Upload 17 files
Browse files- config.json +5 -5
- model.safetensors +2 -2
- optimizer.pt +3 -0
- rng_state_0.pth +3 -0
- rng_state_1.pth +3 -0
- rng_state_2.pth +3 -0
- rng_state_3.pth +3 -0
- rng_state_4.pth +3 -0
- rng_state_5.pth +3 -0
- rng_state_6.pth +3 -0
- rng_state_7.pth +3 -0
- scheduler.pt +3 -0
- trainer_state.json +1204 -0
- training_args.bin +1 -1
config.json
CHANGED
|
@@ -4,7 +4,7 @@
|
|
| 4 |
],
|
| 5 |
"attention_bias": false,
|
| 6 |
"attention_dropout": 0.0,
|
| 7 |
-
"attn_implementation": "
|
| 8 |
"bos_token_id": 50281,
|
| 9 |
"classifier_activation": "gelu",
|
| 10 |
"classifier_bias": false,
|
|
@@ -19,10 +19,10 @@
|
|
| 19 |
"global_rope_theta": 160000.0,
|
| 20 |
"gradient_checkpointing": false,
|
| 21 |
"hidden_activation": "gelu",
|
| 22 |
-
"hidden_size":
|
| 23 |
"initializer_cutoff_factor": 2.0,
|
| 24 |
"initializer_range": 0.02,
|
| 25 |
-
"intermediate_size":
|
| 26 |
"layer_norm_eps": 1e-05,
|
| 27 |
"local_attention": 128,
|
| 28 |
"local_rope_theta": 10000.0,
|
|
@@ -32,8 +32,8 @@
|
|
| 32 |
"model_type": "modernbert",
|
| 33 |
"norm_bias": false,
|
| 34 |
"norm_eps": 1e-05,
|
| 35 |
-
"num_attention_heads":
|
| 36 |
-
"num_hidden_layers":
|
| 37 |
"pad_token_id": 50283,
|
| 38 |
"position_embedding_type": "absolute",
|
| 39 |
"repad_logits_with_grad": false,
|
|
|
|
| 4 |
],
|
| 5 |
"attention_bias": false,
|
| 6 |
"attention_dropout": 0.0,
|
| 7 |
+
"attn_implementation": "flash_attention_2",
|
| 8 |
"bos_token_id": 50281,
|
| 9 |
"classifier_activation": "gelu",
|
| 10 |
"classifier_bias": false,
|
|
|
|
| 19 |
"global_rope_theta": 160000.0,
|
| 20 |
"gradient_checkpointing": false,
|
| 21 |
"hidden_activation": "gelu",
|
| 22 |
+
"hidden_size": 1024,
|
| 23 |
"initializer_cutoff_factor": 2.0,
|
| 24 |
"initializer_range": 0.02,
|
| 25 |
+
"intermediate_size": 2624,
|
| 26 |
"layer_norm_eps": 1e-05,
|
| 27 |
"local_attention": 128,
|
| 28 |
"local_rope_theta": 10000.0,
|
|
|
|
| 32 |
"model_type": "modernbert",
|
| 33 |
"norm_bias": false,
|
| 34 |
"norm_eps": 1e-05,
|
| 35 |
+
"num_attention_heads": 16,
|
| 36 |
+
"num_hidden_layers": 28,
|
| 37 |
"pad_token_id": 50283,
|
| 38 |
"position_embedding_type": "absolute",
|
| 39 |
"repad_logits_with_grad": false,
|
model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2fcb90912bf8a5e13d4b275773666d2c047446e5f2ab572b02420914e69cdc0a
|
| 3 |
+
size 1583544840
|
optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:29407eeb5877166477ccc802ca378d2427f6fc12b556ed65811b48d021cf923a
|
| 3 |
+
size 3167201739
|
rng_state_0.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bb8a92b05dc429a8daac57b03a7d725340f4368c8d3919ffd0fd6461ec882bed
|
| 3 |
+
size 16389
|
rng_state_1.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fd8b4544d9db71d3771f2b43d1a991ec09df88bab20455fa07341f09fb5538fe
|
| 3 |
+
size 16389
|
rng_state_2.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e2b5d1541f4d0714658a1d3c66ffb5f1175ce8f63e0e4dcb34193aced9be633c
|
| 3 |
+
size 16389
|
rng_state_3.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b90e92c618d316503e87a988aa0b4478ba069da501147e52ba1b999af50ad487
|
| 3 |
+
size 16389
|
rng_state_4.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c07e35c11a1223db0ef9e1b618af1d424de1a1bed22dd8adc63d677c685e9af4
|
| 3 |
+
size 16389
|
rng_state_5.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d0428266ef742db978b38c88ba497179f7022ed373fc70e615cd6da6e572e4d5
|
| 3 |
+
size 16389
|
rng_state_6.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9fa8b65aa847f86c3a7acabdb428b8eecab5f8470bf084835f79e37dbbd08991
|
| 3 |
+
size 16389
|
rng_state_7.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6ed360a8586302b4eab66d003f6bc0ec76b55101b54d825ee659f340e81d4afd
|
| 3 |
+
size 16389
|
scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e7a9c66bbb17dafafd22a57810bb9248986283807454f0cb603b90488f8220f2
|
| 3 |
+
size 1465
|
trainer_state.json
ADDED
|
@@ -0,0 +1,1204 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.4771220000954244,
|
| 6 |
+
"eval_steps": 1000,
|
| 7 |
+
"global_step": 15000,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"epoch": 0.003180813333969496,
|
| 14 |
+
"grad_norm": 1.571864128112793,
|
| 15 |
+
"learning_rate": 2.5142857142857143e-05,
|
| 16 |
+
"loss": 9.1183,
|
| 17 |
+
"step": 100
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 0.006361626667938992,
|
| 21 |
+
"grad_norm": 3.1469738483428955,
|
| 22 |
+
"learning_rate": 5.053968253968254e-05,
|
| 23 |
+
"loss": 6.8369,
|
| 24 |
+
"step": 200
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.009542440001908488,
|
| 28 |
+
"grad_norm": 1.7220861911773682,
|
| 29 |
+
"learning_rate": 7.593650793650794e-05,
|
| 30 |
+
"loss": 5.6821,
|
| 31 |
+
"step": 300
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"epoch": 0.012723253335877984,
|
| 35 |
+
"grad_norm": 1.6516789197921753,
|
| 36 |
+
"learning_rate": 7.999856221185598e-05,
|
| 37 |
+
"loss": 4.8522,
|
| 38 |
+
"step": 400
|
| 39 |
+
},
|
| 40 |
+
{
|
| 41 |
+
"epoch": 0.01590406666984748,
|
| 42 |
+
"grad_norm": 1.4884405136108398,
|
| 43 |
+
"learning_rate": 7.99931013821163e-05,
|
| 44 |
+
"loss": 4.4132,
|
| 45 |
+
"step": 500
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.019084880003816976,
|
| 49 |
+
"grad_norm": 1.4667552709579468,
|
| 50 |
+
"learning_rate": 7.998356590220083e-05,
|
| 51 |
+
"loss": 4.1364,
|
| 52 |
+
"step": 600
|
| 53 |
+
},
|
| 54 |
+
{
|
| 55 |
+
"epoch": 0.022265693337786472,
|
| 56 |
+
"grad_norm": 1.3577353954315186,
|
| 57 |
+
"learning_rate": 7.996995674362075e-05,
|
| 58 |
+
"loss": 3.9333,
|
| 59 |
+
"step": 700
|
| 60 |
+
},
|
| 61 |
+
{
|
| 62 |
+
"epoch": 0.025446506671755968,
|
| 63 |
+
"grad_norm": 1.5057123899459839,
|
| 64 |
+
"learning_rate": 7.995227529292917e-05,
|
| 65 |
+
"loss": 3.7778,
|
| 66 |
+
"step": 800
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"epoch": 0.028627320005725463,
|
| 70 |
+
"grad_norm": 1.2254068851470947,
|
| 71 |
+
"learning_rate": 7.993052335157994e-05,
|
| 72 |
+
"loss": 3.6373,
|
| 73 |
+
"step": 900
|
| 74 |
+
},
|
| 75 |
+
{
|
| 76 |
+
"epoch": 0.03180813333969496,
|
| 77 |
+
"grad_norm": 1.2805088758468628,
|
| 78 |
+
"learning_rate": 7.99047031357441e-05,
|
| 79 |
+
"loss": 3.5242,
|
| 80 |
+
"step": 1000
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"epoch": 0.03180813333969496,
|
| 84 |
+
"eval_loss": 3.466144561767578,
|
| 85 |
+
"eval_runtime": 83.4397,
|
| 86 |
+
"eval_samples_per_second": 193.11,
|
| 87 |
+
"eval_steps_per_second": 0.384,
|
| 88 |
+
"step": 1000
|
| 89 |
+
},
|
| 90 |
+
{
|
| 91 |
+
"epoch": 0.03498894667366446,
|
| 92 |
+
"grad_norm": 1.2110868692398071,
|
| 93 |
+
"learning_rate": 7.987481727608398e-05,
|
| 94 |
+
"loss": 3.4105,
|
| 95 |
+
"step": 1100
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"epoch": 0.03816976000763395,
|
| 99 |
+
"grad_norm": 1.2545825242996216,
|
| 100 |
+
"learning_rate": 7.984086881748532e-05,
|
| 101 |
+
"loss": 3.3196,
|
| 102 |
+
"step": 1200
|
| 103 |
+
},
|
| 104 |
+
{
|
| 105 |
+
"epoch": 0.04135057334160345,
|
| 106 |
+
"grad_norm": 1.193214774131775,
|
| 107 |
+
"learning_rate": 7.980286121874697e-05,
|
| 108 |
+
"loss": 3.2374,
|
| 109 |
+
"step": 1300
|
| 110 |
+
},
|
| 111 |
+
{
|
| 112 |
+
"epoch": 0.044531386675572944,
|
| 113 |
+
"grad_norm": 1.2139487266540527,
|
| 114 |
+
"learning_rate": 7.976079835222848e-05,
|
| 115 |
+
"loss": 3.1653,
|
| 116 |
+
"step": 1400
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"epoch": 0.04771220000954244,
|
| 120 |
+
"grad_norm": 1.1818259954452515,
|
| 121 |
+
"learning_rate": 7.971468450345564e-05,
|
| 122 |
+
"loss": 3.1029,
|
| 123 |
+
"step": 1500
|
| 124 |
+
},
|
| 125 |
+
{
|
| 126 |
+
"epoch": 0.050893013343511935,
|
| 127 |
+
"grad_norm": 1.1244895458221436,
|
| 128 |
+
"learning_rate": 7.966452437068377e-05,
|
| 129 |
+
"loss": 3.0479,
|
| 130 |
+
"step": 1600
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 0.05407382667748143,
|
| 134 |
+
"grad_norm": 1.1834404468536377,
|
| 135 |
+
"learning_rate": 7.961032306441911e-05,
|
| 136 |
+
"loss": 2.9967,
|
| 137 |
+
"step": 1700
|
| 138 |
+
},
|
| 139 |
+
{
|
| 140 |
+
"epoch": 0.057254640011450926,
|
| 141 |
+
"grad_norm": 1.0691077709197998,
|
| 142 |
+
"learning_rate": 7.95520861068981e-05,
|
| 143 |
+
"loss": 2.9565,
|
| 144 |
+
"step": 1800
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"epoch": 0.06043545334542042,
|
| 148 |
+
"grad_norm": 1.0717471837997437,
|
| 149 |
+
"learning_rate": 7.948981943152477e-05,
|
| 150 |
+
"loss": 2.9167,
|
| 151 |
+
"step": 1900
|
| 152 |
+
},
|
| 153 |
+
{
|
| 154 |
+
"epoch": 0.06361626667938992,
|
| 155 |
+
"grad_norm": 1.0454130172729492,
|
| 156 |
+
"learning_rate": 7.942352938226626e-05,
|
| 157 |
+
"loss": 2.8795,
|
| 158 |
+
"step": 2000
|
| 159 |
+
},
|
| 160 |
+
{
|
| 161 |
+
"epoch": 0.06361626667938992,
|
| 162 |
+
"eval_loss": 2.8664157390594482,
|
| 163 |
+
"eval_runtime": 74.1422,
|
| 164 |
+
"eval_samples_per_second": 217.326,
|
| 165 |
+
"eval_steps_per_second": 0.432,
|
| 166 |
+
"step": 2000
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"epoch": 0.06679708001335942,
|
| 170 |
+
"grad_norm": 1.0212640762329102,
|
| 171 |
+
"learning_rate": 7.935322271300637e-05,
|
| 172 |
+
"loss": 2.8447,
|
| 173 |
+
"step": 2100
|
| 174 |
+
},
|
| 175 |
+
{
|
| 176 |
+
"epoch": 0.06997789334732891,
|
| 177 |
+
"grad_norm": 0.9942176342010498,
|
| 178 |
+
"learning_rate": 7.927890658685757e-05,
|
| 179 |
+
"loss": 2.8107,
|
| 180 |
+
"step": 2200
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"epoch": 0.07315870668129841,
|
| 184 |
+
"grad_norm": 1.0249791145324707,
|
| 185 |
+
"learning_rate": 7.92005885754311e-05,
|
| 186 |
+
"loss": 2.7803,
|
| 187 |
+
"step": 2300
|
| 188 |
+
},
|
| 189 |
+
{
|
| 190 |
+
"epoch": 0.0763395200152679,
|
| 191 |
+
"grad_norm": 0.9678166508674622,
|
| 192 |
+
"learning_rate": 7.911827665806558e-05,
|
| 193 |
+
"loss": 2.7587,
|
| 194 |
+
"step": 2400
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
"epoch": 0.0795203333492374,
|
| 198 |
+
"grad_norm": 0.9619715213775635,
|
| 199 |
+
"learning_rate": 7.90319792210141e-05,
|
| 200 |
+
"loss": 2.7303,
|
| 201 |
+
"step": 2500
|
| 202 |
+
},
|
| 203 |
+
{
|
| 204 |
+
"epoch": 0.0827011466832069,
|
| 205 |
+
"grad_norm": 0.9805331826210022,
|
| 206 |
+
"learning_rate": 7.894170505658968e-05,
|
| 207 |
+
"loss": 2.706,
|
| 208 |
+
"step": 2600
|
| 209 |
+
},
|
| 210 |
+
{
|
| 211 |
+
"epoch": 0.08588196001717639,
|
| 212 |
+
"grad_norm": 1.0143381357192993,
|
| 213 |
+
"learning_rate": 7.884746336226957e-05,
|
| 214 |
+
"loss": 2.6882,
|
| 215 |
+
"step": 2700
|
| 216 |
+
},
|
| 217 |
+
{
|
| 218 |
+
"epoch": 0.08906277335114589,
|
| 219 |
+
"grad_norm": 0.9475836753845215,
|
| 220 |
+
"learning_rate": 7.874926373975814e-05,
|
| 221 |
+
"loss": 2.6674,
|
| 222 |
+
"step": 2800
|
| 223 |
+
},
|
| 224 |
+
{
|
| 225 |
+
"epoch": 0.09224358668511538,
|
| 226 |
+
"grad_norm": 0.9616538882255554,
|
| 227 |
+
"learning_rate": 7.864711619400865e-05,
|
| 228 |
+
"loss": 2.6405,
|
| 229 |
+
"step": 2900
|
| 230 |
+
},
|
| 231 |
+
{
|
| 232 |
+
"epoch": 0.09542440001908488,
|
| 233 |
+
"grad_norm": 0.9427627325057983,
|
| 234 |
+
"learning_rate": 7.854103113220384e-05,
|
| 235 |
+
"loss": 2.6261,
|
| 236 |
+
"step": 3000
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 0.09542440001908488,
|
| 240 |
+
"eval_loss": 2.6171655654907227,
|
| 241 |
+
"eval_runtime": 74.2572,
|
| 242 |
+
"eval_samples_per_second": 216.989,
|
| 243 |
+
"eval_steps_per_second": 0.431,
|
| 244 |
+
"step": 3000
|
| 245 |
+
},
|
| 246 |
+
{
|
| 247 |
+
"epoch": 0.09860521335305437,
|
| 248 |
+
"grad_norm": 0.9843456149101257,
|
| 249 |
+
"learning_rate": 7.843101936269565e-05,
|
| 250 |
+
"loss": 2.6075,
|
| 251 |
+
"step": 3100
|
| 252 |
+
},
|
| 253 |
+
{
|
| 254 |
+
"epoch": 0.10178602668702387,
|
| 255 |
+
"grad_norm": 0.9504939913749695,
|
| 256 |
+
"learning_rate": 7.831709209390408e-05,
|
| 257 |
+
"loss": 2.5906,
|
| 258 |
+
"step": 3200
|
| 259 |
+
},
|
| 260 |
+
{
|
| 261 |
+
"epoch": 0.10496684002099337,
|
| 262 |
+
"grad_norm": 0.9153879880905151,
|
| 263 |
+
"learning_rate": 7.819926093317513e-05,
|
| 264 |
+
"loss": 2.578,
|
| 265 |
+
"step": 3300
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.10814765335496286,
|
| 269 |
+
"grad_norm": 0.9085161685943604,
|
| 270 |
+
"learning_rate": 7.807753788559824e-05,
|
| 271 |
+
"loss": 2.56,
|
| 272 |
+
"step": 3400
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"epoch": 0.11132846668893236,
|
| 276 |
+
"grad_norm": 0.9289081692695618,
|
| 277 |
+
"learning_rate": 7.795193535278322e-05,
|
| 278 |
+
"loss": 2.5432,
|
| 279 |
+
"step": 3500
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.11450928002290185,
|
| 283 |
+
"grad_norm": 0.9010806083679199,
|
| 284 |
+
"learning_rate": 7.782246613159663e-05,
|
| 285 |
+
"loss": 2.5272,
|
| 286 |
+
"step": 3600
|
| 287 |
+
},
|
| 288 |
+
{
|
| 289 |
+
"epoch": 0.11769009335687135,
|
| 290 |
+
"grad_norm": 0.9009876847267151,
|
| 291 |
+
"learning_rate": 7.768914341285804e-05,
|
| 292 |
+
"loss": 2.5154,
|
| 293 |
+
"step": 3700
|
| 294 |
+
},
|
| 295 |
+
{
|
| 296 |
+
"epoch": 0.12087090669084084,
|
| 297 |
+
"grad_norm": 0.920699417591095,
|
| 298 |
+
"learning_rate": 7.755198077999615e-05,
|
| 299 |
+
"loss": 2.503,
|
| 300 |
+
"step": 3800
|
| 301 |
+
},
|
| 302 |
+
{
|
| 303 |
+
"epoch": 0.12405172002481034,
|
| 304 |
+
"grad_norm": 0.8694368600845337,
|
| 305 |
+
"learning_rate": 7.74109922076647e-05,
|
| 306 |
+
"loss": 2.4917,
|
| 307 |
+
"step": 3900
|
| 308 |
+
},
|
| 309 |
+
{
|
| 310 |
+
"epoch": 0.12723253335877985,
|
| 311 |
+
"grad_norm": 0.8826000094413757,
|
| 312 |
+
"learning_rate": 7.726619206031888e-05,
|
| 313 |
+
"loss": 2.4808,
|
| 314 |
+
"step": 4000
|
| 315 |
+
},
|
| 316 |
+
{
|
| 317 |
+
"epoch": 0.12723253335877985,
|
| 318 |
+
"eval_loss": 2.4718618392944336,
|
| 319 |
+
"eval_runtime": 74.2591,
|
| 320 |
+
"eval_samples_per_second": 216.984,
|
| 321 |
+
"eval_steps_per_second": 0.431,
|
| 322 |
+
"step": 4000
|
| 323 |
+
},
|
| 324 |
+
{
|
| 325 |
+
"epoch": 0.13041334669274934,
|
| 326 |
+
"grad_norm": 0.920994222164154,
|
| 327 |
+
"learning_rate": 7.711759509075166e-05,
|
| 328 |
+
"loss": 2.4688,
|
| 329 |
+
"step": 4100
|
| 330 |
+
},
|
| 331 |
+
{
|
| 332 |
+
"epoch": 0.13359416002671884,
|
| 333 |
+
"grad_norm": 0.8877164125442505,
|
| 334 |
+
"learning_rate": 7.696521643859082e-05,
|
| 335 |
+
"loss": 2.463,
|
| 336 |
+
"step": 4200
|
| 337 |
+
},
|
| 338 |
+
{
|
| 339 |
+
"epoch": 0.13677497336068833,
|
| 340 |
+
"grad_norm": 0.8722163438796997,
|
| 341 |
+
"learning_rate": 7.680907162875641e-05,
|
| 342 |
+
"loss": 2.4462,
|
| 343 |
+
"step": 4300
|
| 344 |
+
},
|
| 345 |
+
{
|
| 346 |
+
"epoch": 0.13995578669465783,
|
| 347 |
+
"grad_norm": 0.8788607716560364,
|
| 348 |
+
"learning_rate": 7.664917656987906e-05,
|
| 349 |
+
"loss": 2.435,
|
| 350 |
+
"step": 4400
|
| 351 |
+
},
|
| 352 |
+
{
|
| 353 |
+
"epoch": 0.14313660002862733,
|
| 354 |
+
"grad_norm": 0.8874791264533997,
|
| 355 |
+
"learning_rate": 7.648554755267907e-05,
|
| 356 |
+
"loss": 2.4276,
|
| 357 |
+
"step": 4500
|
| 358 |
+
},
|
| 359 |
+
{
|
| 360 |
+
"epoch": 0.14631741336259682,
|
| 361 |
+
"grad_norm": 0.921125054359436,
|
| 362 |
+
"learning_rate": 7.631820124830674e-05,
|
| 363 |
+
"loss": 2.4179,
|
| 364 |
+
"step": 4600
|
| 365 |
+
},
|
| 366 |
+
{
|
| 367 |
+
"epoch": 0.14949822669656632,
|
| 368 |
+
"grad_norm": 0.8657805323600769,
|
| 369 |
+
"learning_rate": 7.614715470664379e-05,
|
| 370 |
+
"loss": 2.4098,
|
| 371 |
+
"step": 4700
|
| 372 |
+
},
|
| 373 |
+
{
|
| 374 |
+
"epoch": 0.1526790400305358,
|
| 375 |
+
"grad_norm": 0.863875150680542,
|
| 376 |
+
"learning_rate": 7.597242535456632e-05,
|
| 377 |
+
"loss": 2.3982,
|
| 378 |
+
"step": 4800
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"epoch": 0.1558598533645053,
|
| 382 |
+
"grad_norm": 0.8500732779502869,
|
| 383 |
+
"learning_rate": 7.579403099416917e-05,
|
| 384 |
+
"loss": 2.3869,
|
| 385 |
+
"step": 4900
|
| 386 |
+
},
|
| 387 |
+
{
|
| 388 |
+
"epoch": 0.1590406666984748,
|
| 389 |
+
"grad_norm": 0.8550170063972473,
|
| 390 |
+
"learning_rate": 7.561198980095229e-05,
|
| 391 |
+
"loss": 2.3803,
|
| 392 |
+
"step": 5000
|
| 393 |
+
},
|
| 394 |
+
{
|
| 395 |
+
"epoch": 0.1590406666984748,
|
| 396 |
+
"eval_loss": 2.3734896183013916,
|
| 397 |
+
"eval_runtime": 74.3836,
|
| 398 |
+
"eval_samples_per_second": 216.62,
|
| 399 |
+
"eval_steps_per_second": 0.43,
|
| 400 |
+
"step": 5000
|
| 401 |
+
},
|
| 402 |
+
{
|
| 403 |
+
"epoch": 0.1622214800324443,
|
| 404 |
+
"grad_norm": 0.8349749445915222,
|
| 405 |
+
"learning_rate": 7.54263203219689e-05,
|
| 406 |
+
"loss": 2.3696,
|
| 407 |
+
"step": 5100
|
| 408 |
+
},
|
| 409 |
+
{
|
| 410 |
+
"epoch": 0.1654022933664138,
|
| 411 |
+
"grad_norm": 0.8674113750457764,
|
| 412 |
+
"learning_rate": 7.523704147393588e-05,
|
| 413 |
+
"loss": 2.364,
|
| 414 |
+
"step": 5200
|
| 415 |
+
},
|
| 416 |
+
{
|
| 417 |
+
"epoch": 0.1685831067003833,
|
| 418 |
+
"grad_norm": 0.865531861782074,
|
| 419 |
+
"learning_rate": 7.50441725413064e-05,
|
| 420 |
+
"loss": 2.3548,
|
| 421 |
+
"step": 5300
|
| 422 |
+
},
|
| 423 |
+
{
|
| 424 |
+
"epoch": 0.17176392003435278,
|
| 425 |
+
"grad_norm": 0.848044753074646,
|
| 426 |
+
"learning_rate": 7.484773317430521e-05,
|
| 427 |
+
"loss": 2.3488,
|
| 428 |
+
"step": 5400
|
| 429 |
+
},
|
| 430 |
+
{
|
| 431 |
+
"epoch": 0.17494473336832228,
|
| 432 |
+
"grad_norm": 0.8770771026611328,
|
| 433 |
+
"learning_rate": 7.464774338692655e-05,
|
| 434 |
+
"loss": 2.3418,
|
| 435 |
+
"step": 5500
|
| 436 |
+
},
|
| 437 |
+
{
|
| 438 |
+
"epoch": 0.17812554670229178,
|
| 439 |
+
"grad_norm": 0.8298258781433105,
|
| 440 |
+
"learning_rate": 7.44442235548951e-05,
|
| 441 |
+
"loss": 2.3362,
|
| 442 |
+
"step": 5600
|
| 443 |
+
},
|
| 444 |
+
{
|
| 445 |
+
"epoch": 0.18130636003626127,
|
| 446 |
+
"grad_norm": 0.844881534576416,
|
| 447 |
+
"learning_rate": 7.423719441358998e-05,
|
| 448 |
+
"loss": 2.3303,
|
| 449 |
+
"step": 5700
|
| 450 |
+
},
|
| 451 |
+
{
|
| 452 |
+
"epoch": 0.18448717337023077,
|
| 453 |
+
"grad_norm": 0.8413675427436829,
|
| 454 |
+
"learning_rate": 7.402667705593215e-05,
|
| 455 |
+
"loss": 2.3192,
|
| 456 |
+
"step": 5800
|
| 457 |
+
},
|
| 458 |
+
{
|
| 459 |
+
"epoch": 0.18766798670420026,
|
| 460 |
+
"grad_norm": 0.8078741431236267,
|
| 461 |
+
"learning_rate": 7.381269293023544e-05,
|
| 462 |
+
"loss": 2.3108,
|
| 463 |
+
"step": 5900
|
| 464 |
+
},
|
| 465 |
+
{
|
| 466 |
+
"epoch": 0.19084880003816976,
|
| 467 |
+
"grad_norm": 0.8276827931404114,
|
| 468 |
+
"learning_rate": 7.359526383802122e-05,
|
| 469 |
+
"loss": 2.3051,
|
| 470 |
+
"step": 6000
|
| 471 |
+
},
|
| 472 |
+
{
|
| 473 |
+
"epoch": 0.19084880003816976,
|
| 474 |
+
"eval_loss": 2.304142951965332,
|
| 475 |
+
"eval_runtime": 74.2683,
|
| 476 |
+
"eval_samples_per_second": 216.957,
|
| 477 |
+
"eval_steps_per_second": 0.431,
|
| 478 |
+
"step": 6000
|
| 479 |
+
},
|
| 480 |
+
{
|
| 481 |
+
"epoch": 0.19402961337213925,
|
| 482 |
+
"grad_norm": 0.86152184009552,
|
| 483 |
+
"learning_rate": 7.337441193179726e-05,
|
| 484 |
+
"loss": 2.2977,
|
| 485 |
+
"step": 6100
|
| 486 |
+
},
|
| 487 |
+
{
|
| 488 |
+
"epoch": 0.19721042670610875,
|
| 489 |
+
"grad_norm": 0.8484801650047302,
|
| 490 |
+
"learning_rate": 7.31501597128007e-05,
|
| 491 |
+
"loss": 2.2929,
|
| 492 |
+
"step": 6200
|
| 493 |
+
},
|
| 494 |
+
{
|
| 495 |
+
"epoch": 0.20039124004007824,
|
| 496 |
+
"grad_norm": 0.8117874264717102,
|
| 497 |
+
"learning_rate": 7.292253002870558e-05,
|
| 498 |
+
"loss": 2.2837,
|
| 499 |
+
"step": 6300
|
| 500 |
+
},
|
| 501 |
+
{
|
| 502 |
+
"epoch": 0.20357205337404774,
|
| 503 |
+
"grad_norm": 0.8349144458770752,
|
| 504 |
+
"learning_rate": 7.269154607129499e-05,
|
| 505 |
+
"loss": 2.2795,
|
| 506 |
+
"step": 6400
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.20675286670801724,
|
| 510 |
+
"grad_norm": 0.8227733373641968,
|
| 511 |
+
"learning_rate": 7.24572313740982e-05,
|
| 512 |
+
"loss": 2.2722,
|
| 513 |
+
"step": 6500
|
| 514 |
+
},
|
| 515 |
+
{
|
| 516 |
+
"epoch": 0.20993368004198673,
|
| 517 |
+
"grad_norm": 0.7939384579658508,
|
| 518 |
+
"learning_rate": 7.221960980999302e-05,
|
| 519 |
+
"loss": 2.271,
|
| 520 |
+
"step": 6600
|
| 521 |
+
},
|
| 522 |
+
{
|
| 523 |
+
"epoch": 0.21311449337595623,
|
| 524 |
+
"grad_norm": 0.8117570877075195,
|
| 525 |
+
"learning_rate": 7.19787055887735e-05,
|
| 526 |
+
"loss": 2.2645,
|
| 527 |
+
"step": 6700
|
| 528 |
+
},
|
| 529 |
+
{
|
| 530 |
+
"epoch": 0.21629530670992572,
|
| 531 |
+
"grad_norm": 0.8136454224586487,
|
| 532 |
+
"learning_rate": 7.173454325468337e-05,
|
| 533 |
+
"loss": 2.2569,
|
| 534 |
+
"step": 6800
|
| 535 |
+
},
|
| 536 |
+
{
|
| 537 |
+
"epoch": 0.21947612004389522,
|
| 538 |
+
"grad_norm": 0.7970232367515564,
|
| 539 |
+
"learning_rate": 7.148714768391532e-05,
|
| 540 |
+
"loss": 2.2515,
|
| 541 |
+
"step": 6900
|
| 542 |
+
},
|
| 543 |
+
{
|
| 544 |
+
"epoch": 0.2226569333778647,
|
| 545 |
+
"grad_norm": 0.8086152672767639,
|
| 546 |
+
"learning_rate": 7.123654408207664e-05,
|
| 547 |
+
"loss": 2.246,
|
| 548 |
+
"step": 7000
|
| 549 |
+
},
|
| 550 |
+
{
|
| 551 |
+
"epoch": 0.2226569333778647,
|
| 552 |
+
"eval_loss": 2.243811845779419,
|
| 553 |
+
"eval_runtime": 74.3461,
|
| 554 |
+
"eval_samples_per_second": 216.729,
|
| 555 |
+
"eval_steps_per_second": 0.43,
|
| 556 |
+
"step": 7000
|
| 557 |
+
},
|
| 558 |
+
{
|
| 559 |
+
"epoch": 0.2258377467118342,
|
| 560 |
+
"grad_norm": 0.8161323666572571,
|
| 561 |
+
"learning_rate": 7.098275798162101e-05,
|
| 562 |
+
"loss": 2.2429,
|
| 563 |
+
"step": 7100
|
| 564 |
+
},
|
| 565 |
+
{
|
| 566 |
+
"epoch": 0.2290185600458037,
|
| 567 |
+
"grad_norm": 0.804811418056488,
|
| 568 |
+
"learning_rate": 7.072581523924731e-05,
|
| 569 |
+
"loss": 2.2357,
|
| 570 |
+
"step": 7200
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 0.2321993733797732,
|
| 574 |
+
"grad_norm": 0.8152891993522644,
|
| 575 |
+
"learning_rate": 7.046574203326515e-05,
|
| 576 |
+
"loss": 2.229,
|
| 577 |
+
"step": 7300
|
| 578 |
+
},
|
| 579 |
+
{
|
| 580 |
+
"epoch": 0.2353801867137427,
|
| 581 |
+
"grad_norm": 0.8120368719100952,
|
| 582 |
+
"learning_rate": 7.020256486092776e-05,
|
| 583 |
+
"loss": 2.2261,
|
| 584 |
+
"step": 7400
|
| 585 |
+
},
|
| 586 |
+
{
|
| 587 |
+
"epoch": 0.2385610000477122,
|
| 588 |
+
"grad_norm": 0.7821657061576843,
|
| 589 |
+
"learning_rate": 6.99363105357323e-05,
|
| 590 |
+
"loss": 2.2212,
|
| 591 |
+
"step": 7500
|
| 592 |
+
},
|
| 593 |
+
{
|
| 594 |
+
"epoch": 0.24174181338168169,
|
| 595 |
+
"grad_norm": 0.7894716858863831,
|
| 596 |
+
"learning_rate": 6.96670061846881e-05,
|
| 597 |
+
"loss": 2.217,
|
| 598 |
+
"step": 7600
|
| 599 |
+
},
|
| 600 |
+
{
|
| 601 |
+
"epoch": 0.24492262671565118,
|
| 602 |
+
"grad_norm": 0.7840438485145569,
|
| 603 |
+
"learning_rate": 6.939467924555274e-05,
|
| 604 |
+
"loss": 2.2139,
|
| 605 |
+
"step": 7700
|
| 606 |
+
},
|
| 607 |
+
{
|
| 608 |
+
"epoch": 0.24810344004962068,
|
| 609 |
+
"grad_norm": 0.7956035137176514,
|
| 610 |
+
"learning_rate": 6.911935746403668e-05,
|
| 611 |
+
"loss": 2.2013,
|
| 612 |
+
"step": 7800
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.2512842533835902,
|
| 616 |
+
"grad_norm": 0.7977150082588196,
|
| 617 |
+
"learning_rate": 6.884106889097633e-05,
|
| 618 |
+
"loss": 2.2011,
|
| 619 |
+
"step": 7900
|
| 620 |
+
},
|
| 621 |
+
{
|
| 622 |
+
"epoch": 0.2544650667175597,
|
| 623 |
+
"grad_norm": 0.7777612805366516,
|
| 624 |
+
"learning_rate": 6.85598418794762e-05,
|
| 625 |
+
"loss": 2.1963,
|
| 626 |
+
"step": 8000
|
| 627 |
+
},
|
| 628 |
+
{
|
| 629 |
+
"epoch": 0.2544650667175597,
|
| 630 |
+
"eval_loss": 2.1914148330688477,
|
| 631 |
+
"eval_runtime": 74.2078,
|
| 632 |
+
"eval_samples_per_second": 217.134,
|
| 633 |
+
"eval_steps_per_second": 0.431,
|
| 634 |
+
"step": 8000
|
| 635 |
+
},
|
| 636 |
+
{
|
| 637 |
+
"epoch": 0.25764588005152916,
|
| 638 |
+
"grad_norm": 0.8219434022903442,
|
| 639 |
+
"learning_rate": 6.827570508202017e-05,
|
| 640 |
+
"loss": 2.1922,
|
| 641 |
+
"step": 8100
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"epoch": 0.2608266933854987,
|
| 645 |
+
"grad_norm": 0.8048291802406311,
|
| 646 |
+
"learning_rate": 6.798868744755221e-05,
|
| 647 |
+
"loss": 2.1835,
|
| 648 |
+
"step": 8200
|
| 649 |
+
},
|
| 650 |
+
{
|
| 651 |
+
"epoch": 0.26400750671946815,
|
| 652 |
+
"grad_norm": 0.8045642971992493,
|
| 653 |
+
"learning_rate": 6.7698818218527e-05,
|
| 654 |
+
"loss": 2.18,
|
| 655 |
+
"step": 8300
|
| 656 |
+
},
|
| 657 |
+
{
|
| 658 |
+
"epoch": 0.2671883200534377,
|
| 659 |
+
"grad_norm": 0.7891286611557007,
|
| 660 |
+
"learning_rate": 6.740612692793061e-05,
|
| 661 |
+
"loss": 2.1796,
|
| 662 |
+
"step": 8400
|
| 663 |
+
},
|
| 664 |
+
{
|
| 665 |
+
"epoch": 0.27036913338740715,
|
| 666 |
+
"grad_norm": 0.7675222158432007,
|
| 667 |
+
"learning_rate": 6.711064339627148e-05,
|
| 668 |
+
"loss": 2.175,
|
| 669 |
+
"step": 8500
|
| 670 |
+
},
|
| 671 |
+
{
|
| 672 |
+
"epoch": 0.27354994672137667,
|
| 673 |
+
"grad_norm": 0.7776646018028259,
|
| 674 |
+
"learning_rate": 6.68123977285423e-05,
|
| 675 |
+
"loss": 2.1669,
|
| 676 |
+
"step": 8600
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 0.27673076005534614,
|
| 680 |
+
"grad_norm": 0.7806198000907898,
|
| 681 |
+
"learning_rate": 6.651142031115273e-05,
|
| 682 |
+
"loss": 2.163,
|
| 683 |
+
"step": 8700
|
| 684 |
+
},
|
| 685 |
+
{
|
| 686 |
+
"epoch": 0.27991157338931566,
|
| 687 |
+
"grad_norm": 0.7961702942848206,
|
| 688 |
+
"learning_rate": 6.620774180883357e-05,
|
| 689 |
+
"loss": 2.1603,
|
| 690 |
+
"step": 8800
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"epoch": 0.2830923867232851,
|
| 694 |
+
"grad_norm": 0.7584977149963379,
|
| 695 |
+
"learning_rate": 6.590139316151244e-05,
|
| 696 |
+
"loss": 2.1585,
|
| 697 |
+
"step": 8900
|
| 698 |
+
},
|
| 699 |
+
{
|
| 700 |
+
"epoch": 0.28627320005725465,
|
| 701 |
+
"grad_norm": 0.7702009677886963,
|
| 702 |
+
"learning_rate": 6.559240558116156e-05,
|
| 703 |
+
"loss": 2.1564,
|
| 704 |
+
"step": 9000
|
| 705 |
+
},
|
| 706 |
+
{
|
| 707 |
+
"epoch": 0.28627320005725465,
|
| 708 |
+
"eval_loss": 2.152564525604248,
|
| 709 |
+
"eval_runtime": 74.312,
|
| 710 |
+
"eval_samples_per_second": 216.829,
|
| 711 |
+
"eval_steps_per_second": 0.431,
|
| 712 |
+
"step": 9000
|
| 713 |
+
},
|
| 714 |
+
{
|
| 715 |
+
"epoch": 0.2894540133912241,
|
| 716 |
+
"grad_norm": 0.764265239238739,
|
| 717 |
+
"learning_rate": 6.528081054861775e-05,
|
| 718 |
+
"loss": 2.1482,
|
| 719 |
+
"step": 9100
|
| 720 |
+
},
|
| 721 |
+
{
|
| 722 |
+
"epoch": 0.29263482672519364,
|
| 723 |
+
"grad_norm": 0.7961864471435547,
|
| 724 |
+
"learning_rate": 6.496663981037506e-05,
|
| 725 |
+
"loss": 2.1442,
|
| 726 |
+
"step": 9200
|
| 727 |
+
},
|
| 728 |
+
{
|
| 729 |
+
"epoch": 0.2958156400591631,
|
| 730 |
+
"grad_norm": 0.7772250175476074,
|
| 731 |
+
"learning_rate": 6.464992537535025e-05,
|
| 732 |
+
"loss": 2.1456,
|
| 733 |
+
"step": 9300
|
| 734 |
+
},
|
| 735 |
+
{
|
| 736 |
+
"epoch": 0.29899645339313263,
|
| 737 |
+
"grad_norm": 0.7651770114898682,
|
| 738 |
+
"learning_rate": 6.433069951162168e-05,
|
| 739 |
+
"loss": 2.1428,
|
| 740 |
+
"step": 9400
|
| 741 |
+
},
|
| 742 |
+
{
|
| 743 |
+
"epoch": 0.3021772667271021,
|
| 744 |
+
"grad_norm": 0.7702131867408752,
|
| 745 |
+
"learning_rate": 6.400899474314163e-05,
|
| 746 |
+
"loss": 2.1336,
|
| 747 |
+
"step": 9500
|
| 748 |
+
},
|
| 749 |
+
{
|
| 750 |
+
"epoch": 0.3053580800610716,
|
| 751 |
+
"grad_norm": 0.7733346819877625,
|
| 752 |
+
"learning_rate": 6.368484384642269e-05,
|
| 753 |
+
"loss": 2.1321,
|
| 754 |
+
"step": 9600
|
| 755 |
+
},
|
| 756 |
+
{
|
| 757 |
+
"epoch": 0.3085388933950411,
|
| 758 |
+
"grad_norm": 0.7849894762039185,
|
| 759 |
+
"learning_rate": 6.335827984719839e-05,
|
| 760 |
+
"loss": 2.1319,
|
| 761 |
+
"step": 9700
|
| 762 |
+
},
|
| 763 |
+
{
|
| 764 |
+
"epoch": 0.3117197067290106,
|
| 765 |
+
"grad_norm": 0.7571319341659546,
|
| 766 |
+
"learning_rate": 6.302933601705836e-05,
|
| 767 |
+
"loss": 2.1239,
|
| 768 |
+
"step": 9800
|
| 769 |
+
},
|
| 770 |
+
{
|
| 771 |
+
"epoch": 0.3149005200629801,
|
| 772 |
+
"grad_norm": 0.749380886554718,
|
| 773 |
+
"learning_rate": 6.269804587005846e-05,
|
| 774 |
+
"loss": 2.1172,
|
| 775 |
+
"step": 9900
|
| 776 |
+
},
|
| 777 |
+
{
|
| 778 |
+
"epoch": 0.3180813333969496,
|
| 779 |
+
"grad_norm": 0.7648999691009521,
|
| 780 |
+
"learning_rate": 6.236444315930636e-05,
|
| 781 |
+
"loss": 2.1154,
|
| 782 |
+
"step": 10000
|
| 783 |
+
},
|
| 784 |
+
{
|
| 785 |
+
"epoch": 0.3180813333969496,
|
| 786 |
+
"eval_loss": 2.114206075668335,
|
| 787 |
+
"eval_runtime": 74.2204,
|
| 788 |
+
"eval_samples_per_second": 217.097,
|
| 789 |
+
"eval_steps_per_second": 0.431,
|
| 790 |
+
"step": 10000
|
| 791 |
+
},
|
| 792 |
+
{
|
| 793 |
+
"epoch": 0.3212621467309191,
|
| 794 |
+
"grad_norm": 0.7476580142974854,
|
| 795 |
+
"learning_rate": 6.202856187352252e-05,
|
| 796 |
+
"loss": 2.1124,
|
| 797 |
+
"step": 10100
|
| 798 |
+
},
|
| 799 |
+
{
|
| 800 |
+
"epoch": 0.3244429600648886,
|
| 801 |
+
"grad_norm": 0.8032057285308838,
|
| 802 |
+
"learning_rate": 6.169043623357739e-05,
|
| 803 |
+
"loss": 2.111,
|
| 804 |
+
"step": 10200
|
| 805 |
+
},
|
| 806 |
+
{
|
| 807 |
+
"epoch": 0.32762377339885806,
|
| 808 |
+
"grad_norm": 0.7571981549263,
|
| 809 |
+
"learning_rate": 6.135010068900474e-05,
|
| 810 |
+
"loss": 2.1051,
|
| 811 |
+
"step": 10300
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"epoch": 0.3308045867328276,
|
| 815 |
+
"grad_norm": 0.7669061422348022,
|
| 816 |
+
"learning_rate": 6.100758991449196e-05,
|
| 817 |
+
"loss": 2.107,
|
| 818 |
+
"step": 10400
|
| 819 |
+
},
|
| 820 |
+
{
|
| 821 |
+
"epoch": 0.33398540006679706,
|
| 822 |
+
"grad_norm": 0.7569290399551392,
|
| 823 |
+
"learning_rate": 6.0662938806347125e-05,
|
| 824 |
+
"loss": 2.1036,
|
| 825 |
+
"step": 10500
|
| 826 |
+
},
|
| 827 |
+
{
|
| 828 |
+
"epoch": 0.3371662134007666,
|
| 829 |
+
"grad_norm": 0.764243483543396,
|
| 830 |
+
"learning_rate": 6.03161824789437e-05,
|
| 831 |
+
"loss": 2.0988,
|
| 832 |
+
"step": 10600
|
| 833 |
+
},
|
| 834 |
+
{
|
| 835 |
+
"epoch": 0.34034702673473605,
|
| 836 |
+
"grad_norm": 0.7787636518478394,
|
| 837 |
+
"learning_rate": 5.9967356261142986e-05,
|
| 838 |
+
"loss": 2.0955,
|
| 839 |
+
"step": 10700
|
| 840 |
+
},
|
| 841 |
+
{
|
| 842 |
+
"epoch": 0.34352784006870557,
|
| 843 |
+
"grad_norm": 0.7693697810173035,
|
| 844 |
+
"learning_rate": 5.961649569269457e-05,
|
| 845 |
+
"loss": 2.0928,
|
| 846 |
+
"step": 10800
|
| 847 |
+
},
|
| 848 |
+
{
|
| 849 |
+
"epoch": 0.34670865340267504,
|
| 850 |
+
"grad_norm": 0.7563138604164124,
|
| 851 |
+
"learning_rate": 5.9263636520615524e-05,
|
| 852 |
+
"loss": 2.0909,
|
| 853 |
+
"step": 10900
|
| 854 |
+
},
|
| 855 |
+
{
|
| 856 |
+
"epoch": 0.34988946673664456,
|
| 857 |
+
"grad_norm": 0.7634861469268799,
|
| 858 |
+
"learning_rate": 5.8908814695548266e-05,
|
| 859 |
+
"loss": 2.0856,
|
| 860 |
+
"step": 11000
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"epoch": 0.34988946673664456,
|
| 864 |
+
"eval_loss": 2.0881729125976562,
|
| 865 |
+
"eval_runtime": 74.282,
|
| 866 |
+
"eval_samples_per_second": 216.917,
|
| 867 |
+
"eval_steps_per_second": 0.431,
|
| 868 |
+
"step": 11000
|
| 869 |
+
},
|
| 870 |
+
{
|
| 871 |
+
"epoch": 0.35307028007061403,
|
| 872 |
+
"grad_norm": 0.751215934753418,
|
| 873 |
+
"learning_rate": 5.8552066368097823e-05,
|
| 874 |
+
"loss": 2.0809,
|
| 875 |
+
"step": 11100
|
| 876 |
+
},
|
| 877 |
+
{
|
| 878 |
+
"epoch": 0.35625109340458355,
|
| 879 |
+
"grad_norm": 0.743963897228241,
|
| 880 |
+
"learning_rate": 5.819342788514867e-05,
|
| 881 |
+
"loss": 2.0829,
|
| 882 |
+
"step": 11200
|
| 883 |
+
},
|
| 884 |
+
{
|
| 885 |
+
"epoch": 0.3594319067385531,
|
| 886 |
+
"grad_norm": 0.7820264101028442,
|
| 887 |
+
"learning_rate": 5.783293578616153e-05,
|
| 888 |
+
"loss": 2.0766,
|
| 889 |
+
"step": 11300
|
| 890 |
+
},
|
| 891 |
+
{
|
| 892 |
+
"epoch": 0.36261272007252254,
|
| 893 |
+
"grad_norm": 0.765643298625946,
|
| 894 |
+
"learning_rate": 5.747062679945063e-05,
|
| 895 |
+
"loss": 2.0698,
|
| 896 |
+
"step": 11400
|
| 897 |
+
},
|
| 898 |
+
{
|
| 899 |
+
"epoch": 0.36579353340649207,
|
| 900 |
+
"grad_norm": 0.7529120445251465,
|
| 901 |
+
"learning_rate": 5.710653783844168e-05,
|
| 902 |
+
"loss": 2.0716,
|
| 903 |
+
"step": 11500
|
| 904 |
+
},
|
| 905 |
+
{
|
| 906 |
+
"epoch": 0.36897434674046153,
|
| 907 |
+
"grad_norm": 0.7667486667633057,
|
| 908 |
+
"learning_rate": 5.6740705997910966e-05,
|
| 909 |
+
"loss": 2.0664,
|
| 910 |
+
"step": 11600
|
| 911 |
+
},
|
| 912 |
+
{
|
| 913 |
+
"epoch": 0.37215516007443106,
|
| 914 |
+
"grad_norm": 0.7487459182739258,
|
| 915 |
+
"learning_rate": 5.6373168550206016e-05,
|
| 916 |
+
"loss": 2.0637,
|
| 917 |
+
"step": 11700
|
| 918 |
+
},
|
| 919 |
+
{
|
| 920 |
+
"epoch": 0.3753359734084005,
|
| 921 |
+
"grad_norm": 0.7455624341964722,
|
| 922 |
+
"learning_rate": 5.600396294144813e-05,
|
| 923 |
+
"loss": 2.0648,
|
| 924 |
+
"step": 11800
|
| 925 |
+
},
|
| 926 |
+
{
|
| 927 |
+
"epoch": 0.37851678674237005,
|
| 928 |
+
"grad_norm": 0.7680267691612244,
|
| 929 |
+
"learning_rate": 5.5633126787717255e-05,
|
| 930 |
+
"loss": 2.0616,
|
| 931 |
+
"step": 11900
|
| 932 |
+
},
|
| 933 |
+
{
|
| 934 |
+
"epoch": 0.3816976000763395,
|
| 935 |
+
"grad_norm": 0.7449765205383301,
|
| 936 |
+
"learning_rate": 5.52606978712195e-05,
|
| 937 |
+
"loss": 2.0581,
|
| 938 |
+
"step": 12000
|
| 939 |
+
},
|
| 940 |
+
{
|
| 941 |
+
"epoch": 0.3816976000763395,
|
| 942 |
+
"eval_loss": 2.061724901199341,
|
| 943 |
+
"eval_runtime": 74.3191,
|
| 944 |
+
"eval_samples_per_second": 216.808,
|
| 945 |
+
"eval_steps_per_second": 0.431,
|
| 946 |
+
"step": 12000
|
| 947 |
+
},
|
| 948 |
+
{
|
| 949 |
+
"epoch": 0.38487841341030904,
|
| 950 |
+
"grad_norm": 0.7986348867416382,
|
| 951 |
+
"learning_rate": 5.4886714136437705e-05,
|
| 952 |
+
"loss": 2.0562,
|
| 953 |
+
"step": 12100
|
| 954 |
+
},
|
| 955 |
+
{
|
| 956 |
+
"epoch": 0.3880592267442785,
|
| 957 |
+
"grad_norm": 0.7683764100074768,
|
| 958 |
+
"learning_rate": 5.451121368626559e-05,
|
| 959 |
+
"loss": 2.0566,
|
| 960 |
+
"step": 12200
|
| 961 |
+
},
|
| 962 |
+
{
|
| 963 |
+
"epoch": 0.39124004007824803,
|
| 964 |
+
"grad_norm": 0.7579689621925354,
|
| 965 |
+
"learning_rate": 5.413423477812561e-05,
|
| 966 |
+
"loss": 2.0502,
|
| 967 |
+
"step": 12300
|
| 968 |
+
},
|
| 969 |
+
{
|
| 970 |
+
"epoch": 0.3944208534122175,
|
| 971 |
+
"grad_norm": 0.7543194890022278,
|
| 972 |
+
"learning_rate": 5.375581582007121e-05,
|
| 973 |
+
"loss": 2.0442,
|
| 974 |
+
"step": 12400
|
| 975 |
+
},
|
| 976 |
+
{
|
| 977 |
+
"epoch": 0.397601666746187,
|
| 978 |
+
"grad_norm": 0.7310113906860352,
|
| 979 |
+
"learning_rate": 5.337599536687357e-05,
|
| 980 |
+
"loss": 2.0436,
|
| 981 |
+
"step": 12500
|
| 982 |
+
},
|
| 983 |
+
{
|
| 984 |
+
"epoch": 0.4007824800801565,
|
| 985 |
+
"grad_norm": 0.7546029090881348,
|
| 986 |
+
"learning_rate": 5.2994812116093606e-05,
|
| 987 |
+
"loss": 2.0397,
|
| 988 |
+
"step": 12600
|
| 989 |
+
},
|
| 990 |
+
{
|
| 991 |
+
"epoch": 0.403963293414126,
|
| 992 |
+
"grad_norm": 0.7738301753997803,
|
| 993 |
+
"learning_rate": 5.261230490413925e-05,
|
| 994 |
+
"loss": 2.0403,
|
| 995 |
+
"step": 12700
|
| 996 |
+
},
|
| 997 |
+
{
|
| 998 |
+
"epoch": 0.4071441067480955,
|
| 999 |
+
"grad_norm": 0.7298979759216309,
|
| 1000 |
+
"learning_rate": 5.2228512702308684e-05,
|
| 1001 |
+
"loss": 2.0343,
|
| 1002 |
+
"step": 12800
|
| 1003 |
+
},
|
| 1004 |
+
{
|
| 1005 |
+
"epoch": 0.410324920082065,
|
| 1006 |
+
"grad_norm": 0.7821021676063538,
|
| 1007 |
+
"learning_rate": 5.1843474612819726e-05,
|
| 1008 |
+
"loss": 2.0361,
|
| 1009 |
+
"step": 12900
|
| 1010 |
+
},
|
| 1011 |
+
{
|
| 1012 |
+
"epoch": 0.41350573341603447,
|
| 1013 |
+
"grad_norm": 0.7907674312591553,
|
| 1014 |
+
"learning_rate": 5.145722986482604e-05,
|
| 1015 |
+
"loss": 2.0343,
|
| 1016 |
+
"step": 13000
|
| 1017 |
+
},
|
| 1018 |
+
{
|
| 1019 |
+
"epoch": 0.41350573341603447,
|
| 1020 |
+
"eval_loss": 2.0375359058380127,
|
| 1021 |
+
"eval_runtime": 74.169,
|
| 1022 |
+
"eval_samples_per_second": 217.247,
|
| 1023 |
+
"eval_steps_per_second": 0.431,
|
| 1024 |
+
"step": 13000
|
| 1025 |
+
},
|
| 1026 |
+
{
|
| 1027 |
+
"epoch": 0.416686546750004,
|
| 1028 |
+
"grad_norm": 0.7722309231758118,
|
| 1029 |
+
"learning_rate": 5.106981781042026e-05,
|
| 1030 |
+
"loss": 2.0349,
|
| 1031 |
+
"step": 13100
|
| 1032 |
+
},
|
| 1033 |
+
{
|
| 1034 |
+
"epoch": 0.41986736008397346,
|
| 1035 |
+
"grad_norm": 0.7431825995445251,
|
| 1036 |
+
"learning_rate": 5.068127792062466e-05,
|
| 1037 |
+
"loss": 2.0277,
|
| 1038 |
+
"step": 13200
|
| 1039 |
+
},
|
| 1040 |
+
{
|
| 1041 |
+
"epoch": 0.423048173417943,
|
| 1042 |
+
"grad_norm": 0.739189624786377,
|
| 1043 |
+
"learning_rate": 5.0291649781369725e-05,
|
| 1044 |
+
"loss": 2.0269,
|
| 1045 |
+
"step": 13300
|
| 1046 |
+
},
|
| 1047 |
+
{
|
| 1048 |
+
"epoch": 0.42622898675191245,
|
| 1049 |
+
"grad_norm": 0.7400371432304382,
|
| 1050 |
+
"learning_rate": 4.990097308946089e-05,
|
| 1051 |
+
"loss": 2.0251,
|
| 1052 |
+
"step": 13400
|
| 1053 |
+
},
|
| 1054 |
+
{
|
| 1055 |
+
"epoch": 0.429409800085882,
|
| 1056 |
+
"grad_norm": 0.7629276514053345,
|
| 1057 |
+
"learning_rate": 4.950928764853423e-05,
|
| 1058 |
+
"loss": 2.0235,
|
| 1059 |
+
"step": 13500
|
| 1060 |
+
},
|
| 1061 |
+
{
|
| 1062 |
+
"epoch": 0.43259061341985144,
|
| 1063 |
+
"grad_norm": 0.7603936195373535,
|
| 1064 |
+
"learning_rate": 4.911663336500097e-05,
|
| 1065 |
+
"loss": 2.0175,
|
| 1066 |
+
"step": 13600
|
| 1067 |
+
},
|
| 1068 |
+
{
|
| 1069 |
+
"epoch": 0.43577142675382097,
|
| 1070 |
+
"grad_norm": 0.7549597024917603,
|
| 1071 |
+
"learning_rate": 4.872305024398175e-05,
|
| 1072 |
+
"loss": 2.018,
|
| 1073 |
+
"step": 13700
|
| 1074 |
+
},
|
| 1075 |
+
{
|
| 1076 |
+
"epoch": 0.43895224008779044,
|
| 1077 |
+
"grad_norm": 0.7367106676101685,
|
| 1078 |
+
"learning_rate": 4.8328578385230744e-05,
|
| 1079 |
+
"loss": 2.016,
|
| 1080 |
+
"step": 13800
|
| 1081 |
+
},
|
| 1082 |
+
{
|
| 1083 |
+
"epoch": 0.44213305342175996,
|
| 1084 |
+
"grad_norm": 0.7753973007202148,
|
| 1085 |
+
"learning_rate": 4.7933257979050066e-05,
|
| 1086 |
+
"loss": 2.0103,
|
| 1087 |
+
"step": 13900
|
| 1088 |
+
},
|
| 1089 |
+
{
|
| 1090 |
+
"epoch": 0.4453138667557294,
|
| 1091 |
+
"grad_norm": 0.7437424063682556,
|
| 1092 |
+
"learning_rate": 4.7537129302195154e-05,
|
| 1093 |
+
"loss": 2.008,
|
| 1094 |
+
"step": 14000
|
| 1095 |
+
},
|
| 1096 |
+
{
|
| 1097 |
+
"epoch": 0.4453138667557294,
|
| 1098 |
+
"eval_loss": 2.004824161529541,
|
| 1099 |
+
"eval_runtime": 74.2767,
|
| 1100 |
+
"eval_samples_per_second": 216.932,
|
| 1101 |
+
"eval_steps_per_second": 0.431,
|
| 1102 |
+
"step": 14000
|
| 1103 |
+
},
|
| 1104 |
+
{
|
| 1105 |
+
"epoch": 0.44849468008969895,
|
| 1106 |
+
"grad_norm": 0.74482262134552,
|
| 1107 |
+
"learning_rate": 4.714023271377113e-05,
|
| 1108 |
+
"loss": 2.0059,
|
| 1109 |
+
"step": 14100
|
| 1110 |
+
},
|
| 1111 |
+
{
|
| 1112 |
+
"epoch": 0.4516754934236684,
|
| 1113 |
+
"grad_norm": 0.7460625767707825,
|
| 1114 |
+
"learning_rate": 4.674260865112088e-05,
|
| 1115 |
+
"loss": 2.0087,
|
| 1116 |
+
"step": 14200
|
| 1117 |
+
},
|
| 1118 |
+
{
|
| 1119 |
+
"epoch": 0.45485630675763794,
|
| 1120 |
+
"grad_norm": 0.7695781588554382,
|
| 1121 |
+
"learning_rate": 4.6344297625705144e-05,
|
| 1122 |
+
"loss": 2.0011,
|
| 1123 |
+
"step": 14300
|
| 1124 |
+
},
|
| 1125 |
+
{
|
| 1126 |
+
"epoch": 0.4580371200916074,
|
| 1127 |
+
"grad_norm": 0.7401705980300903,
|
| 1128 |
+
"learning_rate": 4.594534021897507e-05,
|
| 1129 |
+
"loss": 2.0021,
|
| 1130 |
+
"step": 14400
|
| 1131 |
+
},
|
| 1132 |
+
{
|
| 1133 |
+
"epoch": 0.46121793342557693,
|
| 1134 |
+
"grad_norm": 0.7603822946548462,
|
| 1135 |
+
"learning_rate": 4.554577707823759e-05,
|
| 1136 |
+
"loss": 1.9973,
|
| 1137 |
+
"step": 14500
|
| 1138 |
+
},
|
| 1139 |
+
{
|
| 1140 |
+
"epoch": 0.4643987467595464,
|
| 1141 |
+
"grad_norm": 0.7397817969322205,
|
| 1142 |
+
"learning_rate": 4.514564891251417e-05,
|
| 1143 |
+
"loss": 1.9998,
|
| 1144 |
+
"step": 14600
|
| 1145 |
+
},
|
| 1146 |
+
{
|
| 1147 |
+
"epoch": 0.4675795600935159,
|
| 1148 |
+
"grad_norm": 0.7768462300300598,
|
| 1149 |
+
"learning_rate": 4.474499648839313e-05,
|
| 1150 |
+
"loss": 1.9943,
|
| 1151 |
+
"step": 14700
|
| 1152 |
+
},
|
| 1153 |
+
{
|
| 1154 |
+
"epoch": 0.4707603734274854,
|
| 1155 |
+
"grad_norm": 0.7363395094871521,
|
| 1156 |
+
"learning_rate": 4.434386062587629e-05,
|
| 1157 |
+
"loss": 1.9959,
|
| 1158 |
+
"step": 14800
|
| 1159 |
+
},
|
| 1160 |
+
{
|
| 1161 |
+
"epoch": 0.4739411867614549,
|
| 1162 |
+
"grad_norm": 0.7700894474983215,
|
| 1163 |
+
"learning_rate": 4.3942282194219984e-05,
|
| 1164 |
+
"loss": 1.9934,
|
| 1165 |
+
"step": 14900
|
| 1166 |
+
},
|
| 1167 |
+
{
|
| 1168 |
+
"epoch": 0.4771220000954244,
|
| 1169 |
+
"grad_norm": 0.7271220088005066,
|
| 1170 |
+
"learning_rate": 4.3540302107771206e-05,
|
| 1171 |
+
"loss": 1.9899,
|
| 1172 |
+
"step": 15000
|
| 1173 |
+
},
|
| 1174 |
+
{
|
| 1175 |
+
"epoch": 0.4771220000954244,
|
| 1176 |
+
"eval_loss": 1.9911503791809082,
|
| 1177 |
+
"eval_runtime": 74.3028,
|
| 1178 |
+
"eval_samples_per_second": 216.856,
|
| 1179 |
+
"eval_steps_per_second": 0.431,
|
| 1180 |
+
"step": 15000
|
| 1181 |
+
}
|
| 1182 |
+
],
|
| 1183 |
+
"logging_steps": 100,
|
| 1184 |
+
"max_steps": 31439,
|
| 1185 |
+
"num_input_tokens_seen": 0,
|
| 1186 |
+
"num_train_epochs": 1,
|
| 1187 |
+
"save_steps": 5000,
|
| 1188 |
+
"stateful_callbacks": {
|
| 1189 |
+
"TrainerControl": {
|
| 1190 |
+
"args": {
|
| 1191 |
+
"should_epoch_stop": false,
|
| 1192 |
+
"should_evaluate": false,
|
| 1193 |
+
"should_log": false,
|
| 1194 |
+
"should_save": true,
|
| 1195 |
+
"should_training_stop": false
|
| 1196 |
+
},
|
| 1197 |
+
"attributes": {}
|
| 1198 |
+
}
|
| 1199 |
+
},
|
| 1200 |
+
"total_flos": 8.123170431172608e+18,
|
| 1201 |
+
"train_batch_size": 32,
|
| 1202 |
+
"trial_name": null,
|
| 1203 |
+
"trial_params": null
|
| 1204 |
+
}
|
training_args.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 5841
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ecf959d2b9b962920cb00233e188da75df99308a37612910a391c7651f6d60fb
|
| 3 |
size 5841
|