Uploaded checkpoint-25000
Browse files- model.safetensors +1 -1
- optimizer.pt +1 -1
- rng_state.pth +1 -1
- scheduler.pt +1 -1
- trainer_state.json +1795 -5
model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 2692969128
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a1a61d604982debabd50f305447a20e849e6d906fb944a90bfb88ee03a35dd98
|
| 3 |
size 2692969128
|
optimizer.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 5386075202
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:63f96e1ef451a36c123151a3e8a6afe5e5f10261ae5123a4626d1fb00336a925
|
| 3 |
size 5386075202
|
rng_state.pth
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 14244
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0dc3ee4325da0f20e64010c8e1fb9c1567edc642dd9ab4a2d4367d1009e4383e
|
| 3 |
size 14244
|
scheduler.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 1064
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a0769304697a92d05b3f54a364ac1e52204140fdb95fd093b56a8d6138f45860
|
| 3 |
size 1064
|
trainer_state.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
{
|
| 2 |
-
"best_metric": 0.
|
| 3 |
-
"best_model_checkpoint": "runs/deepseek_CMU-AIR2/math-deepseek_FULL_HardArith_Interm_20240424-065814/checkpoint-
|
| 4 |
-
"epoch": 0.
|
| 5 |
"eval_steps": 500,
|
| 6 |
-
"global_step":
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
|
@@ -1797,6 +1797,1796 @@
|
|
| 1797 |
"eval_samples_per_second": 26.192,
|
| 1798 |
"eval_steps_per_second": 26.192,
|
| 1799 |
"step": 2500
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1800 |
}
|
| 1801 |
],
|
| 1802 |
"logging_steps": 10,
|
|
@@ -1804,7 +3594,7 @@
|
|
| 1804 |
"num_input_tokens_seen": 0,
|
| 1805 |
"num_train_epochs": 1,
|
| 1806 |
"save_steps": 2500,
|
| 1807 |
-
"total_flos":
|
| 1808 |
"train_batch_size": 1,
|
| 1809 |
"trial_name": null,
|
| 1810 |
"trial_params": null
|
|
|
|
| 1 |
{
|
| 2 |
+
"best_metric": 0.6027244925498962,
|
| 3 |
+
"best_model_checkpoint": "runs/deepseek_CMU-AIR2/math-deepseek_FULL_HardArith_Interm_20240424-065814/checkpoint-5000",
|
| 4 |
+
"epoch": 0.125,
|
| 5 |
"eval_steps": 500,
|
| 6 |
+
"global_step": 5000,
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
|
|
|
| 1797 |
"eval_samples_per_second": 26.192,
|
| 1798 |
"eval_steps_per_second": 26.192,
|
| 1799 |
"step": 2500
|
| 1800 |
+
},
|
| 1801 |
+
{
|
| 1802 |
+
"epoch": 0.06,
|
| 1803 |
+
"grad_norm": 5.25,
|
| 1804 |
+
"learning_rate": 1.576842105263158e-05,
|
| 1805 |
+
"loss": 0.6227,
|
| 1806 |
+
"step": 2510
|
| 1807 |
+
},
|
| 1808 |
+
{
|
| 1809 |
+
"epoch": 0.06,
|
| 1810 |
+
"grad_norm": 3.796875,
|
| 1811 |
+
"learning_rate": 1.5747368421052635e-05,
|
| 1812 |
+
"loss": 0.5951,
|
| 1813 |
+
"step": 2520
|
| 1814 |
+
},
|
| 1815 |
+
{
|
| 1816 |
+
"epoch": 0.06,
|
| 1817 |
+
"grad_norm": 4.96875,
|
| 1818 |
+
"learning_rate": 1.5726315789473684e-05,
|
| 1819 |
+
"loss": 0.6014,
|
| 1820 |
+
"step": 2530
|
| 1821 |
+
},
|
| 1822 |
+
{
|
| 1823 |
+
"epoch": 0.06,
|
| 1824 |
+
"grad_norm": 5.90625,
|
| 1825 |
+
"learning_rate": 1.570526315789474e-05,
|
| 1826 |
+
"loss": 0.6057,
|
| 1827 |
+
"step": 2540
|
| 1828 |
+
},
|
| 1829 |
+
{
|
| 1830 |
+
"epoch": 0.06,
|
| 1831 |
+
"grad_norm": 4.875,
|
| 1832 |
+
"learning_rate": 1.568421052631579e-05,
|
| 1833 |
+
"loss": 0.6255,
|
| 1834 |
+
"step": 2550
|
| 1835 |
+
},
|
| 1836 |
+
{
|
| 1837 |
+
"epoch": 0.06,
|
| 1838 |
+
"grad_norm": 3.734375,
|
| 1839 |
+
"learning_rate": 1.5663157894736842e-05,
|
| 1840 |
+
"loss": 0.6768,
|
| 1841 |
+
"step": 2560
|
| 1842 |
+
},
|
| 1843 |
+
{
|
| 1844 |
+
"epoch": 0.06,
|
| 1845 |
+
"grad_norm": 7.40625,
|
| 1846 |
+
"learning_rate": 1.5642105263157898e-05,
|
| 1847 |
+
"loss": 0.6953,
|
| 1848 |
+
"step": 2570
|
| 1849 |
+
},
|
| 1850 |
+
{
|
| 1851 |
+
"epoch": 0.06,
|
| 1852 |
+
"grad_norm": 4.8125,
|
| 1853 |
+
"learning_rate": 1.5621052631578947e-05,
|
| 1854 |
+
"loss": 0.6471,
|
| 1855 |
+
"step": 2580
|
| 1856 |
+
},
|
| 1857 |
+
{
|
| 1858 |
+
"epoch": 0.06,
|
| 1859 |
+
"grad_norm": 4.15625,
|
| 1860 |
+
"learning_rate": 1.5600000000000003e-05,
|
| 1861 |
+
"loss": 0.623,
|
| 1862 |
+
"step": 2590
|
| 1863 |
+
},
|
| 1864 |
+
{
|
| 1865 |
+
"epoch": 0.07,
|
| 1866 |
+
"grad_norm": 3.34375,
|
| 1867 |
+
"learning_rate": 1.5578947368421052e-05,
|
| 1868 |
+
"loss": 0.6352,
|
| 1869 |
+
"step": 2600
|
| 1870 |
+
},
|
| 1871 |
+
{
|
| 1872 |
+
"epoch": 0.07,
|
| 1873 |
+
"grad_norm": 3.46875,
|
| 1874 |
+
"learning_rate": 1.555789473684211e-05,
|
| 1875 |
+
"loss": 0.478,
|
| 1876 |
+
"step": 2610
|
| 1877 |
+
},
|
| 1878 |
+
{
|
| 1879 |
+
"epoch": 0.07,
|
| 1880 |
+
"grad_norm": 2.140625,
|
| 1881 |
+
"learning_rate": 1.5536842105263158e-05,
|
| 1882 |
+
"loss": 0.553,
|
| 1883 |
+
"step": 2620
|
| 1884 |
+
},
|
| 1885 |
+
{
|
| 1886 |
+
"epoch": 0.07,
|
| 1887 |
+
"grad_norm": 3.875,
|
| 1888 |
+
"learning_rate": 1.551578947368421e-05,
|
| 1889 |
+
"loss": 0.5354,
|
| 1890 |
+
"step": 2630
|
| 1891 |
+
},
|
| 1892 |
+
{
|
| 1893 |
+
"epoch": 0.07,
|
| 1894 |
+
"grad_norm": 10.0,
|
| 1895 |
+
"learning_rate": 1.5494736842105263e-05,
|
| 1896 |
+
"loss": 0.6313,
|
| 1897 |
+
"step": 2640
|
| 1898 |
+
},
|
| 1899 |
+
{
|
| 1900 |
+
"epoch": 0.07,
|
| 1901 |
+
"grad_norm": 4.09375,
|
| 1902 |
+
"learning_rate": 1.5473684210526316e-05,
|
| 1903 |
+
"loss": 0.6313,
|
| 1904 |
+
"step": 2650
|
| 1905 |
+
},
|
| 1906 |
+
{
|
| 1907 |
+
"epoch": 0.07,
|
| 1908 |
+
"grad_norm": 4.84375,
|
| 1909 |
+
"learning_rate": 1.545263157894737e-05,
|
| 1910 |
+
"loss": 0.5761,
|
| 1911 |
+
"step": 2660
|
| 1912 |
+
},
|
| 1913 |
+
{
|
| 1914 |
+
"epoch": 0.07,
|
| 1915 |
+
"grad_norm": 3.046875,
|
| 1916 |
+
"learning_rate": 1.543157894736842e-05,
|
| 1917 |
+
"loss": 0.6076,
|
| 1918 |
+
"step": 2670
|
| 1919 |
+
},
|
| 1920 |
+
{
|
| 1921 |
+
"epoch": 0.07,
|
| 1922 |
+
"grad_norm": 4.875,
|
| 1923 |
+
"learning_rate": 1.5410526315789477e-05,
|
| 1924 |
+
"loss": 0.5827,
|
| 1925 |
+
"step": 2680
|
| 1926 |
+
},
|
| 1927 |
+
{
|
| 1928 |
+
"epoch": 0.07,
|
| 1929 |
+
"grad_norm": 3.890625,
|
| 1930 |
+
"learning_rate": 1.5389473684210526e-05,
|
| 1931 |
+
"loss": 0.5284,
|
| 1932 |
+
"step": 2690
|
| 1933 |
+
},
|
| 1934 |
+
{
|
| 1935 |
+
"epoch": 0.07,
|
| 1936 |
+
"grad_norm": 3.875,
|
| 1937 |
+
"learning_rate": 1.536842105263158e-05,
|
| 1938 |
+
"loss": 0.6868,
|
| 1939 |
+
"step": 2700
|
| 1940 |
+
},
|
| 1941 |
+
{
|
| 1942 |
+
"epoch": 0.07,
|
| 1943 |
+
"grad_norm": 2.90625,
|
| 1944 |
+
"learning_rate": 1.534736842105263e-05,
|
| 1945 |
+
"loss": 0.6856,
|
| 1946 |
+
"step": 2710
|
| 1947 |
+
},
|
| 1948 |
+
{
|
| 1949 |
+
"epoch": 0.07,
|
| 1950 |
+
"grad_norm": 2.8125,
|
| 1951 |
+
"learning_rate": 1.5326315789473684e-05,
|
| 1952 |
+
"loss": 0.6687,
|
| 1953 |
+
"step": 2720
|
| 1954 |
+
},
|
| 1955 |
+
{
|
| 1956 |
+
"epoch": 0.07,
|
| 1957 |
+
"grad_norm": 4.34375,
|
| 1958 |
+
"learning_rate": 1.530526315789474e-05,
|
| 1959 |
+
"loss": 0.5599,
|
| 1960 |
+
"step": 2730
|
| 1961 |
+
},
|
| 1962 |
+
{
|
| 1963 |
+
"epoch": 0.07,
|
| 1964 |
+
"grad_norm": 4.21875,
|
| 1965 |
+
"learning_rate": 1.528421052631579e-05,
|
| 1966 |
+
"loss": 0.5339,
|
| 1967 |
+
"step": 2740
|
| 1968 |
+
},
|
| 1969 |
+
{
|
| 1970 |
+
"epoch": 0.07,
|
| 1971 |
+
"grad_norm": 8.1875,
|
| 1972 |
+
"learning_rate": 1.5263157894736846e-05,
|
| 1973 |
+
"loss": 0.5145,
|
| 1974 |
+
"step": 2750
|
| 1975 |
+
},
|
| 1976 |
+
{
|
| 1977 |
+
"epoch": 0.07,
|
| 1978 |
+
"grad_norm": 2.75,
|
| 1979 |
+
"learning_rate": 1.5242105263157897e-05,
|
| 1980 |
+
"loss": 0.559,
|
| 1981 |
+
"step": 2760
|
| 1982 |
+
},
|
| 1983 |
+
{
|
| 1984 |
+
"epoch": 0.07,
|
| 1985 |
+
"grad_norm": 3.796875,
|
| 1986 |
+
"learning_rate": 1.5221052631578948e-05,
|
| 1987 |
+
"loss": 0.6415,
|
| 1988 |
+
"step": 2770
|
| 1989 |
+
},
|
| 1990 |
+
{
|
| 1991 |
+
"epoch": 0.07,
|
| 1992 |
+
"grad_norm": 4.8125,
|
| 1993 |
+
"learning_rate": 1.5200000000000002e-05,
|
| 1994 |
+
"loss": 0.6152,
|
| 1995 |
+
"step": 2780
|
| 1996 |
+
},
|
| 1997 |
+
{
|
| 1998 |
+
"epoch": 0.07,
|
| 1999 |
+
"grad_norm": 2.71875,
|
| 2000 |
+
"learning_rate": 1.5178947368421053e-05,
|
| 2001 |
+
"loss": 0.5812,
|
| 2002 |
+
"step": 2790
|
| 2003 |
+
},
|
| 2004 |
+
{
|
| 2005 |
+
"epoch": 0.07,
|
| 2006 |
+
"grad_norm": 5.0625,
|
| 2007 |
+
"learning_rate": 1.5157894736842107e-05,
|
| 2008 |
+
"loss": 0.5348,
|
| 2009 |
+
"step": 2800
|
| 2010 |
+
},
|
| 2011 |
+
{
|
| 2012 |
+
"epoch": 0.07,
|
| 2013 |
+
"grad_norm": 3.546875,
|
| 2014 |
+
"learning_rate": 1.5136842105263158e-05,
|
| 2015 |
+
"loss": 0.5899,
|
| 2016 |
+
"step": 2810
|
| 2017 |
+
},
|
| 2018 |
+
{
|
| 2019 |
+
"epoch": 0.07,
|
| 2020 |
+
"grad_norm": 3.578125,
|
| 2021 |
+
"learning_rate": 1.5115789473684212e-05,
|
| 2022 |
+
"loss": 0.5686,
|
| 2023 |
+
"step": 2820
|
| 2024 |
+
},
|
| 2025 |
+
{
|
| 2026 |
+
"epoch": 0.07,
|
| 2027 |
+
"grad_norm": 3.59375,
|
| 2028 |
+
"learning_rate": 1.5094736842105263e-05,
|
| 2029 |
+
"loss": 0.4602,
|
| 2030 |
+
"step": 2830
|
| 2031 |
+
},
|
| 2032 |
+
{
|
| 2033 |
+
"epoch": 0.07,
|
| 2034 |
+
"grad_norm": 5.4375,
|
| 2035 |
+
"learning_rate": 1.5073684210526316e-05,
|
| 2036 |
+
"loss": 0.6561,
|
| 2037 |
+
"step": 2840
|
| 2038 |
+
},
|
| 2039 |
+
{
|
| 2040 |
+
"epoch": 0.07,
|
| 2041 |
+
"grad_norm": 8.75,
|
| 2042 |
+
"learning_rate": 1.505263157894737e-05,
|
| 2043 |
+
"loss": 0.5948,
|
| 2044 |
+
"step": 2850
|
| 2045 |
+
},
|
| 2046 |
+
{
|
| 2047 |
+
"epoch": 0.07,
|
| 2048 |
+
"grad_norm": 4.1875,
|
| 2049 |
+
"learning_rate": 1.5031578947368421e-05,
|
| 2050 |
+
"loss": 0.6053,
|
| 2051 |
+
"step": 2860
|
| 2052 |
+
},
|
| 2053 |
+
{
|
| 2054 |
+
"epoch": 0.07,
|
| 2055 |
+
"grad_norm": 4.0625,
|
| 2056 |
+
"learning_rate": 1.5010526315789476e-05,
|
| 2057 |
+
"loss": 0.5915,
|
| 2058 |
+
"step": 2870
|
| 2059 |
+
},
|
| 2060 |
+
{
|
| 2061 |
+
"epoch": 0.07,
|
| 2062 |
+
"grad_norm": 3.671875,
|
| 2063 |
+
"learning_rate": 1.4989473684210527e-05,
|
| 2064 |
+
"loss": 0.6329,
|
| 2065 |
+
"step": 2880
|
| 2066 |
+
},
|
| 2067 |
+
{
|
| 2068 |
+
"epoch": 0.07,
|
| 2069 |
+
"grad_norm": 3.421875,
|
| 2070 |
+
"learning_rate": 1.4968421052631581e-05,
|
| 2071 |
+
"loss": 0.6849,
|
| 2072 |
+
"step": 2890
|
| 2073 |
+
},
|
| 2074 |
+
{
|
| 2075 |
+
"epoch": 0.07,
|
| 2076 |
+
"grad_norm": 11.0625,
|
| 2077 |
+
"learning_rate": 1.4947368421052632e-05,
|
| 2078 |
+
"loss": 0.6583,
|
| 2079 |
+
"step": 2900
|
| 2080 |
+
},
|
| 2081 |
+
{
|
| 2082 |
+
"epoch": 0.07,
|
| 2083 |
+
"grad_norm": 3.25,
|
| 2084 |
+
"learning_rate": 1.4926315789473686e-05,
|
| 2085 |
+
"loss": 0.6508,
|
| 2086 |
+
"step": 2910
|
| 2087 |
+
},
|
| 2088 |
+
{
|
| 2089 |
+
"epoch": 0.07,
|
| 2090 |
+
"grad_norm": 4.5,
|
| 2091 |
+
"learning_rate": 1.4905263157894739e-05,
|
| 2092 |
+
"loss": 0.5244,
|
| 2093 |
+
"step": 2920
|
| 2094 |
+
},
|
| 2095 |
+
{
|
| 2096 |
+
"epoch": 0.07,
|
| 2097 |
+
"grad_norm": 4.03125,
|
| 2098 |
+
"learning_rate": 1.488421052631579e-05,
|
| 2099 |
+
"loss": 0.6146,
|
| 2100 |
+
"step": 2930
|
| 2101 |
+
},
|
| 2102 |
+
{
|
| 2103 |
+
"epoch": 0.07,
|
| 2104 |
+
"grad_norm": 3.1875,
|
| 2105 |
+
"learning_rate": 1.4863157894736844e-05,
|
| 2106 |
+
"loss": 0.6245,
|
| 2107 |
+
"step": 2940
|
| 2108 |
+
},
|
| 2109 |
+
{
|
| 2110 |
+
"epoch": 0.07,
|
| 2111 |
+
"grad_norm": 3.8125,
|
| 2112 |
+
"learning_rate": 1.4842105263157895e-05,
|
| 2113 |
+
"loss": 0.6862,
|
| 2114 |
+
"step": 2950
|
| 2115 |
+
},
|
| 2116 |
+
{
|
| 2117 |
+
"epoch": 0.07,
|
| 2118 |
+
"grad_norm": 4.40625,
|
| 2119 |
+
"learning_rate": 1.482105263157895e-05,
|
| 2120 |
+
"loss": 0.5638,
|
| 2121 |
+
"step": 2960
|
| 2122 |
+
},
|
| 2123 |
+
{
|
| 2124 |
+
"epoch": 0.07,
|
| 2125 |
+
"grad_norm": 3.0625,
|
| 2126 |
+
"learning_rate": 1.48e-05,
|
| 2127 |
+
"loss": 0.4932,
|
| 2128 |
+
"step": 2970
|
| 2129 |
+
},
|
| 2130 |
+
{
|
| 2131 |
+
"epoch": 0.07,
|
| 2132 |
+
"grad_norm": 4.3125,
|
| 2133 |
+
"learning_rate": 1.4778947368421055e-05,
|
| 2134 |
+
"loss": 0.626,
|
| 2135 |
+
"step": 2980
|
| 2136 |
+
},
|
| 2137 |
+
{
|
| 2138 |
+
"epoch": 0.07,
|
| 2139 |
+
"grad_norm": 5.125,
|
| 2140 |
+
"learning_rate": 1.4757894736842106e-05,
|
| 2141 |
+
"loss": 0.5843,
|
| 2142 |
+
"step": 2990
|
| 2143 |
+
},
|
| 2144 |
+
{
|
| 2145 |
+
"epoch": 0.07,
|
| 2146 |
+
"grad_norm": 9.1875,
|
| 2147 |
+
"learning_rate": 1.4736842105263159e-05,
|
| 2148 |
+
"loss": 0.5773,
|
| 2149 |
+
"step": 3000
|
| 2150 |
+
},
|
| 2151 |
+
{
|
| 2152 |
+
"epoch": 0.07,
|
| 2153 |
+
"eval_loss": 0.613276481628418,
|
| 2154 |
+
"eval_runtime": 38.1471,
|
| 2155 |
+
"eval_samples_per_second": 26.214,
|
| 2156 |
+
"eval_steps_per_second": 26.214,
|
| 2157 |
+
"step": 3000
|
| 2158 |
+
},
|
| 2159 |
+
{
|
| 2160 |
+
"epoch": 0.08,
|
| 2161 |
+
"grad_norm": 6.8125,
|
| 2162 |
+
"learning_rate": 1.4715789473684213e-05,
|
| 2163 |
+
"loss": 0.475,
|
| 2164 |
+
"step": 3010
|
| 2165 |
+
},
|
| 2166 |
+
{
|
| 2167 |
+
"epoch": 0.08,
|
| 2168 |
+
"grad_norm": 3.21875,
|
| 2169 |
+
"learning_rate": 1.4694736842105264e-05,
|
| 2170 |
+
"loss": 0.592,
|
| 2171 |
+
"step": 3020
|
| 2172 |
+
},
|
| 2173 |
+
{
|
| 2174 |
+
"epoch": 0.08,
|
| 2175 |
+
"grad_norm": 5.78125,
|
| 2176 |
+
"learning_rate": 1.4673684210526318e-05,
|
| 2177 |
+
"loss": 0.6149,
|
| 2178 |
+
"step": 3030
|
| 2179 |
+
},
|
| 2180 |
+
{
|
| 2181 |
+
"epoch": 0.08,
|
| 2182 |
+
"grad_norm": 3.859375,
|
| 2183 |
+
"learning_rate": 1.465263157894737e-05,
|
| 2184 |
+
"loss": 0.5379,
|
| 2185 |
+
"step": 3040
|
| 2186 |
+
},
|
| 2187 |
+
{
|
| 2188 |
+
"epoch": 0.08,
|
| 2189 |
+
"grad_norm": 3.109375,
|
| 2190 |
+
"learning_rate": 1.4631578947368424e-05,
|
| 2191 |
+
"loss": 0.5233,
|
| 2192 |
+
"step": 3050
|
| 2193 |
+
},
|
| 2194 |
+
{
|
| 2195 |
+
"epoch": 0.08,
|
| 2196 |
+
"grad_norm": 4.375,
|
| 2197 |
+
"learning_rate": 1.4610526315789474e-05,
|
| 2198 |
+
"loss": 0.6428,
|
| 2199 |
+
"step": 3060
|
| 2200 |
+
},
|
| 2201 |
+
{
|
| 2202 |
+
"epoch": 0.08,
|
| 2203 |
+
"grad_norm": 5.46875,
|
| 2204 |
+
"learning_rate": 1.4589473684210527e-05,
|
| 2205 |
+
"loss": 0.56,
|
| 2206 |
+
"step": 3070
|
| 2207 |
+
},
|
| 2208 |
+
{
|
| 2209 |
+
"epoch": 0.08,
|
| 2210 |
+
"grad_norm": 13.0625,
|
| 2211 |
+
"learning_rate": 1.456842105263158e-05,
|
| 2212 |
+
"loss": 0.6643,
|
| 2213 |
+
"step": 3080
|
| 2214 |
+
},
|
| 2215 |
+
{
|
| 2216 |
+
"epoch": 0.08,
|
| 2217 |
+
"grad_norm": 9.5625,
|
| 2218 |
+
"learning_rate": 1.4547368421052632e-05,
|
| 2219 |
+
"loss": 0.6606,
|
| 2220 |
+
"step": 3090
|
| 2221 |
+
},
|
| 2222 |
+
{
|
| 2223 |
+
"epoch": 0.08,
|
| 2224 |
+
"grad_norm": 2.953125,
|
| 2225 |
+
"learning_rate": 1.4526315789473687e-05,
|
| 2226 |
+
"loss": 0.6734,
|
| 2227 |
+
"step": 3100
|
| 2228 |
+
},
|
| 2229 |
+
{
|
| 2230 |
+
"epoch": 0.08,
|
| 2231 |
+
"grad_norm": 3.328125,
|
| 2232 |
+
"learning_rate": 1.4505263157894738e-05,
|
| 2233 |
+
"loss": 0.6948,
|
| 2234 |
+
"step": 3110
|
| 2235 |
+
},
|
| 2236 |
+
{
|
| 2237 |
+
"epoch": 0.08,
|
| 2238 |
+
"grad_norm": 5.8125,
|
| 2239 |
+
"learning_rate": 1.4484210526315792e-05,
|
| 2240 |
+
"loss": 0.6735,
|
| 2241 |
+
"step": 3120
|
| 2242 |
+
},
|
| 2243 |
+
{
|
| 2244 |
+
"epoch": 0.08,
|
| 2245 |
+
"grad_norm": 3.96875,
|
| 2246 |
+
"learning_rate": 1.4463157894736843e-05,
|
| 2247 |
+
"loss": 0.6107,
|
| 2248 |
+
"step": 3130
|
| 2249 |
+
},
|
| 2250 |
+
{
|
| 2251 |
+
"epoch": 0.08,
|
| 2252 |
+
"grad_norm": 3.09375,
|
| 2253 |
+
"learning_rate": 1.4442105263157896e-05,
|
| 2254 |
+
"loss": 0.5645,
|
| 2255 |
+
"step": 3140
|
| 2256 |
+
},
|
| 2257 |
+
{
|
| 2258 |
+
"epoch": 0.08,
|
| 2259 |
+
"grad_norm": 4.28125,
|
| 2260 |
+
"learning_rate": 1.4421052631578948e-05,
|
| 2261 |
+
"loss": 0.6415,
|
| 2262 |
+
"step": 3150
|
| 2263 |
+
},
|
| 2264 |
+
{
|
| 2265 |
+
"epoch": 0.08,
|
| 2266 |
+
"grad_norm": 2.875,
|
| 2267 |
+
"learning_rate": 1.4400000000000001e-05,
|
| 2268 |
+
"loss": 0.4961,
|
| 2269 |
+
"step": 3160
|
| 2270 |
+
},
|
| 2271 |
+
{
|
| 2272 |
+
"epoch": 0.08,
|
| 2273 |
+
"grad_norm": 3.484375,
|
| 2274 |
+
"learning_rate": 1.4378947368421054e-05,
|
| 2275 |
+
"loss": 0.609,
|
| 2276 |
+
"step": 3170
|
| 2277 |
+
},
|
| 2278 |
+
{
|
| 2279 |
+
"epoch": 0.08,
|
| 2280 |
+
"grad_norm": 5.59375,
|
| 2281 |
+
"learning_rate": 1.4357894736842106e-05,
|
| 2282 |
+
"loss": 0.6139,
|
| 2283 |
+
"step": 3180
|
| 2284 |
+
},
|
| 2285 |
+
{
|
| 2286 |
+
"epoch": 0.08,
|
| 2287 |
+
"grad_norm": 7.5,
|
| 2288 |
+
"learning_rate": 1.433684210526316e-05,
|
| 2289 |
+
"loss": 0.701,
|
| 2290 |
+
"step": 3190
|
| 2291 |
+
},
|
| 2292 |
+
{
|
| 2293 |
+
"epoch": 0.08,
|
| 2294 |
+
"grad_norm": 4.4375,
|
| 2295 |
+
"learning_rate": 1.4315789473684212e-05,
|
| 2296 |
+
"loss": 0.5645,
|
| 2297 |
+
"step": 3200
|
| 2298 |
+
},
|
| 2299 |
+
{
|
| 2300 |
+
"epoch": 0.08,
|
| 2301 |
+
"grad_norm": 3.0,
|
| 2302 |
+
"learning_rate": 1.4294736842105263e-05,
|
| 2303 |
+
"loss": 0.6149,
|
| 2304 |
+
"step": 3210
|
| 2305 |
+
},
|
| 2306 |
+
{
|
| 2307 |
+
"epoch": 0.08,
|
| 2308 |
+
"grad_norm": 5.25,
|
| 2309 |
+
"learning_rate": 1.4273684210526317e-05,
|
| 2310 |
+
"loss": 0.6544,
|
| 2311 |
+
"step": 3220
|
| 2312 |
+
},
|
| 2313 |
+
{
|
| 2314 |
+
"epoch": 0.08,
|
| 2315 |
+
"grad_norm": 3.21875,
|
| 2316 |
+
"learning_rate": 1.425263157894737e-05,
|
| 2317 |
+
"loss": 0.7131,
|
| 2318 |
+
"step": 3230
|
| 2319 |
+
},
|
| 2320 |
+
{
|
| 2321 |
+
"epoch": 0.08,
|
| 2322 |
+
"grad_norm": 3.359375,
|
| 2323 |
+
"learning_rate": 1.4231578947368422e-05,
|
| 2324 |
+
"loss": 0.5974,
|
| 2325 |
+
"step": 3240
|
| 2326 |
+
},
|
| 2327 |
+
{
|
| 2328 |
+
"epoch": 0.08,
|
| 2329 |
+
"grad_norm": 2.109375,
|
| 2330 |
+
"learning_rate": 1.4210526315789475e-05,
|
| 2331 |
+
"loss": 0.5418,
|
| 2332 |
+
"step": 3250
|
| 2333 |
+
},
|
| 2334 |
+
{
|
| 2335 |
+
"epoch": 0.08,
|
| 2336 |
+
"grad_norm": 3.96875,
|
| 2337 |
+
"learning_rate": 1.418947368421053e-05,
|
| 2338 |
+
"loss": 0.5797,
|
| 2339 |
+
"step": 3260
|
| 2340 |
+
},
|
| 2341 |
+
{
|
| 2342 |
+
"epoch": 0.08,
|
| 2343 |
+
"grad_norm": 2.5,
|
| 2344 |
+
"learning_rate": 1.416842105263158e-05,
|
| 2345 |
+
"loss": 0.5284,
|
| 2346 |
+
"step": 3270
|
| 2347 |
+
},
|
| 2348 |
+
{
|
| 2349 |
+
"epoch": 0.08,
|
| 2350 |
+
"grad_norm": 4.21875,
|
| 2351 |
+
"learning_rate": 1.4147368421052631e-05,
|
| 2352 |
+
"loss": 0.6928,
|
| 2353 |
+
"step": 3280
|
| 2354 |
+
},
|
| 2355 |
+
{
|
| 2356 |
+
"epoch": 0.08,
|
| 2357 |
+
"grad_norm": 5.90625,
|
| 2358 |
+
"learning_rate": 1.4126315789473686e-05,
|
| 2359 |
+
"loss": 0.5994,
|
| 2360 |
+
"step": 3290
|
| 2361 |
+
},
|
| 2362 |
+
{
|
| 2363 |
+
"epoch": 0.08,
|
| 2364 |
+
"grad_norm": 7.09375,
|
| 2365 |
+
"learning_rate": 1.4105263157894738e-05,
|
| 2366 |
+
"loss": 0.4793,
|
| 2367 |
+
"step": 3300
|
| 2368 |
+
},
|
| 2369 |
+
{
|
| 2370 |
+
"epoch": 0.08,
|
| 2371 |
+
"grad_norm": 3.984375,
|
| 2372 |
+
"learning_rate": 1.4084210526315791e-05,
|
| 2373 |
+
"loss": 0.6549,
|
| 2374 |
+
"step": 3310
|
| 2375 |
+
},
|
| 2376 |
+
{
|
| 2377 |
+
"epoch": 0.08,
|
| 2378 |
+
"grad_norm": 2.078125,
|
| 2379 |
+
"learning_rate": 1.4063157894736844e-05,
|
| 2380 |
+
"loss": 0.6774,
|
| 2381 |
+
"step": 3320
|
| 2382 |
+
},
|
| 2383 |
+
{
|
| 2384 |
+
"epoch": 0.08,
|
| 2385 |
+
"grad_norm": 18.5,
|
| 2386 |
+
"learning_rate": 1.4042105263157896e-05,
|
| 2387 |
+
"loss": 0.7274,
|
| 2388 |
+
"step": 3330
|
| 2389 |
+
},
|
| 2390 |
+
{
|
| 2391 |
+
"epoch": 0.08,
|
| 2392 |
+
"grad_norm": 4.0625,
|
| 2393 |
+
"learning_rate": 1.4021052631578949e-05,
|
| 2394 |
+
"loss": 0.625,
|
| 2395 |
+
"step": 3340
|
| 2396 |
+
},
|
| 2397 |
+
{
|
| 2398 |
+
"epoch": 0.08,
|
| 2399 |
+
"grad_norm": 4.5625,
|
| 2400 |
+
"learning_rate": 1.4e-05,
|
| 2401 |
+
"loss": 0.6704,
|
| 2402 |
+
"step": 3350
|
| 2403 |
+
},
|
| 2404 |
+
{
|
| 2405 |
+
"epoch": 0.08,
|
| 2406 |
+
"grad_norm": 3.71875,
|
| 2407 |
+
"learning_rate": 1.3978947368421054e-05,
|
| 2408 |
+
"loss": 0.717,
|
| 2409 |
+
"step": 3360
|
| 2410 |
+
},
|
| 2411 |
+
{
|
| 2412 |
+
"epoch": 0.08,
|
| 2413 |
+
"grad_norm": 5.53125,
|
| 2414 |
+
"learning_rate": 1.3957894736842105e-05,
|
| 2415 |
+
"loss": 0.5063,
|
| 2416 |
+
"step": 3370
|
| 2417 |
+
},
|
| 2418 |
+
{
|
| 2419 |
+
"epoch": 0.08,
|
| 2420 |
+
"grad_norm": 4.875,
|
| 2421 |
+
"learning_rate": 1.393684210526316e-05,
|
| 2422 |
+
"loss": 0.5717,
|
| 2423 |
+
"step": 3380
|
| 2424 |
+
},
|
| 2425 |
+
{
|
| 2426 |
+
"epoch": 0.08,
|
| 2427 |
+
"grad_norm": 4.03125,
|
| 2428 |
+
"learning_rate": 1.3915789473684212e-05,
|
| 2429 |
+
"loss": 0.5619,
|
| 2430 |
+
"step": 3390
|
| 2431 |
+
},
|
| 2432 |
+
{
|
| 2433 |
+
"epoch": 0.09,
|
| 2434 |
+
"grad_norm": 5.75,
|
| 2435 |
+
"learning_rate": 1.3894736842105265e-05,
|
| 2436 |
+
"loss": 0.5526,
|
| 2437 |
+
"step": 3400
|
| 2438 |
+
},
|
| 2439 |
+
{
|
| 2440 |
+
"epoch": 0.09,
|
| 2441 |
+
"grad_norm": 4.875,
|
| 2442 |
+
"learning_rate": 1.3873684210526317e-05,
|
| 2443 |
+
"loss": 0.5975,
|
| 2444 |
+
"step": 3410
|
| 2445 |
+
},
|
| 2446 |
+
{
|
| 2447 |
+
"epoch": 0.09,
|
| 2448 |
+
"grad_norm": 3.890625,
|
| 2449 |
+
"learning_rate": 1.3852631578947368e-05,
|
| 2450 |
+
"loss": 0.6742,
|
| 2451 |
+
"step": 3420
|
| 2452 |
+
},
|
| 2453 |
+
{
|
| 2454 |
+
"epoch": 0.09,
|
| 2455 |
+
"grad_norm": 4.4375,
|
| 2456 |
+
"learning_rate": 1.3831578947368423e-05,
|
| 2457 |
+
"loss": 0.5027,
|
| 2458 |
+
"step": 3430
|
| 2459 |
+
},
|
| 2460 |
+
{
|
| 2461 |
+
"epoch": 0.09,
|
| 2462 |
+
"grad_norm": 6.8125,
|
| 2463 |
+
"learning_rate": 1.3810526315789474e-05,
|
| 2464 |
+
"loss": 0.6286,
|
| 2465 |
+
"step": 3440
|
| 2466 |
+
},
|
| 2467 |
+
{
|
| 2468 |
+
"epoch": 0.09,
|
| 2469 |
+
"grad_norm": 5.1875,
|
| 2470 |
+
"learning_rate": 1.3789473684210528e-05,
|
| 2471 |
+
"loss": 0.5871,
|
| 2472 |
+
"step": 3450
|
| 2473 |
+
},
|
| 2474 |
+
{
|
| 2475 |
+
"epoch": 0.09,
|
| 2476 |
+
"grad_norm": 8.0625,
|
| 2477 |
+
"learning_rate": 1.3768421052631579e-05,
|
| 2478 |
+
"loss": 0.5191,
|
| 2479 |
+
"step": 3460
|
| 2480 |
+
},
|
| 2481 |
+
{
|
| 2482 |
+
"epoch": 0.09,
|
| 2483 |
+
"grad_norm": 8.4375,
|
| 2484 |
+
"learning_rate": 1.3747368421052633e-05,
|
| 2485 |
+
"loss": 0.6679,
|
| 2486 |
+
"step": 3470
|
| 2487 |
+
},
|
| 2488 |
+
{
|
| 2489 |
+
"epoch": 0.09,
|
| 2490 |
+
"grad_norm": 42.5,
|
| 2491 |
+
"learning_rate": 1.3726315789473686e-05,
|
| 2492 |
+
"loss": 0.6627,
|
| 2493 |
+
"step": 3480
|
| 2494 |
+
},
|
| 2495 |
+
{
|
| 2496 |
+
"epoch": 0.09,
|
| 2497 |
+
"grad_norm": 3.375,
|
| 2498 |
+
"learning_rate": 1.3705263157894737e-05,
|
| 2499 |
+
"loss": 0.5305,
|
| 2500 |
+
"step": 3490
|
| 2501 |
+
},
|
| 2502 |
+
{
|
| 2503 |
+
"epoch": 0.09,
|
| 2504 |
+
"grad_norm": 3.6875,
|
| 2505 |
+
"learning_rate": 1.3684210526315791e-05,
|
| 2506 |
+
"loss": 0.6408,
|
| 2507 |
+
"step": 3500
|
| 2508 |
+
},
|
| 2509 |
+
{
|
| 2510 |
+
"epoch": 0.09,
|
| 2511 |
+
"eval_loss": 0.6289076209068298,
|
| 2512 |
+
"eval_runtime": 38.1752,
|
| 2513 |
+
"eval_samples_per_second": 26.195,
|
| 2514 |
+
"eval_steps_per_second": 26.195,
|
| 2515 |
+
"step": 3500
|
| 2516 |
+
},
|
| 2517 |
+
{
|
| 2518 |
+
"epoch": 0.09,
|
| 2519 |
+
"grad_norm": 4.0625,
|
| 2520 |
+
"learning_rate": 1.3663157894736842e-05,
|
| 2521 |
+
"loss": 0.6025,
|
| 2522 |
+
"step": 3510
|
| 2523 |
+
},
|
| 2524 |
+
{
|
| 2525 |
+
"epoch": 0.09,
|
| 2526 |
+
"grad_norm": 11.875,
|
| 2527 |
+
"learning_rate": 1.3642105263157897e-05,
|
| 2528 |
+
"loss": 0.6137,
|
| 2529 |
+
"step": 3520
|
| 2530 |
+
},
|
| 2531 |
+
{
|
| 2532 |
+
"epoch": 0.09,
|
| 2533 |
+
"grad_norm": 22.25,
|
| 2534 |
+
"learning_rate": 1.3621052631578948e-05,
|
| 2535 |
+
"loss": 0.5813,
|
| 2536 |
+
"step": 3530
|
| 2537 |
+
},
|
| 2538 |
+
{
|
| 2539 |
+
"epoch": 0.09,
|
| 2540 |
+
"grad_norm": 4.125,
|
| 2541 |
+
"learning_rate": 1.3600000000000002e-05,
|
| 2542 |
+
"loss": 0.4769,
|
| 2543 |
+
"step": 3540
|
| 2544 |
+
},
|
| 2545 |
+
{
|
| 2546 |
+
"epoch": 0.09,
|
| 2547 |
+
"grad_norm": 7.03125,
|
| 2548 |
+
"learning_rate": 1.3578947368421055e-05,
|
| 2549 |
+
"loss": 0.7036,
|
| 2550 |
+
"step": 3550
|
| 2551 |
+
},
|
| 2552 |
+
{
|
| 2553 |
+
"epoch": 0.09,
|
| 2554 |
+
"grad_norm": 4.5625,
|
| 2555 |
+
"learning_rate": 1.3557894736842106e-05,
|
| 2556 |
+
"loss": 0.6237,
|
| 2557 |
+
"step": 3560
|
| 2558 |
+
},
|
| 2559 |
+
{
|
| 2560 |
+
"epoch": 0.09,
|
| 2561 |
+
"grad_norm": 2.796875,
|
| 2562 |
+
"learning_rate": 1.353684210526316e-05,
|
| 2563 |
+
"loss": 0.5743,
|
| 2564 |
+
"step": 3570
|
| 2565 |
+
},
|
| 2566 |
+
{
|
| 2567 |
+
"epoch": 0.09,
|
| 2568 |
+
"grad_norm": 4.34375,
|
| 2569 |
+
"learning_rate": 1.3515789473684211e-05,
|
| 2570 |
+
"loss": 0.6427,
|
| 2571 |
+
"step": 3580
|
| 2572 |
+
},
|
| 2573 |
+
{
|
| 2574 |
+
"epoch": 0.09,
|
| 2575 |
+
"grad_norm": 12.5,
|
| 2576 |
+
"learning_rate": 1.3494736842105265e-05,
|
| 2577 |
+
"loss": 0.5552,
|
| 2578 |
+
"step": 3590
|
| 2579 |
+
},
|
| 2580 |
+
{
|
| 2581 |
+
"epoch": 0.09,
|
| 2582 |
+
"grad_norm": 4.0625,
|
| 2583 |
+
"learning_rate": 1.3473684210526316e-05,
|
| 2584 |
+
"loss": 0.6391,
|
| 2585 |
+
"step": 3600
|
| 2586 |
+
},
|
| 2587 |
+
{
|
| 2588 |
+
"epoch": 0.09,
|
| 2589 |
+
"grad_norm": 6.125,
|
| 2590 |
+
"learning_rate": 1.345263157894737e-05,
|
| 2591 |
+
"loss": 0.5062,
|
| 2592 |
+
"step": 3610
|
| 2593 |
+
},
|
| 2594 |
+
{
|
| 2595 |
+
"epoch": 0.09,
|
| 2596 |
+
"grad_norm": 4.375,
|
| 2597 |
+
"learning_rate": 1.3431578947368421e-05,
|
| 2598 |
+
"loss": 0.6708,
|
| 2599 |
+
"step": 3620
|
| 2600 |
+
},
|
| 2601 |
+
{
|
| 2602 |
+
"epoch": 0.09,
|
| 2603 |
+
"grad_norm": 2.078125,
|
| 2604 |
+
"learning_rate": 1.3410526315789474e-05,
|
| 2605 |
+
"loss": 0.5664,
|
| 2606 |
+
"step": 3630
|
| 2607 |
+
},
|
| 2608 |
+
{
|
| 2609 |
+
"epoch": 0.09,
|
| 2610 |
+
"grad_norm": 5.90625,
|
| 2611 |
+
"learning_rate": 1.3389473684210528e-05,
|
| 2612 |
+
"loss": 0.6353,
|
| 2613 |
+
"step": 3640
|
| 2614 |
+
},
|
| 2615 |
+
{
|
| 2616 |
+
"epoch": 0.09,
|
| 2617 |
+
"grad_norm": 3.0,
|
| 2618 |
+
"learning_rate": 1.336842105263158e-05,
|
| 2619 |
+
"loss": 0.7542,
|
| 2620 |
+
"step": 3650
|
| 2621 |
+
},
|
| 2622 |
+
{
|
| 2623 |
+
"epoch": 0.09,
|
| 2624 |
+
"grad_norm": 1.859375,
|
| 2625 |
+
"learning_rate": 1.3347368421052634e-05,
|
| 2626 |
+
"loss": 0.5818,
|
| 2627 |
+
"step": 3660
|
| 2628 |
+
},
|
| 2629 |
+
{
|
| 2630 |
+
"epoch": 0.09,
|
| 2631 |
+
"grad_norm": 4.03125,
|
| 2632 |
+
"learning_rate": 1.3326315789473685e-05,
|
| 2633 |
+
"loss": 0.5839,
|
| 2634 |
+
"step": 3670
|
| 2635 |
+
},
|
| 2636 |
+
{
|
| 2637 |
+
"epoch": 0.09,
|
| 2638 |
+
"grad_norm": 2.421875,
|
| 2639 |
+
"learning_rate": 1.3305263157894739e-05,
|
| 2640 |
+
"loss": 0.5676,
|
| 2641 |
+
"step": 3680
|
| 2642 |
+
},
|
| 2643 |
+
{
|
| 2644 |
+
"epoch": 0.09,
|
| 2645 |
+
"grad_norm": 4.9375,
|
| 2646 |
+
"learning_rate": 1.328421052631579e-05,
|
| 2647 |
+
"loss": 0.6292,
|
| 2648 |
+
"step": 3690
|
| 2649 |
+
},
|
| 2650 |
+
{
|
| 2651 |
+
"epoch": 0.09,
|
| 2652 |
+
"grad_norm": 5.34375,
|
| 2653 |
+
"learning_rate": 1.3263157894736843e-05,
|
| 2654 |
+
"loss": 0.5808,
|
| 2655 |
+
"step": 3700
|
| 2656 |
+
},
|
| 2657 |
+
{
|
| 2658 |
+
"epoch": 0.09,
|
| 2659 |
+
"grad_norm": 4.125,
|
| 2660 |
+
"learning_rate": 1.3242105263157895e-05,
|
| 2661 |
+
"loss": 0.7451,
|
| 2662 |
+
"step": 3710
|
| 2663 |
+
},
|
| 2664 |
+
{
|
| 2665 |
+
"epoch": 0.09,
|
| 2666 |
+
"grad_norm": 6.15625,
|
| 2667 |
+
"learning_rate": 1.3221052631578948e-05,
|
| 2668 |
+
"loss": 0.4822,
|
| 2669 |
+
"step": 3720
|
| 2670 |
+
},
|
| 2671 |
+
{
|
| 2672 |
+
"epoch": 0.09,
|
| 2673 |
+
"grad_norm": 3.671875,
|
| 2674 |
+
"learning_rate": 1.3200000000000002e-05,
|
| 2675 |
+
"loss": 0.5966,
|
| 2676 |
+
"step": 3730
|
| 2677 |
+
},
|
| 2678 |
+
{
|
| 2679 |
+
"epoch": 0.09,
|
| 2680 |
+
"grad_norm": 7.71875,
|
| 2681 |
+
"learning_rate": 1.3178947368421053e-05,
|
| 2682 |
+
"loss": 0.6074,
|
| 2683 |
+
"step": 3740
|
| 2684 |
+
},
|
| 2685 |
+
{
|
| 2686 |
+
"epoch": 0.09,
|
| 2687 |
+
"grad_norm": 3.703125,
|
| 2688 |
+
"learning_rate": 1.3157894736842108e-05,
|
| 2689 |
+
"loss": 0.5465,
|
| 2690 |
+
"step": 3750
|
| 2691 |
+
},
|
| 2692 |
+
{
|
| 2693 |
+
"epoch": 0.09,
|
| 2694 |
+
"grad_norm": 5.71875,
|
| 2695 |
+
"learning_rate": 1.3136842105263159e-05,
|
| 2696 |
+
"loss": 0.6226,
|
| 2697 |
+
"step": 3760
|
| 2698 |
+
},
|
| 2699 |
+
{
|
| 2700 |
+
"epoch": 0.09,
|
| 2701 |
+
"grad_norm": 2.9375,
|
| 2702 |
+
"learning_rate": 1.3115789473684211e-05,
|
| 2703 |
+
"loss": 0.6031,
|
| 2704 |
+
"step": 3770
|
| 2705 |
+
},
|
| 2706 |
+
{
|
| 2707 |
+
"epoch": 0.09,
|
| 2708 |
+
"grad_norm": 2.703125,
|
| 2709 |
+
"learning_rate": 1.3094736842105264e-05,
|
| 2710 |
+
"loss": 0.6693,
|
| 2711 |
+
"step": 3780
|
| 2712 |
+
},
|
| 2713 |
+
{
|
| 2714 |
+
"epoch": 0.09,
|
| 2715 |
+
"grad_norm": 4.5,
|
| 2716 |
+
"learning_rate": 1.3073684210526317e-05,
|
| 2717 |
+
"loss": 0.6126,
|
| 2718 |
+
"step": 3790
|
| 2719 |
+
},
|
| 2720 |
+
{
|
| 2721 |
+
"epoch": 0.1,
|
| 2722 |
+
"grad_norm": 3.96875,
|
| 2723 |
+
"learning_rate": 1.305263157894737e-05,
|
| 2724 |
+
"loss": 0.615,
|
| 2725 |
+
"step": 3800
|
| 2726 |
+
},
|
| 2727 |
+
{
|
| 2728 |
+
"epoch": 0.1,
|
| 2729 |
+
"grad_norm": 4.78125,
|
| 2730 |
+
"learning_rate": 1.3031578947368422e-05,
|
| 2731 |
+
"loss": 0.6213,
|
| 2732 |
+
"step": 3810
|
| 2733 |
+
},
|
| 2734 |
+
{
|
| 2735 |
+
"epoch": 0.1,
|
| 2736 |
+
"grad_norm": 7.9375,
|
| 2737 |
+
"learning_rate": 1.3010526315789476e-05,
|
| 2738 |
+
"loss": 0.5922,
|
| 2739 |
+
"step": 3820
|
| 2740 |
+
},
|
| 2741 |
+
{
|
| 2742 |
+
"epoch": 0.1,
|
| 2743 |
+
"grad_norm": 3.453125,
|
| 2744 |
+
"learning_rate": 1.2989473684210527e-05,
|
| 2745 |
+
"loss": 0.5783,
|
| 2746 |
+
"step": 3830
|
| 2747 |
+
},
|
| 2748 |
+
{
|
| 2749 |
+
"epoch": 0.1,
|
| 2750 |
+
"grad_norm": 5.03125,
|
| 2751 |
+
"learning_rate": 1.2968421052631578e-05,
|
| 2752 |
+
"loss": 0.5909,
|
| 2753 |
+
"step": 3840
|
| 2754 |
+
},
|
| 2755 |
+
{
|
| 2756 |
+
"epoch": 0.1,
|
| 2757 |
+
"grad_norm": 3.1875,
|
| 2758 |
+
"learning_rate": 1.2947368421052633e-05,
|
| 2759 |
+
"loss": 0.6543,
|
| 2760 |
+
"step": 3850
|
| 2761 |
+
},
|
| 2762 |
+
{
|
| 2763 |
+
"epoch": 0.1,
|
| 2764 |
+
"grad_norm": 3.625,
|
| 2765 |
+
"learning_rate": 1.2926315789473685e-05,
|
| 2766 |
+
"loss": 0.5652,
|
| 2767 |
+
"step": 3860
|
| 2768 |
+
},
|
| 2769 |
+
{
|
| 2770 |
+
"epoch": 0.1,
|
| 2771 |
+
"grad_norm": 11.75,
|
| 2772 |
+
"learning_rate": 1.2905263157894738e-05,
|
| 2773 |
+
"loss": 0.6078,
|
| 2774 |
+
"step": 3870
|
| 2775 |
+
},
|
| 2776 |
+
{
|
| 2777 |
+
"epoch": 0.1,
|
| 2778 |
+
"grad_norm": 3.828125,
|
| 2779 |
+
"learning_rate": 1.288421052631579e-05,
|
| 2780 |
+
"loss": 0.6246,
|
| 2781 |
+
"step": 3880
|
| 2782 |
+
},
|
| 2783 |
+
{
|
| 2784 |
+
"epoch": 0.1,
|
| 2785 |
+
"grad_norm": 4.15625,
|
| 2786 |
+
"learning_rate": 1.2863157894736845e-05,
|
| 2787 |
+
"loss": 0.5988,
|
| 2788 |
+
"step": 3890
|
| 2789 |
+
},
|
| 2790 |
+
{
|
| 2791 |
+
"epoch": 0.1,
|
| 2792 |
+
"grad_norm": 4.21875,
|
| 2793 |
+
"learning_rate": 1.2842105263157896e-05,
|
| 2794 |
+
"loss": 0.7113,
|
| 2795 |
+
"step": 3900
|
| 2796 |
+
},
|
| 2797 |
+
{
|
| 2798 |
+
"epoch": 0.1,
|
| 2799 |
+
"grad_norm": 7.15625,
|
| 2800 |
+
"learning_rate": 1.2821052631578947e-05,
|
| 2801 |
+
"loss": 0.5497,
|
| 2802 |
+
"step": 3910
|
| 2803 |
+
},
|
| 2804 |
+
{
|
| 2805 |
+
"epoch": 0.1,
|
| 2806 |
+
"grad_norm": 5.09375,
|
| 2807 |
+
"learning_rate": 1.2800000000000001e-05,
|
| 2808 |
+
"loss": 0.6188,
|
| 2809 |
+
"step": 3920
|
| 2810 |
+
},
|
| 2811 |
+
{
|
| 2812 |
+
"epoch": 0.1,
|
| 2813 |
+
"grad_norm": 4.3125,
|
| 2814 |
+
"learning_rate": 1.2778947368421054e-05,
|
| 2815 |
+
"loss": 0.6139,
|
| 2816 |
+
"step": 3930
|
| 2817 |
+
},
|
| 2818 |
+
{
|
| 2819 |
+
"epoch": 0.1,
|
| 2820 |
+
"grad_norm": 3.859375,
|
| 2821 |
+
"learning_rate": 1.2757894736842106e-05,
|
| 2822 |
+
"loss": 0.6298,
|
| 2823 |
+
"step": 3940
|
| 2824 |
+
},
|
| 2825 |
+
{
|
| 2826 |
+
"epoch": 0.1,
|
| 2827 |
+
"grad_norm": 2.953125,
|
| 2828 |
+
"learning_rate": 1.2736842105263159e-05,
|
| 2829 |
+
"loss": 0.5542,
|
| 2830 |
+
"step": 3950
|
| 2831 |
+
},
|
| 2832 |
+
{
|
| 2833 |
+
"epoch": 0.1,
|
| 2834 |
+
"grad_norm": 3.734375,
|
| 2835 |
+
"learning_rate": 1.2715789473684212e-05,
|
| 2836 |
+
"loss": 0.4915,
|
| 2837 |
+
"step": 3960
|
| 2838 |
+
},
|
| 2839 |
+
{
|
| 2840 |
+
"epoch": 0.1,
|
| 2841 |
+
"grad_norm": 6.8125,
|
| 2842 |
+
"learning_rate": 1.2694736842105264e-05,
|
| 2843 |
+
"loss": 0.6733,
|
| 2844 |
+
"step": 3970
|
| 2845 |
+
},
|
| 2846 |
+
{
|
| 2847 |
+
"epoch": 0.1,
|
| 2848 |
+
"grad_norm": 8.1875,
|
| 2849 |
+
"learning_rate": 1.2673684210526315e-05,
|
| 2850 |
+
"loss": 0.5441,
|
| 2851 |
+
"step": 3980
|
| 2852 |
+
},
|
| 2853 |
+
{
|
| 2854 |
+
"epoch": 0.1,
|
| 2855 |
+
"grad_norm": 4.28125,
|
| 2856 |
+
"learning_rate": 1.265263157894737e-05,
|
| 2857 |
+
"loss": 0.6336,
|
| 2858 |
+
"step": 3990
|
| 2859 |
+
},
|
| 2860 |
+
{
|
| 2861 |
+
"epoch": 0.1,
|
| 2862 |
+
"grad_norm": 4.09375,
|
| 2863 |
+
"learning_rate": 1.263157894736842e-05,
|
| 2864 |
+
"loss": 0.5396,
|
| 2865 |
+
"step": 4000
|
| 2866 |
+
},
|
| 2867 |
+
{
|
| 2868 |
+
"epoch": 0.1,
|
| 2869 |
+
"eval_loss": 0.6335766315460205,
|
| 2870 |
+
"eval_runtime": 38.1481,
|
| 2871 |
+
"eval_samples_per_second": 26.214,
|
| 2872 |
+
"eval_steps_per_second": 26.214,
|
| 2873 |
+
"step": 4000
|
| 2874 |
+
},
|
| 2875 |
+
{
|
| 2876 |
+
"epoch": 0.1,
|
| 2877 |
+
"grad_norm": 3.015625,
|
| 2878 |
+
"learning_rate": 1.2610526315789475e-05,
|
| 2879 |
+
"loss": 0.5444,
|
| 2880 |
+
"step": 4010
|
| 2881 |
+
},
|
| 2882 |
+
{
|
| 2883 |
+
"epoch": 0.1,
|
| 2884 |
+
"grad_norm": 4.8125,
|
| 2885 |
+
"learning_rate": 1.2589473684210528e-05,
|
| 2886 |
+
"loss": 0.5945,
|
| 2887 |
+
"step": 4020
|
| 2888 |
+
},
|
| 2889 |
+
{
|
| 2890 |
+
"epoch": 0.1,
|
| 2891 |
+
"grad_norm": 3.90625,
|
| 2892 |
+
"learning_rate": 1.256842105263158e-05,
|
| 2893 |
+
"loss": 0.5489,
|
| 2894 |
+
"step": 4030
|
| 2895 |
+
},
|
| 2896 |
+
{
|
| 2897 |
+
"epoch": 0.1,
|
| 2898 |
+
"grad_norm": 4.8125,
|
| 2899 |
+
"learning_rate": 1.2547368421052633e-05,
|
| 2900 |
+
"loss": 0.6619,
|
| 2901 |
+
"step": 4040
|
| 2902 |
+
},
|
| 2903 |
+
{
|
| 2904 |
+
"epoch": 0.1,
|
| 2905 |
+
"grad_norm": 4.15625,
|
| 2906 |
+
"learning_rate": 1.2526315789473684e-05,
|
| 2907 |
+
"loss": 0.6392,
|
| 2908 |
+
"step": 4050
|
| 2909 |
+
},
|
| 2910 |
+
{
|
| 2911 |
+
"epoch": 0.1,
|
| 2912 |
+
"grad_norm": 2.515625,
|
| 2913 |
+
"learning_rate": 1.2505263157894738e-05,
|
| 2914 |
+
"loss": 0.6275,
|
| 2915 |
+
"step": 4060
|
| 2916 |
+
},
|
| 2917 |
+
{
|
| 2918 |
+
"epoch": 0.1,
|
| 2919 |
+
"grad_norm": 4.0,
|
| 2920 |
+
"learning_rate": 1.248421052631579e-05,
|
| 2921 |
+
"loss": 0.6209,
|
| 2922 |
+
"step": 4070
|
| 2923 |
+
},
|
| 2924 |
+
{
|
| 2925 |
+
"epoch": 0.1,
|
| 2926 |
+
"grad_norm": 5.78125,
|
| 2927 |
+
"learning_rate": 1.2463157894736844e-05,
|
| 2928 |
+
"loss": 0.6417,
|
| 2929 |
+
"step": 4080
|
| 2930 |
+
},
|
| 2931 |
+
{
|
| 2932 |
+
"epoch": 0.1,
|
| 2933 |
+
"grad_norm": 6.59375,
|
| 2934 |
+
"learning_rate": 1.2442105263157895e-05,
|
| 2935 |
+
"loss": 0.5294,
|
| 2936 |
+
"step": 4090
|
| 2937 |
+
},
|
| 2938 |
+
{
|
| 2939 |
+
"epoch": 0.1,
|
| 2940 |
+
"grad_norm": 2.921875,
|
| 2941 |
+
"learning_rate": 1.2421052631578949e-05,
|
| 2942 |
+
"loss": 0.7045,
|
| 2943 |
+
"step": 4100
|
| 2944 |
+
},
|
| 2945 |
+
{
|
| 2946 |
+
"epoch": 0.1,
|
| 2947 |
+
"grad_norm": 2.328125,
|
| 2948 |
+
"learning_rate": 1.2400000000000002e-05,
|
| 2949 |
+
"loss": 0.5133,
|
| 2950 |
+
"step": 4110
|
| 2951 |
+
},
|
| 2952 |
+
{
|
| 2953 |
+
"epoch": 0.1,
|
| 2954 |
+
"grad_norm": 4.21875,
|
| 2955 |
+
"learning_rate": 1.2378947368421053e-05,
|
| 2956 |
+
"loss": 0.5812,
|
| 2957 |
+
"step": 4120
|
| 2958 |
+
},
|
| 2959 |
+
{
|
| 2960 |
+
"epoch": 0.1,
|
| 2961 |
+
"grad_norm": 4.21875,
|
| 2962 |
+
"learning_rate": 1.2357894736842107e-05,
|
| 2963 |
+
"loss": 0.5628,
|
| 2964 |
+
"step": 4130
|
| 2965 |
+
},
|
| 2966 |
+
{
|
| 2967 |
+
"epoch": 0.1,
|
| 2968 |
+
"grad_norm": 3.828125,
|
| 2969 |
+
"learning_rate": 1.2336842105263158e-05,
|
| 2970 |
+
"loss": 0.5895,
|
| 2971 |
+
"step": 4140
|
| 2972 |
+
},
|
| 2973 |
+
{
|
| 2974 |
+
"epoch": 0.1,
|
| 2975 |
+
"grad_norm": 3.734375,
|
| 2976 |
+
"learning_rate": 1.2315789473684212e-05,
|
| 2977 |
+
"loss": 0.6542,
|
| 2978 |
+
"step": 4150
|
| 2979 |
+
},
|
| 2980 |
+
{
|
| 2981 |
+
"epoch": 0.1,
|
| 2982 |
+
"grad_norm": 5.3125,
|
| 2983 |
+
"learning_rate": 1.2294736842105263e-05,
|
| 2984 |
+
"loss": 0.6245,
|
| 2985 |
+
"step": 4160
|
| 2986 |
+
},
|
| 2987 |
+
{
|
| 2988 |
+
"epoch": 0.1,
|
| 2989 |
+
"grad_norm": 2.90625,
|
| 2990 |
+
"learning_rate": 1.2273684210526317e-05,
|
| 2991 |
+
"loss": 0.6157,
|
| 2992 |
+
"step": 4170
|
| 2993 |
+
},
|
| 2994 |
+
{
|
| 2995 |
+
"epoch": 0.1,
|
| 2996 |
+
"grad_norm": 5.65625,
|
| 2997 |
+
"learning_rate": 1.225263157894737e-05,
|
| 2998 |
+
"loss": 0.602,
|
| 2999 |
+
"step": 4180
|
| 3000 |
+
},
|
| 3001 |
+
{
|
| 3002 |
+
"epoch": 0.1,
|
| 3003 |
+
"grad_norm": 4.875,
|
| 3004 |
+
"learning_rate": 1.2231578947368421e-05,
|
| 3005 |
+
"loss": 0.6496,
|
| 3006 |
+
"step": 4190
|
| 3007 |
+
},
|
| 3008 |
+
{
|
| 3009 |
+
"epoch": 0.1,
|
| 3010 |
+
"grad_norm": 2.34375,
|
| 3011 |
+
"learning_rate": 1.2210526315789475e-05,
|
| 3012 |
+
"loss": 0.6249,
|
| 3013 |
+
"step": 4200
|
| 3014 |
+
},
|
| 3015 |
+
{
|
| 3016 |
+
"epoch": 0.11,
|
| 3017 |
+
"grad_norm": 4.3125,
|
| 3018 |
+
"learning_rate": 1.2189473684210526e-05,
|
| 3019 |
+
"loss": 0.6666,
|
| 3020 |
+
"step": 4210
|
| 3021 |
+
},
|
| 3022 |
+
{
|
| 3023 |
+
"epoch": 0.11,
|
| 3024 |
+
"grad_norm": 4.6875,
|
| 3025 |
+
"learning_rate": 1.216842105263158e-05,
|
| 3026 |
+
"loss": 0.6132,
|
| 3027 |
+
"step": 4220
|
| 3028 |
+
},
|
| 3029 |
+
{
|
| 3030 |
+
"epoch": 0.11,
|
| 3031 |
+
"grad_norm": 5.59375,
|
| 3032 |
+
"learning_rate": 1.2147368421052632e-05,
|
| 3033 |
+
"loss": 0.5799,
|
| 3034 |
+
"step": 4230
|
| 3035 |
+
},
|
| 3036 |
+
{
|
| 3037 |
+
"epoch": 0.11,
|
| 3038 |
+
"grad_norm": 4.5625,
|
| 3039 |
+
"learning_rate": 1.2126315789473686e-05,
|
| 3040 |
+
"loss": 0.6745,
|
| 3041 |
+
"step": 4240
|
| 3042 |
+
},
|
| 3043 |
+
{
|
| 3044 |
+
"epoch": 0.11,
|
| 3045 |
+
"grad_norm": 3.140625,
|
| 3046 |
+
"learning_rate": 1.2105263157894737e-05,
|
| 3047 |
+
"loss": 0.6618,
|
| 3048 |
+
"step": 4250
|
| 3049 |
+
},
|
| 3050 |
+
{
|
| 3051 |
+
"epoch": 0.11,
|
| 3052 |
+
"grad_norm": 4.84375,
|
| 3053 |
+
"learning_rate": 1.208421052631579e-05,
|
| 3054 |
+
"loss": 0.5125,
|
| 3055 |
+
"step": 4260
|
| 3056 |
+
},
|
| 3057 |
+
{
|
| 3058 |
+
"epoch": 0.11,
|
| 3059 |
+
"grad_norm": 2.75,
|
| 3060 |
+
"learning_rate": 1.2063157894736844e-05,
|
| 3061 |
+
"loss": 0.6907,
|
| 3062 |
+
"step": 4270
|
| 3063 |
+
},
|
| 3064 |
+
{
|
| 3065 |
+
"epoch": 0.11,
|
| 3066 |
+
"grad_norm": 6.28125,
|
| 3067 |
+
"learning_rate": 1.2042105263157895e-05,
|
| 3068 |
+
"loss": 0.5876,
|
| 3069 |
+
"step": 4280
|
| 3070 |
+
},
|
| 3071 |
+
{
|
| 3072 |
+
"epoch": 0.11,
|
| 3073 |
+
"grad_norm": 3.96875,
|
| 3074 |
+
"learning_rate": 1.202105263157895e-05,
|
| 3075 |
+
"loss": 0.5825,
|
| 3076 |
+
"step": 4290
|
| 3077 |
+
},
|
| 3078 |
+
{
|
| 3079 |
+
"epoch": 0.11,
|
| 3080 |
+
"grad_norm": 2.421875,
|
| 3081 |
+
"learning_rate": 1.2e-05,
|
| 3082 |
+
"loss": 0.624,
|
| 3083 |
+
"step": 4300
|
| 3084 |
+
},
|
| 3085 |
+
{
|
| 3086 |
+
"epoch": 0.11,
|
| 3087 |
+
"grad_norm": 3.0625,
|
| 3088 |
+
"learning_rate": 1.1978947368421055e-05,
|
| 3089 |
+
"loss": 0.6119,
|
| 3090 |
+
"step": 4310
|
| 3091 |
+
},
|
| 3092 |
+
{
|
| 3093 |
+
"epoch": 0.11,
|
| 3094 |
+
"grad_norm": 5.8125,
|
| 3095 |
+
"learning_rate": 1.1957894736842106e-05,
|
| 3096 |
+
"loss": 0.6331,
|
| 3097 |
+
"step": 4320
|
| 3098 |
+
},
|
| 3099 |
+
{
|
| 3100 |
+
"epoch": 0.11,
|
| 3101 |
+
"grad_norm": 4.59375,
|
| 3102 |
+
"learning_rate": 1.1936842105263158e-05,
|
| 3103 |
+
"loss": 0.5641,
|
| 3104 |
+
"step": 4330
|
| 3105 |
+
},
|
| 3106 |
+
{
|
| 3107 |
+
"epoch": 0.11,
|
| 3108 |
+
"grad_norm": 3.578125,
|
| 3109 |
+
"learning_rate": 1.1915789473684211e-05,
|
| 3110 |
+
"loss": 0.4946,
|
| 3111 |
+
"step": 4340
|
| 3112 |
+
},
|
| 3113 |
+
{
|
| 3114 |
+
"epoch": 0.11,
|
| 3115 |
+
"grad_norm": 4.84375,
|
| 3116 |
+
"learning_rate": 1.1894736842105264e-05,
|
| 3117 |
+
"loss": 0.5547,
|
| 3118 |
+
"step": 4350
|
| 3119 |
+
},
|
| 3120 |
+
{
|
| 3121 |
+
"epoch": 0.11,
|
| 3122 |
+
"grad_norm": 18.5,
|
| 3123 |
+
"learning_rate": 1.1873684210526318e-05,
|
| 3124 |
+
"loss": 0.8141,
|
| 3125 |
+
"step": 4360
|
| 3126 |
+
},
|
| 3127 |
+
{
|
| 3128 |
+
"epoch": 0.11,
|
| 3129 |
+
"grad_norm": 3.390625,
|
| 3130 |
+
"learning_rate": 1.1852631578947369e-05,
|
| 3131 |
+
"loss": 0.5756,
|
| 3132 |
+
"step": 4370
|
| 3133 |
+
},
|
| 3134 |
+
{
|
| 3135 |
+
"epoch": 0.11,
|
| 3136 |
+
"grad_norm": 3.65625,
|
| 3137 |
+
"learning_rate": 1.1831578947368423e-05,
|
| 3138 |
+
"loss": 0.5353,
|
| 3139 |
+
"step": 4380
|
| 3140 |
+
},
|
| 3141 |
+
{
|
| 3142 |
+
"epoch": 0.11,
|
| 3143 |
+
"grad_norm": 5.15625,
|
| 3144 |
+
"learning_rate": 1.1810526315789474e-05,
|
| 3145 |
+
"loss": 0.7419,
|
| 3146 |
+
"step": 4390
|
| 3147 |
+
},
|
| 3148 |
+
{
|
| 3149 |
+
"epoch": 0.11,
|
| 3150 |
+
"grad_norm": 4.5,
|
| 3151 |
+
"learning_rate": 1.1789473684210527e-05,
|
| 3152 |
+
"loss": 0.72,
|
| 3153 |
+
"step": 4400
|
| 3154 |
+
},
|
| 3155 |
+
{
|
| 3156 |
+
"epoch": 0.11,
|
| 3157 |
+
"grad_norm": 3.84375,
|
| 3158 |
+
"learning_rate": 1.176842105263158e-05,
|
| 3159 |
+
"loss": 0.6064,
|
| 3160 |
+
"step": 4410
|
| 3161 |
+
},
|
| 3162 |
+
{
|
| 3163 |
+
"epoch": 0.11,
|
| 3164 |
+
"grad_norm": 3.46875,
|
| 3165 |
+
"learning_rate": 1.1747368421052632e-05,
|
| 3166 |
+
"loss": 0.5641,
|
| 3167 |
+
"step": 4420
|
| 3168 |
+
},
|
| 3169 |
+
{
|
| 3170 |
+
"epoch": 0.11,
|
| 3171 |
+
"grad_norm": 9.125,
|
| 3172 |
+
"learning_rate": 1.1726315789473685e-05,
|
| 3173 |
+
"loss": 0.6017,
|
| 3174 |
+
"step": 4430
|
| 3175 |
+
},
|
| 3176 |
+
{
|
| 3177 |
+
"epoch": 0.11,
|
| 3178 |
+
"grad_norm": 6.65625,
|
| 3179 |
+
"learning_rate": 1.1705263157894737e-05,
|
| 3180 |
+
"loss": 0.5863,
|
| 3181 |
+
"step": 4440
|
| 3182 |
+
},
|
| 3183 |
+
{
|
| 3184 |
+
"epoch": 0.11,
|
| 3185 |
+
"grad_norm": 4.03125,
|
| 3186 |
+
"learning_rate": 1.1684210526315792e-05,
|
| 3187 |
+
"loss": 0.6267,
|
| 3188 |
+
"step": 4450
|
| 3189 |
+
},
|
| 3190 |
+
{
|
| 3191 |
+
"epoch": 0.11,
|
| 3192 |
+
"grad_norm": 3.421875,
|
| 3193 |
+
"learning_rate": 1.1663157894736843e-05,
|
| 3194 |
+
"loss": 0.5873,
|
| 3195 |
+
"step": 4460
|
| 3196 |
+
},
|
| 3197 |
+
{
|
| 3198 |
+
"epoch": 0.11,
|
| 3199 |
+
"grad_norm": 9.25,
|
| 3200 |
+
"learning_rate": 1.1642105263157897e-05,
|
| 3201 |
+
"loss": 0.5499,
|
| 3202 |
+
"step": 4470
|
| 3203 |
+
},
|
| 3204 |
+
{
|
| 3205 |
+
"epoch": 0.11,
|
| 3206 |
+
"grad_norm": 4.09375,
|
| 3207 |
+
"learning_rate": 1.1621052631578948e-05,
|
| 3208 |
+
"loss": 0.621,
|
| 3209 |
+
"step": 4480
|
| 3210 |
+
},
|
| 3211 |
+
{
|
| 3212 |
+
"epoch": 0.11,
|
| 3213 |
+
"grad_norm": 5.28125,
|
| 3214 |
+
"learning_rate": 1.16e-05,
|
| 3215 |
+
"loss": 0.5808,
|
| 3216 |
+
"step": 4490
|
| 3217 |
+
},
|
| 3218 |
+
{
|
| 3219 |
+
"epoch": 0.11,
|
| 3220 |
+
"grad_norm": 3.46875,
|
| 3221 |
+
"learning_rate": 1.1578947368421053e-05,
|
| 3222 |
+
"loss": 0.7199,
|
| 3223 |
+
"step": 4500
|
| 3224 |
+
},
|
| 3225 |
+
{
|
| 3226 |
+
"epoch": 0.11,
|
| 3227 |
+
"eval_loss": 0.6083159446716309,
|
| 3228 |
+
"eval_runtime": 38.1826,
|
| 3229 |
+
"eval_samples_per_second": 26.19,
|
| 3230 |
+
"eval_steps_per_second": 26.19,
|
| 3231 |
+
"step": 4500
|
| 3232 |
+
},
|
| 3233 |
+
{
|
| 3234 |
+
"epoch": 0.11,
|
| 3235 |
+
"grad_norm": 2.078125,
|
| 3236 |
+
"learning_rate": 1.1557894736842106e-05,
|
| 3237 |
+
"loss": 0.5804,
|
| 3238 |
+
"step": 4510
|
| 3239 |
+
},
|
| 3240 |
+
{
|
| 3241 |
+
"epoch": 0.11,
|
| 3242 |
+
"grad_norm": 4.65625,
|
| 3243 |
+
"learning_rate": 1.153684210526316e-05,
|
| 3244 |
+
"loss": 0.7276,
|
| 3245 |
+
"step": 4520
|
| 3246 |
+
},
|
| 3247 |
+
{
|
| 3248 |
+
"epoch": 0.11,
|
| 3249 |
+
"grad_norm": 2.90625,
|
| 3250 |
+
"learning_rate": 1.1515789473684211e-05,
|
| 3251 |
+
"loss": 0.6052,
|
| 3252 |
+
"step": 4530
|
| 3253 |
+
},
|
| 3254 |
+
{
|
| 3255 |
+
"epoch": 0.11,
|
| 3256 |
+
"grad_norm": 14.6875,
|
| 3257 |
+
"learning_rate": 1.1494736842105266e-05,
|
| 3258 |
+
"loss": 0.6354,
|
| 3259 |
+
"step": 4540
|
| 3260 |
+
},
|
| 3261 |
+
{
|
| 3262 |
+
"epoch": 0.11,
|
| 3263 |
+
"grad_norm": 5.03125,
|
| 3264 |
+
"learning_rate": 1.1473684210526317e-05,
|
| 3265 |
+
"loss": 0.7017,
|
| 3266 |
+
"step": 4550
|
| 3267 |
+
},
|
| 3268 |
+
{
|
| 3269 |
+
"epoch": 0.11,
|
| 3270 |
+
"grad_norm": 23.875,
|
| 3271 |
+
"learning_rate": 1.145263157894737e-05,
|
| 3272 |
+
"loss": 0.6074,
|
| 3273 |
+
"step": 4560
|
| 3274 |
+
},
|
| 3275 |
+
{
|
| 3276 |
+
"epoch": 0.11,
|
| 3277 |
+
"grad_norm": 5.5,
|
| 3278 |
+
"learning_rate": 1.1431578947368422e-05,
|
| 3279 |
+
"loss": 0.395,
|
| 3280 |
+
"step": 4570
|
| 3281 |
+
},
|
| 3282 |
+
{
|
| 3283 |
+
"epoch": 0.11,
|
| 3284 |
+
"grad_norm": 7.0,
|
| 3285 |
+
"learning_rate": 1.1410526315789475e-05,
|
| 3286 |
+
"loss": 0.5674,
|
| 3287 |
+
"step": 4580
|
| 3288 |
+
},
|
| 3289 |
+
{
|
| 3290 |
+
"epoch": 0.11,
|
| 3291 |
+
"grad_norm": 2.859375,
|
| 3292 |
+
"learning_rate": 1.1389473684210527e-05,
|
| 3293 |
+
"loss": 0.5567,
|
| 3294 |
+
"step": 4590
|
| 3295 |
+
},
|
| 3296 |
+
{
|
| 3297 |
+
"epoch": 0.12,
|
| 3298 |
+
"grad_norm": 2.640625,
|
| 3299 |
+
"learning_rate": 1.136842105263158e-05,
|
| 3300 |
+
"loss": 0.607,
|
| 3301 |
+
"step": 4600
|
| 3302 |
+
},
|
| 3303 |
+
{
|
| 3304 |
+
"epoch": 0.12,
|
| 3305 |
+
"grad_norm": 3.703125,
|
| 3306 |
+
"learning_rate": 1.1347368421052634e-05,
|
| 3307 |
+
"loss": 0.6487,
|
| 3308 |
+
"step": 4610
|
| 3309 |
+
},
|
| 3310 |
+
{
|
| 3311 |
+
"epoch": 0.12,
|
| 3312 |
+
"grad_norm": 3.421875,
|
| 3313 |
+
"learning_rate": 1.1326315789473685e-05,
|
| 3314 |
+
"loss": 0.5341,
|
| 3315 |
+
"step": 4620
|
| 3316 |
+
},
|
| 3317 |
+
{
|
| 3318 |
+
"epoch": 0.12,
|
| 3319 |
+
"grad_norm": 3.78125,
|
| 3320 |
+
"learning_rate": 1.1305263157894736e-05,
|
| 3321 |
+
"loss": 0.6164,
|
| 3322 |
+
"step": 4630
|
| 3323 |
+
},
|
| 3324 |
+
{
|
| 3325 |
+
"epoch": 0.12,
|
| 3326 |
+
"grad_norm": 5.375,
|
| 3327 |
+
"learning_rate": 1.128421052631579e-05,
|
| 3328 |
+
"loss": 0.4827,
|
| 3329 |
+
"step": 4640
|
| 3330 |
+
},
|
| 3331 |
+
{
|
| 3332 |
+
"epoch": 0.12,
|
| 3333 |
+
"grad_norm": 3.28125,
|
| 3334 |
+
"learning_rate": 1.1263157894736843e-05,
|
| 3335 |
+
"loss": 0.647,
|
| 3336 |
+
"step": 4650
|
| 3337 |
+
},
|
| 3338 |
+
{
|
| 3339 |
+
"epoch": 0.12,
|
| 3340 |
+
"grad_norm": 7.9375,
|
| 3341 |
+
"learning_rate": 1.1242105263157896e-05,
|
| 3342 |
+
"loss": 0.5681,
|
| 3343 |
+
"step": 4660
|
| 3344 |
+
},
|
| 3345 |
+
{
|
| 3346 |
+
"epoch": 0.12,
|
| 3347 |
+
"grad_norm": 3.890625,
|
| 3348 |
+
"learning_rate": 1.1221052631578949e-05,
|
| 3349 |
+
"loss": 0.5795,
|
| 3350 |
+
"step": 4670
|
| 3351 |
+
},
|
| 3352 |
+
{
|
| 3353 |
+
"epoch": 0.12,
|
| 3354 |
+
"grad_norm": 4.0,
|
| 3355 |
+
"learning_rate": 1.1200000000000001e-05,
|
| 3356 |
+
"loss": 0.6383,
|
| 3357 |
+
"step": 4680
|
| 3358 |
+
},
|
| 3359 |
+
{
|
| 3360 |
+
"epoch": 0.12,
|
| 3361 |
+
"grad_norm": 4.84375,
|
| 3362 |
+
"learning_rate": 1.1178947368421054e-05,
|
| 3363 |
+
"loss": 0.5925,
|
| 3364 |
+
"step": 4690
|
| 3365 |
+
},
|
| 3366 |
+
{
|
| 3367 |
+
"epoch": 0.12,
|
| 3368 |
+
"grad_norm": 10.625,
|
| 3369 |
+
"learning_rate": 1.1157894736842105e-05,
|
| 3370 |
+
"loss": 0.5283,
|
| 3371 |
+
"step": 4700
|
| 3372 |
+
},
|
| 3373 |
+
{
|
| 3374 |
+
"epoch": 0.12,
|
| 3375 |
+
"grad_norm": 2.671875,
|
| 3376 |
+
"learning_rate": 1.1136842105263159e-05,
|
| 3377 |
+
"loss": 0.7346,
|
| 3378 |
+
"step": 4710
|
| 3379 |
+
},
|
| 3380 |
+
{
|
| 3381 |
+
"epoch": 0.12,
|
| 3382 |
+
"grad_norm": 2.734375,
|
| 3383 |
+
"learning_rate": 1.111578947368421e-05,
|
| 3384 |
+
"loss": 0.6452,
|
| 3385 |
+
"step": 4720
|
| 3386 |
+
},
|
| 3387 |
+
{
|
| 3388 |
+
"epoch": 0.12,
|
| 3389 |
+
"grad_norm": 3.546875,
|
| 3390 |
+
"learning_rate": 1.1094736842105264e-05,
|
| 3391 |
+
"loss": 0.6553,
|
| 3392 |
+
"step": 4730
|
| 3393 |
+
},
|
| 3394 |
+
{
|
| 3395 |
+
"epoch": 0.12,
|
| 3396 |
+
"grad_norm": 3.046875,
|
| 3397 |
+
"learning_rate": 1.1073684210526317e-05,
|
| 3398 |
+
"loss": 0.5828,
|
| 3399 |
+
"step": 4740
|
| 3400 |
+
},
|
| 3401 |
+
{
|
| 3402 |
+
"epoch": 0.12,
|
| 3403 |
+
"grad_norm": 4.6875,
|
| 3404 |
+
"learning_rate": 1.105263157894737e-05,
|
| 3405 |
+
"loss": 0.5143,
|
| 3406 |
+
"step": 4750
|
| 3407 |
+
},
|
| 3408 |
+
{
|
| 3409 |
+
"epoch": 0.12,
|
| 3410 |
+
"grad_norm": 4.21875,
|
| 3411 |
+
"learning_rate": 1.1031578947368422e-05,
|
| 3412 |
+
"loss": 0.6772,
|
| 3413 |
+
"step": 4760
|
| 3414 |
+
},
|
| 3415 |
+
{
|
| 3416 |
+
"epoch": 0.12,
|
| 3417 |
+
"grad_norm": 3.421875,
|
| 3418 |
+
"learning_rate": 1.1010526315789473e-05,
|
| 3419 |
+
"loss": 0.6578,
|
| 3420 |
+
"step": 4770
|
| 3421 |
+
},
|
| 3422 |
+
{
|
| 3423 |
+
"epoch": 0.12,
|
| 3424 |
+
"grad_norm": 4.1875,
|
| 3425 |
+
"learning_rate": 1.0989473684210528e-05,
|
| 3426 |
+
"loss": 0.5947,
|
| 3427 |
+
"step": 4780
|
| 3428 |
+
},
|
| 3429 |
+
{
|
| 3430 |
+
"epoch": 0.12,
|
| 3431 |
+
"grad_norm": 4.9375,
|
| 3432 |
+
"learning_rate": 1.0968421052631579e-05,
|
| 3433 |
+
"loss": 0.6039,
|
| 3434 |
+
"step": 4790
|
| 3435 |
+
},
|
| 3436 |
+
{
|
| 3437 |
+
"epoch": 0.12,
|
| 3438 |
+
"grad_norm": 2.34375,
|
| 3439 |
+
"learning_rate": 1.0947368421052633e-05,
|
| 3440 |
+
"loss": 0.6658,
|
| 3441 |
+
"step": 4800
|
| 3442 |
+
},
|
| 3443 |
+
{
|
| 3444 |
+
"epoch": 0.12,
|
| 3445 |
+
"grad_norm": 8.0625,
|
| 3446 |
+
"learning_rate": 1.0926315789473686e-05,
|
| 3447 |
+
"loss": 0.4581,
|
| 3448 |
+
"step": 4810
|
| 3449 |
+
},
|
| 3450 |
+
{
|
| 3451 |
+
"epoch": 0.12,
|
| 3452 |
+
"grad_norm": 4.84375,
|
| 3453 |
+
"learning_rate": 1.0905263157894738e-05,
|
| 3454 |
+
"loss": 0.5935,
|
| 3455 |
+
"step": 4820
|
| 3456 |
+
},
|
| 3457 |
+
{
|
| 3458 |
+
"epoch": 0.12,
|
| 3459 |
+
"grad_norm": 4.25,
|
| 3460 |
+
"learning_rate": 1.0884210526315791e-05,
|
| 3461 |
+
"loss": 0.6028,
|
| 3462 |
+
"step": 4830
|
| 3463 |
+
},
|
| 3464 |
+
{
|
| 3465 |
+
"epoch": 0.12,
|
| 3466 |
+
"grad_norm": 3.859375,
|
| 3467 |
+
"learning_rate": 1.0863157894736842e-05,
|
| 3468 |
+
"loss": 0.6113,
|
| 3469 |
+
"step": 4840
|
| 3470 |
+
},
|
| 3471 |
+
{
|
| 3472 |
+
"epoch": 0.12,
|
| 3473 |
+
"grad_norm": 21.5,
|
| 3474 |
+
"learning_rate": 1.0842105263157896e-05,
|
| 3475 |
+
"loss": 0.575,
|
| 3476 |
+
"step": 4850
|
| 3477 |
+
},
|
| 3478 |
+
{
|
| 3479 |
+
"epoch": 0.12,
|
| 3480 |
+
"grad_norm": 6.0,
|
| 3481 |
+
"learning_rate": 1.0821052631578947e-05,
|
| 3482 |
+
"loss": 0.5985,
|
| 3483 |
+
"step": 4860
|
| 3484 |
+
},
|
| 3485 |
+
{
|
| 3486 |
+
"epoch": 0.12,
|
| 3487 |
+
"grad_norm": 3.140625,
|
| 3488 |
+
"learning_rate": 1.0800000000000002e-05,
|
| 3489 |
+
"loss": 0.643,
|
| 3490 |
+
"step": 4870
|
| 3491 |
+
},
|
| 3492 |
+
{
|
| 3493 |
+
"epoch": 0.12,
|
| 3494 |
+
"grad_norm": 10.6875,
|
| 3495 |
+
"learning_rate": 1.0778947368421053e-05,
|
| 3496 |
+
"loss": 0.6964,
|
| 3497 |
+
"step": 4880
|
| 3498 |
+
},
|
| 3499 |
+
{
|
| 3500 |
+
"epoch": 0.12,
|
| 3501 |
+
"grad_norm": 5.71875,
|
| 3502 |
+
"learning_rate": 1.0757894736842107e-05,
|
| 3503 |
+
"loss": 0.655,
|
| 3504 |
+
"step": 4890
|
| 3505 |
+
},
|
| 3506 |
+
{
|
| 3507 |
+
"epoch": 0.12,
|
| 3508 |
+
"grad_norm": 5.90625,
|
| 3509 |
+
"learning_rate": 1.073684210526316e-05,
|
| 3510 |
+
"loss": 0.6014,
|
| 3511 |
+
"step": 4900
|
| 3512 |
+
},
|
| 3513 |
+
{
|
| 3514 |
+
"epoch": 0.12,
|
| 3515 |
+
"grad_norm": 4.875,
|
| 3516 |
+
"learning_rate": 1.071578947368421e-05,
|
| 3517 |
+
"loss": 0.6139,
|
| 3518 |
+
"step": 4910
|
| 3519 |
+
},
|
| 3520 |
+
{
|
| 3521 |
+
"epoch": 0.12,
|
| 3522 |
+
"grad_norm": 3.890625,
|
| 3523 |
+
"learning_rate": 1.0694736842105265e-05,
|
| 3524 |
+
"loss": 0.5815,
|
| 3525 |
+
"step": 4920
|
| 3526 |
+
},
|
| 3527 |
+
{
|
| 3528 |
+
"epoch": 0.12,
|
| 3529 |
+
"grad_norm": 4.0,
|
| 3530 |
+
"learning_rate": 1.0673684210526316e-05,
|
| 3531 |
+
"loss": 0.6578,
|
| 3532 |
+
"step": 4930
|
| 3533 |
+
},
|
| 3534 |
+
{
|
| 3535 |
+
"epoch": 0.12,
|
| 3536 |
+
"grad_norm": 3.59375,
|
| 3537 |
+
"learning_rate": 1.065263157894737e-05,
|
| 3538 |
+
"loss": 0.5326,
|
| 3539 |
+
"step": 4940
|
| 3540 |
+
},
|
| 3541 |
+
{
|
| 3542 |
+
"epoch": 0.12,
|
| 3543 |
+
"grad_norm": 4.0,
|
| 3544 |
+
"learning_rate": 1.0631578947368421e-05,
|
| 3545 |
+
"loss": 0.5406,
|
| 3546 |
+
"step": 4950
|
| 3547 |
+
},
|
| 3548 |
+
{
|
| 3549 |
+
"epoch": 0.12,
|
| 3550 |
+
"grad_norm": 4.84375,
|
| 3551 |
+
"learning_rate": 1.0610526315789476e-05,
|
| 3552 |
+
"loss": 0.6739,
|
| 3553 |
+
"step": 4960
|
| 3554 |
+
},
|
| 3555 |
+
{
|
| 3556 |
+
"epoch": 0.12,
|
| 3557 |
+
"grad_norm": 3.8125,
|
| 3558 |
+
"learning_rate": 1.0589473684210526e-05,
|
| 3559 |
+
"loss": 0.5474,
|
| 3560 |
+
"step": 4970
|
| 3561 |
+
},
|
| 3562 |
+
{
|
| 3563 |
+
"epoch": 0.12,
|
| 3564 |
+
"grad_norm": 3.578125,
|
| 3565 |
+
"learning_rate": 1.0568421052631579e-05,
|
| 3566 |
+
"loss": 0.5629,
|
| 3567 |
+
"step": 4980
|
| 3568 |
+
},
|
| 3569 |
+
{
|
| 3570 |
+
"epoch": 0.12,
|
| 3571 |
+
"grad_norm": 3.1875,
|
| 3572 |
+
"learning_rate": 1.0547368421052633e-05,
|
| 3573 |
+
"loss": 0.7017,
|
| 3574 |
+
"step": 4990
|
| 3575 |
+
},
|
| 3576 |
+
{
|
| 3577 |
+
"epoch": 0.12,
|
| 3578 |
+
"grad_norm": 3.984375,
|
| 3579 |
+
"learning_rate": 1.0526315789473684e-05,
|
| 3580 |
+
"loss": 0.4835,
|
| 3581 |
+
"step": 5000
|
| 3582 |
+
},
|
| 3583 |
+
{
|
| 3584 |
+
"epoch": 0.12,
|
| 3585 |
+
"eval_loss": 0.6027244925498962,
|
| 3586 |
+
"eval_runtime": 38.1439,
|
| 3587 |
+
"eval_samples_per_second": 26.216,
|
| 3588 |
+
"eval_steps_per_second": 26.216,
|
| 3589 |
+
"step": 5000
|
| 3590 |
}
|
| 3591 |
],
|
| 3592 |
"logging_steps": 10,
|
|
|
|
| 3594 |
"num_input_tokens_seen": 0,
|
| 3595 |
"num_train_epochs": 1,
|
| 3596 |
"save_steps": 2500,
|
| 3597 |
+
"total_flos": 7.866849165312e+16,
|
| 3598 |
"train_batch_size": 1,
|
| 3599 |
"trial_name": null,
|
| 3600 |
"trial_params": null
|