| 820 |
+
},
|
| 821 |
+
{
|
| 822 |
+
"epoch": 0.06403774856757667,
|
| 823 |
+
"eval_loss": 3.1766083240509033,
|
| 824 |
+
"eval_runtime": 7.9724,
|
| 825 |
+
"eval_samples_per_second": 1194.991,
|
| 826 |
+
"eval_steps_per_second": 0.878,
|
| 827 |
+
"step": 1900
|
| 828 |
+
},
|
| 829 |
+
{
|
| 830 |
+
"epoch": 0.06471183013144591,
|
| 831 |
+
"grad_norm": 1.1953125,
|
| 832 |
+
"learning_rate": 0.0003,
|
| 833 |
+
"loss": 3.1606868743896483,
|
| 834 |
+
"step": 1920
|
| 835 |
+
},
|
| 836 |
+
{
|
| 837 |
+
"epoch": 0.06538591169531513,
|
| 838 |
+
"grad_norm": 1.109375,
|
| 839 |
+
"learning_rate": 0.0003,
|
| 840 |
+
"loss": 3.148599052429199,
|
| 841 |
+
"step": 1940
|
| 842 |
+
},
|
| 843 |
+
{
|
| 844 |
+
"epoch": 0.06605999325918437,
|
| 845 |
+
"grad_norm": 1.2265625,
|
| 846 |
+
"learning_rate": 0.0003,
|
| 847 |
+
"loss": 3.1723419189453126,
|
| 848 |
+
"step": 1960
|
| 849 |
+
},
|
| 850 |
+
{
|
| 851 |
+
"epoch": 0.06673407482305359,
|
| 852 |
+
"grad_norm": 1.234375,
|
| 853 |
+
"learning_rate": 0.0003,
|
| 854 |
+
"loss": 3.184863471984863,
|
| 855 |
+
"step": 1980
|
| 856 |
+
},
|
| 857 |
+
{
|
| 858 |
+
"epoch": 0.06740815638692282,
|
| 859 |
+
"grad_norm": 1.2734375,
|
| 860 |
+
"learning_rate": 0.0003,
|
| 861 |
+
"loss": 3.158551788330078,
|
| 862 |
+
"step": 2000
|
| 863 |
+
},
|
| 864 |
+
{
|
| 865 |
+
"epoch": 0.06740815638692282,
|
| 866 |
+
"eval_loss": 3.154155731201172,
|
| 867 |
+
"eval_runtime": 7.9931,
|
| 868 |
+
"eval_samples_per_second": 1191.905,
|
| 869 |
+
"eval_steps_per_second": 0.876,
|
| 870 |
+
"step": 2000
|
| 871 |
+
}
|
| 872 |
+
],
|
| 873 |
+
"logging_steps": 20,
|
| 874 |
+
"max_steps": 2500,
|
| 875 |
+
"num_input_tokens_seen": 0,
|
| 876 |
+
"num_train_epochs": 1,
|
| 877 |
+
"save_steps": 100,
|
| 878 |
+
"stateful_callbacks": {
|
| 879 |
+
"TrainerControl": {
|
| 880 |
+
"args": {
|
| 881 |
+
"should_epoch_stop": false,
|
| 882 |
+
"should_evaluate": false,
|
| 883 |
+
"should_log": false,
|
| 884 |
+
"should_save": true,
|
| 885 |
+
"should_training_stop": false
|
| 886 |
+
},
|
| 887 |
+
"attributes": {}
|
| 888 |
+
}
|
| 889 |
+
},
|
| 890 |
+
"total_flos": 290388443136000.0,
|
| 891 |
+
"train_batch_size": 32,
|
| 892 |
+
"trial_name": null,
|
| 893 |
+
"trial_params": null
|
| 894 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2000/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a761b578cf2b8c248f28bd1b61a5ec7d0b5d3801c8bd793a6db706df1d3078f6
|
| 3 |
+
size 4920
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/config.json
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "squared_relu",
|
| 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": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 9,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"powlu_m": 3.0,
|
| 25 |
+
"pretraining_tp": 1,
|
| 26 |
+
"rms_norm_eps": 1e-06,
|
| 27 |
+
"rope_parameters": {
|
| 28 |
+
"rope_theta": 10000.0,
|
| 29 |
+
"rope_type": "default"
|
| 30 |
+
},
|
| 31 |
+
"tie_word_embeddings": true,
|
| 32 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 33 |
+
"transformers_version": "5.16.0.dev0",
|
| 34 |
+
"use_cache": false,
|
| 35 |
+
"vocab_size": 4096,
|
| 36 |
+
"waleed_beta": 10.0
|
| 37 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a1aa62fa362d37769fd67fbb46cdfd4c91936ced9a756612b8a489e0bdb3dd4a
|
| 3 |
+
size 4010544
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:56031c0e76c1bce8e7a0a9a3559707b888a6b9c293101948fad70e9c6b531226
|
| 3 |
+
size 8068282
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f7286dee9a5da5b96232ccd668e3561a460616ec53ea433806da321343af8636
|
| 3 |
+
size 14244
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0a05e9cdb3460a35862aac87a0001b7bc3f41bad95490df18a7b51b4f84deeeb
|
| 3 |
+
size 1064
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/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/mlp-squared_relu-9L_run/checkpoint-2100/trainer_state.json
ADDED
|
@@ -0,0 +1,937 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.07077856420626896,
|
| 6 |
+
"eval_steps": 100,
|
| 7 |
+
"global_step": 2100,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0006740815638692282,
|
| 14 |
+
"grad_norm": 2.71875,
|
| 15 |
+
"learning_rate": 1.14e-05,
|
| 16 |
+
"loss": 8.314944458007812,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.0013481631277384564,
|
| 21 |
+
"grad_norm": 1.8515625,
|
| 22 |
+
"learning_rate": 2.34e-05,
|
| 23 |
+
"loss": 8.251383972167968,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.0020222446916076846,
|
| 28 |
+
"grad_norm": 1.3203125,
|
| 29 |
+
"learning_rate": 3.539999999999999e-05,
|
| 30 |
+
"loss": 8.104505920410157,
|
| 31 |
+
"step": 60
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"epoch": 0.002696326255476913,
|
| 35 |
+
"grad_norm": 1.296875,
|
| 36 |
+
"learning_rate": 4.7399999999999993e-05,
|
| 37 |
+
"loss": 7.968361663818359,
|
| 38 |
+
"step": 80
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.003370407819346141,
|
| 42 |
+
"grad_norm": 1.2578125,
|
| 43 |
+
"learning_rate": 5.94e-05,
|
| 44 |
+
"loss": 7.804827880859375,
|
| 45 |
+
"step": 100
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.003370407819346141,
|
| 49 |
+
"eval_loss": 7.703802585601807,
|
| 50 |
+
"eval_runtime": 8.0325,
|
| 51 |
+
"eval_samples_per_second": 1186.059,
|
| 52 |
+
"eval_steps_per_second": 0.871,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.004044489383215369,
|
| 57 |
+
"grad_norm": 1.2734375,
|
| 58 |
+
"learning_rate": 7.139999999999999e-05,
|
| 59 |
+
"loss": 7.600141906738282,
|
| 60 |
+
"step": 120
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"epoch": 0.0047185709470845974,
|
| 64 |
+
"grad_norm": 1.265625,
|
| 65 |
+
"learning_rate": 8.34e-05,
|
| 66 |
+
"loss": 7.368203735351562,
|
| 67 |
+
"step": 140
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"epoch": 0.005392652510953826,
|
| 71 |
+
"grad_norm": 1.265625,
|
| 72 |
+
"learning_rate": 9.539999999999999e-05,
|
| 73 |
+
"loss": 7.133045196533203,
|
| 74 |
+
"step": 160
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.006066734074823054,
|
| 78 |
+
"grad_norm": 1.203125,
|
| 79 |
+
"learning_rate": 0.00010739999999999998,
|
| 80 |
+
"loss": 6.885277557373047,
|
| 81 |
+
"step": 180
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"epoch": 0.006740815638692282,
|
| 85 |
+
"grad_norm": 1.15625,
|
| 86 |
+
"learning_rate": 0.0001194,
|
| 87 |
+
"loss": 6.629558563232422,
|
| 88 |
+
"step": 200
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"epoch": 0.006740815638692282,
|
| 92 |
+
"eval_loss": 6.4997687339782715,
|
| 93 |
+
"eval_runtime": 7.9479,
|
| 94 |
+
"eval_samples_per_second": 1198.683,
|
| 95 |
+
"eval_steps_per_second": 0.881,
|
| 96 |
+
"step": 200
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"epoch": 0.00741489720256151,
|
| 100 |
+
"grad_norm": 1.109375,
|
| 101 |
+
"learning_rate": 0.0001314,
|
| 102 |
+
"loss": 6.391195297241211,
|
| 103 |
+
"step": 220
|
| 104 |
+
},
|
| 105 |
+
{
|
| 106 |
+
"epoch": 0.008088978766430738,
|
| 107 |
+
"grad_norm": 1.21875,
|
| 108 |
+
"learning_rate": 0.0001434,
|
| 109 |
+
"loss": 6.139227676391601,
|
| 110 |
+
"step": 240
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"epoch": 0.008763060330299966,
|
| 114 |
+
"grad_norm": 1.0625,
|
| 115 |
+
"learning_rate": 0.00015539999999999998,
|
| 116 |
+
"loss": 5.917041778564453,
|
| 117 |
+
"step": 260
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"epoch": 0.009437141894169195,
|
| 121 |
+
"grad_norm": 1.1015625,
|
| 122 |
+
"learning_rate": 0.0001674,
|
| 123 |
+
"loss": 5.702054595947265,
|
| 124 |
+
"step": 280
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"epoch": 0.010111223458038422,
|
| 128 |
+
"grad_norm": 0.921875,
|
| 129 |
+
"learning_rate": 0.00017939999999999997,
|
| 130 |
+
"loss": 5.500263214111328,
|
| 131 |
+
"step": 300
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"epoch": 0.010111223458038422,
|
| 135 |
+
"eval_loss": 5.393505573272705,
|
| 136 |
+
"eval_runtime": 8.0449,
|
| 137 |
+
"eval_samples_per_second": 1184.223,
|
| 138 |
+
"eval_steps_per_second": 0.87,
|
| 139 |
+
"step": 300
|
| 140 |
+
},
|
| 141 |
+
{
|
| 142 |
+
"epoch": 0.010785305021907651,
|
| 143 |
+
"grad_norm": 2.5,
|
| 144 |
+
"learning_rate": 0.0001914,
|
| 145 |
+
"loss": 5.319306182861328,
|
| 146 |
+
"step": 320
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.011459386585776879,
|
| 150 |
+
"grad_norm": 1.078125,
|
| 151 |
+
"learning_rate": 0.00020339999999999998,
|
| 152 |
+
"loss": 5.202053070068359,
|
| 153 |
+
"step": 340
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"epoch": 0.012133468149646108,
|
| 157 |
+
"grad_norm": 1.3203125,
|
| 158 |
+
"learning_rate": 0.00021539999999999998,
|
| 159 |
+
"loss": 5.020883941650391,
|
| 160 |
+
"step": 360
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 0.012807549713515335,
|
| 164 |
+
"grad_norm": 1.1171875,
|
| 165 |
+
"learning_rate": 0.00022739999999999997,
|
| 166 |
+
"loss": 4.873059844970703,
|
| 167 |
+
"step": 380
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"epoch": 0.013481631277384564,
|
| 171 |
+
"grad_norm": 1.234375,
|
| 172 |
+
"learning_rate": 0.0002394,
|
| 173 |
+
"loss": 4.741780853271484,
|
| 174 |
+
"step": 400
|
| 175 |
+
},
|
| 176 |
+
{
|
| 177 |
+
"epoch": 0.013481631277384564,
|
| 178 |
+
"eval_loss": 4.676144123077393,
|
| 179 |
+
"eval_runtime": 8.0457,
|
| 180 |
+
"eval_samples_per_second": 1184.113,
|
| 181 |
+
"eval_steps_per_second": 0.87,
|
| 182 |
+
"step": 400
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"epoch": 0.014155712841253791,
|
| 186 |
+
"grad_norm": 1.3046875,
|
| 187 |
+
"learning_rate": 0.0002514,
|
| 188 |
+
"loss": 4.606548690795899,
|
| 189 |
+
"step": 420
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 0.01482979440512302,
|
| 193 |
+
"grad_norm": 1.2578125,
|
| 194 |
+
"learning_rate": 0.00026339999999999995,
|
| 195 |
+
"loss": 4.515376281738281,
|
| 196 |
+
"step": 440
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"epoch": 0.015503875968992248,
|
| 200 |
+
"grad_norm": 1.0625,
|
| 201 |
+
"learning_rate": 0.00027539999999999997,
|
| 202 |
+
"loss": 4.416887664794922,
|
| 203 |
+
"step": 460
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"epoch": 0.016177957532861477,
|
| 207 |
+
"grad_norm": 1.3671875,
|
| 208 |
+
"learning_rate": 0.00028739999999999994,
|
| 209 |
+
"loss": 4.3046520233154295,
|
| 210 |
+
"step": 480
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.016852039096730706,
|
| 214 |
+
"grad_norm": 1.2109375,
|
| 215 |
+
"learning_rate": 0.00029939999999999996,
|
| 216 |
+
"loss": 4.238706970214844,
|
| 217 |
+
"step": 500
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"epoch": 0.016852039096730706,
|
| 221 |
+
"eval_loss": 4.196270942687988,
|
| 222 |
+
"eval_runtime": 7.9301,
|
| 223 |
+
"eval_samples_per_second": 1201.377,
|
| 224 |
+
"eval_steps_per_second": 0.883,
|
| 225 |
+
"step": 500
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"epoch": 0.01752612066059993,
|
| 229 |
+
"grad_norm": 0.97265625,
|
| 230 |
+
"learning_rate": 0.0003,
|
| 231 |
+
"loss": 4.1756237030029295,
|
| 232 |
+
"step": 520
|
| 233 |
+
},
|
| 234 |
+
{
|
| 235 |
+
"epoch": 0.01820020222446916,
|
| 236 |
+
"grad_norm": 0.9140625,
|
| 237 |
+
"learning_rate": 0.0003,
|
| 238 |
+
"loss": 4.0915069580078125,
|
| 239 |
+
"step": 540
|
| 240 |
+
},
|
| 241 |
+
{
|
| 242 |
+
"epoch": 0.01887428378833839,
|
| 243 |
+
"grad_norm": 1.2890625,
|
| 244 |
+
"learning_rate": 0.0003,
|
| 245 |
+
"loss": 4.031630325317383,
|
| 246 |
+
"step": 560
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"epoch": 0.01954836535220762,
|
| 250 |
+
"grad_norm": 1.125,
|
| 251 |
+
"learning_rate": 0.0003,
|
| 252 |
+
"loss": 3.9763336181640625,
|
| 253 |
+
"step": 580
|
| 254 |
+
},
|
| 255 |
+
{
|
| 256 |
+
"epoch": 0.020222446916076844,
|
| 257 |
+
"grad_norm": 1.21875,
|
| 258 |
+
"learning_rate": 0.0003,
|
| 259 |
+
"loss": 3.9347896575927734,
|
| 260 |
+
"step": 600
|
| 261 |
+
},
|
| 262 |
+
{
|
| 263 |
+
"epoch": 0.020222446916076844,
|
| 264 |
+
"eval_loss": 3.9157378673553467,
|
| 265 |
+
"eval_runtime": 8.0606,
|
| 266 |
+
"eval_samples_per_second": 1181.917,
|
| 267 |
+
"eval_steps_per_second": 0.868,
|
| 268 |
+
"step": 600
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.020896528479946073,
|
| 272 |
+
"grad_norm": 1.1640625,
|
| 273 |
+
"learning_rate": 0.0003,
|
| 274 |
+
"loss": 3.878151702880859,
|
| 275 |
+
"step": 620
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.021570610043815303,
|
| 279 |
+
"grad_norm": 1.484375,
|
| 280 |
+
"learning_rate": 0.0003,
|
| 281 |
+
"loss": 3.855076217651367,
|
| 282 |
+
"step": 640
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.022244691607684528,
|
| 286 |
+
"grad_norm": 1.3359375,
|
| 287 |
+
"learning_rate": 0.0003,
|
| 288 |
+
"loss": 3.8232196807861327,
|
| 289 |
+
"step": 660
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.022918773171553757,
|
| 293 |
+
"grad_norm": 1.1640625,
|
| 294 |
+
"learning_rate": 0.0003,
|
| 295 |
+
"loss": 3.782532501220703,
|
| 296 |
+
"step": 680
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.023592854735422986,
|
| 300 |
+
"grad_norm": 1.1328125,
|
| 301 |
+
"learning_rate": 0.0003,
|
| 302 |
+
"loss": 3.7758609771728517,
|
| 303 |
+
"step": 700
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.023592854735422986,
|
| 307 |
+
"eval_loss": 3.7596347332000732,
|
| 308 |
+
"eval_runtime": 7.9743,
|
| 309 |
+
"eval_samples_per_second": 1194.709,
|
| 310 |
+
"eval_steps_per_second": 0.878,
|
| 311 |
+
"step": 700
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"epoch": 0.024266936299292215,
|
| 315 |
+
"grad_norm": 1.25,
|
| 316 |
+
"learning_rate": 0.0003,
|
| 317 |
+
"loss": 3.7433414459228516,
|
| 318 |
+
"step": 720
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 0.02494101786316144,
|
| 322 |
+
"grad_norm": 1.3125,
|
| 323 |
+
"learning_rate": 0.0003,
|
| 324 |
+
"loss": 3.732330322265625,
|
| 325 |
+
"step": 740
|
| 326 |
+
},
|
| 327 |
+
{
|
| 328 |
+
"epoch": 0.02561509942703067,
|
| 329 |
+
"grad_norm": 1.265625,
|
| 330 |
+
"learning_rate": 0.0003,
|
| 331 |
+
"loss": 3.6843395233154297,
|
| 332 |
+
"step": 760
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"epoch": 0.0262891809908999,
|
| 336 |
+
"grad_norm": 1.1640625,
|
| 337 |
+
"learning_rate": 0.0003,
|
| 338 |
+
"loss": 3.6463153839111326,
|
| 339 |
+
"step": 780
|
| 340 |
+
},
|
| 341 |
+
{
|
| 342 |
+
"epoch": 0.026963262554769128,
|
| 343 |
+
"grad_norm": 1.1171875,
|
| 344 |
+
"learning_rate": 0.0003,
|
| 345 |
+
"loss": 3.6544403076171874,
|
| 346 |
+
"step": 800
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.026963262554769128,
|
| 350 |
+
"eval_loss": 3.652294635772705,
|
| 351 |
+
"eval_runtime": 7.9753,
|
| 352 |
+
"eval_samples_per_second": 1194.559,
|
| 353 |
+
"eval_steps_per_second": 0.878,
|
| 354 |
+
"step": 800
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.027637344118638354,
|
| 358 |
+
"grad_norm": 1.1171875,
|
| 359 |
+
"learning_rate": 0.0003,
|
| 360 |
+
"loss": 3.6159641265869142,
|
| 361 |
+
"step": 820
|
| 362 |
+
},
|
| 363 |
+
{
|
| 364 |
+
"epoch": 0.028311425682507583,
|
| 365 |
+
"grad_norm": 1.078125,
|
| 366 |
+
"learning_rate": 0.0003,
|
| 367 |
+
"loss": 3.6099781036376952,
|
| 368 |
+
"step": 840
|
| 369 |
+
},
|
| 370 |
+
{
|
| 371 |
+
"epoch": 0.028985507246376812,
|
| 372 |
+
"grad_norm": 1.2734375,
|
| 373 |
+
"learning_rate": 0.0003,
|
| 374 |
+
"loss": 3.6344444274902346,
|
| 375 |
+
"step": 860
|
| 376 |
+
},
|
| 377 |
+
{
|
| 378 |
+
"epoch": 0.02965958881024604,
|
| 379 |
+
"grad_norm": 1.140625,
|
| 380 |
+
"learning_rate": 0.0003,
|
| 381 |
+
"loss": 3.5855178833007812,
|
| 382 |
+
"step": 880
|
| 383 |
+
},
|
| 384 |
+
{
|
| 385 |
+
"epoch": 0.030333670374115267,
|
| 386 |
+
"grad_norm": 1.109375,
|
| 387 |
+
"learning_rate": 0.0003,
|
| 388 |
+
"loss": 3.5488441467285154,
|
| 389 |
+
"step": 900
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.030333670374115267,
|
| 393 |
+
"eval_loss": 3.5648436546325684,
|
| 394 |
+
"eval_runtime": 8.321,
|
| 395 |
+
"eval_samples_per_second": 1144.931,
|
| 396 |
+
"eval_steps_per_second": 0.841,
|
| 397 |
+
"step": 900
|
| 398 |
+
},
|
| 399 |
+
{
|
| 400 |
+
"epoch": 0.031007751937984496,
|
| 401 |
+
"grad_norm": 1.1015625,
|
| 402 |
+
"learning_rate": 0.0003,
|
| 403 |
+
"loss": 3.5646598815917967,
|
| 404 |
+
"step": 920
|
| 405 |
+
},
|
| 406 |
+
{
|
| 407 |
+
"epoch": 0.031681833501853725,
|
| 408 |
+
"grad_norm": 1.25,
|
| 409 |
+
"learning_rate": 0.0003,
|
| 410 |
+
"loss": 3.5480445861816405,
|
| 411 |
+
"step": 940
|
| 412 |
+
},
|
| 413 |
+
{
|
| 414 |
+
"epoch": 0.032355915065722954,
|
| 415 |
+
"grad_norm": 1.140625,
|
| 416 |
+
"learning_rate": 0.0003,
|
| 417 |
+
"loss": 3.5140933990478516,
|
| 418 |
+
"step": 960
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.03302999662959218,
|
| 422 |
+
"grad_norm": 1.140625,
|
| 423 |
+
"learning_rate": 0.0003,
|
| 424 |
+
"loss": 3.501856231689453,
|
| 425 |
+
"step": 980
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.03370407819346141,
|
| 429 |
+
"grad_norm": 1.359375,
|
| 430 |
+
"learning_rate": 0.0003,
|
| 431 |
+
"loss": 3.500592803955078,
|
| 432 |
+
"step": 1000
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.03370407819346141,
|
| 436 |
+
"eval_loss": 3.4981162548065186,
|
| 437 |
+
"eval_runtime": 8.1013,
|
| 438 |
+
"eval_samples_per_second": 1175.987,
|
| 439 |
+
"eval_steps_per_second": 0.864,
|
| 440 |
+
"step": 1000
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.034378159757330634,
|
| 444 |
+
"grad_norm": 1.1328125,
|
| 445 |
+
"learning_rate": 0.0003,
|
| 446 |
+
"loss": 3.4914066314697267,
|
| 447 |
+
"step": 1020
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.03505224132119986,
|
| 451 |
+
"grad_norm": 1.1171875,
|
| 452 |
+
"learning_rate": 0.0003,
|
| 453 |
+
"loss": 3.474192810058594,
|
| 454 |
+
"step": 1040
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.03572632288506909,
|
| 458 |
+
"grad_norm": 1.28125,
|
| 459 |
+
"learning_rate": 0.0003,
|
| 460 |
+
"loss": 3.4552947998046877,
|
| 461 |
+
"step": 1060
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.03640040444893832,
|
| 465 |
+
"grad_norm": 1.203125,
|
| 466 |
+
"learning_rate": 0.0003,
|
| 467 |
+
"loss": 3.4411731719970704,
|
| 468 |
+
"step": 1080
|
| 469 |
+
},
|
| 470 |
+
{
|
| 471 |
+
"epoch": 0.03707448601280755,
|
| 472 |
+
"grad_norm": 1.2734375,
|
| 473 |
+
"learning_rate": 0.0003,
|
| 474 |
+
"loss": 3.435680389404297,
|
| 475 |
+
"step": 1100
|
| 476 |
+
},
|
| 477 |
+
{
|
| 478 |
+
"epoch": 0.03707448601280755,
|
| 479 |
+
"eval_loss": 3.445117712020874,
|
| 480 |
+
"eval_runtime": 8.1641,
|
| 481 |
+
"eval_samples_per_second": 1166.938,
|
| 482 |
+
"eval_steps_per_second": 0.857,
|
| 483 |
+
"step": 1100
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.03774856757667678,
|
| 487 |
+
"grad_norm": 1.03125,
|
| 488 |
+
"learning_rate": 0.0003,
|
| 489 |
+
"loss": 3.4368762969970703,
|
| 490 |
+
"step": 1120
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.03842264914054601,
|
| 494 |
+
"grad_norm": 1.2421875,
|
| 495 |
+
"learning_rate": 0.0003,
|
| 496 |
+
"loss": 3.410645294189453,
|
| 497 |
+
"step": 1140
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"epoch": 0.03909673070441524,
|
| 501 |
+
"grad_norm": 1.0703125,
|
| 502 |
+
"learning_rate": 0.0003,
|
| 503 |
+
"loss": 3.4069515228271485,
|
| 504 |
+
"step": 1160
|
| 505 |
+
},
|
| 506 |
+
{
|
| 507 |
+
"epoch": 0.03977081226828446,
|
| 508 |
+
"grad_norm": 1.03125,
|
| 509 |
+
"learning_rate": 0.0003,
|
| 510 |
+
"loss": 3.4227649688720705,
|
| 511 |
+
"step": 1180
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"epoch": 0.04044489383215369,
|
| 515 |
+
"grad_norm": 1.203125,
|
| 516 |
+
"learning_rate": 0.0003,
|
| 517 |
+
"loss": 3.3914558410644533,
|
| 518 |
+
"step": 1200
|
| 519 |
+
},
|
| 520 |
+
{
|
| 521 |
+
"epoch": 0.04044489383215369,
|
| 522 |
+
"eval_loss": 3.3945467472076416,
|
| 523 |
+
"eval_runtime": 8.0433,
|
| 524 |
+
"eval_samples_per_second": 1184.469,
|
| 525 |
+
"eval_steps_per_second": 0.87,
|
| 526 |
+
"step": 1200
|
| 527 |
+
},
|
| 528 |
+
{
|
| 529 |
+
"epoch": 0.04111897539602292,
|
| 530 |
+
"grad_norm": 1.0546875,
|
| 531 |
+
"learning_rate": 0.0003,
|
| 532 |
+
"loss": 3.3614086151123046,
|
| 533 |
+
"step": 1220
|
| 534 |
+
},
|
| 535 |
+
{
|
| 536 |
+
"epoch": 0.04179305695989215,
|
| 537 |
+
"grad_norm": 1.09375,
|
| 538 |
+
"learning_rate": 0.0003,
|
| 539 |
+
"loss": 3.3790237426757814,
|
| 540 |
+
"step": 1240
|
| 541 |
+
},
|
| 542 |
+
{
|
| 543 |
+
"epoch": 0.042467138523761376,
|
| 544 |
+
"grad_norm": 1.3515625,
|
| 545 |
+
"learning_rate": 0.0003,
|
| 546 |
+
"loss": 3.38794059753418,
|
| 547 |
+
"step": 1260
|
| 548 |
+
},
|
| 549 |
+
{
|
| 550 |
+
"epoch": 0.043141220087630605,
|
| 551 |
+
"grad_norm": 1.28125,
|
| 552 |
+
"learning_rate": 0.0003,
|
| 553 |
+
"loss": 3.3654109954833986,
|
| 554 |
+
"step": 1280
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.043815301651499834,
|
| 558 |
+
"grad_norm": 1.1796875,
|
| 559 |
+
"learning_rate": 0.0003,
|
| 560 |
+
"loss": 3.3572288513183595,
|
| 561 |
+
"step": 1300
|
| 562 |
+
},
|
| 563 |
+
{
|
| 564 |
+
"epoch": 0.043815301651499834,
|
| 565 |
+
"eval_loss": 3.3501007556915283,
|
| 566 |
+
"eval_runtime": 8.0621,
|
| 567 |
+
"eval_samples_per_second": 1181.702,
|
| 568 |
+
"eval_steps_per_second": 0.868,
|
| 569 |
+
"step": 1300
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 0.044489383215369056,
|
| 573 |
+
"grad_norm": 1.2734375,
|
| 574 |
+
"learning_rate": 0.0003,
|
| 575 |
+
"loss": 3.321734619140625,
|
| 576 |
+
"step": 1320
|
| 577 |
+
},
|
| 578 |
+
{
|
| 579 |
+
"epoch": 0.045163464779238285,
|
| 580 |
+
"grad_norm": 1.2890625,
|
| 581 |
+
"learning_rate": 0.0003,
|
| 582 |
+
"loss": 3.3493595123291016,
|
| 583 |
+
"step": 1340
|
| 584 |
+
},
|
| 585 |
+
{
|
| 586 |
+
"epoch": 0.045837546343107514,
|
| 587 |
+
"grad_norm": 1.1875,
|
| 588 |
+
"learning_rate": 0.0003,
|
| 589 |
+
"loss": 3.3272212982177733,
|
| 590 |
+
"step": 1360
|
| 591 |
+
},
|
| 592 |
+
{
|
| 593 |
+
"epoch": 0.046511627906976744,
|
| 594 |
+
"grad_norm": 1.1328125,
|
| 595 |
+
"learning_rate": 0.0003,
|
| 596 |
+
"loss": 3.3202045440673826,
|
| 597 |
+
"step": 1380
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 0.04718570947084597,
|
| 601 |
+
"grad_norm": 1.1171875,
|
| 602 |
+
"learning_rate": 0.0003,
|
| 603 |
+
"loss": 3.2998809814453125,
|
| 604 |
+
"step": 1400
|
| 605 |
+
},
|
| 606 |
+
{
|
| 607 |
+
"epoch": 0.04718570947084597,
|
| 608 |
+
"eval_loss": 3.311476707458496,
|
| 609 |
+
"eval_runtime": 8.1361,
|
| 610 |
+
"eval_samples_per_second": 1170.957,
|
| 611 |
+
"eval_steps_per_second": 0.86,
|
| 612 |
+
"step": 1400
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.0478597910347152,
|
| 616 |
+
"grad_norm": 1.140625,
|
| 617 |
+
"learning_rate": 0.0003,
|
| 618 |
+
"loss": 3.3057735443115233,
|
| 619 |
+
"step": 1420
|
| 620 |
+
},
|
| 621 |
+
{
|
| 622 |
+
"epoch": 0.04853387259858443,
|
| 623 |
+
"grad_norm": 1.109375,
|
| 624 |
+
"learning_rate": 0.0003,
|
| 625 |
+
"loss": 3.2907379150390623,
|
| 626 |
+
"step": 1440
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.04920795416245366,
|
| 630 |
+
"grad_norm": 1.078125,
|
| 631 |
+
"learning_rate": 0.0003,
|
| 632 |
+
"loss": 3.264795684814453,
|
| 633 |
+
"step": 1460
|
| 634 |
+
},
|
| 635 |
+
{
|
| 636 |
+
"epoch": 0.04988203572632288,
|
| 637 |
+
"grad_norm": 1.3671875,
|
| 638 |
+
"learning_rate": 0.0003,
|
| 639 |
+
"loss": 3.2918243408203125,
|
| 640 |
+
"step": 1480
|
| 641 |
+
},
|
| 642 |
+
{
|
| 643 |
+
"epoch": 0.05055611729019211,
|
| 644 |
+
"grad_norm": 1.2421875,
|
| 645 |
+
"learning_rate": 0.0003,
|
| 646 |
+
"loss": 3.2844181060791016,
|
| 647 |
+
"step": 1500
|
| 648 |
+
},
|
| 649 |
+
{
|
| 650 |
+
"epoch": 0.05055611729019211,
|
| 651 |
+
"eval_loss": 3.277984142303467,
|
| 652 |
+
"eval_runtime": 8.0313,
|
| 653 |
+
"eval_samples_per_second": 1186.24,
|
| 654 |
+
"eval_steps_per_second": 0.872,
|
| 655 |
+
"step": 1500
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 0.05123019885406134,
|
| 659 |
+
"grad_norm": 1.265625,
|
| 660 |
+
"learning_rate": 0.0003,
|
| 661 |
+
"loss": 3.2729434967041016,
|
| 662 |
+
"step": 1520
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 0.05190428041793057,
|
| 666 |
+
"grad_norm": 1.1015625,
|
| 667 |
+
"learning_rate": 0.0003,
|
| 668 |
+
"loss": 3.279766845703125,
|
| 669 |
+
"step": 1540
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 0.0525783619817998,
|
| 673 |
+
"grad_norm": 1.25,
|
| 674 |
+
"learning_rate": 0.0003,
|
| 675 |
+
"loss": 3.281978988647461,
|
| 676 |
+
"step": 1560
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 0.05325244354566903,
|
| 680 |
+
"grad_norm": 1.125,
|
| 681 |
+
"learning_rate": 0.0003,
|
| 682 |
+
"loss": 3.2608470916748047,
|
| 683 |
+
"step": 1580
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 0.053926525109538256,
|
| 687 |
+
"grad_norm": 1.2265625,
|
| 688 |
+
"learning_rate": 0.0003,
|
| 689 |
+
"loss": 3.2503360748291015,
|
| 690 |
+
"step": 1600
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.053926525109538256,
|
| 694 |
+
"eval_loss": 3.2511110305786133,
|
| 695 |
+
"eval_runtime": 8.097,
|
| 696 |
+
"eval_samples_per_second": 1176.602,
|
| 697 |
+
"eval_steps_per_second": 0.865,
|
| 698 |
+
"step": 1600
|
| 699 |
+
},
|
| 700 |
+
{
|
| 701 |
+
"epoch": 0.054600606673407485,
|
| 702 |
+
"grad_norm": 1.234375,
|
| 703 |
+
"learning_rate": 0.0003,
|
| 704 |
+
"loss": 3.2713451385498047,
|
| 705 |
+
"step": 1620
|
| 706 |
+
},
|
| 707 |
+
{
|
| 708 |
+
"epoch": 0.05527468823727671,
|
| 709 |
+
"grad_norm": 1.015625,
|
| 710 |
+
"learning_rate": 0.0003,
|
| 711 |
+
"loss": 3.2468338012695312,
|
| 712 |
+
"step": 1640
|
| 713 |
+
},
|
| 714 |
+
{
|
| 715 |
+
"epoch": 0.05594876980114594,
|
| 716 |
+
"grad_norm": 1.15625,
|
| 717 |
+
"learning_rate": 0.0003,
|
| 718 |
+
"loss": 3.255104827880859,
|
| 719 |
+
"step": 1660
|
| 720 |
+
},
|
| 721 |
+
{
|
| 722 |
+
"epoch": 0.056622851365015166,
|
| 723 |
+
"grad_norm": 1.1875,
|
| 724 |
+
"learning_rate": 0.0003,
|
| 725 |
+
"loss": 3.2280048370361327,
|
| 726 |
+
"step": 1680
|
| 727 |
+
},
|
| 728 |
+
{
|
| 729 |
+
"epoch": 0.057296932928884395,
|
| 730 |
+
"grad_norm": 1.265625,
|
| 731 |
+
"learning_rate": 0.0003,
|
| 732 |
+
"loss": 3.216978073120117,
|
| 733 |
+
"step": 1700
|
| 734 |
+
},
|
| 735 |
+
{
|
| 736 |
+
"epoch": 0.057296932928884395,
|
| 737 |
+
"eval_loss": 3.2238714694976807,
|
| 738 |
+
"eval_runtime": 8.059,
|
| 739 |
+
"eval_samples_per_second": 1182.16,
|
| 740 |
+
"eval_steps_per_second": 0.869,
|
| 741 |
+
"step": 1700
|
| 742 |
+
},
|
| 743 |
+
{
|
| 744 |
+
"epoch": 0.057971014492753624,
|
| 745 |
+
"grad_norm": 1.125,
|
| 746 |
+
"learning_rate": 0.0003,
|
| 747 |
+
"loss": 3.213521194458008,
|
| 748 |
+
"step": 1720
|
| 749 |
+
},
|
| 750 |
+
{
|
| 751 |
+
"epoch": 0.05864509605662285,
|
| 752 |
+
"grad_norm": 1.234375,
|
| 753 |
+
"learning_rate": 0.0003,
|
| 754 |
+
"loss": 3.2087985992431642,
|
| 755 |
+
"step": 1740
|
| 756 |
+
},
|
| 757 |
+
{
|
| 758 |
+
"epoch": 0.05931917762049208,
|
| 759 |
+
"grad_norm": 1.2109375,
|
| 760 |
+
"learning_rate": 0.0003,
|
| 761 |
+
"loss": 3.198755645751953,
|
| 762 |
+
"step": 1760
|
| 763 |
+
},
|
| 764 |
+
{
|
| 765 |
+
"epoch": 0.05999325918436131,
|
| 766 |
+
"grad_norm": 1.4453125,
|
| 767 |
+
"learning_rate": 0.0003,
|
| 768 |
+
"loss": 3.2067222595214844,
|
| 769 |
+
"step": 1780
|
| 770 |
+
},
|
| 771 |
+
{
|
| 772 |
+
"epoch": 0.06066734074823053,
|
| 773 |
+
"grad_norm": 1.140625,
|
| 774 |
+
"learning_rate": 0.0003,
|
| 775 |
+
"loss": 3.195578956604004,
|
| 776 |
+
"step": 1800
|
| 777 |
+
},
|
| 778 |
+
{
|
| 779 |
+
"epoch": 0.06066734074823053,
|
| 780 |
+
"eval_loss": 3.1985833644866943,
|
| 781 |
+
"eval_runtime": 8.2334,
|
| 782 |
+
"eval_samples_per_second": 1157.109,
|
| 783 |
+
"eval_steps_per_second": 0.85,
|
| 784 |
+
"step": 1800
|
| 785 |
+
},
|
| 786 |
+
{
|
| 787 |
+
"epoch": 0.06134142231209976,
|
| 788 |
+
"grad_norm": 1.1640625,
|
| 789 |
+
"learning_rate": 0.0003,
|
| 790 |
+
"loss": 3.186159133911133,
|
| 791 |
+
"step": 1820
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"epoch": 0.06201550387596899,
|
| 795 |
+
"grad_norm": 1.21875,
|
| 796 |
+
"learning_rate": 0.0003,
|
| 797 |
+
"loss": 3.1850618362426757,
|
| 798 |
+
"step": 1840
|
| 799 |
+
},
|
| 800 |
+
{
|
| 801 |
+
"epoch": 0.06268958543983821,
|
| 802 |
+
"grad_norm": 1.3515625,
|
| 803 |
+
"learning_rate": 0.0003,
|
| 804 |
+
"loss": 3.1967243194580077,
|
| 805 |
+
"step": 1860
|
| 806 |
+
},
|
| 807 |
+
{
|
| 808 |
+
"epoch": 0.06336366700370745,
|
| 809 |
+
"grad_norm": 1.3125,
|
| 810 |
+
"learning_rate": 0.0003,
|
| 811 |
+
"loss": 3.1668914794921874,
|
| 812 |
+
"step": 1880
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 0.06403774856757667,
|
| 816 |
+
"grad_norm": 1.1484375,
|
| 817 |
+
"learning_rate": 0.0003,
|
| 818 |
+
"loss": 3.1733448028564455,
|
| 819 |
+
"step": 1900
|
| 820 |
+
},
|
| 821 |
+
{
|
| 822 |
+
"epoch": 0.06403774856757667,
|
| 823 |
+
"eval_loss": 3.1766083240509033,
|
| 824 |
+
"eval_runtime": 7.9724,
|
| 825 |
+
"eval_samples_per_second": 1194.991,
|
| 826 |
+
"eval_steps_per_second": 0.878,
|
| 827 |
+
"step": 1900
|
| 828 |
+
},
|
| 829 |
+
{
|
| 830 |
+
"epoch": 0.06471183013144591,
|
| 831 |
+
"grad_norm": 1.1953125,
|
| 832 |
+
"learning_rate": 0.0003,
|
| 833 |
+
"loss": 3.1606868743896483,
|
| 834 |
+
"step": 1920
|
| 835 |
+
},
|
| 836 |
+
{
|
| 837 |
+
"epoch": 0.06538591169531513,
|
| 838 |
+
"grad_norm": 1.109375,
|
| 839 |
+
"learning_rate": 0.0003,
|
| 840 |
+
"loss": 3.148599052429199,
|
| 841 |
+
"step": 1940
|
| 842 |
+
},
|
| 843 |
+
{
|
| 844 |
+
"epoch": 0.06605999325918437,
|
| 845 |
+
"grad_norm": 1.2265625,
|
| 846 |
+
"learning_rate": 0.0003,
|
| 847 |
+
"loss": 3.1723419189453126,
|
| 848 |
+
"step": 1960
|
| 849 |
+
},
|
| 850 |
+
{
|
| 851 |
+
"epoch": 0.06673407482305359,
|
| 852 |
+
"grad_norm": 1.234375,
|
| 853 |
+
"learning_rate": 0.0003,
|
| 854 |
+
"loss": 3.184863471984863,
|
| 855 |
+
"step": 1980
|
| 856 |
+
},
|
| 857 |
+
{
|
| 858 |
+
"epoch": 0.06740815638692282,
|
| 859 |
+
"grad_norm": 1.2734375,
|
| 860 |
+
"learning_rate": 0.0003,
|
| 861 |
+
"loss": 3.158551788330078,
|
| 862 |
+
"step": 2000
|
| 863 |
+
},
|
| 864 |
+
{
|
| 865 |
+
"epoch": 0.06740815638692282,
|
| 866 |
+
"eval_loss": 3.154155731201172,
|
| 867 |
+
"eval_runtime": 7.9931,
|
| 868 |
+
"eval_samples_per_second": 1191.905,
|
| 869 |
+
"eval_steps_per_second": 0.876,
|
| 870 |
+
"step": 2000
|
| 871 |
+
},
|
| 872 |
+
{
|
| 873 |
+
"epoch": 0.06808223795079205,
|
| 874 |
+
"grad_norm": 1.2734375,
|
| 875 |
+
"learning_rate": 0.0003,
|
| 876 |
+
"loss": 3.147844696044922,
|
| 877 |
+
"step": 2020
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 0.06875631951466127,
|
| 881 |
+
"grad_norm": 1.1796875,
|
| 882 |
+
"learning_rate": 0.0003,
|
| 883 |
+
"loss": 3.149020957946777,
|
| 884 |
+
"step": 2040
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 0.0694304010785305,
|
| 888 |
+
"grad_norm": 1.234375,
|
| 889 |
+
"learning_rate": 0.0003,
|
| 890 |
+
"loss": 3.1382816314697264,
|
| 891 |
+
"step": 2060
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 0.07010448264239973,
|
| 895 |
+
"grad_norm": 1.203125,
|
| 896 |
+
"learning_rate": 0.0003,
|
| 897 |
+
"loss": 3.1440708160400392,
|
| 898 |
+
"step": 2080
|
| 899 |
+
},
|
| 900 |
+
{
|
| 901 |
+
"epoch": 0.07077856420626896,
|
| 902 |
+
"grad_norm": 1.1171875,
|
| 903 |
+
"learning_rate": 0.0003,
|
| 904 |
+
"loss": 3.132351875305176,
|
| 905 |
+
"step": 2100
|
| 906 |
+
},
|
| 907 |
+
{
|
| 908 |
+
"epoch": 0.07077856420626896,
|
| 909 |
+
"eval_loss": 3.1331419944763184,
|
| 910 |
+
"eval_runtime": 8.2456,
|
| 911 |
+
"eval_samples_per_second": 1155.405,
|
| 912 |
+
"eval_steps_per_second": 0.849,
|
| 913 |
+
"step": 2100
|
| 914 |
+
}
|
| 915 |
+
],
|
| 916 |
+
"logging_steps": 20,
|
| 917 |
+
"max_steps": 2500,
|
| 918 |
+
"num_input_tokens_seen": 0,
|
| 919 |
+
"num_train_epochs": 1,
|
| 920 |
+
"save_steps": 100,
|
| 921 |
+
"stateful_callbacks": {
|
| 922 |
+
"TrainerControl": {
|
| 923 |
+
"args": {
|
| 924 |
+
"should_epoch_stop": false,
|
| 925 |
+
"should_evaluate": false,
|
| 926 |
+
"should_log": false,
|
| 927 |
+
"should_save": true,
|
| 928 |
+
"should_training_stop": false
|
| 929 |
+
},
|
| 930 |
+
"attributes": {}
|
| 931 |
+
}
|
| 932 |
+
},
|
| 933 |
+
"total_flos": 304907865292800.0,
|
| 934 |
+
"train_batch_size": 32,
|
| 935 |
+
"trial_name": null,
|
| 936 |
+
"trial_params": null
|
| 937 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2100/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a761b578cf2b8c248f28bd1b61a5ec7d0b5d3801c8bd793a6db706df1d3078f6
|
| 3 |
+
size 4920
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/config.json
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "squared_relu",
|
| 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": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 9,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"powlu_m": 3.0,
|
| 25 |
+
"pretraining_tp": 1,
|
| 26 |
+
"rms_norm_eps": 1e-06,
|
| 27 |
+
"rope_parameters": {
|
| 28 |
+
"rope_theta": 10000.0,
|
| 29 |
+
"rope_type": "default"
|
| 30 |
+
},
|
| 31 |
+
"tie_word_embeddings": true,
|
| 32 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 33 |
+
"transformers_version": "5.16.0.dev0",
|
| 34 |
+
"use_cache": false,
|
| 35 |
+
"vocab_size": 4096,
|
| 36 |
+
"waleed_beta": 10.0
|
| 37 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1922d5c6a57c5d6251629ea17f3a6e37d02614b232e1bcf31294583de754cab6
|
| 3 |
+
size 4010544
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:21fe5755951404a5e1e2d2eecc2b48edc6b7047c7fd478391035983c57ae158e
|
| 3 |
+
size 8068282
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7d7dc5cd1ed4bd6f56fc9ef12090bb8dcd7e4f162336bc05d8c592731dbe5b42
|
| 3 |
+
size 14244
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:95f4343cb1ad65cd3aa916e2c22cfe7d27a8e1e8772a1b42aa25f1b477d39b9e
|
| 3 |
+
size 1064
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/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/mlp-squared_relu-9L_run/checkpoint-2200/trainer_state.json
ADDED
|
@@ -0,0 +1,980 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.0741489720256151,
|
| 6 |
+
"eval_steps": 100,
|
| 7 |
+
"global_step": 2200,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0006740815638692282,
|
| 14 |
+
"grad_norm": 2.71875,
|
| 15 |
+
"learning_rate": 1.14e-05,
|
| 16 |
+
"loss": 8.314944458007812,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.0013481631277384564,
|
| 21 |
+
"grad_norm": 1.8515625,
|
| 22 |
+
"learning_rate": 2.34e-05,
|
| 23 |
+
"loss": 8.251383972167968,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.0020222446916076846,
|
| 28 |
+
"grad_norm": 1.3203125,
|
| 29 |
+
"learning_rate": 3.539999999999999e-05,
|
| 30 |
+
"loss": 8.104505920410157,
|
| 31 |
+
"step": 60
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"epoch": 0.002696326255476913,
|
| 35 |
+
"grad_norm": 1.296875,
|
| 36 |
+
"learning_rate": 4.7399999999999993e-05,
|
| 37 |
+
"loss": 7.968361663818359,
|
| 38 |
+
"step": 80
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.003370407819346141,
|
| 42 |
+
"grad_norm": 1.2578125,
|
| 43 |
+
"learning_rate": 5.94e-05,
|
| 44 |
+
"loss": 7.804827880859375,
|
| 45 |
+
"step": 100
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.003370407819346141,
|
| 49 |
+
"eval_loss": 7.703802585601807,
|
| 50 |
+
"eval_runtime": 8.0325,
|
| 51 |
+
"eval_samples_per_second": 1186.059,
|
| 52 |
+
"eval_steps_per_second": 0.871,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.004044489383215369,
|
| 57 |
+
"grad_norm": 1.2734375,
|
| 58 |
+
"learning_rate": 7.139999999999999e-05,
|
| 59 |
+
"loss": 7.600141906738282,
|
| 60 |
+
"step": 120
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"epoch": 0.0047185709470845974,
|
| 64 |
+
"grad_norm": 1.265625,
|
| 65 |
+
"learning_rate": 8.34e-05,
|
| 66 |
+
"loss": 7.368203735351562,
|
| 67 |
+
"step": 140
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"epoch": 0.005392652510953826,
|
| 71 |
+
"grad_norm": 1.265625,
|
| 72 |
+
"learning_rate": 9.539999999999999e-05,
|
| 73 |
+
"loss": 7.133045196533203,
|
| 74 |
+
"step": 160
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.006066734074823054,
|
| 78 |
+
"grad_norm": 1.203125,
|
| 79 |
+
"learning_rate": 0.00010739999999999998,
|
| 80 |
+
"loss": 6.885277557373047,
|
| 81 |
+
"step": 180
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"epoch": 0.006740815638692282,
|
| 85 |
+
"grad_norm": 1.15625,
|
| 86 |
+
"learning_rate": 0.0001194,
|
| 87 |
+
"loss": 6.629558563232422,
|
| 88 |
+
"step": 200
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"epoch": 0.006740815638692282,
|
| 92 |
+
"eval_loss": 6.4997687339782715,
|
| 93 |
+
"eval_runtime": 7.9479,
|
| 94 |
+
"eval_samples_per_second": 1198.683,
|
| 95 |
+
"eval_steps_per_second": 0.881,
|
| 96 |
+
"step": 200
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"epoch": 0.00741489720256151,
|
| 100 |
+
"grad_norm": 1.109375,
|
| 101 |
+
"learning_rate": 0.0001314,
|
| 102 |
+
"loss": 6.391195297241211,
|
| 103 |
+
"step": 220
|
| 104 |
+
},
|
| 105 |
+
{
|
| 106 |
+
"epoch": 0.008088978766430738,
|
| 107 |
+
"grad_norm": 1.21875,
|
| 108 |
+
"learning_rate": 0.0001434,
|
| 109 |
+
"loss": 6.139227676391601,
|
| 110 |
+
"step": 240
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"epoch": 0.008763060330299966,
|
| 114 |
+
"grad_norm": 1.0625,
|
| 115 |
+
"learning_rate": 0.00015539999999999998,
|
| 116 |
+
"loss": 5.917041778564453,
|
| 117 |
+
"step": 260
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"epoch": 0.009437141894169195,
|
| 121 |
+
"grad_norm": 1.1015625,
|
| 122 |
+
"learning_rate": 0.0001674,
|
| 123 |
+
"loss": 5.702054595947265,
|
| 124 |
+
"step": 280
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"epoch": 0.010111223458038422,
|
| 128 |
+
"grad_norm": 0.921875,
|
| 129 |
+
"learning_rate": 0.00017939999999999997,
|
| 130 |
+
"loss": 5.500263214111328,
|
| 131 |
+
"step": 300
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"epoch": 0.010111223458038422,
|
| 135 |
+
"eval_loss": 5.393505573272705,
|
| 136 |
+
"eval_runtime": 8.0449,
|
| 137 |
+
"eval_samples_per_second": 1184.223,
|
| 138 |
+
"eval_steps_per_second": 0.87,
|
| 139 |
+
"step": 300
|
| 140 |
+
},
|
| 141 |
+
{
|
| 142 |
+
"epoch": 0.010785305021907651,
|
| 143 |
+
"grad_norm": 2.5,
|
| 144 |
+
"learning_rate": 0.0001914,
|
| 145 |
+
"loss": 5.319306182861328,
|
| 146 |
+
"step": 320
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.011459386585776879,
|
| 150 |
+
"grad_norm": 1.078125,
|
| 151 |
+
"learning_rate": 0.00020339999999999998,
|
| 152 |
+
"loss": 5.202053070068359,
|
| 153 |
+
"step": 340
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"epoch": 0.012133468149646108,
|
| 157 |
+
"grad_norm": 1.3203125,
|
| 158 |
+
"learning_rate": 0.00021539999999999998,
|
| 159 |
+
"loss": 5.020883941650391,
|
| 160 |
+
"step": 360
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 0.012807549713515335,
|
| 164 |
+
"grad_norm": 1.1171875,
|
| 165 |
+
"learning_rate": 0.00022739999999999997,
|
| 166 |
+
"loss": 4.873059844970703,
|
| 167 |
+
"step": 380
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"epoch": 0.013481631277384564,
|
| 171 |
+
"grad_norm": 1.234375,
|
| 172 |
+
"learning_rate": 0.0002394,
|
| 173 |
+
"loss": 4.741780853271484,
|
| 174 |
+
"step": 400
|
| 175 |
+
},
|
| 176 |
+
{
|
| 177 |
+
"epoch": 0.013481631277384564,
|
| 178 |
+
"eval_loss": 4.676144123077393,
|
| 179 |
+
"eval_runtime": 8.0457,
|
| 180 |
+
"eval_samples_per_second": 1184.113,
|
| 181 |
+
"eval_steps_per_second": 0.87,
|
| 182 |
+
"step": 400
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"epoch": 0.014155712841253791,
|
| 186 |
+
"grad_norm": 1.3046875,
|
| 187 |
+
"learning_rate": 0.0002514,
|
| 188 |
+
"loss": 4.606548690795899,
|
| 189 |
+
"step": 420
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 0.01482979440512302,
|
| 193 |
+
"grad_norm": 1.2578125,
|
| 194 |
+
"learning_rate": 0.00026339999999999995,
|
| 195 |
+
"loss": 4.515376281738281,
|
| 196 |
+
"step": 440
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"epoch": 0.015503875968992248,
|
| 200 |
+
"grad_norm": 1.0625,
|
| 201 |
+
"learning_rate": 0.00027539999999999997,
|
| 202 |
+
"loss": 4.416887664794922,
|
| 203 |
+
"step": 460
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"epoch": 0.016177957532861477,
|
| 207 |
+
"grad_norm": 1.3671875,
|
| 208 |
+
"learning_rate": 0.00028739999999999994,
|
| 209 |
+
"loss": 4.3046520233154295,
|
| 210 |
+
"step": 480
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.016852039096730706,
|
| 214 |
+
"grad_norm": 1.2109375,
|
| 215 |
+
"learning_rate": 0.00029939999999999996,
|
| 216 |
+
"loss": 4.238706970214844,
|
| 217 |
+
"step": 500
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"epoch": 0.016852039096730706,
|
| 221 |
+
"eval_loss": 4.196270942687988,
|
| 222 |
+
"eval_runtime": 7.9301,
|
| 223 |
+
"eval_samples_per_second": 1201.377,
|
| 224 |
+
"eval_steps_per_second": 0.883,
|
| 225 |
+
"step": 500
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"epoch": 0.01752612066059993,
|
| 229 |
+
"grad_norm": 0.97265625,
|
| 230 |
+
"learning_rate": 0.0003,
|
| 231 |
+
"loss": 4.1756237030029295,
|
| 232 |
+
"step": 520
|
| 233 |
+
},
|
| 234 |
+
{
|
| 235 |
+
"epoch": 0.01820020222446916,
|
| 236 |
+
"grad_norm": 0.9140625,
|
| 237 |
+
"learning_rate": 0.0003,
|
| 238 |
+
"loss": 4.0915069580078125,
|
| 239 |
+
"step": 540
|
| 240 |
+
},
|
| 241 |
+
{
|
| 242 |
+
"epoch": 0.01887428378833839,
|
| 243 |
+
"grad_norm": 1.2890625,
|
| 244 |
+
"learning_rate": 0.0003,
|
| 245 |
+
"loss": 4.031630325317383,
|
| 246 |
+
"step": 560
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"epoch": 0.01954836535220762,
|
| 250 |
+
"grad_norm": 1.125,
|
| 251 |
+
"learning_rate": 0.0003,
|
| 252 |
+
"loss": 3.9763336181640625,
|
| 253 |
+
"step": 580
|
| 254 |
+
},
|
| 255 |
+
{
|
| 256 |
+
"epoch": 0.020222446916076844,
|
| 257 |
+
"grad_norm": 1.21875,
|
| 258 |
+
"learning_rate": 0.0003,
|
| 259 |
+
"loss": 3.9347896575927734,
|
| 260 |
+
"step": 600
|
| 261 |
+
},
|
| 262 |
+
{
|
| 263 |
+
"epoch": 0.020222446916076844,
|
| 264 |
+
"eval_loss": 3.9157378673553467,
|
| 265 |
+
"eval_runtime": 8.0606,
|
| 266 |
+
"eval_samples_per_second": 1181.917,
|
| 267 |
+
"eval_steps_per_second": 0.868,
|
| 268 |
+
"step": 600
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.020896528479946073,
|
| 272 |
+
"grad_norm": 1.1640625,
|
| 273 |
+
"learning_rate": 0.0003,
|
| 274 |
+
"loss": 3.878151702880859,
|
| 275 |
+
"step": 620
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.021570610043815303,
|
| 279 |
+
"grad_norm": 1.484375,
|
| 280 |
+
"learning_rate": 0.0003,
|
| 281 |
+
"loss": 3.855076217651367,
|
| 282 |
+
"step": 640
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.022244691607684528,
|
| 286 |
+
"grad_norm": 1.3359375,
|
| 287 |
+
"learning_rate": 0.0003,
|
| 288 |
+
"loss": 3.8232196807861327,
|
| 289 |
+
"step": 660
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.022918773171553757,
|
| 293 |
+
"grad_norm": 1.1640625,
|
| 294 |
+
"learning_rate": 0.0003,
|
| 295 |
+
"loss": 3.782532501220703,
|
| 296 |
+
"step": 680
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.023592854735422986,
|
| 300 |
+
"grad_norm": 1.1328125,
|
| 301 |
+
"learning_rate": 0.0003,
|
| 302 |
+
"loss": 3.7758609771728517,
|
| 303 |
+
"step": 700
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.023592854735422986,
|
| 307 |
+
"eval_loss": 3.7596347332000732,
|
| 308 |
+
"eval_runtime": 7.9743,
|
| 309 |
+
"eval_samples_per_second": 1194.709,
|
| 310 |
+
"eval_steps_per_second": 0.878,
|
| 311 |
+
"step": 700
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"epoch": 0.024266936299292215,
|
| 315 |
+
"grad_norm": 1.25,
|
| 316 |
+
"learning_rate": 0.0003,
|
| 317 |
+
"loss": 3.7433414459228516,
|
| 318 |
+
"step": 720
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 0.02494101786316144,
|
| 322 |
+
"grad_norm": 1.3125,
|
| 323 |
+
"learning_rate": 0.0003,
|
| 324 |
+
"loss": 3.732330322265625,
|
| 325 |
+
"step": 740
|
| 326 |
+
},
|
| 327 |
+
{
|
| 328 |
+
"epoch": 0.02561509942703067,
|
| 329 |
+
"grad_norm": 1.265625,
|
| 330 |
+
"learning_rate": 0.0003,
|
| 331 |
+
"loss": 3.6843395233154297,
|
| 332 |
+
"step": 760
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"epoch": 0.0262891809908999,
|
| 336 |
+
"grad_norm": 1.1640625,
|
| 337 |
+
"learning_rate": 0.0003,
|
| 338 |
+
"loss": 3.6463153839111326,
|
| 339 |
+
"step": 780
|
| 340 |
+
},
|
| 341 |
+
{
|
| 342 |
+
"epoch": 0.026963262554769128,
|
| 343 |
+
"grad_norm": 1.1171875,
|
| 344 |
+
"learning_rate": 0.0003,
|
| 345 |
+
"loss": 3.6544403076171874,
|
| 346 |
+
"step": 800
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.026963262554769128,
|
| 350 |
+
"eval_loss": 3.652294635772705,
|
| 351 |
+
"eval_runtime": 7.9753,
|
| 352 |
+
"eval_samples_per_second": 1194.559,
|
| 353 |
+
"eval_steps_per_second": 0.878,
|
| 354 |
+
"step": 800
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.027637344118638354,
|
| 358 |
+
"grad_norm": 1.1171875,
|
| 359 |
+
"learning_rate": 0.0003,
|
| 360 |
+
"loss": 3.6159641265869142,
|
| 361 |
+
"step": 820
|
| 362 |
+
},
|
| 363 |
+
{
|
| 364 |
+
"epoch": 0.028311425682507583,
|
| 365 |
+
"grad_norm": 1.078125,
|
| 366 |
+
"learning_rate": 0.0003,
|
| 367 |
+
"loss": 3.6099781036376952,
|
| 368 |
+
"step": 840
|
| 369 |
+
},
|
| 370 |
+
{
|
| 371 |
+
"epoch": 0.028985507246376812,
|
| 372 |
+
"grad_norm": 1.2734375,
|
| 373 |
+
"learning_rate": 0.0003,
|
| 374 |
+
"loss": 3.6344444274902346,
|
| 375 |
+
"step": 860
|
| 376 |
+
},
|
| 377 |
+
{
|
| 378 |
+
"epoch": 0.02965958881024604,
|
| 379 |
+
"grad_norm": 1.140625,
|
| 380 |
+
"learning_rate": 0.0003,
|
| 381 |
+
"loss": 3.5855178833007812,
|
| 382 |
+
"step": 880
|
| 383 |
+
},
|
| 384 |
+
{
|
| 385 |
+
"epoch": 0.030333670374115267,
|
| 386 |
+
"grad_norm": 1.109375,
|
| 387 |
+
"learning_rate": 0.0003,
|
| 388 |
+
"loss": 3.5488441467285154,
|
| 389 |
+
"step": 900
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.030333670374115267,
|
| 393 |
+
"eval_loss": 3.5648436546325684,
|
| 394 |
+
"eval_runtime": 8.321,
|
| 395 |
+
"eval_samples_per_second": 1144.931,
|
| 396 |
+
"eval_steps_per_second": 0.841,
|
| 397 |
+
"step": 900
|
| 398 |
+
},
|
| 399 |
+
{
|
| 400 |
+
"epoch": 0.031007751937984496,
|
| 401 |
+
"grad_norm": 1.1015625,
|
| 402 |
+
"learning_rate": 0.0003,
|
| 403 |
+
"loss": 3.5646598815917967,
|
| 404 |
+
"step": 920
|
| 405 |
+
},
|
| 406 |
+
{
|
| 407 |
+
"epoch": 0.031681833501853725,
|
| 408 |
+
"grad_norm": 1.25,
|
| 409 |
+
"learning_rate": 0.0003,
|
| 410 |
+
"loss": 3.5480445861816405,
|
| 411 |
+
"step": 940
|
| 412 |
+
},
|
| 413 |
+
{
|
| 414 |
+
"epoch": 0.032355915065722954,
|
| 415 |
+
"grad_norm": 1.140625,
|
| 416 |
+
"learning_rate": 0.0003,
|
| 417 |
+
"loss": 3.5140933990478516,
|
| 418 |
+
"step": 960
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.03302999662959218,
|
| 422 |
+
"grad_norm": 1.140625,
|
| 423 |
+
"learning_rate": 0.0003,
|
| 424 |
+
"loss": 3.501856231689453,
|
| 425 |
+
"step": 980
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.03370407819346141,
|
| 429 |
+
"grad_norm": 1.359375,
|
| 430 |
+
"learning_rate": 0.0003,
|
| 431 |
+
"loss": 3.500592803955078,
|
| 432 |
+
"step": 1000
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.03370407819346141,
|
| 436 |
+
"eval_loss": 3.4981162548065186,
|
| 437 |
+
"eval_runtime": 8.1013,
|
| 438 |
+
"eval_samples_per_second": 1175.987,
|
| 439 |
+
"eval_steps_per_second": 0.864,
|
| 440 |
+
"step": 1000
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.034378159757330634,
|
| 444 |
+
"grad_norm": 1.1328125,
|
| 445 |
+
"learning_rate": 0.0003,
|
| 446 |
+
"loss": 3.4914066314697267,
|
| 447 |
+
"step": 1020
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.03505224132119986,
|
| 451 |
+
"grad_norm": 1.1171875,
|
| 452 |
+
"learning_rate": 0.0003,
|
| 453 |
+
"loss": 3.474192810058594,
|
| 454 |
+
"step": 1040
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.03572632288506909,
|
| 458 |
+
"grad_norm": 1.28125,
|
| 459 |
+
"learning_rate": 0.0003,
|
| 460 |
+
"loss": 3.4552947998046877,
|
| 461 |
+
"step": 1060
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.03640040444893832,
|
| 465 |
+
"grad_norm": 1.203125,
|
| 466 |
+
"learning_rate": 0.0003,
|
| 467 |
+
"loss": 3.4411731719970704,
|
| 468 |
+
"step": 1080
|
| 469 |
+
},
|
| 470 |
+
{
|
| 471 |
+
"epoch": 0.03707448601280755,
|
| 472 |
+
"grad_norm": 1.2734375,
|
| 473 |
+
"learning_rate": 0.0003,
|
| 474 |
+
"loss": 3.435680389404297,
|
| 475 |
+
"step": 1100
|
| 476 |
+
},
|
| 477 |
+
{
|
| 478 |
+
"epoch": 0.03707448601280755,
|
| 479 |
+
"eval_loss": 3.445117712020874,
|
| 480 |
+
"eval_runtime": 8.1641,
|
| 481 |
+
"eval_samples_per_second": 1166.938,
|
| 482 |
+
"eval_steps_per_second": 0.857,
|
| 483 |
+
"step": 1100
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.03774856757667678,
|
| 487 |
+
"grad_norm": 1.03125,
|
| 488 |
+
"learning_rate": 0.0003,
|
| 489 |
+
"loss": 3.4368762969970703,
|
| 490 |
+
"step": 1120
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.03842264914054601,
|
| 494 |
+
"grad_norm": 1.2421875,
|
| 495 |
+
"learning_rate": 0.0003,
|
| 496 |
+
"loss": 3.410645294189453,
|
| 497 |
+
"step": 1140
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"epoch": 0.03909673070441524,
|
| 501 |
+
"grad_norm": 1.0703125,
|
| 502 |
+
"learning_rate": 0.0003,
|
| 503 |
+
"loss": 3.4069515228271485,
|
| 504 |
+
"step": 1160
|
| 505 |
+
},
|
| 506 |
+
{
|
| 507 |
+
"epoch": 0.03977081226828446,
|
| 508 |
+
"grad_norm": 1.03125,
|
| 509 |
+
"learning_rate": 0.0003,
|
| 510 |
+
"loss": 3.4227649688720705,
|
| 511 |
+
"step": 1180
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"epoch": 0.04044489383215369,
|
| 515 |
+
"grad_norm": 1.203125,
|
| 516 |
+
"learning_rate": 0.0003,
|
| 517 |
+
"loss": 3.3914558410644533,
|
| 518 |
+
"step": 1200
|
| 519 |
+
},
|
| 520 |
+
{
|
| 521 |
+
"epoch": 0.04044489383215369,
|
| 522 |
+
"eval_loss": 3.3945467472076416,
|
| 523 |
+
"eval_runtime": 8.0433,
|
| 524 |
+
"eval_samples_per_second": 1184.469,
|
| 525 |
+
"eval_steps_per_second": 0.87,
|
| 526 |
+
"step": 1200
|
| 527 |
+
},
|
| 528 |
+
{
|
| 529 |
+
"epoch": 0.04111897539602292,
|
| 530 |
+
"grad_norm": 1.0546875,
|
| 531 |
+
"learning_rate": 0.0003,
|
| 532 |
+
"loss": 3.3614086151123046,
|
| 533 |
+
"step": 1220
|
| 534 |
+
},
|
| 535 |
+
{
|
| 536 |
+
"epoch": 0.04179305695989215,
|
| 537 |
+
"grad_norm": 1.09375,
|
| 538 |
+
"learning_rate": 0.0003,
|
| 539 |
+
"loss": 3.3790237426757814,
|
| 540 |
+
"step": 1240
|
| 541 |
+
},
|
| 542 |
+
{
|
| 543 |
+
"epoch": 0.042467138523761376,
|
| 544 |
+
"grad_norm": 1.3515625,
|
| 545 |
+
"learning_rate": 0.0003,
|
| 546 |
+
"loss": 3.38794059753418,
|
| 547 |
+
"step": 1260
|
| 548 |
+
},
|
| 549 |
+
{
|
| 550 |
+
"epoch": 0.043141220087630605,
|
| 551 |
+
"grad_norm": 1.28125,
|
| 552 |
+
"learning_rate": 0.0003,
|
| 553 |
+
"loss": 3.3654109954833986,
|
| 554 |
+
"step": 1280
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.043815301651499834,
|
| 558 |
+
"grad_norm": 1.1796875,
|
| 559 |
+
"learning_rate": 0.0003,
|
| 560 |
+
"loss": 3.3572288513183595,
|
| 561 |
+
"step": 1300
|
| 562 |
+
},
|
| 563 |
+
{
|
| 564 |
+
"epoch": 0.043815301651499834,
|
| 565 |
+
"eval_loss": 3.3501007556915283,
|
| 566 |
+
"eval_runtime": 8.0621,
|
| 567 |
+
"eval_samples_per_second": 1181.702,
|
| 568 |
+
"eval_steps_per_second": 0.868,
|
| 569 |
+
"step": 1300
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 0.044489383215369056,
|
| 573 |
+
"grad_norm": 1.2734375,
|
| 574 |
+
"learning_rate": 0.0003,
|
| 575 |
+
"loss": 3.321734619140625,
|
| 576 |
+
"step": 1320
|
| 577 |
+
},
|
| 578 |
+
{
|
| 579 |
+
"epoch": 0.045163464779238285,
|
| 580 |
+
"grad_norm": 1.2890625,
|
| 581 |
+
"learning_rate": 0.0003,
|
| 582 |
+
"loss": 3.3493595123291016,
|
| 583 |
+
"step": 1340
|
| 584 |
+
},
|
| 585 |
+
{
|
| 586 |
+
"epoch": 0.045837546343107514,
|
| 587 |
+
"grad_norm": 1.1875,
|
| 588 |
+
"learning_rate": 0.0003,
|
| 589 |
+
"loss": 3.3272212982177733,
|
| 590 |
+
"step": 1360
|
| 591 |
+
},
|
| 592 |
+
{
|
| 593 |
+
"epoch": 0.046511627906976744,
|
| 594 |
+
"grad_norm": 1.1328125,
|
| 595 |
+
"learning_rate": 0.0003,
|
| 596 |
+
"loss": 3.3202045440673826,
|
| 597 |
+
"step": 1380
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 0.04718570947084597,
|
| 601 |
+
"grad_norm": 1.1171875,
|
| 602 |
+
"learning_rate": 0.0003,
|
| 603 |
+
"loss": 3.2998809814453125,
|
| 604 |
+
"step": 1400
|
| 605 |
+
},
|
| 606 |
+
{
|
| 607 |
+
"epoch": 0.04718570947084597,
|
| 608 |
+
"eval_loss": 3.311476707458496,
|
| 609 |
+
"eval_runtime": 8.1361,
|
| 610 |
+
"eval_samples_per_second": 1170.957,
|
| 611 |
+
"eval_steps_per_second": 0.86,
|
| 612 |
+
"step": 1400
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.0478597910347152,
|
| 616 |
+
"grad_norm": 1.140625,
|
| 617 |
+
"learning_rate": 0.0003,
|
| 618 |
+
"loss": 3.3057735443115233,
|
| 619 |
+
"step": 1420
|
| 620 |
+
},
|
| 621 |
+
{
|
| 622 |
+
"epoch": 0.04853387259858443,
|
| 623 |
+
"grad_norm": 1.109375,
|
| 624 |
+
"learning_rate": 0.0003,
|
| 625 |
+
"loss": 3.2907379150390623,
|
| 626 |
+
"step": 1440
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.04920795416245366,
|
| 630 |
+
"grad_norm": 1.078125,
|
| 631 |
+
"learning_rate": 0.0003,
|
| 632 |
+
"loss": 3.264795684814453,
|
| 633 |
+
"step": 1460
|
| 634 |
+
},
|
| 635 |
+
{
|
| 636 |
+
"epoch": 0.04988203572632288,
|
| 637 |
+
"grad_norm": 1.3671875,
|
| 638 |
+
"learning_rate": 0.0003,
|
| 639 |
+
"loss": 3.2918243408203125,
|
| 640 |
+
"step": 1480
|
| 641 |
+
},
|
| 642 |
+
{
|
| 643 |
+
"epoch": 0.05055611729019211,
|
| 644 |
+
"grad_norm": 1.2421875,
|
| 645 |
+
"learning_rate": 0.0003,
|
| 646 |
+
"loss": 3.2844181060791016,
|
| 647 |
+
"step": 1500
|
| 648 |
+
},
|
| 649 |
+
{
|
| 650 |
+
"epoch": 0.05055611729019211,
|
| 651 |
+
"eval_loss": 3.277984142303467,
|
| 652 |
+
"eval_runtime": 8.0313,
|
| 653 |
+
"eval_samples_per_second": 1186.24,
|
| 654 |
+
"eval_steps_per_second": 0.872,
|
| 655 |
+
"step": 1500
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 0.05123019885406134,
|
| 659 |
+
"grad_norm": 1.265625,
|
| 660 |
+
"learning_rate": 0.0003,
|
| 661 |
+
"loss": 3.2729434967041016,
|
| 662 |
+
"step": 1520
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 0.05190428041793057,
|
| 666 |
+
"grad_norm": 1.1015625,
|
| 667 |
+
"learning_rate": 0.0003,
|
| 668 |
+
"loss": 3.279766845703125,
|
| 669 |
+
"step": 1540
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 0.0525783619817998,
|
| 673 |
+
"grad_norm": 1.25,
|
| 674 |
+
"learning_rate": 0.0003,
|
| 675 |
+
"loss": 3.281978988647461,
|
| 676 |
+
"step": 1560
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 0.05325244354566903,
|
| 680 |
+
"grad_norm": 1.125,
|
| 681 |
+
"learning_rate": 0.0003,
|
| 682 |
+
"loss": 3.2608470916748047,
|
| 683 |
+
"step": 1580
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 0.053926525109538256,
|
| 687 |
+
"grad_norm": 1.2265625,
|
| 688 |
+
"learning_rate": 0.0003,
|
| 689 |
+
"loss": 3.2503360748291015,
|
| 690 |
+
"step": 1600
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.053926525109538256,
|
| 694 |
+
"eval_loss": 3.2511110305786133,
|
| 695 |
+
"eval_runtime": 8.097,
|
| 696 |
+
"eval_samples_per_second": 1176.602,
|
| 697 |
+
"eval_steps_per_second": 0.865,
|
| 698 |
+
"step": 1600
|
| 699 |
+
},
|
| 700 |
+
{
|
| 701 |
+
"epoch": 0.054600606673407485,
|
| 702 |
+
"grad_norm": 1.234375,
|
| 703 |
+
"learning_rate": 0.0003,
|
| 704 |
+
"loss": 3.2713451385498047,
|
| 705 |
+
"step": 1620
|
| 706 |
+
},
|
| 707 |
+
{
|
| 708 |
+
"epoch": 0.05527468823727671,
|
| 709 |
+
"grad_norm": 1.015625,
|
| 710 |
+
"learning_rate": 0.0003,
|
| 711 |
+
"loss": 3.2468338012695312,
|
| 712 |
+
"step": 1640
|
| 713 |
+
},
|
| 714 |
+
{
|
| 715 |
+
"epoch": 0.05594876980114594,
|
| 716 |
+
"grad_norm": 1.15625,
|
| 717 |
+
"learning_rate": 0.0003,
|
| 718 |
+
"loss": 3.255104827880859,
|
| 719 |
+
"step": 1660
|
| 720 |
+
},
|
| 721 |
+
{
|
| 722 |
+
"epoch": 0.056622851365015166,
|
| 723 |
+
"grad_norm": 1.1875,
|
| 724 |
+
"learning_rate": 0.0003,
|
| 725 |
+
"loss": 3.2280048370361327,
|
| 726 |
+
"step": 1680
|
| 727 |
+
},
|
| 728 |
+
{
|
| 729 |
+
"epoch": 0.057296932928884395,
|
| 730 |
+
"grad_norm": 1.265625,
|
| 731 |
+
"learning_rate": 0.0003,
|
| 732 |
+
"loss": 3.216978073120117,
|
| 733 |
+
"step": 1700
|
| 734 |
+
},
|
| 735 |
+
{
|
| 736 |
+
"epoch": 0.057296932928884395,
|
| 737 |
+
"eval_loss": 3.2238714694976807,
|
| 738 |
+
"eval_runtime": 8.059,
|
| 739 |
+
"eval_samples_per_second": 1182.16,
|
| 740 |
+
"eval_steps_per_second": 0.869,
|
| 741 |
+
"step": 1700
|
| 742 |
+
},
|
| 743 |
+
{
|
| 744 |
+
"epoch": 0.057971014492753624,
|
| 745 |
+
"grad_norm": 1.125,
|
| 746 |
+
"learning_rate": 0.0003,
|
| 747 |
+
"loss": 3.213521194458008,
|
| 748 |
+
"step": 1720
|
| 749 |
+
},
|
| 750 |
+
{
|
| 751 |
+
"epoch": 0.05864509605662285,
|
| 752 |
+
"grad_norm": 1.234375,
|
| 753 |
+
"learning_rate": 0.0003,
|
| 754 |
+
"loss": 3.2087985992431642,
|
| 755 |
+
"step": 1740
|
| 756 |
+
},
|
| 757 |
+
{
|
| 758 |
+
"epoch": 0.05931917762049208,
|
| 759 |
+
"grad_norm": 1.2109375,
|
| 760 |
+
"learning_rate": 0.0003,
|
| 761 |
+
"loss": 3.198755645751953,
|
| 762 |
+
"step": 1760
|
| 763 |
+
},
|
| 764 |
+
{
|
| 765 |
+
"epoch": 0.05999325918436131,
|
| 766 |
+
"grad_norm": 1.4453125,
|
| 767 |
+
"learning_rate": 0.0003,
|
| 768 |
+
"loss": 3.2067222595214844,
|
| 769 |
+
"step": 1780
|
| 770 |
+
},
|
| 771 |
+
{
|
| 772 |
+
"epoch": 0.06066734074823053,
|
| 773 |
+
"grad_norm": 1.140625,
|
| 774 |
+
"learning_rate": 0.0003,
|
| 775 |
+
"loss": 3.195578956604004,
|
| 776 |
+
"step": 1800
|
| 777 |
+
},
|
| 778 |
+
{
|
| 779 |
+
"epoch": 0.06066734074823053,
|
| 780 |
+
"eval_loss": 3.1985833644866943,
|
| 781 |
+
"eval_runtime": 8.2334,
|
| 782 |
+
"eval_samples_per_second": 1157.109,
|
| 783 |
+
"eval_steps_per_second": 0.85,
|
| 784 |
+
"step": 1800
|
| 785 |
+
},
|
| 786 |
+
{
|
| 787 |
+
"epoch": 0.06134142231209976,
|
| 788 |
+
"grad_norm": 1.1640625,
|
| 789 |
+
"learning_rate": 0.0003,
|
| 790 |
+
"loss": 3.186159133911133,
|
| 791 |
+
"step": 1820
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"epoch": 0.06201550387596899,
|
| 795 |
+
"grad_norm": 1.21875,
|
| 796 |
+
"learning_rate": 0.0003,
|
| 797 |
+
"loss": 3.1850618362426757,
|
| 798 |
+
"step": 1840
|
| 799 |
+
},
|
| 800 |
+
{
|
| 801 |
+
"epoch": 0.06268958543983821,
|
| 802 |
+
"grad_norm": 1.3515625,
|
| 803 |
+
"learning_rate": 0.0003,
|
| 804 |
+
"loss": 3.1967243194580077,
|
| 805 |
+
"step": 1860
|
| 806 |
+
},
|
| 807 |
+
{
|
| 808 |
+
"epoch": 0.06336366700370745,
|
| 809 |
+
"grad_norm": 1.3125,
|
| 810 |
+
"learning_rate": 0.0003,
|
| 811 |
+
"loss": 3.1668914794921874,
|
| 812 |
+
"step": 1880
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 0.06403774856757667,
|
| 816 |
+
"grad_norm": 1.1484375,
|
| 817 |
+
"learning_rate": 0.0003,
|
| 818 |
+
"loss": 3.1733448028564455,
|
| 819 |
+
"step": 1900
|
| 820 |
+
},
|
| 821 |
+
{
|
| 822 |
+
"epoch": 0.06403774856757667,
|
| 823 |
+
"eval_loss": 3.1766083240509033,
|
| 824 |
+
"eval_runtime": 7.9724,
|
| 825 |
+
"eval_samples_per_second": 1194.991,
|
| 826 |
+
"eval_steps_per_second": 0.878,
|
| 827 |
+
"step": 1900
|
| 828 |
+
},
|
| 829 |
+
{
|
| 830 |
+
"epoch": 0.06471183013144591,
|
| 831 |
+
"grad_norm": 1.1953125,
|
| 832 |
+
"learning_rate": 0.0003,
|
| 833 |
+
"loss": 3.1606868743896483,
|
| 834 |
+
"step": 1920
|
| 835 |
+
},
|
| 836 |
+
{
|
| 837 |
+
"epoch": 0.06538591169531513,
|
| 838 |
+
"grad_norm": 1.109375,
|
| 839 |
+
"learning_rate": 0.0003,
|
| 840 |
+
"loss": 3.148599052429199,
|
| 841 |
+
"step": 1940
|
| 842 |
+
},
|
| 843 |
+
{
|
| 844 |
+
"epoch": 0.06605999325918437,
|
| 845 |
+
"grad_norm": 1.2265625,
|
| 846 |
+
"learning_rate": 0.0003,
|
| 847 |
+
"loss": 3.1723419189453126,
|
| 848 |
+
"step": 1960
|
| 849 |
+
},
|
| 850 |
+
{
|
| 851 |
+
"epoch": 0.06673407482305359,
|
| 852 |
+
"grad_norm": 1.234375,
|
| 853 |
+
"learning_rate": 0.0003,
|
| 854 |
+
"loss": 3.184863471984863,
|
| 855 |
+
"step": 1980
|
| 856 |
+
},
|
| 857 |
+
{
|
| 858 |
+
"epoch": 0.06740815638692282,
|
| 859 |
+
"grad_norm": 1.2734375,
|
| 860 |
+
"learning_rate": 0.0003,
|
| 861 |
+
"loss": 3.158551788330078,
|
| 862 |
+
"step": 2000
|
| 863 |
+
},
|
| 864 |
+
{
|
| 865 |
+
"epoch": 0.06740815638692282,
|
| 866 |
+
"eval_loss": 3.154155731201172,
|
| 867 |
+
"eval_runtime": 7.9931,
|
| 868 |
+
"eval_samples_per_second": 1191.905,
|
| 869 |
+
"eval_steps_per_second": 0.876,
|
| 870 |
+
"step": 2000
|
| 871 |
+
},
|
| 872 |
+
{
|
| 873 |
+
"epoch": 0.06808223795079205,
|
| 874 |
+
"grad_norm": 1.2734375,
|
| 875 |
+
"learning_rate": 0.0003,
|
| 876 |
+
"loss": 3.147844696044922,
|
| 877 |
+
"step": 2020
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 0.06875631951466127,
|
| 881 |
+
"grad_norm": 1.1796875,
|
| 882 |
+
"learning_rate": 0.0003,
|
| 883 |
+
"loss": 3.149020957946777,
|
| 884 |
+
"step": 2040
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 0.0694304010785305,
|
| 888 |
+
"grad_norm": 1.234375,
|
| 889 |
+
"learning_rate": 0.0003,
|
| 890 |
+
"loss": 3.1382816314697264,
|
| 891 |
+
"step": 2060
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 0.07010448264239973,
|
| 895 |
+
"grad_norm": 1.203125,
|
| 896 |
+
"learning_rate": 0.0003,
|
| 897 |
+
"loss": 3.1440708160400392,
|
| 898 |
+
"step": 2080
|
| 899 |
+
},
|
| 900 |
+
{
|
| 901 |
+
"epoch": 0.07077856420626896,
|
| 902 |
+
"grad_norm": 1.1171875,
|
| 903 |
+
"learning_rate": 0.0003,
|
| 904 |
+
"loss": 3.132351875305176,
|
| 905 |
+
"step": 2100
|
| 906 |
+
},
|
| 907 |
+
{
|
| 908 |
+
"epoch": 0.07077856420626896,
|
| 909 |
+
"eval_loss": 3.1331419944763184,
|
| 910 |
+
"eval_runtime": 8.2456,
|
| 911 |
+
"eval_samples_per_second": 1155.405,
|
| 912 |
+
"eval_steps_per_second": 0.849,
|
| 913 |
+
"step": 2100
|
| 914 |
+
},
|
| 915 |
+
{
|
| 916 |
+
"epoch": 0.07145264577013818,
|
| 917 |
+
"grad_norm": 1.109375,
|
| 918 |
+
"learning_rate": 0.0003,
|
| 919 |
+
"loss": 3.1341459274291994,
|
| 920 |
+
"step": 2120
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 0.07212672733400742,
|
| 924 |
+
"grad_norm": 1.1640625,
|
| 925 |
+
"learning_rate": 0.0003,
|
| 926 |
+
"loss": 3.1331857681274413,
|
| 927 |
+
"step": 2140
|
| 928 |
+
},
|
| 929 |
+
{
|
| 930 |
+
"epoch": 0.07280080889787664,
|
| 931 |
+
"grad_norm": 1.1796875,
|
| 932 |
+
"learning_rate": 0.0003,
|
| 933 |
+
"loss": 3.1179393768310546,
|
| 934 |
+
"step": 2160
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 0.07347489046174586,
|
| 938 |
+
"grad_norm": 1.546875,
|
| 939 |
+
"learning_rate": 0.0003,
|
| 940 |
+
"loss": 3.1271270751953124,
|
| 941 |
+
"step": 2180
|
| 942 |
+
},
|
| 943 |
+
{
|
| 944 |
+
"epoch": 0.0741489720256151,
|
| 945 |
+
"grad_norm": 1.234375,
|
| 946 |
+
"learning_rate": 0.0003,
|
| 947 |
+
"loss": 3.1252058029174803,
|
| 948 |
+
"step": 2200
|
| 949 |
+
},
|
| 950 |
+
{
|
| 951 |
+
"epoch": 0.0741489720256151,
|
| 952 |
+
"eval_loss": 3.1193113327026367,
|
| 953 |
+
"eval_runtime": 8.0852,
|
| 954 |
+
"eval_samples_per_second": 1178.326,
|
| 955 |
+
"eval_steps_per_second": 0.866,
|
| 956 |
+
"step": 2200
|
| 957 |
+
}
|
| 958 |
+
],
|
| 959 |
+
"logging_steps": 20,
|
| 960 |
+
"max_steps": 2500,
|
| 961 |
+
"num_input_tokens_seen": 0,
|
| 962 |
+
"num_train_epochs": 1,
|
| 963 |
+
"save_steps": 100,
|
| 964 |
+
"stateful_callbacks": {
|
| 965 |
+
"TrainerControl": {
|
| 966 |
+
"args": {
|
| 967 |
+
"should_epoch_stop": false,
|
| 968 |
+
"should_evaluate": false,
|
| 969 |
+
"should_log": false,
|
| 970 |
+
"should_save": true,
|
| 971 |
+
"should_training_stop": false
|
| 972 |
+
},
|
| 973 |
+
"attributes": {}
|
| 974 |
+
}
|
| 975 |
+
},
|
| 976 |
+
"total_flos": 319427287449600.0,
|
| 977 |
+
"train_batch_size": 32,
|
| 978 |
+
"trial_name": null,
|
| 979 |
+
"trial_params": null
|
| 980 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2200/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a761b578cf2b8c248f28bd1b61a5ec7d0b5d3801c8bd793a6db706df1d3078f6
|
| 3 |
+
size 4920
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/config.json
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "squared_relu",
|
| 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": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 9,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"powlu_m": 3.0,
|
| 25 |
+
"pretraining_tp": 1,
|
| 26 |
+
"rms_norm_eps": 1e-06,
|
| 27 |
+
"rope_parameters": {
|
| 28 |
+
"rope_theta": 10000.0,
|
| 29 |
+
"rope_type": "default"
|
| 30 |
+
},
|
| 31 |
+
"tie_word_embeddings": true,
|
| 32 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 33 |
+
"transformers_version": "5.16.0.dev0",
|
| 34 |
+
"use_cache": false,
|
| 35 |
+
"vocab_size": 4096,
|
| 36 |
+
"waleed_beta": 10.0
|
| 37 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e53ea93ef74931e3a08ab9b4c02109ce2ff0565543222e9bde1f1b90d0e12288
|
| 3 |
+
size 4010544
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:56dcf09f0b8195b3b51196279f3ec72e9e7e3fe5a008ca3bc97a6c221907a668
|
| 3 |
+
size 8068282
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ed84324664a63913fe5c29a818cb67e7aa68822c0b51f712812da2a667aa578c
|
| 3 |
+
size 14244
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c15f7cedf7769027fdf6ebd2e8f7385a6ae528697209f7b99f6e4666af50bf85
|
| 3 |
+
size 1064
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/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/mlp-squared_relu-9L_run/checkpoint-2300/trainer_state.json
ADDED
|
@@ -0,0 +1,1023 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.07751937984496124,
|
| 6 |
+
"eval_steps": 100,
|
| 7 |
+
"global_step": 2300,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.0006740815638692282,
|
| 14 |
+
"grad_norm": 2.71875,
|
| 15 |
+
"learning_rate": 1.14e-05,
|
| 16 |
+
"loss": 8.314944458007812,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.0013481631277384564,
|
| 21 |
+
"grad_norm": 1.8515625,
|
| 22 |
+
"learning_rate": 2.34e-05,
|
| 23 |
+
"loss": 8.251383972167968,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.0020222446916076846,
|
| 28 |
+
"grad_norm": 1.3203125,
|
| 29 |
+
"learning_rate": 3.539999999999999e-05,
|
| 30 |
+
"loss": 8.104505920410157,
|
| 31 |
+
"step": 60
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"epoch": 0.002696326255476913,
|
| 35 |
+
"grad_norm": 1.296875,
|
| 36 |
+
"learning_rate": 4.7399999999999993e-05,
|
| 37 |
+
"loss": 7.968361663818359,
|
| 38 |
+
"step": 80
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.003370407819346141,
|
| 42 |
+
"grad_norm": 1.2578125,
|
| 43 |
+
"learning_rate": 5.94e-05,
|
| 44 |
+
"loss": 7.804827880859375,
|
| 45 |
+
"step": 100
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.003370407819346141,
|
| 49 |
+
"eval_loss": 7.703802585601807,
|
| 50 |
+
"eval_runtime": 8.0325,
|
| 51 |
+
"eval_samples_per_second": 1186.059,
|
| 52 |
+
"eval_steps_per_second": 0.871,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.004044489383215369,
|
| 57 |
+
"grad_norm": 1.2734375,
|
| 58 |
+
"learning_rate": 7.139999999999999e-05,
|
| 59 |
+
"loss": 7.600141906738282,
|
| 60 |
+
"step": 120
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"epoch": 0.0047185709470845974,
|
| 64 |
+
"grad_norm": 1.265625,
|
| 65 |
+
"learning_rate": 8.34e-05,
|
| 66 |
+
"loss": 7.368203735351562,
|
| 67 |
+
"step": 140
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"epoch": 0.005392652510953826,
|
| 71 |
+
"grad_norm": 1.265625,
|
| 72 |
+
"learning_rate": 9.539999999999999e-05,
|
| 73 |
+
"loss": 7.133045196533203,
|
| 74 |
+
"step": 160
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.006066734074823054,
|
| 78 |
+
"grad_norm": 1.203125,
|
| 79 |
+
"learning_rate": 0.00010739999999999998,
|
| 80 |
+
"loss": 6.885277557373047,
|
| 81 |
+
"step": 180
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"epoch": 0.006740815638692282,
|
| 85 |
+
"grad_norm": 1.15625,
|
| 86 |
+
"learning_rate": 0.0001194,
|
| 87 |
+
"loss": 6.629558563232422,
|
| 88 |
+
"step": 200
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"epoch": 0.006740815638692282,
|
| 92 |
+
"eval_loss": 6.4997687339782715,
|
| 93 |
+
"eval_runtime": 7.9479,
|
| 94 |
+
"eval_samples_per_second": 1198.683,
|
| 95 |
+
"eval_steps_per_second": 0.881,
|
| 96 |
+
"step": 200
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"epoch": 0.00741489720256151,
|
| 100 |
+
"grad_norm": 1.109375,
|
| 101 |
+
"learning_rate": 0.0001314,
|
| 102 |
+
"loss": 6.391195297241211,
|
| 103 |
+
"step": 220
|
| 104 |
+
},
|
| 105 |
+
{
|
| 106 |
+
"epoch": 0.008088978766430738,
|
| 107 |
+
"grad_norm": 1.21875,
|
| 108 |
+
"learning_rate": 0.0001434,
|
| 109 |
+
"loss": 6.139227676391601,
|
| 110 |
+
"step": 240
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"epoch": 0.008763060330299966,
|
| 114 |
+
"grad_norm": 1.0625,
|
| 115 |
+
"learning_rate": 0.00015539999999999998,
|
| 116 |
+
"loss": 5.917041778564453,
|
| 117 |
+
"step": 260
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"epoch": 0.009437141894169195,
|
| 121 |
+
"grad_norm": 1.1015625,
|
| 122 |
+
"learning_rate": 0.0001674,
|
| 123 |
+
"loss": 5.702054595947265,
|
| 124 |
+
"step": 280
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"epoch": 0.010111223458038422,
|
| 128 |
+
"grad_norm": 0.921875,
|
| 129 |
+
"learning_rate": 0.00017939999999999997,
|
| 130 |
+
"loss": 5.500263214111328,
|
| 131 |
+
"step": 300
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"epoch": 0.010111223458038422,
|
| 135 |
+
"eval_loss": 5.393505573272705,
|
| 136 |
+
"eval_runtime": 8.0449,
|
| 137 |
+
"eval_samples_per_second": 1184.223,
|
| 138 |
+
"eval_steps_per_second": 0.87,
|
| 139 |
+
"step": 300
|
| 140 |
+
},
|
| 141 |
+
{
|
| 142 |
+
"epoch": 0.010785305021907651,
|
| 143 |
+
"grad_norm": 2.5,
|
| 144 |
+
"learning_rate": 0.0001914,
|
| 145 |
+
"loss": 5.319306182861328,
|
| 146 |
+
"step": 320
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.011459386585776879,
|
| 150 |
+
"grad_norm": 1.078125,
|
| 151 |
+
"learning_rate": 0.00020339999999999998,
|
| 152 |
+
"loss": 5.202053070068359,
|
| 153 |
+
"step": 340
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"epoch": 0.012133468149646108,
|
| 157 |
+
"grad_norm": 1.3203125,
|
| 158 |
+
"learning_rate": 0.00021539999999999998,
|
| 159 |
+
"loss": 5.020883941650391,
|
| 160 |
+
"step": 360
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 0.012807549713515335,
|
| 164 |
+
"grad_norm": 1.1171875,
|
| 165 |
+
"learning_rate": 0.00022739999999999997,
|
| 166 |
+
"loss": 4.873059844970703,
|
| 167 |
+
"step": 380
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"epoch": 0.013481631277384564,
|
| 171 |
+
"grad_norm": 1.234375,
|
| 172 |
+
"learning_rate": 0.0002394,
|
| 173 |
+
"loss": 4.741780853271484,
|
| 174 |
+
"step": 400
|
| 175 |
+
},
|
| 176 |
+
{
|
| 177 |
+
"epoch": 0.013481631277384564,
|
| 178 |
+
"eval_loss": 4.676144123077393,
|
| 179 |
+
"eval_runtime": 8.0457,
|
| 180 |
+
"eval_samples_per_second": 1184.113,
|
| 181 |
+
"eval_steps_per_second": 0.87,
|
| 182 |
+
"step": 400
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"epoch": 0.014155712841253791,
|
| 186 |
+
"grad_norm": 1.3046875,
|
| 187 |
+
"learning_rate": 0.0002514,
|
| 188 |
+
"loss": 4.606548690795899,
|
| 189 |
+
"step": 420
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 0.01482979440512302,
|
| 193 |
+
"grad_norm": 1.2578125,
|
| 194 |
+
"learning_rate": 0.00026339999999999995,
|
| 195 |
+
"loss": 4.515376281738281,
|
| 196 |
+
"step": 440
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"epoch": 0.015503875968992248,
|
| 200 |
+
"grad_norm": 1.0625,
|
| 201 |
+
"learning_rate": 0.00027539999999999997,
|
| 202 |
+
"loss": 4.416887664794922,
|
| 203 |
+
"step": 460
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"epoch": 0.016177957532861477,
|
| 207 |
+
"grad_norm": 1.3671875,
|
| 208 |
+
"learning_rate": 0.00028739999999999994,
|
| 209 |
+
"loss": 4.3046520233154295,
|
| 210 |
+
"step": 480
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.016852039096730706,
|
| 214 |
+
"grad_norm": 1.2109375,
|
| 215 |
+
"learning_rate": 0.00029939999999999996,
|
| 216 |
+
"loss": 4.238706970214844,
|
| 217 |
+
"step": 500
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"epoch": 0.016852039096730706,
|
| 221 |
+
"eval_loss": 4.196270942687988,
|
| 222 |
+
"eval_runtime": 7.9301,
|
| 223 |
+
"eval_samples_per_second": 1201.377,
|
| 224 |
+
"eval_steps_per_second": 0.883,
|
| 225 |
+
"step": 500
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"epoch": 0.01752612066059993,
|
| 229 |
+
"grad_norm": 0.97265625,
|
| 230 |
+
"learning_rate": 0.0003,
|
| 231 |
+
"loss": 4.1756237030029295,
|
| 232 |
+
"step": 520
|
| 233 |
+
},
|
| 234 |
+
{
|
| 235 |
+
"epoch": 0.01820020222446916,
|
| 236 |
+
"grad_norm": 0.9140625,
|
| 237 |
+
"learning_rate": 0.0003,
|
| 238 |
+
"loss": 4.0915069580078125,
|
| 239 |
+
"step": 540
|
| 240 |
+
},
|
| 241 |
+
{
|
| 242 |
+
"epoch": 0.01887428378833839,
|
| 243 |
+
"grad_norm": 1.2890625,
|
| 244 |
+
"learning_rate": 0.0003,
|
| 245 |
+
"loss": 4.031630325317383,
|
| 246 |
+
"step": 560
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"epoch": 0.01954836535220762,
|
| 250 |
+
"grad_norm": 1.125,
|
| 251 |
+
"learning_rate": 0.0003,
|
| 252 |
+
"loss": 3.9763336181640625,
|
| 253 |
+
"step": 580
|
| 254 |
+
},
|
| 255 |
+
{
|
| 256 |
+
"epoch": 0.020222446916076844,
|
| 257 |
+
"grad_norm": 1.21875,
|
| 258 |
+
"learning_rate": 0.0003,
|
| 259 |
+
"loss": 3.9347896575927734,
|
| 260 |
+
"step": 600
|
| 261 |
+
},
|
| 262 |
+
{
|
| 263 |
+
"epoch": 0.020222446916076844,
|
| 264 |
+
"eval_loss": 3.9157378673553467,
|
| 265 |
+
"eval_runtime": 8.0606,
|
| 266 |
+
"eval_samples_per_second": 1181.917,
|
| 267 |
+
"eval_steps_per_second": 0.868,
|
| 268 |
+
"step": 600
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.020896528479946073,
|
| 272 |
+
"grad_norm": 1.1640625,
|
| 273 |
+
"learning_rate": 0.0003,
|
| 274 |
+
"loss": 3.878151702880859,
|
| 275 |
+
"step": 620
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.021570610043815303,
|
| 279 |
+
"grad_norm": 1.484375,
|
| 280 |
+
"learning_rate": 0.0003,
|
| 281 |
+
"loss": 3.855076217651367,
|
| 282 |
+
"step": 640
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.022244691607684528,
|
| 286 |
+
"grad_norm": 1.3359375,
|
| 287 |
+
"learning_rate": 0.0003,
|
| 288 |
+
"loss": 3.8232196807861327,
|
| 289 |
+
"step": 660
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.022918773171553757,
|
| 293 |
+
"grad_norm": 1.1640625,
|
| 294 |
+
"learning_rate": 0.0003,
|
| 295 |
+
"loss": 3.782532501220703,
|
| 296 |
+
"step": 680
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.023592854735422986,
|
| 300 |
+
"grad_norm": 1.1328125,
|
| 301 |
+
"learning_rate": 0.0003,
|
| 302 |
+
"loss": 3.7758609771728517,
|
| 303 |
+
"step": 700
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.023592854735422986,
|
| 307 |
+
"eval_loss": 3.7596347332000732,
|
| 308 |
+
"eval_runtime": 7.9743,
|
| 309 |
+
"eval_samples_per_second": 1194.709,
|
| 310 |
+
"eval_steps_per_second": 0.878,
|
| 311 |
+
"step": 700
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"epoch": 0.024266936299292215,
|
| 315 |
+
"grad_norm": 1.25,
|
| 316 |
+
"learning_rate": 0.0003,
|
| 317 |
+
"loss": 3.7433414459228516,
|
| 318 |
+
"step": 720
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 0.02494101786316144,
|
| 322 |
+
"grad_norm": 1.3125,
|
| 323 |
+
"learning_rate": 0.0003,
|
| 324 |
+
"loss": 3.732330322265625,
|
| 325 |
+
"step": 740
|
| 326 |
+
},
|
| 327 |
+
{
|
| 328 |
+
"epoch": 0.02561509942703067,
|
| 329 |
+
"grad_norm": 1.265625,
|
| 330 |
+
"learning_rate": 0.0003,
|
| 331 |
+
"loss": 3.6843395233154297,
|
| 332 |
+
"step": 760
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"epoch": 0.0262891809908999,
|
| 336 |
+
"grad_norm": 1.1640625,
|
| 337 |
+
"learning_rate": 0.0003,
|
| 338 |
+
"loss": 3.6463153839111326,
|
| 339 |
+
"step": 780
|
| 340 |
+
},
|
| 341 |
+
{
|
| 342 |
+
"epoch": 0.026963262554769128,
|
| 343 |
+
"grad_norm": 1.1171875,
|
| 344 |
+
"learning_rate": 0.0003,
|
| 345 |
+
"loss": 3.6544403076171874,
|
| 346 |
+
"step": 800
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.026963262554769128,
|
| 350 |
+
"eval_loss": 3.652294635772705,
|
| 351 |
+
"eval_runtime": 7.9753,
|
| 352 |
+
"eval_samples_per_second": 1194.559,
|
| 353 |
+
"eval_steps_per_second": 0.878,
|
| 354 |
+
"step": 800
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.027637344118638354,
|
| 358 |
+
"grad_norm": 1.1171875,
|
| 359 |
+
"learning_rate": 0.0003,
|
| 360 |
+
"loss": 3.6159641265869142,
|
| 361 |
+
"step": 820
|
| 362 |
+
},
|
| 363 |
+
{
|
| 364 |
+
"epoch": 0.028311425682507583,
|
| 365 |
+
"grad_norm": 1.078125,
|
| 366 |
+
"learning_rate": 0.0003,
|
| 367 |
+
"loss": 3.6099781036376952,
|
| 368 |
+
"step": 840
|
| 369 |
+
},
|
| 370 |
+
{
|
| 371 |
+
"epoch": 0.028985507246376812,
|
| 372 |
+
"grad_norm": 1.2734375,
|
| 373 |
+
"learning_rate": 0.0003,
|
| 374 |
+
"loss": 3.6344444274902346,
|
| 375 |
+
"step": 860
|
| 376 |
+
},
|
| 377 |
+
{
|
| 378 |
+
"epoch": 0.02965958881024604,
|
| 379 |
+
"grad_norm": 1.140625,
|
| 380 |
+
"learning_rate": 0.0003,
|
| 381 |
+
"loss": 3.5855178833007812,
|
| 382 |
+
"step": 880
|
| 383 |
+
},
|
| 384 |
+
{
|
| 385 |
+
"epoch": 0.030333670374115267,
|
| 386 |
+
"grad_norm": 1.109375,
|
| 387 |
+
"learning_rate": 0.0003,
|
| 388 |
+
"loss": 3.5488441467285154,
|
| 389 |
+
"step": 900
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.030333670374115267,
|
| 393 |
+
"eval_loss": 3.5648436546325684,
|
| 394 |
+
"eval_runtime": 8.321,
|
| 395 |
+
"eval_samples_per_second": 1144.931,
|
| 396 |
+
"eval_steps_per_second": 0.841,
|
| 397 |
+
"step": 900
|
| 398 |
+
},
|
| 399 |
+
{
|
| 400 |
+
"epoch": 0.031007751937984496,
|
| 401 |
+
"grad_norm": 1.1015625,
|
| 402 |
+
"learning_rate": 0.0003,
|
| 403 |
+
"loss": 3.5646598815917967,
|
| 404 |
+
"step": 920
|
| 405 |
+
},
|
| 406 |
+
{
|
| 407 |
+
"epoch": 0.031681833501853725,
|
| 408 |
+
"grad_norm": 1.25,
|
| 409 |
+
"learning_rate": 0.0003,
|
| 410 |
+
"loss": 3.5480445861816405,
|
| 411 |
+
"step": 940
|
| 412 |
+
},
|
| 413 |
+
{
|
| 414 |
+
"epoch": 0.032355915065722954,
|
| 415 |
+
"grad_norm": 1.140625,
|
| 416 |
+
"learning_rate": 0.0003,
|
| 417 |
+
"loss": 3.5140933990478516,
|
| 418 |
+
"step": 960
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.03302999662959218,
|
| 422 |
+
"grad_norm": 1.140625,
|
| 423 |
+
"learning_rate": 0.0003,
|
| 424 |
+
"loss": 3.501856231689453,
|
| 425 |
+
"step": 980
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.03370407819346141,
|
| 429 |
+
"grad_norm": 1.359375,
|
| 430 |
+
"learning_rate": 0.0003,
|
| 431 |
+
"loss": 3.500592803955078,
|
| 432 |
+
"step": 1000
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.03370407819346141,
|
| 436 |
+
"eval_loss": 3.4981162548065186,
|
| 437 |
+
"eval_runtime": 8.1013,
|
| 438 |
+
"eval_samples_per_second": 1175.987,
|
| 439 |
+
"eval_steps_per_second": 0.864,
|
| 440 |
+
"step": 1000
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.034378159757330634,
|
| 444 |
+
"grad_norm": 1.1328125,
|
| 445 |
+
"learning_rate": 0.0003,
|
| 446 |
+
"loss": 3.4914066314697267,
|
| 447 |
+
"step": 1020
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.03505224132119986,
|
| 451 |
+
"grad_norm": 1.1171875,
|
| 452 |
+
"learning_rate": 0.0003,
|
| 453 |
+
"loss": 3.474192810058594,
|
| 454 |
+
"step": 1040
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.03572632288506909,
|
| 458 |
+
"grad_norm": 1.28125,
|
| 459 |
+
"learning_rate": 0.0003,
|
| 460 |
+
"loss": 3.4552947998046877,
|
| 461 |
+
"step": 1060
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.03640040444893832,
|
| 465 |
+
"grad_norm": 1.203125,
|
| 466 |
+
"learning_rate": 0.0003,
|
| 467 |
+
"loss": 3.4411731719970704,
|
| 468 |
+
"step": 1080
|
| 469 |
+
},
|
| 470 |
+
{
|
| 471 |
+
"epoch": 0.03707448601280755,
|
| 472 |
+
"grad_norm": 1.2734375,
|
| 473 |
+
"learning_rate": 0.0003,
|
| 474 |
+
"loss": 3.435680389404297,
|
| 475 |
+
"step": 1100
|
| 476 |
+
},
|
| 477 |
+
{
|
| 478 |
+
"epoch": 0.03707448601280755,
|
| 479 |
+
"eval_loss": 3.445117712020874,
|
| 480 |
+
"eval_runtime": 8.1641,
|
| 481 |
+
"eval_samples_per_second": 1166.938,
|
| 482 |
+
"eval_steps_per_second": 0.857,
|
| 483 |
+
"step": 1100
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.03774856757667678,
|
| 487 |
+
"grad_norm": 1.03125,
|
| 488 |
+
"learning_rate": 0.0003,
|
| 489 |
+
"loss": 3.4368762969970703,
|
| 490 |
+
"step": 1120
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.03842264914054601,
|
| 494 |
+
"grad_norm": 1.2421875,
|
| 495 |
+
"learning_rate": 0.0003,
|
| 496 |
+
"loss": 3.410645294189453,
|
| 497 |
+
"step": 1140
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"epoch": 0.03909673070441524,
|
| 501 |
+
"grad_norm": 1.0703125,
|
| 502 |
+
"learning_rate": 0.0003,
|
| 503 |
+
"loss": 3.4069515228271485,
|
| 504 |
+
"step": 1160
|
| 505 |
+
},
|
| 506 |
+
{
|
| 507 |
+
"epoch": 0.03977081226828446,
|
| 508 |
+
"grad_norm": 1.03125,
|
| 509 |
+
"learning_rate": 0.0003,
|
| 510 |
+
"loss": 3.4227649688720705,
|
| 511 |
+
"step": 1180
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"epoch": 0.04044489383215369,
|
| 515 |
+
"grad_norm": 1.203125,
|
| 516 |
+
"learning_rate": 0.0003,
|
| 517 |
+
"loss": 3.3914558410644533,
|
| 518 |
+
"step": 1200
|
| 519 |
+
},
|
| 520 |
+
{
|
| 521 |
+
"epoch": 0.04044489383215369,
|
| 522 |
+
"eval_loss": 3.3945467472076416,
|
| 523 |
+
"eval_runtime": 8.0433,
|
| 524 |
+
"eval_samples_per_second": 1184.469,
|
| 525 |
+
"eval_steps_per_second": 0.87,
|
| 526 |
+
"step": 1200
|
| 527 |
+
},
|
| 528 |
+
{
|
| 529 |
+
"epoch": 0.04111897539602292,
|
| 530 |
+
"grad_norm": 1.0546875,
|
| 531 |
+
"learning_rate": 0.0003,
|
| 532 |
+
"loss": 3.3614086151123046,
|
| 533 |
+
"step": 1220
|
| 534 |
+
},
|
| 535 |
+
{
|
| 536 |
+
"epoch": 0.04179305695989215,
|
| 537 |
+
"grad_norm": 1.09375,
|
| 538 |
+
"learning_rate": 0.0003,
|
| 539 |
+
"loss": 3.3790237426757814,
|
| 540 |
+
"step": 1240
|
| 541 |
+
},
|
| 542 |
+
{
|
| 543 |
+
"epoch": 0.042467138523761376,
|
| 544 |
+
"grad_norm": 1.3515625,
|
| 545 |
+
"learning_rate": 0.0003,
|
| 546 |
+
"loss": 3.38794059753418,
|
| 547 |
+
"step": 1260
|
| 548 |
+
},
|
| 549 |
+
{
|
| 550 |
+
"epoch": 0.043141220087630605,
|
| 551 |
+
"grad_norm": 1.28125,
|
| 552 |
+
"learning_rate": 0.0003,
|
| 553 |
+
"loss": 3.3654109954833986,
|
| 554 |
+
"step": 1280
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.043815301651499834,
|
| 558 |
+
"grad_norm": 1.1796875,
|
| 559 |
+
"learning_rate": 0.0003,
|
| 560 |
+
"loss": 3.3572288513183595,
|
| 561 |
+
"step": 1300
|
| 562 |
+
},
|
| 563 |
+
{
|
| 564 |
+
"epoch": 0.043815301651499834,
|
| 565 |
+
"eval_loss": 3.3501007556915283,
|
| 566 |
+
"eval_runtime": 8.0621,
|
| 567 |
+
"eval_samples_per_second": 1181.702,
|
| 568 |
+
"eval_steps_per_second": 0.868,
|
| 569 |
+
"step": 1300
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 0.044489383215369056,
|
| 573 |
+
"grad_norm": 1.2734375,
|
| 574 |
+
"learning_rate": 0.0003,
|
| 575 |
+
"loss": 3.321734619140625,
|
| 576 |
+
"step": 1320
|
| 577 |
+
},
|
| 578 |
+
{
|
| 579 |
+
"epoch": 0.045163464779238285,
|
| 580 |
+
"grad_norm": 1.2890625,
|
| 581 |
+
"learning_rate": 0.0003,
|
| 582 |
+
"loss": 3.3493595123291016,
|
| 583 |
+
"step": 1340
|
| 584 |
+
},
|
| 585 |
+
{
|
| 586 |
+
"epoch": 0.045837546343107514,
|
| 587 |
+
"grad_norm": 1.1875,
|
| 588 |
+
"learning_rate": 0.0003,
|
| 589 |
+
"loss": 3.3272212982177733,
|
| 590 |
+
"step": 1360
|
| 591 |
+
},
|
| 592 |
+
{
|
| 593 |
+
"epoch": 0.046511627906976744,
|
| 594 |
+
"grad_norm": 1.1328125,
|
| 595 |
+
"learning_rate": 0.0003,
|
| 596 |
+
"loss": 3.3202045440673826,
|
| 597 |
+
"step": 1380
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 0.04718570947084597,
|
| 601 |
+
"grad_norm": 1.1171875,
|
| 602 |
+
"learning_rate": 0.0003,
|
| 603 |
+
"loss": 3.2998809814453125,
|
| 604 |
+
"step": 1400
|
| 605 |
+
},
|
| 606 |
+
{
|
| 607 |
+
"epoch": 0.04718570947084597,
|
| 608 |
+
"eval_loss": 3.311476707458496,
|
| 609 |
+
"eval_runtime": 8.1361,
|
| 610 |
+
"eval_samples_per_second": 1170.957,
|
| 611 |
+
"eval_steps_per_second": 0.86,
|
| 612 |
+
"step": 1400
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.0478597910347152,
|
| 616 |
+
"grad_norm": 1.140625,
|
| 617 |
+
"learning_rate": 0.0003,
|
| 618 |
+
"loss": 3.3057735443115233,
|
| 619 |
+
"step": 1420
|
| 620 |
+
},
|
| 621 |
+
{
|
| 622 |
+
"epoch": 0.04853387259858443,
|
| 623 |
+
"grad_norm": 1.109375,
|
| 624 |
+
"learning_rate": 0.0003,
|
| 625 |
+
"loss": 3.2907379150390623,
|
| 626 |
+
"step": 1440
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.04920795416245366,
|
| 630 |
+
"grad_norm": 1.078125,
|
| 631 |
+
"learning_rate": 0.0003,
|
| 632 |
+
"loss": 3.264795684814453,
|
| 633 |
+
"step": 1460
|
| 634 |
+
},
|
| 635 |
+
{
|
| 636 |
+
"epoch": 0.04988203572632288,
|
| 637 |
+
"grad_norm": 1.3671875,
|
| 638 |
+
"learning_rate": 0.0003,
|
| 639 |
+
"loss": 3.2918243408203125,
|
| 640 |
+
"step": 1480
|
| 641 |
+
},
|
| 642 |
+
{
|
| 643 |
+
"epoch": 0.05055611729019211,
|
| 644 |
+
"grad_norm": 1.2421875,
|
| 645 |
+
"learning_rate": 0.0003,
|
| 646 |
+
"loss": 3.2844181060791016,
|
| 647 |
+
"step": 1500
|
| 648 |
+
},
|
| 649 |
+
{
|
| 650 |
+
"epoch": 0.05055611729019211,
|
| 651 |
+
"eval_loss": 3.277984142303467,
|
| 652 |
+
"eval_runtime": 8.0313,
|
| 653 |
+
"eval_samples_per_second": 1186.24,
|
| 654 |
+
"eval_steps_per_second": 0.872,
|
| 655 |
+
"step": 1500
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 0.05123019885406134,
|
| 659 |
+
"grad_norm": 1.265625,
|
| 660 |
+
"learning_rate": 0.0003,
|
| 661 |
+
"loss": 3.2729434967041016,
|
| 662 |
+
"step": 1520
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 0.05190428041793057,
|
| 666 |
+
"grad_norm": 1.1015625,
|
| 667 |
+
"learning_rate": 0.0003,
|
| 668 |
+
"loss": 3.279766845703125,
|
| 669 |
+
"step": 1540
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 0.0525783619817998,
|
| 673 |
+
"grad_norm": 1.25,
|
| 674 |
+
"learning_rate": 0.0003,
|
| 675 |
+
"loss": 3.281978988647461,
|
| 676 |
+
"step": 1560
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 0.05325244354566903,
|
| 680 |
+
"grad_norm": 1.125,
|
| 681 |
+
"learning_rate": 0.0003,
|
| 682 |
+
"loss": 3.2608470916748047,
|
| 683 |
+
"step": 1580
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 0.053926525109538256,
|
| 687 |
+
"grad_norm": 1.2265625,
|
| 688 |
+
"learning_rate": 0.0003,
|
| 689 |
+
"loss": 3.2503360748291015,
|
| 690 |
+
"step": 1600
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.053926525109538256,
|
| 694 |
+
"eval_loss": 3.2511110305786133,
|
| 695 |
+
"eval_runtime": 8.097,
|
| 696 |
+
"eval_samples_per_second": 1176.602,
|
| 697 |
+
"eval_steps_per_second": 0.865,
|
| 698 |
+
"step": 1600
|
| 699 |
+
},
|
| 700 |
+
{
|
| 701 |
+
"epoch": 0.054600606673407485,
|
| 702 |
+
"grad_norm": 1.234375,
|
| 703 |
+
"learning_rate": 0.0003,
|
| 704 |
+
"loss": 3.2713451385498047,
|
| 705 |
+
"step": 1620
|
| 706 |
+
},
|
| 707 |
+
{
|
| 708 |
+
"epoch": 0.05527468823727671,
|
| 709 |
+
"grad_norm": 1.015625,
|
| 710 |
+
"learning_rate": 0.0003,
|
| 711 |
+
"loss": 3.2468338012695312,
|
| 712 |
+
"step": 1640
|
| 713 |
+
},
|
| 714 |
+
{
|
| 715 |
+
"epoch": 0.05594876980114594,
|
| 716 |
+
"grad_norm": 1.15625,
|
| 717 |
+
"learning_rate": 0.0003,
|
| 718 |
+
"loss": 3.255104827880859,
|
| 719 |
+
"step": 1660
|
| 720 |
+
},
|
| 721 |
+
{
|
| 722 |
+
"epoch": 0.056622851365015166,
|
| 723 |
+
"grad_norm": 1.1875,
|
| 724 |
+
"learning_rate": 0.0003,
|
| 725 |
+
"loss": 3.2280048370361327,
|
| 726 |
+
"step": 1680
|
| 727 |
+
},
|
| 728 |
+
{
|
| 729 |
+
"epoch": 0.057296932928884395,
|
| 730 |
+
"grad_norm": 1.265625,
|
| 731 |
+
"learning_rate": 0.0003,
|
| 732 |
+
"loss": 3.216978073120117,
|
| 733 |
+
"step": 1700
|
| 734 |
+
},
|
| 735 |
+
{
|
| 736 |
+
"epoch": 0.057296932928884395,
|
| 737 |
+
"eval_loss": 3.2238714694976807,
|
| 738 |
+
"eval_runtime": 8.059,
|
| 739 |
+
"eval_samples_per_second": 1182.16,
|
| 740 |
+
"eval_steps_per_second": 0.869,
|
| 741 |
+
"step": 1700
|
| 742 |
+
},
|
| 743 |
+
{
|
| 744 |
+
"epoch": 0.057971014492753624,
|
| 745 |
+
"grad_norm": 1.125,
|
| 746 |
+
"learning_rate": 0.0003,
|
| 747 |
+
"loss": 3.213521194458008,
|
| 748 |
+
"step": 1720
|
| 749 |
+
},
|
| 750 |
+
{
|
| 751 |
+
"epoch": 0.05864509605662285,
|
| 752 |
+
"grad_norm": 1.234375,
|
| 753 |
+
"learning_rate": 0.0003,
|
| 754 |
+
"loss": 3.2087985992431642,
|
| 755 |
+
"step": 1740
|
| 756 |
+
},
|
| 757 |
+
{
|
| 758 |
+
"epoch": 0.05931917762049208,
|
| 759 |
+
"grad_norm": 1.2109375,
|
| 760 |
+
"learning_rate": 0.0003,
|
| 761 |
+
"loss": 3.198755645751953,
|
| 762 |
+
"step": 1760
|
| 763 |
+
},
|
| 764 |
+
{
|
| 765 |
+
"epoch": 0.05999325918436131,
|
| 766 |
+
"grad_norm": 1.4453125,
|
| 767 |
+
"learning_rate": 0.0003,
|
| 768 |
+
"loss": 3.2067222595214844,
|
| 769 |
+
"step": 1780
|
| 770 |
+
},
|
| 771 |
+
{
|
| 772 |
+
"epoch": 0.06066734074823053,
|
| 773 |
+
"grad_norm": 1.140625,
|
| 774 |
+
"learning_rate": 0.0003,
|
| 775 |
+
"loss": 3.195578956604004,
|
| 776 |
+
"step": 1800
|
| 777 |
+
},
|
| 778 |
+
{
|
| 779 |
+
"epoch": 0.06066734074823053,
|
| 780 |
+
"eval_loss": 3.1985833644866943,
|
| 781 |
+
"eval_runtime": 8.2334,
|
| 782 |
+
"eval_samples_per_second": 1157.109,
|
| 783 |
+
"eval_steps_per_second": 0.85,
|
| 784 |
+
"step": 1800
|
| 785 |
+
},
|
| 786 |
+
{
|
| 787 |
+
"epoch": 0.06134142231209976,
|
| 788 |
+
"grad_norm": 1.1640625,
|
| 789 |
+
"learning_rate": 0.0003,
|
| 790 |
+
"loss": 3.186159133911133,
|
| 791 |
+
"step": 1820
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"epoch": 0.06201550387596899,
|
| 795 |
+
"grad_norm": 1.21875,
|
| 796 |
+
"learning_rate": 0.0003,
|
| 797 |
+
"loss": 3.1850618362426757,
|
| 798 |
+
"step": 1840
|
| 799 |
+
},
|
| 800 |
+
{
|
| 801 |
+
"epoch": 0.06268958543983821,
|
| 802 |
+
"grad_norm": 1.3515625,
|
| 803 |
+
"learning_rate": 0.0003,
|
| 804 |
+
"loss": 3.1967243194580077,
|
| 805 |
+
"step": 1860
|
| 806 |
+
},
|
| 807 |
+
{
|
| 808 |
+
"epoch": 0.06336366700370745,
|
| 809 |
+
"grad_norm": 1.3125,
|
| 810 |
+
"learning_rate": 0.0003,
|
| 811 |
+
"loss": 3.1668914794921874,
|
| 812 |
+
"step": 1880
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 0.06403774856757667,
|
| 816 |
+
"grad_norm": 1.1484375,
|
| 817 |
+
"learning_rate": 0.0003,
|
| 818 |
+
"loss": 3.1733448028564455,
|
| 819 |
+
"step": 1900
|
| 820 |
+
},
|
| 821 |
+
{
|
| 822 |
+
"epoch": 0.06403774856757667,
|
| 823 |
+
"eval_loss": 3.1766083240509033,
|
| 824 |
+
"eval_runtime": 7.9724,
|
| 825 |
+
"eval_samples_per_second": 1194.991,
|
| 826 |
+
"eval_steps_per_second": 0.878,
|
| 827 |
+
"step": 1900
|
| 828 |
+
},
|
| 829 |
+
{
|
| 830 |
+
"epoch": 0.06471183013144591,
|
| 831 |
+
"grad_norm": 1.1953125,
|
| 832 |
+
"learning_rate": 0.0003,
|
| 833 |
+
"loss": 3.1606868743896483,
|
| 834 |
+
"step": 1920
|
| 835 |
+
},
|
| 836 |
+
{
|
| 837 |
+
"epoch": 0.06538591169531513,
|
| 838 |
+
"grad_norm": 1.109375,
|
| 839 |
+
"learning_rate": 0.0003,
|
| 840 |
+
"loss": 3.148599052429199,
|
| 841 |
+
"step": 1940
|
| 842 |
+
},
|
| 843 |
+
{
|
| 844 |
+
"epoch": 0.06605999325918437,
|
| 845 |
+
"grad_norm": 1.2265625,
|
| 846 |
+
"learning_rate": 0.0003,
|
| 847 |
+
"loss": 3.1723419189453126,
|
| 848 |
+
"step": 1960
|
| 849 |
+
},
|
| 850 |
+
{
|
| 851 |
+
"epoch": 0.06673407482305359,
|
| 852 |
+
"grad_norm": 1.234375,
|
| 853 |
+
"learning_rate": 0.0003,
|
| 854 |
+
"loss": 3.184863471984863,
|
| 855 |
+
"step": 1980
|
| 856 |
+
},
|
| 857 |
+
{
|
| 858 |
+
"epoch": 0.06740815638692282,
|
| 859 |
+
"grad_norm": 1.2734375,
|
| 860 |
+
"learning_rate": 0.0003,
|
| 861 |
+
"loss": 3.158551788330078,
|
| 862 |
+
"step": 2000
|
| 863 |
+
},
|
| 864 |
+
{
|
| 865 |
+
"epoch": 0.06740815638692282,
|
| 866 |
+
"eval_loss": 3.154155731201172,
|
| 867 |
+
"eval_runtime": 7.9931,
|
| 868 |
+
"eval_samples_per_second": 1191.905,
|
| 869 |
+
"eval_steps_per_second": 0.876,
|
| 870 |
+
"step": 2000
|
| 871 |
+
},
|
| 872 |
+
{
|
| 873 |
+
"epoch": 0.06808223795079205,
|
| 874 |
+
"grad_norm": 1.2734375,
|
| 875 |
+
"learning_rate": 0.0003,
|
| 876 |
+
"loss": 3.147844696044922,
|
| 877 |
+
"step": 2020
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 0.06875631951466127,
|
| 881 |
+
"grad_norm": 1.1796875,
|
| 882 |
+
"learning_rate": 0.0003,
|
| 883 |
+
"loss": 3.149020957946777,
|
| 884 |
+
"step": 2040
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 0.0694304010785305,
|
| 888 |
+
"grad_norm": 1.234375,
|
| 889 |
+
"learning_rate": 0.0003,
|
| 890 |
+
"loss": 3.1382816314697264,
|
| 891 |
+
"step": 2060
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 0.07010448264239973,
|
| 895 |
+
"grad_norm": 1.203125,
|
| 896 |
+
"learning_rate": 0.0003,
|
| 897 |
+
"loss": 3.1440708160400392,
|
| 898 |
+
"step": 2080
|
| 899 |
+
},
|
| 900 |
+
{
|
| 901 |
+
"epoch": 0.07077856420626896,
|
| 902 |
+
"grad_norm": 1.1171875,
|
| 903 |
+
"learning_rate": 0.0003,
|
| 904 |
+
"loss": 3.132351875305176,
|
| 905 |
+
"step": 2100
|
| 906 |
+
},
|
| 907 |
+
{
|
| 908 |
+
"epoch": 0.07077856420626896,
|
| 909 |
+
"eval_loss": 3.1331419944763184,
|
| 910 |
+
"eval_runtime": 8.2456,
|
| 911 |
+
"eval_samples_per_second": 1155.405,
|
| 912 |
+
"eval_steps_per_second": 0.849,
|
| 913 |
+
"step": 2100
|
| 914 |
+
},
|
| 915 |
+
{
|
| 916 |
+
"epoch": 0.07145264577013818,
|
| 917 |
+
"grad_norm": 1.109375,
|
| 918 |
+
"learning_rate": 0.0003,
|
| 919 |
+
"loss": 3.1341459274291994,
|
| 920 |
+
"step": 2120
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 0.07212672733400742,
|
| 924 |
+
"grad_norm": 1.1640625,
|
| 925 |
+
"learning_rate": 0.0003,
|
| 926 |
+
"loss": 3.1331857681274413,
|
| 927 |
+
"step": 2140
|
| 928 |
+
},
|
| 929 |
+
{
|
| 930 |
+
"epoch": 0.07280080889787664,
|
| 931 |
+
"grad_norm": 1.1796875,
|
| 932 |
+
"learning_rate": 0.0003,
|
| 933 |
+
"loss": 3.1179393768310546,
|
| 934 |
+
"step": 2160
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 0.07347489046174586,
|
| 938 |
+
"grad_norm": 1.546875,
|
| 939 |
+
"learning_rate": 0.0003,
|
| 940 |
+
"loss": 3.1271270751953124,
|
| 941 |
+
"step": 2180
|
| 942 |
+
},
|
| 943 |
+
{
|
| 944 |
+
"epoch": 0.0741489720256151,
|
| 945 |
+
"grad_norm": 1.234375,
|
| 946 |
+
"learning_rate": 0.0003,
|
| 947 |
+
"loss": 3.1252058029174803,
|
| 948 |
+
"step": 2200
|
| 949 |
+
},
|
| 950 |
+
{
|
| 951 |
+
"epoch": 0.0741489720256151,
|
| 952 |
+
"eval_loss": 3.1193113327026367,
|
| 953 |
+
"eval_runtime": 8.0852,
|
| 954 |
+
"eval_samples_per_second": 1178.326,
|
| 955 |
+
"eval_steps_per_second": 0.866,
|
| 956 |
+
"step": 2200
|
| 957 |
+
},
|
| 958 |
+
{
|
| 959 |
+
"epoch": 0.07482305358948432,
|
| 960 |
+
"grad_norm": 1.1015625,
|
| 961 |
+
"learning_rate": 0.0003,
|
| 962 |
+
"loss": 3.0879518508911135,
|
| 963 |
+
"step": 2220
|
| 964 |
+
},
|
| 965 |
+
{
|
| 966 |
+
"epoch": 0.07549713515335356,
|
| 967 |
+
"grad_norm": 1.2109375,
|
| 968 |
+
"learning_rate": 0.0003,
|
| 969 |
+
"loss": 3.1103500366210937,
|
| 970 |
+
"step": 2240
|
| 971 |
+
},
|
| 972 |
+
{
|
| 973 |
+
"epoch": 0.07617121671722278,
|
| 974 |
+
"grad_norm": 1.3828125,
|
| 975 |
+
"learning_rate": 0.0003,
|
| 976 |
+
"loss": 3.08082275390625,
|
| 977 |
+
"step": 2260
|
| 978 |
+
},
|
| 979 |
+
{
|
| 980 |
+
"epoch": 0.07684529828109202,
|
| 981 |
+
"grad_norm": 1.3203125,
|
| 982 |
+
"learning_rate": 0.0003,
|
| 983 |
+
"loss": 3.103218650817871,
|
| 984 |
+
"step": 2280
|
| 985 |
+
},
|
| 986 |
+
{
|
| 987 |
+
"epoch": 0.07751937984496124,
|
| 988 |
+
"grad_norm": 1.1953125,
|
| 989 |
+
"learning_rate": 0.0003,
|
| 990 |
+
"loss": 3.0974100112915037,
|
| 991 |
+
"step": 2300
|
| 992 |
+
},
|
| 993 |
+
{
|
| 994 |
+
"epoch": 0.07751937984496124,
|
| 995 |
+
"eval_loss": 3.098923921585083,
|
| 996 |
+
"eval_runtime": 8.2349,
|
| 997 |
+
"eval_samples_per_second": 1156.908,
|
| 998 |
+
"eval_steps_per_second": 0.85,
|
| 999 |
+
"step": 2300
|
| 1000 |
+
}
|
| 1001 |
+
],
|
| 1002 |
+
"logging_steps": 20,
|
| 1003 |
+
"max_steps": 2500,
|
| 1004 |
+
"num_input_tokens_seen": 0,
|
| 1005 |
+
"num_train_epochs": 1,
|
| 1006 |
+
"save_steps": 100,
|
| 1007 |
+
"stateful_callbacks": {
|
| 1008 |
+
"TrainerControl": {
|
| 1009 |
+
"args": {
|
| 1010 |
+
"should_epoch_stop": false,
|
| 1011 |
+
"should_evaluate": false,
|
| 1012 |
+
"should_log": false,
|
| 1013 |
+
"should_save": true,
|
| 1014 |
+
"should_training_stop": false
|
| 1015 |
+
},
|
| 1016 |
+
"attributes": {}
|
| 1017 |
+
}
|
| 1018 |
+
},
|
| 1019 |
+
"total_flos": 333946709606400.0,
|
| 1020 |
+
"train_batch_size": 32,
|
| 1021 |
+
"trial_name": null,
|
| 1022 |
+
"trial_params": null
|
| 1023 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2300/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a761b578cf2b8c248f28bd1b61a5ec7d0b5d3801c8bd793a6db706df1d3078f6
|
| 3 |
+
size 4920
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/config.json
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "squared_relu",
|
| 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": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 9,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"powlu_m": 3.0,
|
| 25 |
+
"pretraining_tp": 1,
|
| 26 |
+
"rms_norm_eps": 1e-06,
|
| 27 |
+
"rope_parameters": {
|
| 28 |
+
"rope_theta": 10000.0,
|
| 29 |
+
"rope_type": "default"
|
| 30 |
+
},
|
| 31 |
+
"tie_word_embeddings": true,
|
| 32 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 33 |
+
"transformers_version": "5.16.0.dev0",
|
| 34 |
+
"use_cache": false,
|
| 35 |
+
"vocab_size": 4096,
|
| 36 |
+
"waleed_beta": 10.0
|
| 37 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:68bbb4bd123c0d4c584e6c619cb6108e43e78be1e5fb5253dba2b264e510633d
|
| 3 |
+
size 4010544
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b0f0473810d2aee510ff64151a5c803de3e68f9be9cb19517b0c7541b29a2150
|
| 3 |
+
size 8068282
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9022ea8893d375dc2079ce97b2df7f3286392f4c409461865c91e578acc3ba10
|
| 3 |
+
size 14244
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1127c4973386b20d3418aeecc3a05ef9e4aba4952a23325133bed0d9a3788847
|
| 3 |
+
size 1064
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
zain/Activation/out/mlp-squared_relu-9L_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/mlp-squared_relu-9L_run/checkpoint-2400/trainer_state.json
ADDED
|
@@ -0,0 +1,1066 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.08088978766430738,
|
| 6 |
+
"eval_steps": 100,
|
| 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": 0.0006740815638692282,
|
| 14 |
+
"grad_norm": 2.71875,
|
| 15 |
+
"learning_rate": 1.14e-05,
|
| 16 |
+
"loss": 8.314944458007812,
|
| 17 |
+
"step": 20
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.0013481631277384564,
|
| 21 |
+
"grad_norm": 1.8515625,
|
| 22 |
+
"learning_rate": 2.34e-05,
|
| 23 |
+
"loss": 8.251383972167968,
|
| 24 |
+
"step": 40
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.0020222446916076846,
|
| 28 |
+
"grad_norm": 1.3203125,
|
| 29 |
+
"learning_rate": 3.539999999999999e-05,
|
| 30 |
+
"loss": 8.104505920410157,
|
| 31 |
+
"step": 60
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"epoch": 0.002696326255476913,
|
| 35 |
+
"grad_norm": 1.296875,
|
| 36 |
+
"learning_rate": 4.7399999999999993e-05,
|
| 37 |
+
"loss": 7.968361663818359,
|
| 38 |
+
"step": 80
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.003370407819346141,
|
| 42 |
+
"grad_norm": 1.2578125,
|
| 43 |
+
"learning_rate": 5.94e-05,
|
| 44 |
+
"loss": 7.804827880859375,
|
| 45 |
+
"step": 100
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.003370407819346141,
|
| 49 |
+
"eval_loss": 7.703802585601807,
|
| 50 |
+
"eval_runtime": 8.0325,
|
| 51 |
+
"eval_samples_per_second": 1186.059,
|
| 52 |
+
"eval_steps_per_second": 0.871,
|
| 53 |
+
"step": 100
|
| 54 |
+
},
|
| 55 |
+
{
|
| 56 |
+
"epoch": 0.004044489383215369,
|
| 57 |
+
"grad_norm": 1.2734375,
|
| 58 |
+
"learning_rate": 7.139999999999999e-05,
|
| 59 |
+
"loss": 7.600141906738282,
|
| 60 |
+
"step": 120
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"epoch": 0.0047185709470845974,
|
| 64 |
+
"grad_norm": 1.265625,
|
| 65 |
+
"learning_rate": 8.34e-05,
|
| 66 |
+
"loss": 7.368203735351562,
|
| 67 |
+
"step": 140
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"epoch": 0.005392652510953826,
|
| 71 |
+
"grad_norm": 1.265625,
|
| 72 |
+
"learning_rate": 9.539999999999999e-05,
|
| 73 |
+
"loss": 7.133045196533203,
|
| 74 |
+
"step": 160
|
| 75 |
+
},
|
| 76 |
+
{
|
| 77 |
+
"epoch": 0.006066734074823054,
|
| 78 |
+
"grad_norm": 1.203125,
|
| 79 |
+
"learning_rate": 0.00010739999999999998,
|
| 80 |
+
"loss": 6.885277557373047,
|
| 81 |
+
"step": 180
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"epoch": 0.006740815638692282,
|
| 85 |
+
"grad_norm": 1.15625,
|
| 86 |
+
"learning_rate": 0.0001194,
|
| 87 |
+
"loss": 6.629558563232422,
|
| 88 |
+
"step": 200
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"epoch": 0.006740815638692282,
|
| 92 |
+
"eval_loss": 6.4997687339782715,
|
| 93 |
+
"eval_runtime": 7.9479,
|
| 94 |
+
"eval_samples_per_second": 1198.683,
|
| 95 |
+
"eval_steps_per_second": 0.881,
|
| 96 |
+
"step": 200
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"epoch": 0.00741489720256151,
|
| 100 |
+
"grad_norm": 1.109375,
|
| 101 |
+
"learning_rate": 0.0001314,
|
| 102 |
+
"loss": 6.391195297241211,
|
| 103 |
+
"step": 220
|
| 104 |
+
},
|
| 105 |
+
{
|
| 106 |
+
"epoch": 0.008088978766430738,
|
| 107 |
+
"grad_norm": 1.21875,
|
| 108 |
+
"learning_rate": 0.0001434,
|
| 109 |
+
"loss": 6.139227676391601,
|
| 110 |
+
"step": 240
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"epoch": 0.008763060330299966,
|
| 114 |
+
"grad_norm": 1.0625,
|
| 115 |
+
"learning_rate": 0.00015539999999999998,
|
| 116 |
+
"loss": 5.917041778564453,
|
| 117 |
+
"step": 260
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"epoch": 0.009437141894169195,
|
| 121 |
+
"grad_norm": 1.1015625,
|
| 122 |
+
"learning_rate": 0.0001674,
|
| 123 |
+
"loss": 5.702054595947265,
|
| 124 |
+
"step": 280
|
| 125 |
+
},
|
| 126 |
+
{
|
| 127 |
+
"epoch": 0.010111223458038422,
|
| 128 |
+
"grad_norm": 0.921875,
|
| 129 |
+
"learning_rate": 0.00017939999999999997,
|
| 130 |
+
"loss": 5.500263214111328,
|
| 131 |
+
"step": 300
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"epoch": 0.010111223458038422,
|
| 135 |
+
"eval_loss": 5.393505573272705,
|
| 136 |
+
"eval_runtime": 8.0449,
|
| 137 |
+
"eval_samples_per_second": 1184.223,
|
| 138 |
+
"eval_steps_per_second": 0.87,
|
| 139 |
+
"step": 300
|
| 140 |
+
},
|
| 141 |
+
{
|
| 142 |
+
"epoch": 0.010785305021907651,
|
| 143 |
+
"grad_norm": 2.5,
|
| 144 |
+
"learning_rate": 0.0001914,
|
| 145 |
+
"loss": 5.319306182861328,
|
| 146 |
+
"step": 320
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.011459386585776879,
|
| 150 |
+
"grad_norm": 1.078125,
|
| 151 |
+
"learning_rate": 0.00020339999999999998,
|
| 152 |
+
"loss": 5.202053070068359,
|
| 153 |
+
"step": 340
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"epoch": 0.012133468149646108,
|
| 157 |
+
"grad_norm": 1.3203125,
|
| 158 |
+
"learning_rate": 0.00021539999999999998,
|
| 159 |
+
"loss": 5.020883941650391,
|
| 160 |
+
"step": 360
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 0.012807549713515335,
|
| 164 |
+
"grad_norm": 1.1171875,
|
| 165 |
+
"learning_rate": 0.00022739999999999997,
|
| 166 |
+
"loss": 4.873059844970703,
|
| 167 |
+
"step": 380
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"epoch": 0.013481631277384564,
|
| 171 |
+
"grad_norm": 1.234375,
|
| 172 |
+
"learning_rate": 0.0002394,
|
| 173 |
+
"loss": 4.741780853271484,
|
| 174 |
+
"step": 400
|
| 175 |
+
},
|
| 176 |
+
{
|
| 177 |
+
"epoch": 0.013481631277384564,
|
| 178 |
+
"eval_loss": 4.676144123077393,
|
| 179 |
+
"eval_runtime": 8.0457,
|
| 180 |
+
"eval_samples_per_second": 1184.113,
|
| 181 |
+
"eval_steps_per_second": 0.87,
|
| 182 |
+
"step": 400
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"epoch": 0.014155712841253791,
|
| 186 |
+
"grad_norm": 1.3046875,
|
| 187 |
+
"learning_rate": 0.0002514,
|
| 188 |
+
"loss": 4.606548690795899,
|
| 189 |
+
"step": 420
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 0.01482979440512302,
|
| 193 |
+
"grad_norm": 1.2578125,
|
| 194 |
+
"learning_rate": 0.00026339999999999995,
|
| 195 |
+
"loss": 4.515376281738281,
|
| 196 |
+
"step": 440
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"epoch": 0.015503875968992248,
|
| 200 |
+
"grad_norm": 1.0625,
|
| 201 |
+
"learning_rate": 0.00027539999999999997,
|
| 202 |
+
"loss": 4.416887664794922,
|
| 203 |
+
"step": 460
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"epoch": 0.016177957532861477,
|
| 207 |
+
"grad_norm": 1.3671875,
|
| 208 |
+
"learning_rate": 0.00028739999999999994,
|
| 209 |
+
"loss": 4.3046520233154295,
|
| 210 |
+
"step": 480
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"epoch": 0.016852039096730706,
|
| 214 |
+
"grad_norm": 1.2109375,
|
| 215 |
+
"learning_rate": 0.00029939999999999996,
|
| 216 |
+
"loss": 4.238706970214844,
|
| 217 |
+
"step": 500
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"epoch": 0.016852039096730706,
|
| 221 |
+
"eval_loss": 4.196270942687988,
|
| 222 |
+
"eval_runtime": 7.9301,
|
| 223 |
+
"eval_samples_per_second": 1201.377,
|
| 224 |
+
"eval_steps_per_second": 0.883,
|
| 225 |
+
"step": 500
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"epoch": 0.01752612066059993,
|
| 229 |
+
"grad_norm": 0.97265625,
|
| 230 |
+
"learning_rate": 0.0003,
|
| 231 |
+
"loss": 4.1756237030029295,
|
| 232 |
+
"step": 520
|
| 233 |
+
},
|
| 234 |
+
{
|
| 235 |
+
"epoch": 0.01820020222446916,
|
| 236 |
+
"grad_norm": 0.9140625,
|
| 237 |
+
"learning_rate": 0.0003,
|
| 238 |
+
"loss": 4.0915069580078125,
|
| 239 |
+
"step": 540
|
| 240 |
+
},
|
| 241 |
+
{
|
| 242 |
+
"epoch": 0.01887428378833839,
|
| 243 |
+
"grad_norm": 1.2890625,
|
| 244 |
+
"learning_rate": 0.0003,
|
| 245 |
+
"loss": 4.031630325317383,
|
| 246 |
+
"step": 560
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"epoch": 0.01954836535220762,
|
| 250 |
+
"grad_norm": 1.125,
|
| 251 |
+
"learning_rate": 0.0003,
|
| 252 |
+
"loss": 3.9763336181640625,
|
| 253 |
+
"step": 580
|
| 254 |
+
},
|
| 255 |
+
{
|
| 256 |
+
"epoch": 0.020222446916076844,
|
| 257 |
+
"grad_norm": 1.21875,
|
| 258 |
+
"learning_rate": 0.0003,
|
| 259 |
+
"loss": 3.9347896575927734,
|
| 260 |
+
"step": 600
|
| 261 |
+
},
|
| 262 |
+
{
|
| 263 |
+
"epoch": 0.020222446916076844,
|
| 264 |
+
"eval_loss": 3.9157378673553467,
|
| 265 |
+
"eval_runtime": 8.0606,
|
| 266 |
+
"eval_samples_per_second": 1181.917,
|
| 267 |
+
"eval_steps_per_second": 0.868,
|
| 268 |
+
"step": 600
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.020896528479946073,
|
| 272 |
+
"grad_norm": 1.1640625,
|
| 273 |
+
"learning_rate": 0.0003,
|
| 274 |
+
"loss": 3.878151702880859,
|
| 275 |
+
"step": 620
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 0.021570610043815303,
|
| 279 |
+
"grad_norm": 1.484375,
|
| 280 |
+
"learning_rate": 0.0003,
|
| 281 |
+
"loss": 3.855076217651367,
|
| 282 |
+
"step": 640
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 0.022244691607684528,
|
| 286 |
+
"grad_norm": 1.3359375,
|
| 287 |
+
"learning_rate": 0.0003,
|
| 288 |
+
"loss": 3.8232196807861327,
|
| 289 |
+
"step": 660
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 0.022918773171553757,
|
| 293 |
+
"grad_norm": 1.1640625,
|
| 294 |
+
"learning_rate": 0.0003,
|
| 295 |
+
"loss": 3.782532501220703,
|
| 296 |
+
"step": 680
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 0.023592854735422986,
|
| 300 |
+
"grad_norm": 1.1328125,
|
| 301 |
+
"learning_rate": 0.0003,
|
| 302 |
+
"loss": 3.7758609771728517,
|
| 303 |
+
"step": 700
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 0.023592854735422986,
|
| 307 |
+
"eval_loss": 3.7596347332000732,
|
| 308 |
+
"eval_runtime": 7.9743,
|
| 309 |
+
"eval_samples_per_second": 1194.709,
|
| 310 |
+
"eval_steps_per_second": 0.878,
|
| 311 |
+
"step": 700
|
| 312 |
+
},
|
| 313 |
+
{
|
| 314 |
+
"epoch": 0.024266936299292215,
|
| 315 |
+
"grad_norm": 1.25,
|
| 316 |
+
"learning_rate": 0.0003,
|
| 317 |
+
"loss": 3.7433414459228516,
|
| 318 |
+
"step": 720
|
| 319 |
+
},
|
| 320 |
+
{
|
| 321 |
+
"epoch": 0.02494101786316144,
|
| 322 |
+
"grad_norm": 1.3125,
|
| 323 |
+
"learning_rate": 0.0003,
|
| 324 |
+
"loss": 3.732330322265625,
|
| 325 |
+
"step": 740
|
| 326 |
+
},
|
| 327 |
+
{
|
| 328 |
+
"epoch": 0.02561509942703067,
|
| 329 |
+
"grad_norm": 1.265625,
|
| 330 |
+
"learning_rate": 0.0003,
|
| 331 |
+
"loss": 3.6843395233154297,
|
| 332 |
+
"step": 760
|
| 333 |
+
},
|
| 334 |
+
{
|
| 335 |
+
"epoch": 0.0262891809908999,
|
| 336 |
+
"grad_norm": 1.1640625,
|
| 337 |
+
"learning_rate": 0.0003,
|
| 338 |
+
"loss": 3.6463153839111326,
|
| 339 |
+
"step": 780
|
| 340 |
+
},
|
| 341 |
+
{
|
| 342 |
+
"epoch": 0.026963262554769128,
|
| 343 |
+
"grad_norm": 1.1171875,
|
| 344 |
+
"learning_rate": 0.0003,
|
| 345 |
+
"loss": 3.6544403076171874,
|
| 346 |
+
"step": 800
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"epoch": 0.026963262554769128,
|
| 350 |
+
"eval_loss": 3.652294635772705,
|
| 351 |
+
"eval_runtime": 7.9753,
|
| 352 |
+
"eval_samples_per_second": 1194.559,
|
| 353 |
+
"eval_steps_per_second": 0.878,
|
| 354 |
+
"step": 800
|
| 355 |
+
},
|
| 356 |
+
{
|
| 357 |
+
"epoch": 0.027637344118638354,
|
| 358 |
+
"grad_norm": 1.1171875,
|
| 359 |
+
"learning_rate": 0.0003,
|
| 360 |
+
"loss": 3.6159641265869142,
|
| 361 |
+
"step": 820
|
| 362 |
+
},
|
| 363 |
+
{
|
| 364 |
+
"epoch": 0.028311425682507583,
|
| 365 |
+
"grad_norm": 1.078125,
|
| 366 |
+
"learning_rate": 0.0003,
|
| 367 |
+
"loss": 3.6099781036376952,
|
| 368 |
+
"step": 840
|
| 369 |
+
},
|
| 370 |
+
{
|
| 371 |
+
"epoch": 0.028985507246376812,
|
| 372 |
+
"grad_norm": 1.2734375,
|
| 373 |
+
"learning_rate": 0.0003,
|
| 374 |
+
"loss": 3.6344444274902346,
|
| 375 |
+
"step": 860
|
| 376 |
+
},
|
| 377 |
+
{
|
| 378 |
+
"epoch": 0.02965958881024604,
|
| 379 |
+
"grad_norm": 1.140625,
|
| 380 |
+
"learning_rate": 0.0003,
|
| 381 |
+
"loss": 3.5855178833007812,
|
| 382 |
+
"step": 880
|
| 383 |
+
},
|
| 384 |
+
{
|
| 385 |
+
"epoch": 0.030333670374115267,
|
| 386 |
+
"grad_norm": 1.109375,
|
| 387 |
+
"learning_rate": 0.0003,
|
| 388 |
+
"loss": 3.5488441467285154,
|
| 389 |
+
"step": 900
|
| 390 |
+
},
|
| 391 |
+
{
|
| 392 |
+
"epoch": 0.030333670374115267,
|
| 393 |
+
"eval_loss": 3.5648436546325684,
|
| 394 |
+
"eval_runtime": 8.321,
|
| 395 |
+
"eval_samples_per_second": 1144.931,
|
| 396 |
+
"eval_steps_per_second": 0.841,
|
| 397 |
+
"step": 900
|
| 398 |
+
},
|
| 399 |
+
{
|
| 400 |
+
"epoch": 0.031007751937984496,
|
| 401 |
+
"grad_norm": 1.1015625,
|
| 402 |
+
"learning_rate": 0.0003,
|
| 403 |
+
"loss": 3.5646598815917967,
|
| 404 |
+
"step": 920
|
| 405 |
+
},
|
| 406 |
+
{
|
| 407 |
+
"epoch": 0.031681833501853725,
|
| 408 |
+
"grad_norm": 1.25,
|
| 409 |
+
"learning_rate": 0.0003,
|
| 410 |
+
"loss": 3.5480445861816405,
|
| 411 |
+
"step": 940
|
| 412 |
+
},
|
| 413 |
+
{
|
| 414 |
+
"epoch": 0.032355915065722954,
|
| 415 |
+
"grad_norm": 1.140625,
|
| 416 |
+
"learning_rate": 0.0003,
|
| 417 |
+
"loss": 3.5140933990478516,
|
| 418 |
+
"step": 960
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 0.03302999662959218,
|
| 422 |
+
"grad_norm": 1.140625,
|
| 423 |
+
"learning_rate": 0.0003,
|
| 424 |
+
"loss": 3.501856231689453,
|
| 425 |
+
"step": 980
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.03370407819346141,
|
| 429 |
+
"grad_norm": 1.359375,
|
| 430 |
+
"learning_rate": 0.0003,
|
| 431 |
+
"loss": 3.500592803955078,
|
| 432 |
+
"step": 1000
|
| 433 |
+
},
|
| 434 |
+
{
|
| 435 |
+
"epoch": 0.03370407819346141,
|
| 436 |
+
"eval_loss": 3.4981162548065186,
|
| 437 |
+
"eval_runtime": 8.1013,
|
| 438 |
+
"eval_samples_per_second": 1175.987,
|
| 439 |
+
"eval_steps_per_second": 0.864,
|
| 440 |
+
"step": 1000
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"epoch": 0.034378159757330634,
|
| 444 |
+
"grad_norm": 1.1328125,
|
| 445 |
+
"learning_rate": 0.0003,
|
| 446 |
+
"loss": 3.4914066314697267,
|
| 447 |
+
"step": 1020
|
| 448 |
+
},
|
| 449 |
+
{
|
| 450 |
+
"epoch": 0.03505224132119986,
|
| 451 |
+
"grad_norm": 1.1171875,
|
| 452 |
+
"learning_rate": 0.0003,
|
| 453 |
+
"loss": 3.474192810058594,
|
| 454 |
+
"step": 1040
|
| 455 |
+
},
|
| 456 |
+
{
|
| 457 |
+
"epoch": 0.03572632288506909,
|
| 458 |
+
"grad_norm": 1.28125,
|
| 459 |
+
"learning_rate": 0.0003,
|
| 460 |
+
"loss": 3.4552947998046877,
|
| 461 |
+
"step": 1060
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.03640040444893832,
|
| 465 |
+
"grad_norm": 1.203125,
|
| 466 |
+
"learning_rate": 0.0003,
|
| 467 |
+
"loss": 3.4411731719970704,
|
| 468 |
+
"step": 1080
|
| 469 |
+
},
|
| 470 |
+
{
|
| 471 |
+
"epoch": 0.03707448601280755,
|
| 472 |
+
"grad_norm": 1.2734375,
|
| 473 |
+
"learning_rate": 0.0003,
|
| 474 |
+
"loss": 3.435680389404297,
|
| 475 |
+
"step": 1100
|
| 476 |
+
},
|
| 477 |
+
{
|
| 478 |
+
"epoch": 0.03707448601280755,
|
| 479 |
+
"eval_loss": 3.445117712020874,
|
| 480 |
+
"eval_runtime": 8.1641,
|
| 481 |
+
"eval_samples_per_second": 1166.938,
|
| 482 |
+
"eval_steps_per_second": 0.857,
|
| 483 |
+
"step": 1100
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"epoch": 0.03774856757667678,
|
| 487 |
+
"grad_norm": 1.03125,
|
| 488 |
+
"learning_rate": 0.0003,
|
| 489 |
+
"loss": 3.4368762969970703,
|
| 490 |
+
"step": 1120
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"epoch": 0.03842264914054601,
|
| 494 |
+
"grad_norm": 1.2421875,
|
| 495 |
+
"learning_rate": 0.0003,
|
| 496 |
+
"loss": 3.410645294189453,
|
| 497 |
+
"step": 1140
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"epoch": 0.03909673070441524,
|
| 501 |
+
"grad_norm": 1.0703125,
|
| 502 |
+
"learning_rate": 0.0003,
|
| 503 |
+
"loss": 3.4069515228271485,
|
| 504 |
+
"step": 1160
|
| 505 |
+
},
|
| 506 |
+
{
|
| 507 |
+
"epoch": 0.03977081226828446,
|
| 508 |
+
"grad_norm": 1.03125,
|
| 509 |
+
"learning_rate": 0.0003,
|
| 510 |
+
"loss": 3.4227649688720705,
|
| 511 |
+
"step": 1180
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"epoch": 0.04044489383215369,
|
| 515 |
+
"grad_norm": 1.203125,
|
| 516 |
+
"learning_rate": 0.0003,
|
| 517 |
+
"loss": 3.3914558410644533,
|
| 518 |
+
"step": 1200
|
| 519 |
+
},
|
| 520 |
+
{
|
| 521 |
+
"epoch": 0.04044489383215369,
|
| 522 |
+
"eval_loss": 3.3945467472076416,
|
| 523 |
+
"eval_runtime": 8.0433,
|
| 524 |
+
"eval_samples_per_second": 1184.469,
|
| 525 |
+
"eval_steps_per_second": 0.87,
|
| 526 |
+
"step": 1200
|
| 527 |
+
},
|
| 528 |
+
{
|
| 529 |
+
"epoch": 0.04111897539602292,
|
| 530 |
+
"grad_norm": 1.0546875,
|
| 531 |
+
"learning_rate": 0.0003,
|
| 532 |
+
"loss": 3.3614086151123046,
|
| 533 |
+
"step": 1220
|
| 534 |
+
},
|
| 535 |
+
{
|
| 536 |
+
"epoch": 0.04179305695989215,
|
| 537 |
+
"grad_norm": 1.09375,
|
| 538 |
+
"learning_rate": 0.0003,
|
| 539 |
+
"loss": 3.3790237426757814,
|
| 540 |
+
"step": 1240
|
| 541 |
+
},
|
| 542 |
+
{
|
| 543 |
+
"epoch": 0.042467138523761376,
|
| 544 |
+
"grad_norm": 1.3515625,
|
| 545 |
+
"learning_rate": 0.0003,
|
| 546 |
+
"loss": 3.38794059753418,
|
| 547 |
+
"step": 1260
|
| 548 |
+
},
|
| 549 |
+
{
|
| 550 |
+
"epoch": 0.043141220087630605,
|
| 551 |
+
"grad_norm": 1.28125,
|
| 552 |
+
"learning_rate": 0.0003,
|
| 553 |
+
"loss": 3.3654109954833986,
|
| 554 |
+
"step": 1280
|
| 555 |
+
},
|
| 556 |
+
{
|
| 557 |
+
"epoch": 0.043815301651499834,
|
| 558 |
+
"grad_norm": 1.1796875,
|
| 559 |
+
"learning_rate": 0.0003,
|
| 560 |
+
"loss": 3.3572288513183595,
|
| 561 |
+
"step": 1300
|
| 562 |
+
},
|
| 563 |
+
{
|
| 564 |
+
"epoch": 0.043815301651499834,
|
| 565 |
+
"eval_loss": 3.3501007556915283,
|
| 566 |
+
"eval_runtime": 8.0621,
|
| 567 |
+
"eval_samples_per_second": 1181.702,
|
| 568 |
+
"eval_steps_per_second": 0.868,
|
| 569 |
+
"step": 1300
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 0.044489383215369056,
|
| 573 |
+
"grad_norm": 1.2734375,
|
| 574 |
+
"learning_rate": 0.0003,
|
| 575 |
+
"loss": 3.321734619140625,
|
| 576 |
+
"step": 1320
|
| 577 |
+
},
|
| 578 |
+
{
|
| 579 |
+
"epoch": 0.045163464779238285,
|
| 580 |
+
"grad_norm": 1.2890625,
|
| 581 |
+
"learning_rate": 0.0003,
|
| 582 |
+
"loss": 3.3493595123291016,
|
| 583 |
+
"step": 1340
|
| 584 |
+
},
|
| 585 |
+
{
|
| 586 |
+
"epoch": 0.045837546343107514,
|
| 587 |
+
"grad_norm": 1.1875,
|
| 588 |
+
"learning_rate": 0.0003,
|
| 589 |
+
"loss": 3.3272212982177733,
|
| 590 |
+
"step": 1360
|
| 591 |
+
},
|
| 592 |
+
{
|
| 593 |
+
"epoch": 0.046511627906976744,
|
| 594 |
+
"grad_norm": 1.1328125,
|
| 595 |
+
"learning_rate": 0.0003,
|
| 596 |
+
"loss": 3.3202045440673826,
|
| 597 |
+
"step": 1380
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 0.04718570947084597,
|
| 601 |
+
"grad_norm": 1.1171875,
|
| 602 |
+
"learning_rate": 0.0003,
|
| 603 |
+
"loss": 3.2998809814453125,
|
| 604 |
+
"step": 1400
|
| 605 |
+
},
|
| 606 |
+
{
|
| 607 |
+
"epoch": 0.04718570947084597,
|
| 608 |
+
"eval_loss": 3.311476707458496,
|
| 609 |
+
"eval_runtime": 8.1361,
|
| 610 |
+
"eval_samples_per_second": 1170.957,
|
| 611 |
+
"eval_steps_per_second": 0.86,
|
| 612 |
+
"step": 1400
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.0478597910347152,
|
| 616 |
+
"grad_norm": 1.140625,
|
| 617 |
+
"learning_rate": 0.0003,
|
| 618 |
+
"loss": 3.3057735443115233,
|
| 619 |
+
"step": 1420
|
| 620 |
+
},
|
| 621 |
+
{
|
| 622 |
+
"epoch": 0.04853387259858443,
|
| 623 |
+
"grad_norm": 1.109375,
|
| 624 |
+
"learning_rate": 0.0003,
|
| 625 |
+
"loss": 3.2907379150390623,
|
| 626 |
+
"step": 1440
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.04920795416245366,
|
| 630 |
+
"grad_norm": 1.078125,
|
| 631 |
+
"learning_rate": 0.0003,
|
| 632 |
+
"loss": 3.264795684814453,
|
| 633 |
+
"step": 1460
|
| 634 |
+
},
|
| 635 |
+
{
|
| 636 |
+
"epoch": 0.04988203572632288,
|
| 637 |
+
"grad_norm": 1.3671875,
|
| 638 |
+
"learning_rate": 0.0003,
|
| 639 |
+
"loss": 3.2918243408203125,
|
| 640 |
+
"step": 1480
|
| 641 |
+
},
|
| 642 |
+
{
|
| 643 |
+
"epoch": 0.05055611729019211,
|
| 644 |
+
"grad_norm": 1.2421875,
|
| 645 |
+
"learning_rate": 0.0003,
|
| 646 |
+
"loss": 3.2844181060791016,
|
| 647 |
+
"step": 1500
|
| 648 |
+
},
|
| 649 |
+
{
|
| 650 |
+
"epoch": 0.05055611729019211,
|
| 651 |
+
"eval_loss": 3.277984142303467,
|
| 652 |
+
"eval_runtime": 8.0313,
|
| 653 |
+
"eval_samples_per_second": 1186.24,
|
| 654 |
+
"eval_steps_per_second": 0.872,
|
| 655 |
+
"step": 1500
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 0.05123019885406134,
|
| 659 |
+
"grad_norm": 1.265625,
|
| 660 |
+
"learning_rate": 0.0003,
|
| 661 |
+
"loss": 3.2729434967041016,
|
| 662 |
+
"step": 1520
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 0.05190428041793057,
|
| 666 |
+
"grad_norm": 1.1015625,
|
| 667 |
+
"learning_rate": 0.0003,
|
| 668 |
+
"loss": 3.279766845703125,
|
| 669 |
+
"step": 1540
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 0.0525783619817998,
|
| 673 |
+
"grad_norm": 1.25,
|
| 674 |
+
"learning_rate": 0.0003,
|
| 675 |
+
"loss": 3.281978988647461,
|
| 676 |
+
"step": 1560
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 0.05325244354566903,
|
| 680 |
+
"grad_norm": 1.125,
|
| 681 |
+
"learning_rate": 0.0003,
|
| 682 |
+
"loss": 3.2608470916748047,
|
| 683 |
+
"step": 1580
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 0.053926525109538256,
|
| 687 |
+
"grad_norm": 1.2265625,
|
| 688 |
+
"learning_rate": 0.0003,
|
| 689 |
+
"loss": 3.2503360748291015,
|
| 690 |
+
"step": 1600
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.053926525109538256,
|
| 694 |
+
"eval_loss": 3.2511110305786133,
|
| 695 |
+
"eval_runtime": 8.097,
|
| 696 |
+
"eval_samples_per_second": 1176.602,
|
| 697 |
+
"eval_steps_per_second": 0.865,
|
| 698 |
+
"step": 1600
|
| 699 |
+
},
|
| 700 |
+
{
|
| 701 |
+
"epoch": 0.054600606673407485,
|
| 702 |
+
"grad_norm": 1.234375,
|
| 703 |
+
"learning_rate": 0.0003,
|
| 704 |
+
"loss": 3.2713451385498047,
|
| 705 |
+
"step": 1620
|
| 706 |
+
},
|
| 707 |
+
{
|
| 708 |
+
"epoch": 0.05527468823727671,
|
| 709 |
+
"grad_norm": 1.015625,
|
| 710 |
+
"learning_rate": 0.0003,
|
| 711 |
+
"loss": 3.2468338012695312,
|
| 712 |
+
"step": 1640
|
| 713 |
+
},
|
| 714 |
+
{
|
| 715 |
+
"epoch": 0.05594876980114594,
|
| 716 |
+
"grad_norm": 1.15625,
|
| 717 |
+
"learning_rate": 0.0003,
|
| 718 |
+
"loss": 3.255104827880859,
|
| 719 |
+
"step": 1660
|
| 720 |
+
},
|
| 721 |
+
{
|
| 722 |
+
"epoch": 0.056622851365015166,
|
| 723 |
+
"grad_norm": 1.1875,
|
| 724 |
+
"learning_rate": 0.0003,
|
| 725 |
+
"loss": 3.2280048370361327,
|
| 726 |
+
"step": 1680
|
| 727 |
+
},
|
| 728 |
+
{
|
| 729 |
+
"epoch": 0.057296932928884395,
|
| 730 |
+
"grad_norm": 1.265625,
|
| 731 |
+
"learning_rate": 0.0003,
|
| 732 |
+
"loss": 3.216978073120117,
|
| 733 |
+
"step": 1700
|
| 734 |
+
},
|
| 735 |
+
{
|
| 736 |
+
"epoch": 0.057296932928884395,
|
| 737 |
+
"eval_loss": 3.2238714694976807,
|
| 738 |
+
"eval_runtime": 8.059,
|
| 739 |
+
"eval_samples_per_second": 1182.16,
|
| 740 |
+
"eval_steps_per_second": 0.869,
|
| 741 |
+
"step": 1700
|
| 742 |
+
},
|
| 743 |
+
{
|
| 744 |
+
"epoch": 0.057971014492753624,
|
| 745 |
+
"grad_norm": 1.125,
|
| 746 |
+
"learning_rate": 0.0003,
|
| 747 |
+
"loss": 3.213521194458008,
|
| 748 |
+
"step": 1720
|
| 749 |
+
},
|
| 750 |
+
{
|
| 751 |
+
"epoch": 0.05864509605662285,
|
| 752 |
+
"grad_norm": 1.234375,
|
| 753 |
+
"learning_rate": 0.0003,
|
| 754 |
+
"loss": 3.2087985992431642,
|
| 755 |
+
"step": 1740
|
| 756 |
+
},
|
| 757 |
+
{
|
| 758 |
+
"epoch": 0.05931917762049208,
|
| 759 |
+
"grad_norm": 1.2109375,
|
| 760 |
+
"learning_rate": 0.0003,
|
| 761 |
+
"loss": 3.198755645751953,
|
| 762 |
+
"step": 1760
|
| 763 |
+
},
|
| 764 |
+
{
|
| 765 |
+
"epoch": 0.05999325918436131,
|
| 766 |
+
"grad_norm": 1.4453125,
|
| 767 |
+
"learning_rate": 0.0003,
|
| 768 |
+
"loss": 3.2067222595214844,
|
| 769 |
+
"step": 1780
|
| 770 |
+
},
|
| 771 |
+
{
|
| 772 |
+
"epoch": 0.06066734074823053,
|
| 773 |
+
"grad_norm": 1.140625,
|
| 774 |
+
"learning_rate": 0.0003,
|
| 775 |
+
"loss": 3.195578956604004,
|
| 776 |
+
"step": 1800
|
| 777 |
+
},
|
| 778 |
+
{
|
| 779 |
+
"epoch": 0.06066734074823053,
|
| 780 |
+
"eval_loss": 3.1985833644866943,
|
| 781 |
+
"eval_runtime": 8.2334,
|
| 782 |
+
"eval_samples_per_second": 1157.109,
|
| 783 |
+
"eval_steps_per_second": 0.85,
|
| 784 |
+
"step": 1800
|
| 785 |
+
},
|
| 786 |
+
{
|
| 787 |
+
"epoch": 0.06134142231209976,
|
| 788 |
+
"grad_norm": 1.1640625,
|
| 789 |
+
"learning_rate": 0.0003,
|
| 790 |
+
"loss": 3.186159133911133,
|
| 791 |
+
"step": 1820
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"epoch": 0.06201550387596899,
|
| 795 |
+
"grad_norm": 1.21875,
|
| 796 |
+
"learning_rate": 0.0003,
|
| 797 |
+
"loss": 3.1850618362426757,
|
| 798 |
+
"step": 1840
|
| 799 |
+
},
|
| 800 |
+
{
|
| 801 |
+
"epoch": 0.06268958543983821,
|
| 802 |
+
"grad_norm": 1.3515625,
|
| 803 |
+
"learning_rate": 0.0003,
|
| 804 |
+
"loss": 3.1967243194580077,
|
| 805 |
+
"step": 1860
|
| 806 |
+
},
|
| 807 |
+
{
|
| 808 |
+
"epoch": 0.06336366700370745,
|
| 809 |
+
"grad_norm": 1.3125,
|
| 810 |
+
"learning_rate": 0.0003,
|
| 811 |
+
"loss": 3.1668914794921874,
|
| 812 |
+
"step": 1880
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 0.06403774856757667,
|
| 816 |
+
"grad_norm": 1.1484375,
|
| 817 |
+
"learning_rate": 0.0003,
|
| 818 |
+
"loss": 3.1733448028564455,
|
| 819 |
+
"step": 1900
|
| 820 |
+
},
|
| 821 |
+
{
|
| 822 |
+
"epoch": 0.06403774856757667,
|
| 823 |
+
"eval_loss": 3.1766083240509033,
|
| 824 |
+
"eval_runtime": 7.9724,
|
| 825 |
+
"eval_samples_per_second": 1194.991,
|
| 826 |
+
"eval_steps_per_second": 0.878,
|
| 827 |
+
"step": 1900
|
| 828 |
+
},
|
| 829 |
+
{
|
| 830 |
+
"epoch": 0.06471183013144591,
|
| 831 |
+
"grad_norm": 1.1953125,
|
| 832 |
+
"learning_rate": 0.0003,
|
| 833 |
+
"loss": 3.1606868743896483,
|
| 834 |
+
"step": 1920
|
| 835 |
+
},
|
| 836 |
+
{
|
| 837 |
+
"epoch": 0.06538591169531513,
|
| 838 |
+
"grad_norm": 1.109375,
|
| 839 |
+
"learning_rate": 0.0003,
|
| 840 |
+
"loss": 3.148599052429199,
|
| 841 |
+
"step": 1940
|
| 842 |
+
},
|
| 843 |
+
{
|
| 844 |
+
"epoch": 0.06605999325918437,
|
| 845 |
+
"grad_norm": 1.2265625,
|
| 846 |
+
"learning_rate": 0.0003,
|
| 847 |
+
"loss": 3.1723419189453126,
|
| 848 |
+
"step": 1960
|
| 849 |
+
},
|
| 850 |
+
{
|
| 851 |
+
"epoch": 0.06673407482305359,
|
| 852 |
+
"grad_norm": 1.234375,
|
| 853 |
+
"learning_rate": 0.0003,
|
| 854 |
+
"loss": 3.184863471984863,
|
| 855 |
+
"step": 1980
|
| 856 |
+
},
|
| 857 |
+
{
|
| 858 |
+
"epoch": 0.06740815638692282,
|
| 859 |
+
"grad_norm": 1.2734375,
|
| 860 |
+
"learning_rate": 0.0003,
|
| 861 |
+
"loss": 3.158551788330078,
|
| 862 |
+
"step": 2000
|
| 863 |
+
},
|
| 864 |
+
{
|
| 865 |
+
"epoch": 0.06740815638692282,
|
| 866 |
+
"eval_loss": 3.154155731201172,
|
| 867 |
+
"eval_runtime": 7.9931,
|
| 868 |
+
"eval_samples_per_second": 1191.905,
|
| 869 |
+
"eval_steps_per_second": 0.876,
|
| 870 |
+
"step": 2000
|
| 871 |
+
},
|
| 872 |
+
{
|
| 873 |
+
"epoch": 0.06808223795079205,
|
| 874 |
+
"grad_norm": 1.2734375,
|
| 875 |
+
"learning_rate": 0.0003,
|
| 876 |
+
"loss": 3.147844696044922,
|
| 877 |
+
"step": 2020
|
| 878 |
+
},
|
| 879 |
+
{
|
| 880 |
+
"epoch": 0.06875631951466127,
|
| 881 |
+
"grad_norm": 1.1796875,
|
| 882 |
+
"learning_rate": 0.0003,
|
| 883 |
+
"loss": 3.149020957946777,
|
| 884 |
+
"step": 2040
|
| 885 |
+
},
|
| 886 |
+
{
|
| 887 |
+
"epoch": 0.0694304010785305,
|
| 888 |
+
"grad_norm": 1.234375,
|
| 889 |
+
"learning_rate": 0.0003,
|
| 890 |
+
"loss": 3.1382816314697264,
|
| 891 |
+
"step": 2060
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"epoch": 0.07010448264239973,
|
| 895 |
+
"grad_norm": 1.203125,
|
| 896 |
+
"learning_rate": 0.0003,
|
| 897 |
+
"loss": 3.1440708160400392,
|
| 898 |
+
"step": 2080
|
| 899 |
+
},
|
| 900 |
+
{
|
| 901 |
+
"epoch": 0.07077856420626896,
|
| 902 |
+
"grad_norm": 1.1171875,
|
| 903 |
+
"learning_rate": 0.0003,
|
| 904 |
+
"loss": 3.132351875305176,
|
| 905 |
+
"step": 2100
|
| 906 |
+
},
|
| 907 |
+
{
|
| 908 |
+
"epoch": 0.07077856420626896,
|
| 909 |
+
"eval_loss": 3.1331419944763184,
|
| 910 |
+
"eval_runtime": 8.2456,
|
| 911 |
+
"eval_samples_per_second": 1155.405,
|
| 912 |
+
"eval_steps_per_second": 0.849,
|
| 913 |
+
"step": 2100
|
| 914 |
+
},
|
| 915 |
+
{
|
| 916 |
+
"epoch": 0.07145264577013818,
|
| 917 |
+
"grad_norm": 1.109375,
|
| 918 |
+
"learning_rate": 0.0003,
|
| 919 |
+
"loss": 3.1341459274291994,
|
| 920 |
+
"step": 2120
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"epoch": 0.07212672733400742,
|
| 924 |
+
"grad_norm": 1.1640625,
|
| 925 |
+
"learning_rate": 0.0003,
|
| 926 |
+
"loss": 3.1331857681274413,
|
| 927 |
+
"step": 2140
|
| 928 |
+
},
|
| 929 |
+
{
|
| 930 |
+
"epoch": 0.07280080889787664,
|
| 931 |
+
"grad_norm": 1.1796875,
|
| 932 |
+
"learning_rate": 0.0003,
|
| 933 |
+
"loss": 3.1179393768310546,
|
| 934 |
+
"step": 2160
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 0.07347489046174586,
|
| 938 |
+
"grad_norm": 1.546875,
|
| 939 |
+
"learning_rate": 0.0003,
|
| 940 |
+
"loss": 3.1271270751953124,
|
| 941 |
+
"step": 2180
|
| 942 |
+
},
|
| 943 |
+
{
|
| 944 |
+
"epoch": 0.0741489720256151,
|
| 945 |
+
"grad_norm": 1.234375,
|
| 946 |
+
"learning_rate": 0.0003,
|
| 947 |
+
"loss": 3.1252058029174803,
|
| 948 |
+
"step": 2200
|
| 949 |
+
},
|
| 950 |
+
{
|
| 951 |
+
"epoch": 0.0741489720256151,
|
| 952 |
+
"eval_loss": 3.1193113327026367,
|
| 953 |
+
"eval_runtime": 8.0852,
|
| 954 |
+
"eval_samples_per_second": 1178.326,
|
| 955 |
+
"eval_steps_per_second": 0.866,
|
| 956 |
+
"step": 2200
|
| 957 |
+
},
|
| 958 |
+
{
|
| 959 |
+
"epoch": 0.07482305358948432,
|
| 960 |
+
"grad_norm": 1.1015625,
|
| 961 |
+
"learning_rate": 0.0003,
|
| 962 |
+
"loss": 3.0879518508911135,
|
| 963 |
+
"step": 2220
|
| 964 |
+
},
|
| 965 |
+
{
|
| 966 |
+
"epoch": 0.07549713515335356,
|
| 967 |
+
"grad_norm": 1.2109375,
|
| 968 |
+
"learning_rate": 0.0003,
|
| 969 |
+
"loss": 3.1103500366210937,
|
| 970 |
+
"step": 2240
|
| 971 |
+
},
|
| 972 |
+
{
|
| 973 |
+
"epoch": 0.07617121671722278,
|
| 974 |
+
"grad_norm": 1.3828125,
|
| 975 |
+
"learning_rate": 0.0003,
|
| 976 |
+
"loss": 3.08082275390625,
|
| 977 |
+
"step": 2260
|
| 978 |
+
},
|
| 979 |
+
{
|
| 980 |
+
"epoch": 0.07684529828109202,
|
| 981 |
+
"grad_norm": 1.3203125,
|
| 982 |
+
"learning_rate": 0.0003,
|
| 983 |
+
"loss": 3.103218650817871,
|
| 984 |
+
"step": 2280
|
| 985 |
+
},
|
| 986 |
+
{
|
| 987 |
+
"epoch": 0.07751937984496124,
|
| 988 |
+
"grad_norm": 1.1953125,
|
| 989 |
+
"learning_rate": 0.0003,
|
| 990 |
+
"loss": 3.0974100112915037,
|
| 991 |
+
"step": 2300
|
| 992 |
+
},
|
| 993 |
+
{
|
| 994 |
+
"epoch": 0.07751937984496124,
|
| 995 |
+
"eval_loss": 3.098923921585083,
|
| 996 |
+
"eval_runtime": 8.2349,
|
| 997 |
+
"eval_samples_per_second": 1156.908,
|
| 998 |
+
"eval_steps_per_second": 0.85,
|
| 999 |
+
"step": 2300
|
| 1000 |
+
},
|
| 1001 |
+
{
|
| 1002 |
+
"epoch": 0.07819346140883048,
|
| 1003 |
+
"grad_norm": 1.3515625,
|
| 1004 |
+
"learning_rate": 0.0003,
|
| 1005 |
+
"loss": 3.095122718811035,
|
| 1006 |
+
"step": 2320
|
| 1007 |
+
},
|
| 1008 |
+
{
|
| 1009 |
+
"epoch": 0.0788675429726997,
|
| 1010 |
+
"grad_norm": 1.140625,
|
| 1011 |
+
"learning_rate": 0.0003,
|
| 1012 |
+
"loss": 3.084772491455078,
|
| 1013 |
+
"step": 2340
|
| 1014 |
+
},
|
| 1015 |
+
{
|
| 1016 |
+
"epoch": 0.07954162453656892,
|
| 1017 |
+
"grad_norm": 1.34375,
|
| 1018 |
+
"learning_rate": 0.0003,
|
| 1019 |
+
"loss": 3.110363006591797,
|
| 1020 |
+
"step": 2360
|
| 1021 |
+
},
|
| 1022 |
+
{
|
| 1023 |
+
"epoch": 0.08021570610043816,
|
| 1024 |
+
"grad_norm": 1.2734375,
|
| 1025 |
+
"learning_rate": 0.0003,
|
| 1026 |
+
"loss": 3.095768165588379,
|
| 1027 |
+
"step": 2380
|
| 1028 |
+
},
|
| 1029 |
+
{
|
| 1030 |
+
"epoch": 0.08088978766430738,
|
| 1031 |
+
"grad_norm": 1.203125,
|
| 1032 |
+
"learning_rate": 0.0003,
|
| 1033 |
+
"loss": 3.0969696044921875,
|
| 1034 |
+
"step": 2400
|
| 1035 |
+
},
|
| 1036 |
+
{
|
| 1037 |
+
"epoch": 0.08088978766430738,
|
| 1038 |
+
"eval_loss": 3.086376190185547,
|
| 1039 |
+
"eval_runtime": 7.9898,
|
| 1040 |
+
"eval_samples_per_second": 1192.401,
|
| 1041 |
+
"eval_steps_per_second": 0.876,
|
| 1042 |
+
"step": 2400
|
| 1043 |
+
}
|
| 1044 |
+
],
|
| 1045 |
+
"logging_steps": 20,
|
| 1046 |
+
"max_steps": 2500,
|
| 1047 |
+
"num_input_tokens_seen": 0,
|
| 1048 |
+
"num_train_epochs": 1,
|
| 1049 |
+
"save_steps": 100,
|
| 1050 |
+
"stateful_callbacks": {
|
| 1051 |
+
"TrainerControl": {
|
| 1052 |
+
"args": {
|
| 1053 |
+
"should_epoch_stop": false,
|
| 1054 |
+
"should_evaluate": false,
|
| 1055 |
+
"should_log": false,
|
| 1056 |
+
"should_save": true,
|
| 1057 |
+
"should_training_stop": false
|
| 1058 |
+
},
|
| 1059 |
+
"attributes": {}
|
| 1060 |
+
}
|
| 1061 |
+
},
|
| 1062 |
+
"total_flos": 348466131763200.0,
|
| 1063 |
+
"train_batch_size": 32,
|
| 1064 |
+
"trial_name": null,
|
| 1065 |
+
"trial_params": null
|
| 1066 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2400/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a761b578cf2b8c248f28bd1b61a5ec7d0b5d3801c8bd793a6db706df1d3078f6
|
| 3 |
+
size 4920
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/config.json
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"activation": "squared_relu",
|
| 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": "mlp",
|
| 19 |
+
"model_type": "tiny_llama",
|
| 20 |
+
"num_attention_heads": 4,
|
| 21 |
+
"num_hidden_layers": 9,
|
| 22 |
+
"num_key_value_heads": 4,
|
| 23 |
+
"pad_token_id": 0,
|
| 24 |
+
"powlu_m": 3.0,
|
| 25 |
+
"pretraining_tp": 1,
|
| 26 |
+
"rms_norm_eps": 1e-06,
|
| 27 |
+
"rope_parameters": {
|
| 28 |
+
"rope_theta": 10000.0,
|
| 29 |
+
"rope_type": "default"
|
| 30 |
+
},
|
| 31 |
+
"tie_word_embeddings": true,
|
| 32 |
+
"tokenizer_name": "w-ahmad/tiny-stories-tokenizer",
|
| 33 |
+
"transformers_version": "5.16.0.dev0",
|
| 34 |
+
"use_cache": false,
|
| 35 |
+
"vocab_size": 4096,
|
| 36 |
+
"waleed_beta": 10.0
|
| 37 |
+
}
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:596c994cea0d2f8807f3f0d68ce17434c600caf3e81579468e68327cf22b2d3f
|
| 3 |
+
size 4010544
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ff4a94c8b8a33b8d91e473f837301b92ff0ef1475360e236aca24f41be92eef9
|
| 3 |
+
size 8068282
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bfcfb0f88b0d9bcf40424586deee61f72e199a42add1edb92a1b9b3c9e5d5ec6
|
| 3 |
+
size 14244
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e149eeb4e00ed73fd6ed80f14daecda1fa4e464fbe9ea4f3c613128f28cf8a55
|
| 3 |
+
size 1064
|
zain/Activation/out/mlp-squared_relu-9L_run/checkpoint-2500/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|