hugodk-sch
commited on
Commit
•
c080a55
1
Parent(s):
4831ad3
Model save
Browse files- README.md +32 -22
- adapter_model.safetensors +1 -1
- all_results.json +5 -5
- train_results.json +5 -5
- trainer_state.json +2030 -90
README.md
CHANGED
@@ -1,13 +1,11 @@
|
|
1 |
---
|
2 |
library_name: peft
|
3 |
tags:
|
4 |
-
- alignment-handbook
|
5 |
- trl
|
6 |
- dpo
|
|
|
7 |
- generated_from_trainer
|
8 |
base_model: NbAiLab/nb-gpt-j-6B-v2
|
9 |
-
datasets:
|
10 |
-
- hugodk-sch/aftonposten_title_prefs
|
11 |
model-index:
|
12 |
- name: aftonposten-6b-align-scan
|
13 |
results: []
|
@@ -18,17 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
18 |
|
19 |
# aftonposten-6b-align-scan
|
20 |
|
21 |
-
This model is a fine-tuned version of [
|
22 |
It achieves the following results on the evaluation set:
|
23 |
-
- Loss: 0.
|
24 |
-
- Rewards/chosen: -0.
|
25 |
-
- Rewards/rejected: -0.
|
26 |
-
- Rewards/accuracies: 0.
|
27 |
-
- Rewards/margins: 0.
|
28 |
-
- Logps/rejected: -
|
29 |
-
- Logps/chosen: -34.
|
30 |
-
- Logits/rejected: -2.
|
31 |
-
- Logits/chosen: -2.
|
32 |
|
33 |
## Model description
|
34 |
|
@@ -57,21 +55,33 @@ The following hyperparameters were used during training:
|
|
57 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
58 |
- lr_scheduler_type: cosine
|
59 |
- lr_scheduler_warmup_ratio: 0.1
|
60 |
-
- num_epochs:
|
61 |
|
62 |
### Training results
|
63 |
|
64 |
-
| Training Loss | Epoch | Step |
|
65 |
-
|
66 |
-
| 0.6464 | 0.26 | 100 |
|
67 |
-
| 0.5931 | 0.52 | 200 |
|
68 |
-
| 0.5301 | 0.78 | 300 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
|
71 |
### Framework versions
|
72 |
|
73 |
-
- PEFT 0.
|
74 |
-
- Transformers 4.
|
75 |
- Pytorch 2.1.2+cu121
|
76 |
-
- Datasets 2.
|
77 |
- Tokenizers 0.15.1
|
|
|
1 |
---
|
2 |
library_name: peft
|
3 |
tags:
|
|
|
4 |
- trl
|
5 |
- dpo
|
6 |
+
- alignment-handbook
|
7 |
- generated_from_trainer
|
8 |
base_model: NbAiLab/nb-gpt-j-6B-v2
|
|
|
|
|
9 |
model-index:
|
10 |
- name: aftonposten-6b-align-scan
|
11 |
results: []
|
|
|
16 |
|
17 |
# aftonposten-6b-align-scan
|
18 |
|
19 |
+
This model is a fine-tuned version of [NbAiLab/nb-gpt-j-6B-v2](https://huggingface.co/NbAiLab/nb-gpt-j-6B-v2) on the None dataset.
|
20 |
It achieves the following results on the evaluation set:
|
21 |
+
- Loss: 0.6943
|
22 |
+
- Rewards/chosen: -0.2880
|
23 |
+
- Rewards/rejected: -0.4573
|
24 |
+
- Rewards/accuracies: 0.5718
|
25 |
+
- Rewards/margins: 0.1693
|
26 |
+
- Logps/rejected: -38.0247
|
27 |
+
- Logps/chosen: -34.3545
|
28 |
+
- Logits/rejected: -2.1371
|
29 |
+
- Logits/chosen: -2.1419
|
30 |
|
31 |
## Model description
|
32 |
|
|
|
55 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
56 |
- lr_scheduler_type: cosine
|
57 |
- lr_scheduler_warmup_ratio: 0.1
|
58 |
+
- num_epochs: 4
|
59 |
|
60 |
### Training results
|
61 |
|
62 |
+
| Training Loss | Epoch | Step | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected |
|
63 |
+
|:-------------:|:-----:|:----:|:-------------:|:---------------:|:------------:|:--------------:|:---------------:|:------------------:|:--------------:|:---------------:|:----------------:|
|
64 |
+
| 0.6464 | 0.26 | 100 | -2.2340 | -2.2291 | -34.0405 | -37.5500 | 0.6903 | 0.5685 | -0.0054 | 0.0246 | -0.0300 |
|
65 |
+
| 0.5931 | 0.52 | 200 | -2.2316 | -2.2267 | -34.0730 | -37.5769 | 0.6980 | 0.5158 | -0.0346 | 0.0196 | -0.0543 |
|
66 |
+
| 0.5301 | 0.78 | 300 | -2.2292 | -2.2243 | -34.0962 | -37.6000 | 0.6973 | 0.5390 | -0.0555 | 0.0195 | -0.0750 |
|
67 |
+
| 0.389 | 1.04 | 400 | 0.6933 | -0.1201 | -0.1849 | 0.5507 | 0.0649 | -37.7221 | -34.1680 | -2.1983 | -2.2032 |
|
68 |
+
| 0.322 | 1.3 | 500 | 0.7055 | -0.2815 | -0.3556 | 0.5515 | 0.0741 | -37.9118 | -34.3473 | -2.1969 | -2.2017 |
|
69 |
+
| 0.327 | 1.56 | 600 | 0.6703 | -0.1443 | -0.2944 | 0.5806 | 0.1500 | -37.8437 | -34.1949 | -2.1819 | -2.1867 |
|
70 |
+
| 0.3034 | 1.82 | 700 | 0.6868 | -0.1851 | -0.3175 | 0.5656 | 0.1323 | -37.8694 | -34.2402 | -2.1701 | -2.1749 |
|
71 |
+
| 0.1649 | 2.08 | 800 | 0.6812 | -0.2229 | -0.3850 | 0.5951 | 0.1621 | -37.9443 | -34.2822 | -2.1594 | -2.1642 |
|
72 |
+
| 0.1691 | 2.34 | 900 | 0.6881 | -0.2514 | -0.4183 | 0.5831 | 0.1669 | -37.9814 | -34.3138 | -2.1476 | -2.1524 |
|
73 |
+
| 0.1953 | 2.6 | 1000 | 0.6957 | -0.2986 | -0.4680 | 0.5918 | 0.1694 | -38.0366 | -34.3663 | -2.1400 | -2.1447 |
|
74 |
+
| 0.1463 | 2.86 | 1100 | 0.7010 | -0.3003 | -0.4559 | 0.5714 | 0.1555 | -38.0231 | -34.3682 | -2.1379 | -2.1427 |
|
75 |
+
| 0.1796 | 3.12 | 1200 | 0.6908 | -0.2876 | -0.4581 | 0.5748 | 0.1705 | -38.0257 | -34.3541 | -2.1376 | -2.1423 |
|
76 |
+
| 0.1264 | 3.38 | 1300 | 0.6911 | -0.2772 | -0.4526 | 0.5893 | 0.1755 | -38.0196 | -34.3425 | -2.1374 | -2.1422 |
|
77 |
+
| 0.1206 | 3.64 | 1400 | 0.6924 | -0.2868 | -0.4582 | 0.5918 | 0.1714 | -38.0257 | -34.3532 | -2.1371 | -2.1419 |
|
78 |
+
| 0.1645 | 3.9 | 1500 | 0.6943 | -0.2880 | -0.4573 | 0.5718 | 0.1693 | -38.0247 | -34.3545 | -2.1371 | -2.1419 |
|
79 |
|
80 |
|
81 |
### Framework versions
|
82 |
|
83 |
+
- PEFT 0.10.0
|
84 |
+
- Transformers 4.39.0.dev0
|
85 |
- Pytorch 2.1.2+cu121
|
86 |
+
- Datasets 2.14.6
|
87 |
- Tokenizers 0.15.1
|
adapter_model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 176183216
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c936c50283e025007c0f0e9e6d277cc7e1147636cbbc341fe588c812d4f7c88d
|
3 |
size 176183216
|
all_results.json
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
{
|
2 |
-
"epoch":
|
3 |
"eval_logits/chosen": -2.22967529296875,
|
4 |
"eval_logits/rejected": -2.2248213291168213,
|
5 |
"eval_logps/chosen": -34.096519470214844,
|
@@ -13,9 +13,9 @@
|
|
13 |
"eval_samples": 343,
|
14 |
"eval_samples_per_second": 2.355,
|
15 |
"eval_steps_per_second": 0.295,
|
16 |
-
"train_loss": 0.
|
17 |
-
"train_runtime":
|
18 |
"train_samples": 3079,
|
19 |
-
"train_samples_per_second":
|
20 |
-
"train_steps_per_second": 0.
|
21 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 4.0,
|
3 |
"eval_logits/chosen": -2.22967529296875,
|
4 |
"eval_logits/rejected": -2.2248213291168213,
|
5 |
"eval_logps/chosen": -34.096519470214844,
|
|
|
13 |
"eval_samples": 343,
|
14 |
"eval_samples_per_second": 2.355,
|
15 |
"eval_steps_per_second": 0.295,
|
16 |
+
"train_loss": 0.19153660760297403,
|
17 |
+
"train_runtime": 10767.2812,
|
18 |
"train_samples": 3079,
|
19 |
+
"train_samples_per_second": 1.144,
|
20 |
+
"train_steps_per_second": 0.143
|
21 |
}
|
train_results.json
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
{
|
2 |
-
"epoch":
|
3 |
-
"train_loss": 0.
|
4 |
-
"train_runtime":
|
5 |
"train_samples": 3079,
|
6 |
-
"train_samples_per_second":
|
7 |
-
"train_steps_per_second": 0.
|
8 |
}
|
|
|
1 |
{
|
2 |
+
"epoch": 4.0,
|
3 |
+
"train_loss": 0.19153660760297403,
|
4 |
+
"train_runtime": 10767.2812,
|
5 |
"train_samples": 3079,
|
6 |
+
"train_samples_per_second": 1.144,
|
7 |
+
"train_steps_per_second": 0.143
|
8 |
}
|
trainer_state.json
CHANGED
@@ -1,9 +1,9 @@
|
|
1 |
{
|
2 |
"best_metric": null,
|
3 |
"best_model_checkpoint": null,
|
4 |
-
"epoch":
|
5 |
"eval_steps": 100,
|
6 |
-
"global_step":
|
7 |
"is_hyper_param_search": false,
|
8 |
"is_local_process_zero": true,
|
9 |
"is_world_process_zero": true,
|
@@ -492,130 +492,2070 @@
|
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.81,
|
495 |
-
"
|
496 |
-
"
|
497 |
-
"logits/
|
498 |
-
"
|
499 |
-
"logps/
|
500 |
-
"
|
501 |
-
"
|
502 |
-
"rewards/
|
503 |
-
"rewards/
|
504 |
-
"rewards/
|
|
|
505 |
"step": 310
|
506 |
},
|
507 |
{
|
508 |
"epoch": 0.83,
|
509 |
-
"
|
510 |
-
"
|
511 |
-
"logits/
|
512 |
-
"
|
513 |
-
"logps/
|
514 |
-
"
|
515 |
-
"
|
516 |
-
"rewards/
|
517 |
-
"rewards/
|
518 |
-
"rewards/
|
|
|
519 |
"step": 320
|
520 |
},
|
521 |
{
|
522 |
"epoch": 0.86,
|
523 |
-
"
|
524 |
-
"
|
525 |
-
"logits/
|
526 |
-
"
|
527 |
-
"logps/
|
528 |
-
"
|
529 |
-
"
|
530 |
-
"rewards/
|
531 |
-
"rewards/
|
532 |
-
"rewards/
|
|
|
533 |
"step": 330
|
534 |
},
|
535 |
{
|
536 |
"epoch": 0.88,
|
537 |
-
"
|
538 |
-
"
|
539 |
-
"logits/
|
540 |
-
"
|
541 |
-
"logps/
|
542 |
-
"
|
543 |
-
"
|
544 |
-
"rewards/
|
545 |
-
"rewards/
|
546 |
-
"rewards/
|
|
|
547 |
"step": 340
|
548 |
},
|
549 |
{
|
550 |
"epoch": 0.91,
|
551 |
-
"
|
552 |
-
"
|
553 |
-
"logits/
|
554 |
-
"
|
555 |
-
"logps/
|
556 |
-
"
|
557 |
-
"
|
558 |
-
"rewards/
|
559 |
-
"rewards/
|
560 |
-
"rewards/
|
|
|
561 |
"step": 350
|
562 |
},
|
563 |
{
|
564 |
"epoch": 0.94,
|
565 |
-
"
|
566 |
-
"
|
567 |
-
"logits/
|
568 |
-
"
|
569 |
-
"logps/
|
570 |
-
"
|
571 |
-
"
|
572 |
-
"rewards/
|
573 |
-
"rewards/
|
574 |
-
"rewards/
|
|
|
575 |
"step": 360
|
576 |
},
|
577 |
{
|
578 |
"epoch": 0.96,
|
579 |
-
"
|
580 |
-
"
|
581 |
-
"logits/
|
582 |
-
"
|
583 |
-
"logps/
|
584 |
-
"
|
585 |
-
"
|
586 |
-
"rewards/
|
587 |
-
"rewards/
|
588 |
-
"rewards/
|
|
|
589 |
"step": 370
|
590 |
},
|
591 |
{
|
592 |
"epoch": 0.99,
|
593 |
-
"
|
594 |
-
"
|
595 |
-
"logits/
|
596 |
-
"
|
597 |
-
"logps/
|
598 |
-
"
|
599 |
-
"
|
600 |
-
"rewards/
|
601 |
-
"rewards/
|
602 |
-
"rewards/
|
|
|
603 |
"step": 380
|
604 |
},
|
605 |
{
|
606 |
-
"epoch": 1.
|
607 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
608 |
"total_flos": 0.0,
|
609 |
-
"train_loss": 0.
|
610 |
-
"train_runtime":
|
611 |
-
"train_samples_per_second":
|
612 |
-
"train_steps_per_second": 0.
|
613 |
}
|
614 |
],
|
615 |
"logging_steps": 10,
|
616 |
-
"max_steps":
|
617 |
"num_input_tokens_seen": 0,
|
618 |
-
"num_train_epochs":
|
619 |
"save_steps": 100,
|
620 |
"total_flos": 0.0,
|
621 |
"train_batch_size": 4,
|
|
|
1 |
{
|
2 |
"best_metric": null,
|
3 |
"best_model_checkpoint": null,
|
4 |
+
"epoch": 4.0,
|
5 |
"eval_steps": 100,
|
6 |
+
"global_step": 1540,
|
7 |
"is_hyper_param_search": false,
|
8 |
"is_local_process_zero": true,
|
9 |
"is_world_process_zero": true,
|
|
|
492 |
},
|
493 |
{
|
494 |
"epoch": 0.81,
|
495 |
+
"grad_norm": 11.75,
|
496 |
+
"learning_rate": 4.84533120650964e-06,
|
497 |
+
"logits/chosen": -2.0636165142059326,
|
498 |
+
"logits/rejected": -2.0508041381835938,
|
499 |
+
"logps/chosen": -32.113487243652344,
|
500 |
+
"logps/rejected": -32.89537811279297,
|
501 |
+
"loss": 0.4684,
|
502 |
+
"rewards/accuracies": 0.9125000238418579,
|
503 |
+
"rewards/chosen": 0.3734120726585388,
|
504 |
+
"rewards/margins": 0.6023003458976746,
|
505 |
+
"rewards/rejected": -0.2288883477449417,
|
506 |
"step": 310
|
507 |
},
|
508 |
{
|
509 |
"epoch": 0.83,
|
510 |
+
"grad_norm": 10.875,
|
511 |
+
"learning_rate": 4.825108134172131e-06,
|
512 |
+
"logits/chosen": -1.9748560190200806,
|
513 |
+
"logits/rejected": -1.9662139415740967,
|
514 |
+
"logps/chosen": -31.80029296875,
|
515 |
+
"logps/rejected": -30.449291229248047,
|
516 |
+
"loss": 0.463,
|
517 |
+
"rewards/accuracies": 0.824999988079071,
|
518 |
+
"rewards/chosen": 0.49922245740890503,
|
519 |
+
"rewards/margins": 0.6863331198692322,
|
520 |
+
"rewards/rejected": -0.18711069226264954,
|
521 |
"step": 320
|
522 |
},
|
523 |
{
|
524 |
"epoch": 0.86,
|
525 |
+
"grad_norm": 12.875,
|
526 |
+
"learning_rate": 4.80369052967602e-06,
|
527 |
+
"logits/chosen": -1.910094976425171,
|
528 |
+
"logits/rejected": -1.9221827983856201,
|
529 |
+
"logps/chosen": -29.87582778930664,
|
530 |
+
"logps/rejected": -33.66598129272461,
|
531 |
+
"loss": 0.4237,
|
532 |
+
"rewards/accuracies": 0.8374999761581421,
|
533 |
+
"rewards/chosen": 0.5430954694747925,
|
534 |
+
"rewards/margins": 0.7804504632949829,
|
535 |
+
"rewards/rejected": -0.23735502362251282,
|
536 |
"step": 330
|
537 |
},
|
538 |
{
|
539 |
"epoch": 0.88,
|
540 |
+
"grad_norm": 18.375,
|
541 |
+
"learning_rate": 4.781089396387968e-06,
|
542 |
+
"logits/chosen": -1.8735644817352295,
|
543 |
+
"logits/rejected": -1.8643461465835571,
|
544 |
+
"logps/chosen": -34.02741241455078,
|
545 |
+
"logps/rejected": -36.179935455322266,
|
546 |
+
"loss": 0.4051,
|
547 |
+
"rewards/accuracies": 0.925000011920929,
|
548 |
+
"rewards/chosen": 0.6043617725372314,
|
549 |
+
"rewards/margins": 0.89524906873703,
|
550 |
+
"rewards/rejected": -0.29088738560676575,
|
551 |
"step": 340
|
552 |
},
|
553 |
{
|
554 |
"epoch": 0.91,
|
555 |
+
"grad_norm": 12.25,
|
556 |
+
"learning_rate": 4.757316345716554e-06,
|
557 |
+
"logits/chosen": -1.9254121780395508,
|
558 |
+
"logits/rejected": -1.9260631799697876,
|
559 |
+
"logps/chosen": -33.68886947631836,
|
560 |
+
"logps/rejected": -34.135963439941406,
|
561 |
+
"loss": 0.4149,
|
562 |
+
"rewards/accuracies": 0.8374999761581421,
|
563 |
+
"rewards/chosen": 0.6447720527648926,
|
564 |
+
"rewards/margins": 0.9030619859695435,
|
565 |
+
"rewards/rejected": -0.2582899332046509,
|
566 |
"step": 350
|
567 |
},
|
568 |
{
|
569 |
"epoch": 0.94,
|
570 |
+
"grad_norm": 14.9375,
|
571 |
+
"learning_rate": 4.73238359114687e-06,
|
572 |
+
"logits/chosen": -2.052173376083374,
|
573 |
+
"logits/rejected": -2.0583267211914062,
|
574 |
+
"logps/chosen": -31.09401512145996,
|
575 |
+
"logps/rejected": -33.012630462646484,
|
576 |
+
"loss": 0.4478,
|
577 |
+
"rewards/accuracies": 0.875,
|
578 |
+
"rewards/chosen": 0.43130677938461304,
|
579 |
+
"rewards/margins": 0.7412186861038208,
|
580 |
+
"rewards/rejected": -0.30991190671920776,
|
581 |
"step": 360
|
582 |
},
|
583 |
{
|
584 |
"epoch": 0.96,
|
585 |
+
"grad_norm": 28.0,
|
586 |
+
"learning_rate": 4.706303941965804e-06,
|
587 |
+
"logits/chosen": -1.9808381795883179,
|
588 |
+
"logits/rejected": -1.9804092645645142,
|
589 |
+
"logps/chosen": -32.843143463134766,
|
590 |
+
"logps/rejected": -36.34934616088867,
|
591 |
+
"loss": 0.4406,
|
592 |
+
"rewards/accuracies": 0.875,
|
593 |
+
"rewards/chosen": 0.5416162610054016,
|
594 |
+
"rewards/margins": 0.8112291097640991,
|
595 |
+
"rewards/rejected": -0.2696128487586975,
|
596 |
"step": 370
|
597 |
},
|
598 |
{
|
599 |
"epoch": 0.99,
|
600 |
+
"grad_norm": 10.5,
|
601 |
+
"learning_rate": 4.679090796681225e-06,
|
602 |
+
"logits/chosen": -2.012341022491455,
|
603 |
+
"logits/rejected": -2.0077567100524902,
|
604 |
+
"logps/chosen": -30.083026885986328,
|
605 |
+
"logps/rejected": -29.55636215209961,
|
606 |
+
"loss": 0.411,
|
607 |
+
"rewards/accuracies": 0.9125000238418579,
|
608 |
+
"rewards/chosen": 0.5648170709609985,
|
609 |
+
"rewards/margins": 0.873041033744812,
|
610 |
+
"rewards/rejected": -0.30822402238845825,
|
611 |
"step": 380
|
612 |
},
|
613 |
{
|
614 |
+
"epoch": 1.01,
|
615 |
+
"grad_norm": 13.75,
|
616 |
+
"learning_rate": 4.650758136138454e-06,
|
617 |
+
"logits/chosen": -1.784257173538208,
|
618 |
+
"logits/rejected": -1.7905915975570679,
|
619 |
+
"logps/chosen": -31.67917823791504,
|
620 |
+
"logps/rejected": -36.660545349121094,
|
621 |
+
"loss": 0.3793,
|
622 |
+
"rewards/accuracies": 0.925000011920929,
|
623 |
+
"rewards/chosen": 0.635884165763855,
|
624 |
+
"rewards/margins": 1.1239113807678223,
|
625 |
+
"rewards/rejected": -0.4880271553993225,
|
626 |
+
"step": 390
|
627 |
+
},
|
628 |
+
{
|
629 |
+
"epoch": 1.04,
|
630 |
+
"grad_norm": 10.625,
|
631 |
+
"learning_rate": 4.621320516337559e-06,
|
632 |
+
"logits/chosen": -1.937954306602478,
|
633 |
+
"logits/rejected": -1.931673288345337,
|
634 |
+
"logps/chosen": -33.02653121948242,
|
635 |
+
"logps/rejected": -32.67657470703125,
|
636 |
+
"loss": 0.389,
|
637 |
+
"rewards/accuracies": 0.824999988079071,
|
638 |
+
"rewards/chosen": 0.7255499958992004,
|
639 |
+
"rewards/margins": 1.0564748048782349,
|
640 |
+
"rewards/rejected": -0.33092474937438965,
|
641 |
+
"step": 400
|
642 |
+
},
|
643 |
+
{
|
644 |
+
"epoch": 1.04,
|
645 |
+
"eval_logits/chosen": -2.2031846046447754,
|
646 |
+
"eval_logits/rejected": -2.198335647583008,
|
647 |
+
"eval_logps/chosen": -34.16795349121094,
|
648 |
+
"eval_logps/rejected": -37.72208786010742,
|
649 |
+
"eval_loss": 0.6933022737503052,
|
650 |
+
"eval_rewards/accuracies": 0.550664484500885,
|
651 |
+
"eval_rewards/chosen": -0.12006273865699768,
|
652 |
+
"eval_rewards/margins": 0.06485801190137863,
|
653 |
+
"eval_rewards/rejected": -0.1849207729101181,
|
654 |
+
"eval_runtime": 145.5175,
|
655 |
+
"eval_samples_per_second": 2.357,
|
656 |
+
"eval_steps_per_second": 0.295,
|
657 |
+
"step": 400
|
658 |
+
},
|
659 |
+
{
|
660 |
+
"epoch": 1.06,
|
661 |
+
"grad_norm": 13.375,
|
662 |
+
"learning_rate": 4.590793060955158e-06,
|
663 |
+
"logits/chosen": -1.9355392456054688,
|
664 |
+
"logits/rejected": -1.9427944421768188,
|
665 |
+
"logps/chosen": -28.549734115600586,
|
666 |
+
"logps/rejected": -29.689483642578125,
|
667 |
+
"loss": 0.3677,
|
668 |
+
"rewards/accuracies": 0.925000011920929,
|
669 |
+
"rewards/chosen": 0.5135748982429504,
|
670 |
+
"rewards/margins": 1.0457103252410889,
|
671 |
+
"rewards/rejected": -0.5321354866027832,
|
672 |
+
"step": 410
|
673 |
+
},
|
674 |
+
{
|
675 |
+
"epoch": 1.09,
|
676 |
+
"grad_norm": 11.375,
|
677 |
+
"learning_rate": 4.559191453574582e-06,
|
678 |
+
"logits/chosen": -1.9551494121551514,
|
679 |
+
"logits/rejected": -1.9541441202163696,
|
680 |
+
"logps/chosen": -33.60606002807617,
|
681 |
+
"logps/rejected": -31.240774154663086,
|
682 |
+
"loss": 0.4587,
|
683 |
+
"rewards/accuracies": 0.7875000238418579,
|
684 |
+
"rewards/chosen": 0.5801833868026733,
|
685 |
+
"rewards/margins": 0.8481400609016418,
|
686 |
+
"rewards/rejected": -0.2679567039012909,
|
687 |
+
"step": 420
|
688 |
+
},
|
689 |
+
{
|
690 |
+
"epoch": 1.12,
|
691 |
+
"grad_norm": 15.125,
|
692 |
+
"learning_rate": 4.52653192962838e-06,
|
693 |
+
"logits/chosen": -1.9499473571777344,
|
694 |
+
"logits/rejected": -1.932682752609253,
|
695 |
+
"logps/chosen": -30.44845199584961,
|
696 |
+
"logps/rejected": -33.599735260009766,
|
697 |
+
"loss": 0.3784,
|
698 |
+
"rewards/accuracies": 0.887499988079071,
|
699 |
+
"rewards/chosen": 0.4758743345737457,
|
700 |
+
"rewards/margins": 1.0819809436798096,
|
701 |
+
"rewards/rejected": -0.6061066389083862,
|
702 |
+
"step": 430
|
703 |
+
},
|
704 |
+
{
|
705 |
+
"epoch": 1.14,
|
706 |
+
"grad_norm": 14.1875,
|
707 |
+
"learning_rate": 4.492831268057307e-06,
|
708 |
+
"logits/chosen": -1.9813868999481201,
|
709 |
+
"logits/rejected": -1.9835193157196045,
|
710 |
+
"logps/chosen": -35.81322479248047,
|
711 |
+
"logps/rejected": -35.490821838378906,
|
712 |
+
"loss": 0.313,
|
713 |
+
"rewards/accuracies": 0.925000011920929,
|
714 |
+
"rewards/chosen": 0.7192217111587524,
|
715 |
+
"rewards/margins": 1.2869927883148193,
|
716 |
+
"rewards/rejected": -0.5677711367607117,
|
717 |
+
"step": 440
|
718 |
+
},
|
719 |
+
{
|
720 |
+
"epoch": 1.17,
|
721 |
+
"grad_norm": 12.3125,
|
722 |
+
"learning_rate": 4.458106782690094e-06,
|
723 |
+
"logits/chosen": -2.05714750289917,
|
724 |
+
"logits/rejected": -2.0569212436676025,
|
725 |
+
"logps/chosen": -31.911890029907227,
|
726 |
+
"logps/rejected": -34.006473541259766,
|
727 |
+
"loss": 0.3759,
|
728 |
+
"rewards/accuracies": 0.875,
|
729 |
+
"rewards/chosen": 0.6493128538131714,
|
730 |
+
"rewards/margins": 1.0510342121124268,
|
731 |
+
"rewards/rejected": -0.40172141790390015,
|
732 |
+
"step": 450
|
733 |
+
},
|
734 |
+
{
|
735 |
+
"epoch": 1.19,
|
736 |
+
"grad_norm": 11.9375,
|
737 |
+
"learning_rate": 4.422376313348405e-06,
|
738 |
+
"logits/chosen": -2.001530885696411,
|
739 |
+
"logits/rejected": -1.9940494298934937,
|
740 |
+
"logps/chosen": -31.56293296813965,
|
741 |
+
"logps/rejected": -36.774314880371094,
|
742 |
+
"loss": 0.3025,
|
743 |
+
"rewards/accuracies": 0.8999999761581421,
|
744 |
+
"rewards/chosen": 0.7587249279022217,
|
745 |
+
"rewards/margins": 1.4069726467132568,
|
746 |
+
"rewards/rejected": -0.6482478380203247,
|
747 |
+
"step": 460
|
748 |
+
},
|
749 |
+
{
|
750 |
+
"epoch": 1.22,
|
751 |
+
"grad_norm": 20.375,
|
752 |
+
"learning_rate": 4.3856582166815696e-06,
|
753 |
+
"logits/chosen": -1.9044301509857178,
|
754 |
+
"logits/rejected": -1.900957465171814,
|
755 |
+
"logps/chosen": -33.156280517578125,
|
756 |
+
"logps/rejected": -33.48976516723633,
|
757 |
+
"loss": 0.3626,
|
758 |
+
"rewards/accuracies": 0.875,
|
759 |
+
"rewards/chosen": 0.6984156966209412,
|
760 |
+
"rewards/margins": 1.2077140808105469,
|
761 |
+
"rewards/rejected": -0.5092984437942505,
|
762 |
+
"step": 470
|
763 |
+
},
|
764 |
+
{
|
765 |
+
"epoch": 1.25,
|
766 |
+
"grad_norm": 14.0625,
|
767 |
+
"learning_rate": 4.347971356735789e-06,
|
768 |
+
"logits/chosen": -2.0298960208892822,
|
769 |
+
"logits/rejected": -2.0229077339172363,
|
770 |
+
"logps/chosen": -30.285165786743164,
|
771 |
+
"logps/rejected": -32.622947692871094,
|
772 |
+
"loss": 0.4059,
|
773 |
+
"rewards/accuracies": 0.824999988079071,
|
774 |
+
"rewards/chosen": 0.48903241753578186,
|
775 |
+
"rewards/margins": 1.0555496215820312,
|
776 |
+
"rewards/rejected": -0.566517174243927,
|
777 |
+
"step": 480
|
778 |
+
},
|
779 |
+
{
|
780 |
+
"epoch": 1.27,
|
781 |
+
"grad_norm": 9.4375,
|
782 |
+
"learning_rate": 4.309335095262675e-06,
|
783 |
+
"logits/chosen": -1.9742801189422607,
|
784 |
+
"logits/rejected": -1.975847840309143,
|
785 |
+
"logps/chosen": -34.49077606201172,
|
786 |
+
"logps/rejected": -34.25959396362305,
|
787 |
+
"loss": 0.3167,
|
788 |
+
"rewards/accuracies": 0.887499988079071,
|
789 |
+
"rewards/chosen": 0.7761706113815308,
|
790 |
+
"rewards/margins": 1.3324440717697144,
|
791 |
+
"rewards/rejected": -0.5562735795974731,
|
792 |
+
"step": 490
|
793 |
+
},
|
794 |
+
{
|
795 |
+
"epoch": 1.3,
|
796 |
+
"grad_norm": 20.25,
|
797 |
+
"learning_rate": 4.269769281772082e-06,
|
798 |
+
"logits/chosen": -1.8631584644317627,
|
799 |
+
"logits/rejected": -1.860769510269165,
|
800 |
+
"logps/chosen": -32.35368347167969,
|
801 |
+
"logps/rejected": -37.259803771972656,
|
802 |
+
"loss": 0.322,
|
803 |
+
"rewards/accuracies": 0.887499988079071,
|
804 |
+
"rewards/chosen": 0.7367446422576904,
|
805 |
+
"rewards/margins": 1.452678918838501,
|
806 |
+
"rewards/rejected": -0.715934157371521,
|
807 |
+
"step": 500
|
808 |
+
},
|
809 |
+
{
|
810 |
+
"epoch": 1.3,
|
811 |
+
"eval_logits/chosen": -2.201748847961426,
|
812 |
+
"eval_logits/rejected": -2.1969025135040283,
|
813 |
+
"eval_logps/chosen": -34.34733200073242,
|
814 |
+
"eval_logps/rejected": -37.91175842285156,
|
815 |
+
"eval_loss": 0.7055429816246033,
|
816 |
+
"eval_rewards/accuracies": 0.5514950156211853,
|
817 |
+
"eval_rewards/chosen": -0.281506210565567,
|
818 |
+
"eval_rewards/margins": 0.07411985099315643,
|
819 |
+
"eval_rewards/rejected": -0.35562604665756226,
|
820 |
+
"eval_runtime": 145.2329,
|
821 |
+
"eval_samples_per_second": 2.362,
|
822 |
+
"eval_steps_per_second": 0.296,
|
823 |
+
"step": 500
|
824 |
+
},
|
825 |
+
{
|
826 |
+
"epoch": 1.32,
|
827 |
+
"grad_norm": 14.9375,
|
828 |
+
"learning_rate": 4.22929424333435e-06,
|
829 |
+
"logits/chosen": -1.9645631313323975,
|
830 |
+
"logits/rejected": -1.9692989587783813,
|
831 |
+
"logps/chosen": -32.608741760253906,
|
832 |
+
"logps/rejected": -32.16291809082031,
|
833 |
+
"loss": 0.3561,
|
834 |
+
"rewards/accuracies": 0.8500000238418579,
|
835 |
+
"rewards/chosen": 0.7637556791305542,
|
836 |
+
"rewards/margins": 1.2288849353790283,
|
837 |
+
"rewards/rejected": -0.46512943506240845,
|
838 |
+
"step": 510
|
839 |
+
},
|
840 |
+
{
|
841 |
+
"epoch": 1.35,
|
842 |
+
"grad_norm": 14.375,
|
843 |
+
"learning_rate": 4.1879307741372085e-06,
|
844 |
+
"logits/chosen": -1.9946361780166626,
|
845 |
+
"logits/rejected": -2.005610466003418,
|
846 |
+
"logps/chosen": -30.554733276367188,
|
847 |
+
"logps/rejected": -32.16284942626953,
|
848 |
+
"loss": 0.358,
|
849 |
+
"rewards/accuracies": 0.862500011920929,
|
850 |
+
"rewards/chosen": 0.7521130442619324,
|
851 |
+
"rewards/margins": 1.3417600393295288,
|
852 |
+
"rewards/rejected": -0.589647114276886,
|
853 |
+
"step": 520
|
854 |
+
},
|
855 |
+
{
|
856 |
+
"epoch": 1.38,
|
857 |
+
"grad_norm": 6.53125,
|
858 |
+
"learning_rate": 4.145700124802693e-06,
|
859 |
+
"logits/chosen": -1.9223169088363647,
|
860 |
+
"logits/rejected": -1.9189828634262085,
|
861 |
+
"logps/chosen": -31.747081756591797,
|
862 |
+
"logps/rejected": -33.04930877685547,
|
863 |
+
"loss": 0.349,
|
864 |
+
"rewards/accuracies": 0.925000011920929,
|
865 |
+
"rewards/chosen": 0.6060883402824402,
|
866 |
+
"rewards/margins": 1.2239677906036377,
|
867 |
+
"rewards/rejected": -0.6178793907165527,
|
868 |
+
"step": 530
|
869 |
+
},
|
870 |
+
{
|
871 |
+
"epoch": 1.4,
|
872 |
+
"grad_norm": 11.0,
|
873 |
+
"learning_rate": 4.102623991469562e-06,
|
874 |
+
"logits/chosen": -1.7875115871429443,
|
875 |
+
"logits/rejected": -1.7967026233673096,
|
876 |
+
"logps/chosen": -31.79451560974121,
|
877 |
+
"logps/rejected": -32.5256233215332,
|
878 |
+
"loss": 0.3606,
|
879 |
+
"rewards/accuracies": 0.862500011920929,
|
880 |
+
"rewards/chosen": 0.697385847568512,
|
881 |
+
"rewards/margins": 1.3000586032867432,
|
882 |
+
"rewards/rejected": -0.602672815322876,
|
883 |
+
"step": 540
|
884 |
+
},
|
885 |
+
{
|
886 |
+
"epoch": 1.43,
|
887 |
+
"grad_norm": 13.1875,
|
888 |
+
"learning_rate": 4.058724504646834e-06,
|
889 |
+
"logits/chosen": -1.8841272592544556,
|
890 |
+
"logits/rejected": -1.8778330087661743,
|
891 |
+
"logps/chosen": -32.83342742919922,
|
892 |
+
"logps/rejected": -31.528995513916016,
|
893 |
+
"loss": 0.3722,
|
894 |
+
"rewards/accuracies": 0.8374999761581421,
|
895 |
+
"rewards/chosen": 0.7873939275741577,
|
896 |
+
"rewards/margins": 1.2712757587432861,
|
897 |
+
"rewards/rejected": -0.48388180136680603,
|
898 |
+
"step": 550
|
899 |
+
},
|
900 |
+
{
|
901 |
+
"epoch": 1.45,
|
902 |
+
"grad_norm": 10.125,
|
903 |
+
"learning_rate": 4.014024217844167e-06,
|
904 |
+
"logits/chosen": -1.9727208614349365,
|
905 |
+
"logits/rejected": -1.9707006216049194,
|
906 |
+
"logps/chosen": -33.61963653564453,
|
907 |
+
"logps/rejected": -31.988178253173828,
|
908 |
+
"loss": 0.3617,
|
909 |
+
"rewards/accuracies": 0.875,
|
910 |
+
"rewards/chosen": 0.8564074635505676,
|
911 |
+
"rewards/margins": 1.298119306564331,
|
912 |
+
"rewards/rejected": -0.4417116641998291,
|
913 |
+
"step": 560
|
914 |
+
},
|
915 |
+
{
|
916 |
+
"epoch": 1.48,
|
917 |
+
"grad_norm": 12.625,
|
918 |
+
"learning_rate": 3.968546095984911e-06,
|
919 |
+
"logits/chosen": -1.8058189153671265,
|
920 |
+
"logits/rejected": -1.8036988973617554,
|
921 |
+
"logps/chosen": -31.945751190185547,
|
922 |
+
"logps/rejected": -31.496994018554688,
|
923 |
+
"loss": 0.3948,
|
924 |
+
"rewards/accuracies": 0.862500011920929,
|
925 |
+
"rewards/chosen": 0.7655945420265198,
|
926 |
+
"rewards/margins": 1.210815191268921,
|
927 |
+
"rewards/rejected": -0.44522079825401306,
|
928 |
+
"step": 570
|
929 |
+
},
|
930 |
+
{
|
931 |
+
"epoch": 1.51,
|
932 |
+
"grad_norm": 13.125,
|
933 |
+
"learning_rate": 3.922313503607806e-06,
|
934 |
+
"logits/chosen": -1.9404058456420898,
|
935 |
+
"logits/rejected": -1.9370429515838623,
|
936 |
+
"logps/chosen": -30.20537757873535,
|
937 |
+
"logps/rejected": -35.37580871582031,
|
938 |
+
"loss": 0.32,
|
939 |
+
"rewards/accuracies": 0.875,
|
940 |
+
"rewards/chosen": 0.8093468546867371,
|
941 |
+
"rewards/margins": 1.4840278625488281,
|
942 |
+
"rewards/rejected": -0.6746810078620911,
|
943 |
+
"step": 580
|
944 |
+
},
|
945 |
+
{
|
946 |
+
"epoch": 1.53,
|
947 |
+
"grad_norm": 9.6875,
|
948 |
+
"learning_rate": 3.875350192863368e-06,
|
949 |
+
"logits/chosen": -1.8791511058807373,
|
950 |
+
"logits/rejected": -1.8827145099639893,
|
951 |
+
"logps/chosen": -28.8929386138916,
|
952 |
+
"logps/rejected": -31.133419036865234,
|
953 |
+
"loss": 0.3795,
|
954 |
+
"rewards/accuracies": 0.8500000238418579,
|
955 |
+
"rewards/chosen": 0.6498847603797913,
|
956 |
+
"rewards/margins": 1.0769810676574707,
|
957 |
+
"rewards/rejected": -0.4270961880683899,
|
958 |
+
"step": 590
|
959 |
+
},
|
960 |
+
{
|
961 |
+
"epoch": 1.56,
|
962 |
+
"grad_norm": 15.0625,
|
963 |
+
"learning_rate": 3.8276802913111436e-06,
|
964 |
+
"logits/chosen": -1.915834665298462,
|
965 |
+
"logits/rejected": -1.9157018661499023,
|
966 |
+
"logps/chosen": -31.168197631835938,
|
967 |
+
"logps/rejected": -31.740047454833984,
|
968 |
+
"loss": 0.327,
|
969 |
+
"rewards/accuracies": 0.875,
|
970 |
+
"rewards/chosen": 0.9480020403862,
|
971 |
+
"rewards/margins": 1.4403297901153564,
|
972 |
+
"rewards/rejected": -0.49232783913612366,
|
973 |
+
"step": 600
|
974 |
+
},
|
975 |
+
{
|
976 |
+
"epoch": 1.56,
|
977 |
+
"eval_logits/chosen": -2.1867270469665527,
|
978 |
+
"eval_logits/rejected": -2.1818978786468506,
|
979 |
+
"eval_logps/chosen": -34.19493103027344,
|
980 |
+
"eval_logps/rejected": -37.843685150146484,
|
981 |
+
"eval_loss": 0.6703336834907532,
|
982 |
+
"eval_rewards/accuracies": 0.5805647969245911,
|
983 |
+
"eval_rewards/chosen": -0.14434270560741425,
|
984 |
+
"eval_rewards/margins": 0.15001599490642548,
|
985 |
+
"eval_rewards/rejected": -0.2943587005138397,
|
986 |
+
"eval_runtime": 145.4291,
|
987 |
+
"eval_samples_per_second": 2.359,
|
988 |
+
"eval_steps_per_second": 0.296,
|
989 |
+
"step": 600
|
990 |
+
},
|
991 |
+
{
|
992 |
+
"epoch": 1.58,
|
993 |
+
"grad_norm": 13.75,
|
994 |
+
"learning_rate": 3.7793282895240927e-06,
|
995 |
+
"logits/chosen": -1.9787023067474365,
|
996 |
+
"logits/rejected": -1.9794394969940186,
|
997 |
+
"logps/chosen": -33.91895294189453,
|
998 |
+
"logps/rejected": -33.77281951904297,
|
999 |
+
"loss": 0.2961,
|
1000 |
+
"rewards/accuracies": 0.875,
|
1001 |
+
"rewards/chosen": 1.0000724792480469,
|
1002 |
+
"rewards/margins": 1.6198437213897705,
|
1003 |
+
"rewards/rejected": -0.6197710037231445,
|
1004 |
+
"step": 610
|
1005 |
+
},
|
1006 |
+
{
|
1007 |
+
"epoch": 1.61,
|
1008 |
+
"grad_norm": 10.25,
|
1009 |
+
"learning_rate": 3.730319028506478e-06,
|
1010 |
+
"logits/chosen": -1.9425113201141357,
|
1011 |
+
"logits/rejected": -1.9399850368499756,
|
1012 |
+
"logps/chosen": -32.121673583984375,
|
1013 |
+
"logps/rejected": -32.70948791503906,
|
1014 |
+
"loss": 0.3125,
|
1015 |
+
"rewards/accuracies": 0.8999999761581421,
|
1016 |
+
"rewards/chosen": 0.9033487439155579,
|
1017 |
+
"rewards/margins": 1.4789619445800781,
|
1018 |
+
"rewards/rejected": -0.5756131410598755,
|
1019 |
+
"step": 620
|
1020 |
+
},
|
1021 |
+
{
|
1022 |
+
"epoch": 1.64,
|
1023 |
+
"grad_norm": 30.75,
|
1024 |
+
"learning_rate": 3.6806776869317074e-06,
|
1025 |
+
"logits/chosen": -1.953460693359375,
|
1026 |
+
"logits/rejected": -1.9444434642791748,
|
1027 |
+
"logps/chosen": -31.709529876708984,
|
1028 |
+
"logps/rejected": -31.5223388671875,
|
1029 |
+
"loss": 0.3619,
|
1030 |
+
"rewards/accuracies": 0.8374999761581421,
|
1031 |
+
"rewards/chosen": 0.9599300622940063,
|
1032 |
+
"rewards/margins": 1.488205909729004,
|
1033 |
+
"rewards/rejected": -0.5282759070396423,
|
1034 |
+
"step": 630
|
1035 |
+
},
|
1036 |
+
{
|
1037 |
+
"epoch": 1.66,
|
1038 |
+
"grad_norm": 17.25,
|
1039 |
+
"learning_rate": 3.6304297682067146e-06,
|
1040 |
+
"logits/chosen": -1.9538257122039795,
|
1041 |
+
"logits/rejected": -1.9505916833877563,
|
1042 |
+
"logps/chosen": -31.257614135742188,
|
1043 |
+
"logps/rejected": -33.02531814575195,
|
1044 |
+
"loss": 0.3186,
|
1045 |
+
"rewards/accuracies": 0.9375,
|
1046 |
+
"rewards/chosen": 0.8467265367507935,
|
1047 |
+
"rewards/margins": 1.3920328617095947,
|
1048 |
+
"rewards/rejected": -0.5453063249588013,
|
1049 |
+
"step": 640
|
1050 |
+
},
|
1051 |
+
{
|
1052 |
+
"epoch": 1.69,
|
1053 |
+
"grad_norm": 9.4375,
|
1054 |
+
"learning_rate": 3.579601087369492e-06,
|
1055 |
+
"logits/chosen": -1.9619709253311157,
|
1056 |
+
"logits/rejected": -1.9643146991729736,
|
1057 |
+
"logps/chosen": -32.65868377685547,
|
1058 |
+
"logps/rejected": -34.36846160888672,
|
1059 |
+
"loss": 0.2785,
|
1060 |
+
"rewards/accuracies": 0.925000011920929,
|
1061 |
+
"rewards/chosen": 0.8955224752426147,
|
1062 |
+
"rewards/margins": 1.512078881263733,
|
1063 |
+
"rewards/rejected": -0.6165562868118286,
|
1064 |
+
"step": 650
|
1065 |
+
},
|
1066 |
+
{
|
1067 |
+
"epoch": 1.71,
|
1068 |
+
"grad_norm": 19.625,
|
1069 |
+
"learning_rate": 3.5282177578265295e-06,
|
1070 |
+
"logits/chosen": -1.8742033243179321,
|
1071 |
+
"logits/rejected": -1.8746894598007202,
|
1072 |
+
"logps/chosen": -32.91667175292969,
|
1073 |
+
"logps/rejected": -32.129493713378906,
|
1074 |
+
"loss": 0.328,
|
1075 |
+
"rewards/accuracies": 0.887499988079071,
|
1076 |
+
"rewards/chosen": 0.9271462559700012,
|
1077 |
+
"rewards/margins": 1.3971627950668335,
|
1078 |
+
"rewards/rejected": -0.4700165390968323,
|
1079 |
+
"step": 660
|
1080 |
+
},
|
1081 |
+
{
|
1082 |
+
"epoch": 1.74,
|
1083 |
+
"grad_norm": 14.1875,
|
1084 |
+
"learning_rate": 3.476306177936961e-06,
|
1085 |
+
"logits/chosen": -1.9133888483047485,
|
1086 |
+
"logits/rejected": -1.9037227630615234,
|
1087 |
+
"logps/chosen": -32.933998107910156,
|
1088 |
+
"logps/rejected": -33.00373077392578,
|
1089 |
+
"loss": 0.2651,
|
1090 |
+
"rewards/accuracies": 0.9125000238418579,
|
1091 |
+
"rewards/chosen": 0.9605558514595032,
|
1092 |
+
"rewards/margins": 1.586902379989624,
|
1093 |
+
"rewards/rejected": -0.6263464689254761,
|
1094 |
+
"step": 670
|
1095 |
+
},
|
1096 |
+
{
|
1097 |
+
"epoch": 1.77,
|
1098 |
+
"grad_norm": 10.1875,
|
1099 |
+
"learning_rate": 3.423893017450324e-06,
|
1100 |
+
"logits/chosen": -1.8113712072372437,
|
1101 |
+
"logits/rejected": -1.8081716299057007,
|
1102 |
+
"logps/chosen": -30.349456787109375,
|
1103 |
+
"logps/rejected": -34.96870422363281,
|
1104 |
+
"loss": 0.2808,
|
1105 |
+
"rewards/accuracies": 0.925000011920929,
|
1106 |
+
"rewards/chosen": 1.0003511905670166,
|
1107 |
+
"rewards/margins": 1.5854613780975342,
|
1108 |
+
"rewards/rejected": -0.5851101875305176,
|
1109 |
+
"step": 680
|
1110 |
+
},
|
1111 |
+
{
|
1112 |
+
"epoch": 1.79,
|
1113 |
+
"grad_norm": 8.4375,
|
1114 |
+
"learning_rate": 3.3710052038048794e-06,
|
1115 |
+
"logits/chosen": -1.8722549676895142,
|
1116 |
+
"logits/rejected": -1.8714803457260132,
|
1117 |
+
"logps/chosen": -34.03639602661133,
|
1118 |
+
"logps/rejected": -36.12696075439453,
|
1119 |
+
"loss": 0.2342,
|
1120 |
+
"rewards/accuracies": 0.9375,
|
1121 |
+
"rewards/chosen": 1.110505223274231,
|
1122 |
+
"rewards/margins": 1.8063312768936157,
|
1123 |
+
"rewards/rejected": -0.6958259344100952,
|
1124 |
+
"step": 690
|
1125 |
+
},
|
1126 |
+
{
|
1127 |
+
"epoch": 1.82,
|
1128 |
+
"grad_norm": 6.90625,
|
1129 |
+
"learning_rate": 3.3176699082935546e-06,
|
1130 |
+
"logits/chosen": -1.8479416370391846,
|
1131 |
+
"logits/rejected": -1.8507680892944336,
|
1132 |
+
"logps/chosen": -31.403573989868164,
|
1133 |
+
"logps/rejected": -36.341434478759766,
|
1134 |
+
"loss": 0.3034,
|
1135 |
+
"rewards/accuracies": 0.875,
|
1136 |
+
"rewards/chosen": 1.079347014427185,
|
1137 |
+
"rewards/margins": 1.7091315984725952,
|
1138 |
+
"rewards/rejected": -0.6297845840454102,
|
1139 |
+
"step": 700
|
1140 |
+
},
|
1141 |
+
{
|
1142 |
+
"epoch": 1.82,
|
1143 |
+
"eval_logits/chosen": -2.174877166748047,
|
1144 |
+
"eval_logits/rejected": -2.1700735092163086,
|
1145 |
+
"eval_logps/chosen": -34.240230560302734,
|
1146 |
+
"eval_logps/rejected": -37.869354248046875,
|
1147 |
+
"eval_loss": 0.6868197917938232,
|
1148 |
+
"eval_rewards/accuracies": 0.565614640712738,
|
1149 |
+
"eval_rewards/chosen": -0.18511110544204712,
|
1150 |
+
"eval_rewards/margins": 0.13234683871269226,
|
1151 |
+
"eval_rewards/rejected": -0.31745797395706177,
|
1152 |
+
"eval_runtime": 145.1401,
|
1153 |
+
"eval_samples_per_second": 2.363,
|
1154 |
+
"eval_steps_per_second": 0.296,
|
1155 |
+
"step": 700
|
1156 |
+
},
|
1157 |
+
{
|
1158 |
+
"epoch": 1.84,
|
1159 |
+
"grad_norm": 15.9375,
|
1160 |
+
"learning_rate": 3.2639145321045933e-06,
|
1161 |
+
"logits/chosen": -1.9555838108062744,
|
1162 |
+
"logits/rejected": -1.958298683166504,
|
1163 |
+
"logps/chosen": -33.82456970214844,
|
1164 |
+
"logps/rejected": -34.84859085083008,
|
1165 |
+
"loss": 0.3553,
|
1166 |
+
"rewards/accuracies": 0.862500011920929,
|
1167 |
+
"rewards/chosen": 0.9639171361923218,
|
1168 |
+
"rewards/margins": 1.4845099449157715,
|
1169 |
+
"rewards/rejected": -0.5205925703048706,
|
1170 |
+
"step": 710
|
1171 |
+
},
|
1172 |
+
{
|
1173 |
+
"epoch": 1.87,
|
1174 |
+
"grad_norm": 14.375,
|
1175 |
+
"learning_rate": 3.2097666922441107e-06,
|
1176 |
+
"logits/chosen": -1.8090355396270752,
|
1177 |
+
"logits/rejected": -1.8031476736068726,
|
1178 |
+
"logps/chosen": -33.71675491333008,
|
1179 |
+
"logps/rejected": -32.93245315551758,
|
1180 |
+
"loss": 0.323,
|
1181 |
+
"rewards/accuracies": 0.862500011920929,
|
1182 |
+
"rewards/chosen": 1.0048266649246216,
|
1183 |
+
"rewards/margins": 1.527376413345337,
|
1184 |
+
"rewards/rejected": -0.5225496292114258,
|
1185 |
+
"step": 720
|
1186 |
+
},
|
1187 |
+
{
|
1188 |
+
"epoch": 1.9,
|
1189 |
+
"grad_norm": 7.46875,
|
1190 |
+
"learning_rate": 3.1552542073477554e-06,
|
1191 |
+
"logits/chosen": -1.9834911823272705,
|
1192 |
+
"logits/rejected": -1.9804458618164062,
|
1193 |
+
"logps/chosen": -29.5953426361084,
|
1194 |
+
"logps/rejected": -32.31734085083008,
|
1195 |
+
"loss": 0.2567,
|
1196 |
+
"rewards/accuracies": 0.949999988079071,
|
1197 |
+
"rewards/chosen": 0.9518804550170898,
|
1198 |
+
"rewards/margins": 1.7341070175170898,
|
1199 |
+
"rewards/rejected": -0.7822265028953552,
|
1200 |
+
"step": 730
|
1201 |
+
},
|
1202 |
+
{
|
1203 |
+
"epoch": 1.92,
|
1204 |
+
"grad_norm": 7.28125,
|
1205 |
+
"learning_rate": 3.100405083388799e-06,
|
1206 |
+
"logits/chosen": -1.8222984075546265,
|
1207 |
+
"logits/rejected": -1.8223203420639038,
|
1208 |
+
"logps/chosen": -32.201148986816406,
|
1209 |
+
"logps/rejected": -38.2669792175293,
|
1210 |
+
"loss": 0.2789,
|
1211 |
+
"rewards/accuracies": 0.887499988079071,
|
1212 |
+
"rewards/chosen": 1.1358217000961304,
|
1213 |
+
"rewards/margins": 1.7110786437988281,
|
1214 |
+
"rewards/rejected": -0.5752568244934082,
|
1215 |
+
"step": 740
|
1216 |
+
},
|
1217 |
+
{
|
1218 |
+
"epoch": 1.95,
|
1219 |
+
"grad_norm": 6.75,
|
1220 |
+
"learning_rate": 3.0452474992899645e-06,
|
1221 |
+
"logits/chosen": -1.7060989141464233,
|
1222 |
+
"logits/rejected": -1.711395025253296,
|
1223 |
+
"logps/chosen": -35.7166748046875,
|
1224 |
+
"logps/rejected": -34.62081527709961,
|
1225 |
+
"loss": 0.3267,
|
1226 |
+
"rewards/accuracies": 0.875,
|
1227 |
+
"rewards/chosen": 1.09420907497406,
|
1228 |
+
"rewards/margins": 1.6778392791748047,
|
1229 |
+
"rewards/rejected": -0.5836302638053894,
|
1230 |
+
"step": 750
|
1231 |
+
},
|
1232 |
+
{
|
1233 |
+
"epoch": 1.97,
|
1234 |
+
"grad_norm": 9.0,
|
1235 |
+
"learning_rate": 2.989809792446417e-06,
|
1236 |
+
"logits/chosen": -1.9088201522827148,
|
1237 |
+
"logits/rejected": -1.910348892211914,
|
1238 |
+
"logps/chosen": -31.48809814453125,
|
1239 |
+
"logps/rejected": -33.39298629760742,
|
1240 |
+
"loss": 0.2839,
|
1241 |
+
"rewards/accuracies": 0.925000011920929,
|
1242 |
+
"rewards/chosen": 1.005824327468872,
|
1243 |
+
"rewards/margins": 1.5582213401794434,
|
1244 |
+
"rewards/rejected": -0.5523970723152161,
|
1245 |
+
"step": 760
|
1246 |
+
},
|
1247 |
+
{
|
1248 |
+
"epoch": 2.0,
|
1249 |
+
"grad_norm": 14.625,
|
1250 |
+
"learning_rate": 2.9341204441673267e-06,
|
1251 |
+
"logits/chosen": -1.882775068283081,
|
1252 |
+
"logits/rejected": -1.8818010091781616,
|
1253 |
+
"logps/chosen": -30.975149154663086,
|
1254 |
+
"logps/rejected": -35.45886993408203,
|
1255 |
+
"loss": 0.3076,
|
1256 |
+
"rewards/accuracies": 0.908333420753479,
|
1257 |
+
"rewards/chosen": 1.0804139375686646,
|
1258 |
+
"rewards/margins": 1.4680591821670532,
|
1259 |
+
"rewards/rejected": -0.3876451253890991,
|
1260 |
+
"step": 770
|
1261 |
+
},
|
1262 |
+
{
|
1263 |
+
"epoch": 2.03,
|
1264 |
+
"grad_norm": 6.53125,
|
1265 |
+
"learning_rate": 2.878208065043501e-06,
|
1266 |
+
"logits/chosen": -1.882063627243042,
|
1267 |
+
"logits/rejected": -1.8813728094100952,
|
1268 |
+
"logps/chosen": -33.322509765625,
|
1269 |
+
"logps/rejected": -32.38517379760742,
|
1270 |
+
"loss": 0.1884,
|
1271 |
+
"rewards/accuracies": 0.949999988079071,
|
1272 |
+
"rewards/chosen": 1.193357229232788,
|
1273 |
+
"rewards/margins": 2.062441349029541,
|
1274 |
+
"rewards/rejected": -0.8690838813781738,
|
1275 |
+
"step": 780
|
1276 |
+
},
|
1277 |
+
{
|
1278 |
+
"epoch": 2.05,
|
1279 |
+
"grad_norm": 10.0625,
|
1280 |
+
"learning_rate": 2.8221013802485974e-06,
|
1281 |
+
"logits/chosen": -1.9113022089004517,
|
1282 |
+
"logits/rejected": -1.909850835800171,
|
1283 |
+
"logps/chosen": -28.325769424438477,
|
1284 |
+
"logps/rejected": -33.69379425048828,
|
1285 |
+
"loss": 0.1718,
|
1286 |
+
"rewards/accuracies": 0.9750000238418579,
|
1287 |
+
"rewards/chosen": 1.1136726140975952,
|
1288 |
+
"rewards/margins": 2.0654101371765137,
|
1289 |
+
"rewards/rejected": -0.9517375826835632,
|
1290 |
+
"step": 790
|
1291 |
+
},
|
1292 |
+
{
|
1293 |
+
"epoch": 2.08,
|
1294 |
+
"grad_norm": 4.375,
|
1295 |
+
"learning_rate": 2.76582921478147e-06,
|
1296 |
+
"logits/chosen": -1.9627540111541748,
|
1297 |
+
"logits/rejected": -1.959313988685608,
|
1298 |
+
"logps/chosen": -31.032363891601562,
|
1299 |
+
"logps/rejected": -35.28049087524414,
|
1300 |
+
"loss": 0.1649,
|
1301 |
+
"rewards/accuracies": 0.9375,
|
1302 |
+
"rewards/chosen": 1.2553597688674927,
|
1303 |
+
"rewards/margins": 2.2286696434020996,
|
1304 |
+
"rewards/rejected": -0.9733098745346069,
|
1305 |
+
"step": 800
|
1306 |
+
},
|
1307 |
+
{
|
1308 |
+
"epoch": 2.08,
|
1309 |
+
"eval_logits/chosen": -2.164193868637085,
|
1310 |
+
"eval_logits/rejected": -2.1594185829162598,
|
1311 |
+
"eval_logps/chosen": -34.28217315673828,
|
1312 |
+
"eval_logps/rejected": -37.94434356689453,
|
1313 |
+
"eval_loss": 0.6811564564704895,
|
1314 |
+
"eval_rewards/accuracies": 0.595099687576294,
|
1315 |
+
"eval_rewards/chosen": -0.22285737097263336,
|
1316 |
+
"eval_rewards/margins": 0.16209454834461212,
|
1317 |
+
"eval_rewards/rejected": -0.3849518895149231,
|
1318 |
+
"eval_runtime": 145.2348,
|
1319 |
+
"eval_samples_per_second": 2.362,
|
1320 |
+
"eval_steps_per_second": 0.296,
|
1321 |
+
"step": 800
|
1322 |
+
},
|
1323 |
+
{
|
1324 |
+
"epoch": 2.1,
|
1325 |
+
"grad_norm": 7.375,
|
1326 |
+
"learning_rate": 2.7094204786572254e-06,
|
1327 |
+
"logits/chosen": -1.793891191482544,
|
1328 |
+
"logits/rejected": -1.7861675024032593,
|
1329 |
+
"logps/chosen": -32.827354431152344,
|
1330 |
+
"logps/rejected": -35.588600158691406,
|
1331 |
+
"loss": 0.1606,
|
1332 |
+
"rewards/accuracies": 0.9375,
|
1333 |
+
"rewards/chosen": 1.465995192527771,
|
1334 |
+
"rewards/margins": 2.391630172729492,
|
1335 |
+
"rewards/rejected": -0.9256349802017212,
|
1336 |
+
"step": 810
|
1337 |
+
},
|
1338 |
+
{
|
1339 |
+
"epoch": 2.13,
|
1340 |
+
"grad_norm": 8.75,
|
1341 |
+
"learning_rate": 2.6529041520546072e-06,
|
1342 |
+
"logits/chosen": -1.8692048788070679,
|
1343 |
+
"logits/rejected": -1.879642128944397,
|
1344 |
+
"logps/chosen": -34.127784729003906,
|
1345 |
+
"logps/rejected": -33.262062072753906,
|
1346 |
+
"loss": 0.1935,
|
1347 |
+
"rewards/accuracies": 0.949999988079071,
|
1348 |
+
"rewards/chosen": 1.4111144542694092,
|
1349 |
+
"rewards/margins": 2.158674955368042,
|
1350 |
+
"rewards/rejected": -0.7475605607032776,
|
1351 |
+
"step": 820
|
1352 |
+
},
|
1353 |
+
{
|
1354 |
+
"epoch": 2.16,
|
1355 |
+
"grad_norm": 11.375,
|
1356 |
+
"learning_rate": 2.5963092704273302e-06,
|
1357 |
+
"logits/chosen": -1.918723702430725,
|
1358 |
+
"logits/rejected": -1.9233121871948242,
|
1359 |
+
"logps/chosen": -33.50402069091797,
|
1360 |
+
"logps/rejected": -30.132360458374023,
|
1361 |
+
"loss": 0.1956,
|
1362 |
+
"rewards/accuracies": 0.9624999761581421,
|
1363 |
+
"rewards/chosen": 1.2588849067687988,
|
1364 |
+
"rewards/margins": 2.036588430404663,
|
1365 |
+
"rewards/rejected": -0.777703583240509,
|
1366 |
+
"step": 830
|
1367 |
+
},
|
1368 |
+
{
|
1369 |
+
"epoch": 2.18,
|
1370 |
+
"grad_norm": 6.65625,
|
1371 |
+
"learning_rate": 2.53966490958702e-06,
|
1372 |
+
"logits/chosen": -1.9170500040054321,
|
1373 |
+
"logits/rejected": -1.9250987768173218,
|
1374 |
+
"logps/chosen": -33.287925720214844,
|
1375 |
+
"logps/rejected": -30.88974952697754,
|
1376 |
+
"loss": 0.1648,
|
1377 |
+
"rewards/accuracies": 0.9750000238418579,
|
1378 |
+
"rewards/chosen": 1.350099802017212,
|
1379 |
+
"rewards/margins": 2.28794264793396,
|
1380 |
+
"rewards/rejected": -0.937842845916748,
|
1381 |
+
"step": 840
|
1382 |
+
},
|
1383 |
+
{
|
1384 |
+
"epoch": 2.21,
|
1385 |
+
"grad_norm": 9.1875,
|
1386 |
+
"learning_rate": 2.4830001707654135e-06,
|
1387 |
+
"logits/chosen": -1.8453495502471924,
|
1388 |
+
"logits/rejected": -1.835889458656311,
|
1389 |
+
"logps/chosen": -30.441539764404297,
|
1390 |
+
"logps/rejected": -32.993690490722656,
|
1391 |
+
"loss": 0.1814,
|
1392 |
+
"rewards/accuracies": 0.987500011920929,
|
1393 |
+
"rewards/chosen": 1.2164894342422485,
|
1394 |
+
"rewards/margins": 2.098388195037842,
|
1395 |
+
"rewards/rejected": -0.8818984031677246,
|
1396 |
+
"step": 850
|
1397 |
+
},
|
1398 |
+
{
|
1399 |
+
"epoch": 2.23,
|
1400 |
+
"grad_norm": 5.25,
|
1401 |
+
"learning_rate": 2.4263441656635054e-06,
|
1402 |
+
"logits/chosen": -1.983048677444458,
|
1403 |
+
"logits/rejected": -1.973232626914978,
|
1404 |
+
"logps/chosen": -24.832033157348633,
|
1405 |
+
"logps/rejected": -30.95058250427246,
|
1406 |
+
"loss": 0.1791,
|
1407 |
+
"rewards/accuracies": 0.9375,
|
1408 |
+
"rewards/chosen": 1.2528350353240967,
|
1409 |
+
"rewards/margins": 2.2232518196105957,
|
1410 |
+
"rewards/rejected": -0.9704168438911438,
|
1411 |
+
"step": 860
|
1412 |
+
},
|
1413 |
+
{
|
1414 |
+
"epoch": 2.26,
|
1415 |
+
"grad_norm": 7.1875,
|
1416 |
+
"learning_rate": 2.3697260014953107e-06,
|
1417 |
+
"logits/chosen": -1.8357470035552979,
|
1418 |
+
"logits/rejected": -1.8370872735977173,
|
1419 |
+
"logps/chosen": -32.829017639160156,
|
1420 |
+
"logps/rejected": -30.976612091064453,
|
1421 |
+
"loss": 0.1644,
|
1422 |
+
"rewards/accuracies": 0.9624999761581421,
|
1423 |
+
"rewards/chosen": 1.353589653968811,
|
1424 |
+
"rewards/margins": 2.2482855319976807,
|
1425 |
+
"rewards/rejected": -0.8946956396102905,
|
1426 |
+
"step": 870
|
1427 |
+
},
|
1428 |
+
{
|
1429 |
+
"epoch": 2.29,
|
1430 |
+
"grad_norm": 6.46875,
|
1431 |
+
"learning_rate": 2.3131747660339396e-06,
|
1432 |
+
"logits/chosen": -1.8432958126068115,
|
1433 |
+
"logits/rejected": -1.8441736698150635,
|
1434 |
+
"logps/chosen": -31.128490447998047,
|
1435 |
+
"logps/rejected": -34.13156509399414,
|
1436 |
+
"loss": 0.2286,
|
1437 |
+
"rewards/accuracies": 0.8999999761581421,
|
1438 |
+
"rewards/chosen": 1.1346783638000488,
|
1439 |
+
"rewards/margins": 2.107259750366211,
|
1440 |
+
"rewards/rejected": -0.9725813865661621,
|
1441 |
+
"step": 880
|
1442 |
+
},
|
1443 |
+
{
|
1444 |
+
"epoch": 2.31,
|
1445 |
+
"grad_norm": 6.4375,
|
1446 |
+
"learning_rate": 2.256719512667651e-06,
|
1447 |
+
"logits/chosen": -1.747079849243164,
|
1448 |
+
"logits/rejected": -1.7454869747161865,
|
1449 |
+
"logps/chosen": -34.19733810424805,
|
1450 |
+
"logps/rejected": -37.23557662963867,
|
1451 |
+
"loss": 0.1614,
|
1452 |
+
"rewards/accuracies": 0.949999988079071,
|
1453 |
+
"rewards/chosen": 1.3208175897598267,
|
1454 |
+
"rewards/margins": 2.543323040008545,
|
1455 |
+
"rewards/rejected": -1.2225055694580078,
|
1456 |
+
"step": 890
|
1457 |
+
},
|
1458 |
+
{
|
1459 |
+
"epoch": 2.34,
|
1460 |
+
"grad_norm": 6.0,
|
1461 |
+
"learning_rate": 2.2003892454735786e-06,
|
1462 |
+
"logits/chosen": -1.8969894647598267,
|
1463 |
+
"logits/rejected": -1.8902244567871094,
|
1464 |
+
"logps/chosen": -30.53342056274414,
|
1465 |
+
"logps/rejected": -33.870323181152344,
|
1466 |
+
"loss": 0.1691,
|
1467 |
+
"rewards/accuracies": 0.9750000238418579,
|
1468 |
+
"rewards/chosen": 1.3222969770431519,
|
1469 |
+
"rewards/margins": 2.3233275413513184,
|
1470 |
+
"rewards/rejected": -1.0010308027267456,
|
1471 |
+
"step": 900
|
1472 |
+
},
|
1473 |
+
{
|
1474 |
+
"epoch": 2.34,
|
1475 |
+
"eval_logits/chosen": -2.15236496925354,
|
1476 |
+
"eval_logits/rejected": -2.1475820541381836,
|
1477 |
+
"eval_logps/chosen": -34.313846588134766,
|
1478 |
+
"eval_logps/rejected": -37.98138427734375,
|
1479 |
+
"eval_loss": 0.6880638599395752,
|
1480 |
+
"eval_rewards/accuracies": 0.5830564498901367,
|
1481 |
+
"eval_rewards/chosen": -0.2513664662837982,
|
1482 |
+
"eval_rewards/margins": 0.16692043840885162,
|
1483 |
+
"eval_rewards/rejected": -0.41828688979148865,
|
1484 |
+
"eval_runtime": 145.3621,
|
1485 |
+
"eval_samples_per_second": 2.36,
|
1486 |
+
"eval_steps_per_second": 0.296,
|
1487 |
+
"step": 900
|
1488 |
+
},
|
1489 |
+
{
|
1490 |
+
"epoch": 2.36,
|
1491 |
+
"grad_norm": 12.0625,
|
1492 |
+
"learning_rate": 2.1442129043167877e-06,
|
1493 |
+
"logits/chosen": -1.9361755847930908,
|
1494 |
+
"logits/rejected": -1.9315325021743774,
|
1495 |
+
"logps/chosen": -32.13105010986328,
|
1496 |
+
"logps/rejected": -35.70573425292969,
|
1497 |
+
"loss": 0.1596,
|
1498 |
+
"rewards/accuracies": 0.9750000238418579,
|
1499 |
+
"rewards/chosen": 1.2839645147323608,
|
1500 |
+
"rewards/margins": 2.3433659076690674,
|
1501 |
+
"rewards/rejected": -1.0594011545181274,
|
1502 |
+
"step": 910
|
1503 |
+
},
|
1504 |
+
{
|
1505 |
+
"epoch": 2.39,
|
1506 |
+
"grad_norm": 6.5,
|
1507 |
+
"learning_rate": 2.088219349982323e-06,
|
1508 |
+
"logits/chosen": -1.8552948236465454,
|
1509 |
+
"logits/rejected": -1.860346794128418,
|
1510 |
+
"logps/chosen": -33.7907829284668,
|
1511 |
+
"logps/rejected": -33.559757232666016,
|
1512 |
+
"loss": 0.1777,
|
1513 |
+
"rewards/accuracies": 0.949999988079071,
|
1514 |
+
"rewards/chosen": 1.4020235538482666,
|
1515 |
+
"rewards/margins": 2.3558895587921143,
|
1516 |
+
"rewards/rejected": -0.9538658261299133,
|
1517 |
+
"step": 920
|
1518 |
+
},
|
1519 |
+
{
|
1520 |
+
"epoch": 2.42,
|
1521 |
+
"grad_norm": 6.53125,
|
1522 |
+
"learning_rate": 2.0324373493478803e-06,
|
1523 |
+
"logits/chosen": -1.953168272972107,
|
1524 |
+
"logits/rejected": -1.943918228149414,
|
1525 |
+
"logps/chosen": -30.902240753173828,
|
1526 |
+
"logps/rejected": -35.58992385864258,
|
1527 |
+
"loss": 0.1581,
|
1528 |
+
"rewards/accuracies": 1.0,
|
1529 |
+
"rewards/chosen": 1.2444217205047607,
|
1530 |
+
"rewards/margins": 2.276834011077881,
|
1531 |
+
"rewards/rejected": -1.032412052154541,
|
1532 |
+
"step": 930
|
1533 |
+
},
|
1534 |
+
{
|
1535 |
+
"epoch": 2.44,
|
1536 |
+
"grad_norm": 8.5,
|
1537 |
+
"learning_rate": 1.976895560604729e-06,
|
1538 |
+
"logits/chosen": -1.879119873046875,
|
1539 |
+
"logits/rejected": -1.8759711980819702,
|
1540 |
+
"logps/chosen": -30.023090362548828,
|
1541 |
+
"logps/rejected": -33.454833984375,
|
1542 |
+
"loss": 0.2017,
|
1543 |
+
"rewards/accuracies": 0.9750000238418579,
|
1544 |
+
"rewards/chosen": 1.1388328075408936,
|
1545 |
+
"rewards/margins": 2.029343366622925,
|
1546 |
+
"rewards/rejected": -0.8905106782913208,
|
1547 |
+
"step": 940
|
1548 |
+
},
|
1549 |
+
{
|
1550 |
+
"epoch": 2.47,
|
1551 |
+
"grad_norm": 4.625,
|
1552 |
+
"learning_rate": 1.921622518534466e-06,
|
1553 |
+
"logits/chosen": -1.8175594806671143,
|
1554 |
+
"logits/rejected": -1.8249561786651611,
|
1555 |
+
"logps/chosen": -31.570093154907227,
|
1556 |
+
"logps/rejected": -36.65325164794922,
|
1557 |
+
"loss": 0.2168,
|
1558 |
+
"rewards/accuracies": 0.9125000238418579,
|
1559 |
+
"rewards/chosen": 1.1708660125732422,
|
1560 |
+
"rewards/margins": 2.163855791091919,
|
1561 |
+
"rewards/rejected": -0.9929895401000977,
|
1562 |
+
"step": 950
|
1563 |
+
},
|
1564 |
+
{
|
1565 |
+
"epoch": 2.49,
|
1566 |
+
"grad_norm": 3.953125,
|
1567 |
+
"learning_rate": 1.8666466198491794e-06,
|
1568 |
+
"logits/chosen": -1.8278528451919556,
|
1569 |
+
"logits/rejected": -1.8209831714630127,
|
1570 |
+
"logps/chosen": -32.37910842895508,
|
1571 |
+
"logps/rejected": -37.90230178833008,
|
1572 |
+
"loss": 0.1494,
|
1573 |
+
"rewards/accuracies": 0.9750000238418579,
|
1574 |
+
"rewards/chosen": 1.3044663667678833,
|
1575 |
+
"rewards/margins": 2.4776108264923096,
|
1576 |
+
"rewards/rejected": -1.1731446981430054,
|
1577 |
+
"step": 960
|
1578 |
+
},
|
1579 |
+
{
|
1580 |
+
"epoch": 2.52,
|
1581 |
+
"grad_norm": 5.375,
|
1582 |
+
"learning_rate": 1.8119961086025376e-06,
|
1583 |
+
"logits/chosen": -1.8297231197357178,
|
1584 |
+
"logits/rejected": -1.8303101062774658,
|
1585 |
+
"logps/chosen": -29.768722534179688,
|
1586 |
+
"logps/rejected": -33.98983383178711,
|
1587 |
+
"loss": 0.1847,
|
1588 |
+
"rewards/accuracies": 0.9375,
|
1589 |
+
"rewards/chosen": 1.3209145069122314,
|
1590 |
+
"rewards/margins": 2.2286314964294434,
|
1591 |
+
"rewards/rejected": -0.9077168703079224,
|
1592 |
+
"step": 970
|
1593 |
+
},
|
1594 |
+
{
|
1595 |
+
"epoch": 2.55,
|
1596 |
+
"grad_norm": 7.9375,
|
1597 |
+
"learning_rate": 1.7576990616793139e-06,
|
1598 |
+
"logits/chosen": -1.8326854705810547,
|
1599 |
+
"logits/rejected": -1.8439195156097412,
|
1600 |
+
"logps/chosen": -31.450542449951172,
|
1601 |
+
"logps/rejected": -35.19971466064453,
|
1602 |
+
"loss": 0.168,
|
1603 |
+
"rewards/accuracies": 0.9125000238418579,
|
1604 |
+
"rewards/chosen": 1.4635357856750488,
|
1605 |
+
"rewards/margins": 2.4251866340637207,
|
1606 |
+
"rewards/rejected": -0.9616511464118958,
|
1607 |
+
"step": 980
|
1608 |
+
},
|
1609 |
+
{
|
1610 |
+
"epoch": 2.57,
|
1611 |
+
"grad_norm": 6.0,
|
1612 |
+
"learning_rate": 1.7037833743707892e-06,
|
1613 |
+
"logits/chosen": -1.903786301612854,
|
1614 |
+
"logits/rejected": -1.901314377784729,
|
1615 |
+
"logps/chosen": -34.712928771972656,
|
1616 |
+
"logps/rejected": -33.3343620300293,
|
1617 |
+
"loss": 0.2409,
|
1618 |
+
"rewards/accuracies": 0.8999999761581421,
|
1619 |
+
"rewards/chosen": 1.2673081159591675,
|
1620 |
+
"rewards/margins": 2.1089327335357666,
|
1621 |
+
"rewards/rejected": -0.8416244387626648,
|
1622 |
+
"step": 990
|
1623 |
+
},
|
1624 |
+
{
|
1625 |
+
"epoch": 2.6,
|
1626 |
+
"grad_norm": 9.5,
|
1627 |
+
"learning_rate": 1.6502767460434588e-06,
|
1628 |
+
"logits/chosen": -1.875101089477539,
|
1629 |
+
"logits/rejected": -1.8791959285736084,
|
1630 |
+
"logps/chosen": -33.045143127441406,
|
1631 |
+
"logps/rejected": -35.46049880981445,
|
1632 |
+
"loss": 0.1953,
|
1633 |
+
"rewards/accuracies": 0.949999988079071,
|
1634 |
+
"rewards/chosen": 1.2763327360153198,
|
1635 |
+
"rewards/margins": 2.153337001800537,
|
1636 |
+
"rewards/rejected": -0.8770040273666382,
|
1637 |
+
"step": 1000
|
1638 |
+
},
|
1639 |
+
{
|
1640 |
+
"epoch": 2.6,
|
1641 |
+
"eval_logits/chosen": -2.1447298526763916,
|
1642 |
+
"eval_logits/rejected": -2.139974594116211,
|
1643 |
+
"eval_logps/chosen": -34.36631774902344,
|
1644 |
+
"eval_logps/rejected": -38.036582946777344,
|
1645 |
+
"eval_loss": 0.695711076259613,
|
1646 |
+
"eval_rewards/accuracies": 0.5917773842811584,
|
1647 |
+
"eval_rewards/chosen": -0.2985913157463074,
|
1648 |
+
"eval_rewards/margins": 0.16937348246574402,
|
1649 |
+
"eval_rewards/rejected": -0.4679647982120514,
|
1650 |
+
"eval_runtime": 145.3859,
|
1651 |
+
"eval_samples_per_second": 2.359,
|
1652 |
+
"eval_steps_per_second": 0.296,
|
1653 |
+
"step": 1000
|
1654 |
+
},
|
1655 |
+
{
|
1656 |
+
"epoch": 2.62,
|
1657 |
+
"grad_norm": 20.25,
|
1658 |
+
"learning_rate": 1.5972066659083796e-06,
|
1659 |
+
"logits/chosen": -1.8841043710708618,
|
1660 |
+
"logits/rejected": -1.8890196084976196,
|
1661 |
+
"logps/chosen": -31.524646759033203,
|
1662 |
+
"logps/rejected": -33.430824279785156,
|
1663 |
+
"loss": 0.2343,
|
1664 |
+
"rewards/accuracies": 0.925000011920929,
|
1665 |
+
"rewards/chosen": 1.1482000350952148,
|
1666 |
+
"rewards/margins": 1.9668105840682983,
|
1667 |
+
"rewards/rejected": -0.8186105489730835,
|
1668 |
+
"step": 1010
|
1669 |
+
},
|
1670 |
+
{
|
1671 |
+
"epoch": 2.65,
|
1672 |
+
"grad_norm": 15.5625,
|
1673 |
+
"learning_rate": 1.5446003988985041e-06,
|
1674 |
+
"logits/chosen": -1.843640685081482,
|
1675 |
+
"logits/rejected": -1.8461425304412842,
|
1676 |
+
"logps/chosen": -29.06394386291504,
|
1677 |
+
"logps/rejected": -32.841705322265625,
|
1678 |
+
"loss": 0.1766,
|
1679 |
+
"rewards/accuracies": 0.949999988079071,
|
1680 |
+
"rewards/chosen": 1.2435271739959717,
|
1681 |
+
"rewards/margins": 2.2718143463134766,
|
1682 |
+
"rewards/rejected": -1.0282870531082153,
|
1683 |
+
"step": 1020
|
1684 |
+
},
|
1685 |
+
{
|
1686 |
+
"epoch": 2.68,
|
1687 |
+
"grad_norm": 4.625,
|
1688 |
+
"learning_rate": 1.4924849716612211e-06,
|
1689 |
+
"logits/chosen": -1.8973257541656494,
|
1690 |
+
"logits/rejected": -1.8918126821517944,
|
1691 |
+
"logps/chosen": -32.64158248901367,
|
1692 |
+
"logps/rejected": -34.6368293762207,
|
1693 |
+
"loss": 0.1812,
|
1694 |
+
"rewards/accuracies": 0.987500011920929,
|
1695 |
+
"rewards/chosen": 1.2127745151519775,
|
1696 |
+
"rewards/margins": 2.2197234630584717,
|
1697 |
+
"rewards/rejected": -1.0069488286972046,
|
1698 |
+
"step": 1030
|
1699 |
+
},
|
1700 |
+
{
|
1701 |
+
"epoch": 2.7,
|
1702 |
+
"grad_norm": 17.875,
|
1703 |
+
"learning_rate": 1.440887158673332e-06,
|
1704 |
+
"logits/chosen": -1.8320951461791992,
|
1705 |
+
"logits/rejected": -1.836050033569336,
|
1706 |
+
"logps/chosen": -34.43301010131836,
|
1707 |
+
"logps/rejected": -35.46531677246094,
|
1708 |
+
"loss": 0.1938,
|
1709 |
+
"rewards/accuracies": 0.9624999761581421,
|
1710 |
+
"rewards/chosen": 1.1886460781097412,
|
1711 |
+
"rewards/margins": 2.1298069953918457,
|
1712 |
+
"rewards/rejected": -0.9411608576774597,
|
1713 |
+
"step": 1040
|
1714 |
+
},
|
1715 |
+
{
|
1716 |
+
"epoch": 2.73,
|
1717 |
+
"grad_norm": 7.09375,
|
1718 |
+
"learning_rate": 1.3898334684855647e-06,
|
1719 |
+
"logits/chosen": -1.8290131092071533,
|
1720 |
+
"logits/rejected": -1.841897964477539,
|
1721 |
+
"logps/chosen": -30.590591430664062,
|
1722 |
+
"logps/rejected": -34.06149673461914,
|
1723 |
+
"loss": 0.2044,
|
1724 |
+
"rewards/accuracies": 0.9375,
|
1725 |
+
"rewards/chosen": 1.1450647115707397,
|
1726 |
+
"rewards/margins": 2.082374095916748,
|
1727 |
+
"rewards/rejected": -0.9373094439506531,
|
1728 |
+
"step": 1050
|
1729 |
+
},
|
1730 |
+
{
|
1731 |
+
"epoch": 2.75,
|
1732 |
+
"grad_norm": 19.25,
|
1733 |
+
"learning_rate": 1.3393501301037245e-06,
|
1734 |
+
"logits/chosen": -1.8185697793960571,
|
1735 |
+
"logits/rejected": -1.8119618892669678,
|
1736 |
+
"logps/chosen": -30.647253036499023,
|
1737 |
+
"logps/rejected": -34.33002471923828,
|
1738 |
+
"loss": 0.1727,
|
1739 |
+
"rewards/accuracies": 0.9750000238418579,
|
1740 |
+
"rewards/chosen": 1.3608683347702026,
|
1741 |
+
"rewards/margins": 2.2717270851135254,
|
1742 |
+
"rewards/rejected": -0.9108586311340332,
|
1743 |
+
"step": 1060
|
1744 |
+
},
|
1745 |
+
{
|
1746 |
+
"epoch": 2.78,
|
1747 |
+
"grad_norm": 5.1875,
|
1748 |
+
"learning_rate": 1.2894630795134454e-06,
|
1749 |
+
"logits/chosen": -1.944183349609375,
|
1750 |
+
"logits/rejected": -1.944902777671814,
|
1751 |
+
"logps/chosen": -32.107669830322266,
|
1752 |
+
"logps/rejected": -33.907928466796875,
|
1753 |
+
"loss": 0.1656,
|
1754 |
+
"rewards/accuracies": 0.9750000238418579,
|
1755 |
+
"rewards/chosen": 1.2954630851745605,
|
1756 |
+
"rewards/margins": 2.2523720264434814,
|
1757 |
+
"rewards/rejected": -0.9569088816642761,
|
1758 |
+
"step": 1070
|
1759 |
+
},
|
1760 |
+
{
|
1761 |
+
"epoch": 2.81,
|
1762 |
+
"grad_norm": 10.0,
|
1763 |
+
"learning_rate": 1.2401979463554984e-06,
|
1764 |
+
"logits/chosen": -1.8745800256729126,
|
1765 |
+
"logits/rejected": -1.8734228610992432,
|
1766 |
+
"logps/chosen": -32.74195861816406,
|
1767 |
+
"logps/rejected": -34.108428955078125,
|
1768 |
+
"loss": 0.2386,
|
1769 |
+
"rewards/accuracies": 0.887499988079071,
|
1770 |
+
"rewards/chosen": 1.2130850553512573,
|
1771 |
+
"rewards/margins": 2.0380234718322754,
|
1772 |
+
"rewards/rejected": -0.8249381184577942,
|
1773 |
+
"step": 1080
|
1774 |
+
},
|
1775 |
+
{
|
1776 |
+
"epoch": 2.83,
|
1777 |
+
"grad_norm": 4.96875,
|
1778 |
+
"learning_rate": 1.1915800407584705e-06,
|
1779 |
+
"logits/chosen": -1.8999583721160889,
|
1780 |
+
"logits/rejected": -1.8924366235733032,
|
1781 |
+
"logps/chosen": -32.777381896972656,
|
1782 |
+
"logps/rejected": -31.974105834960938,
|
1783 |
+
"loss": 0.1764,
|
1784 |
+
"rewards/accuracies": 0.9624999761581421,
|
1785 |
+
"rewards/chosen": 1.3036444187164307,
|
1786 |
+
"rewards/margins": 2.142960548400879,
|
1787 |
+
"rewards/rejected": -0.8393163681030273,
|
1788 |
+
"step": 1090
|
1789 |
+
},
|
1790 |
+
{
|
1791 |
+
"epoch": 2.86,
|
1792 |
+
"grad_norm": 5.125,
|
1793 |
+
"learning_rate": 1.1436343403356019e-06,
|
1794 |
+
"logits/chosen": -1.8712406158447266,
|
1795 |
+
"logits/rejected": -1.870269775390625,
|
1796 |
+
"logps/chosen": -33.90606689453125,
|
1797 |
+
"logps/rejected": -37.61457061767578,
|
1798 |
+
"loss": 0.1463,
|
1799 |
+
"rewards/accuracies": 1.0,
|
1800 |
+
"rewards/chosen": 1.361112117767334,
|
1801 |
+
"rewards/margins": 2.496788263320923,
|
1802 |
+
"rewards/rejected": -1.1356757879257202,
|
1803 |
+
"step": 1100
|
1804 |
+
},
|
1805 |
+
{
|
1806 |
+
"epoch": 2.86,
|
1807 |
+
"eval_logits/chosen": -2.142699956893921,
|
1808 |
+
"eval_logits/rejected": -2.1379306316375732,
|
1809 |
+
"eval_logps/chosen": -34.36824035644531,
|
1810 |
+
"eval_logps/rejected": -38.02313995361328,
|
1811 |
+
"eval_loss": 0.7009721994400024,
|
1812 |
+
"eval_rewards/accuracies": 0.5714285373687744,
|
1813 |
+
"eval_rewards/chosen": -0.30031847953796387,
|
1814 |
+
"eval_rewards/margins": 0.15554992854595184,
|
1815 |
+
"eval_rewards/rejected": -0.4558684229850769,
|
1816 |
+
"eval_runtime": 145.3955,
|
1817 |
+
"eval_samples_per_second": 2.359,
|
1818 |
+
"eval_steps_per_second": 0.296,
|
1819 |
+
"step": 1100
|
1820 |
+
},
|
1821 |
+
{
|
1822 |
+
"epoch": 2.88,
|
1823 |
+
"grad_norm": 6.53125,
|
1824 |
+
"learning_rate": 1.0963854773524548e-06,
|
1825 |
+
"logits/chosen": -1.9193534851074219,
|
1826 |
+
"logits/rejected": -1.9247900247573853,
|
1827 |
+
"logps/chosen": -34.15703582763672,
|
1828 |
+
"logps/rejected": -36.63218688964844,
|
1829 |
+
"loss": 0.2075,
|
1830 |
+
"rewards/accuracies": 0.9125000238418579,
|
1831 |
+
"rewards/chosen": 1.179071307182312,
|
1832 |
+
"rewards/margins": 2.15881085395813,
|
1833 |
+
"rewards/rejected": -0.9797393679618835,
|
1834 |
+
"step": 1110
|
1835 |
+
},
|
1836 |
+
{
|
1837 |
+
"epoch": 2.91,
|
1838 |
+
"grad_norm": 4.625,
|
1839 |
+
"learning_rate": 1.049857726072005e-06,
|
1840 |
+
"logits/chosen": -1.888942003250122,
|
1841 |
+
"logits/rejected": -1.8869132995605469,
|
1842 |
+
"logps/chosen": -31.29522132873535,
|
1843 |
+
"logps/rejected": -34.09874725341797,
|
1844 |
+
"loss": 0.1873,
|
1845 |
+
"rewards/accuracies": 0.9624999761581421,
|
1846 |
+
"rewards/chosen": 1.1888843774795532,
|
1847 |
+
"rewards/margins": 2.193281650543213,
|
1848 |
+
"rewards/rejected": -1.0043971538543701,
|
1849 |
+
"step": 1120
|
1850 |
+
},
|
1851 |
+
{
|
1852 |
+
"epoch": 2.94,
|
1853 |
+
"grad_norm": 5.90625,
|
1854 |
+
"learning_rate": 1.0040749902836508e-06,
|
1855 |
+
"logits/chosen": -1.7918437719345093,
|
1856 |
+
"logits/rejected": -1.7945436239242554,
|
1857 |
+
"logps/chosen": -28.33380126953125,
|
1858 |
+
"logps/rejected": -31.490909576416016,
|
1859 |
+
"loss": 0.156,
|
1860 |
+
"rewards/accuracies": 0.9624999761581421,
|
1861 |
+
"rewards/chosen": 1.273350715637207,
|
1862 |
+
"rewards/margins": 2.394632577896118,
|
1863 |
+
"rewards/rejected": -1.1212818622589111,
|
1864 |
+
"step": 1130
|
1865 |
+
},
|
1866 |
+
{
|
1867 |
+
"epoch": 2.96,
|
1868 |
+
"grad_norm": 5.8125,
|
1869 |
+
"learning_rate": 9.59060791022566e-07,
|
1870 |
+
"logits/chosen": -1.8893934488296509,
|
1871 |
+
"logits/rejected": -1.8862323760986328,
|
1872 |
+
"logps/chosen": -31.71720314025879,
|
1873 |
+
"logps/rejected": -33.38925552368164,
|
1874 |
+
"loss": 0.2101,
|
1875 |
+
"rewards/accuracies": 0.949999988079071,
|
1876 |
+
"rewards/chosen": 1.337890863418579,
|
1877 |
+
"rewards/margins": 2.114030122756958,
|
1878 |
+
"rewards/rejected": -0.7761393189430237,
|
1879 |
+
"step": 1140
|
1880 |
+
},
|
1881 |
+
{
|
1882 |
+
"epoch": 2.99,
|
1883 |
+
"grad_norm": 8.8125,
|
1884 |
+
"learning_rate": 9.148382544856885e-07,
|
1885 |
+
"logits/chosen": -1.8082507848739624,
|
1886 |
+
"logits/rejected": -1.8019250631332397,
|
1887 |
+
"logps/chosen": -26.8853702545166,
|
1888 |
+
"logps/rejected": -33.35715103149414,
|
1889 |
+
"loss": 0.1765,
|
1890 |
+
"rewards/accuracies": 0.9750000238418579,
|
1891 |
+
"rewards/chosen": 1.1273572444915771,
|
1892 |
+
"rewards/margins": 2.2153234481811523,
|
1893 |
+
"rewards/rejected": -1.0879663228988647,
|
1894 |
+
"step": 1150
|
1895 |
+
},
|
1896 |
+
{
|
1897 |
+
"epoch": 3.01,
|
1898 |
+
"grad_norm": 3.359375,
|
1899 |
+
"learning_rate": 8.714301001505568e-07,
|
1900 |
+
"logits/chosen": -1.922569990158081,
|
1901 |
+
"logits/rejected": -1.920064926147461,
|
1902 |
+
"logps/chosen": -31.649404525756836,
|
1903 |
+
"logps/rejected": -36.346885681152344,
|
1904 |
+
"loss": 0.1692,
|
1905 |
+
"rewards/accuracies": 0.9624999761581421,
|
1906 |
+
"rewards/chosen": 1.3019969463348389,
|
1907 |
+
"rewards/margins": 2.431814670562744,
|
1908 |
+
"rewards/rejected": -1.1298176050186157,
|
1909 |
+
"step": 1160
|
1910 |
+
},
|
1911 |
+
{
|
1912 |
+
"epoch": 3.04,
|
1913 |
+
"grad_norm": 4.625,
|
1914 |
+
"learning_rate": 8.288586291031025e-07,
|
1915 |
+
"logits/chosen": -1.9079128503799438,
|
1916 |
+
"logits/rejected": -1.9058338403701782,
|
1917 |
+
"logps/chosen": -30.213918685913086,
|
1918 |
+
"logps/rejected": -33.71696472167969,
|
1919 |
+
"loss": 0.1383,
|
1920 |
+
"rewards/accuracies": 0.9750000238418579,
|
1921 |
+
"rewards/chosen": 1.2604410648345947,
|
1922 |
+
"rewards/margins": 2.478085517883301,
|
1923 |
+
"rewards/rejected": -1.217644453048706,
|
1924 |
+
"step": 1170
|
1925 |
+
},
|
1926 |
+
{
|
1927 |
+
"epoch": 3.06,
|
1928 |
+
"grad_norm": 4.96875,
|
1929 |
+
"learning_rate": 7.871457125803897e-07,
|
1930 |
+
"logits/chosen": -1.9004592895507812,
|
1931 |
+
"logits/rejected": -1.8882315158843994,
|
1932 |
+
"logps/chosen": -34.54669189453125,
|
1933 |
+
"logps/rejected": -35.38771057128906,
|
1934 |
+
"loss": 0.1478,
|
1935 |
+
"rewards/accuracies": 0.9750000238418579,
|
1936 |
+
"rewards/chosen": 1.348745584487915,
|
1937 |
+
"rewards/margins": 2.5333569049835205,
|
1938 |
+
"rewards/rejected": -1.1846110820770264,
|
1939 |
+
"step": 1180
|
1940 |
+
},
|
1941 |
+
{
|
1942 |
+
"epoch": 3.09,
|
1943 |
+
"grad_norm": 6.75,
|
1944 |
+
"learning_rate": 7.463127807341966e-07,
|
1945 |
+
"logits/chosen": -1.7955455780029297,
|
1946 |
+
"logits/rejected": -1.790220022201538,
|
1947 |
+
"logps/chosen": -33.658992767333984,
|
1948 |
+
"logps/rejected": -35.314979553222656,
|
1949 |
+
"loss": 0.1451,
|
1950 |
+
"rewards/accuracies": 0.9624999761581421,
|
1951 |
+
"rewards/chosen": 1.52162504196167,
|
1952 |
+
"rewards/margins": 2.4910683631896973,
|
1953 |
+
"rewards/rejected": -0.9694433212280273,
|
1954 |
+
"step": 1190
|
1955 |
+
},
|
1956 |
+
{
|
1957 |
+
"epoch": 3.12,
|
1958 |
+
"grad_norm": 6.90625,
|
1959 |
+
"learning_rate": 7.063808116212021e-07,
|
1960 |
+
"logits/chosen": -1.8400506973266602,
|
1961 |
+
"logits/rejected": -1.8432636260986328,
|
1962 |
+
"logps/chosen": -31.264917373657227,
|
1963 |
+
"logps/rejected": -32.59490203857422,
|
1964 |
+
"loss": 0.1796,
|
1965 |
+
"rewards/accuracies": 0.9624999761581421,
|
1966 |
+
"rewards/chosen": 1.2151604890823364,
|
1967 |
+
"rewards/margins": 2.21301007270813,
|
1968 |
+
"rewards/rejected": -0.9978495836257935,
|
1969 |
+
"step": 1200
|
1970 |
+
},
|
1971 |
+
{
|
1972 |
+
"epoch": 3.12,
|
1973 |
+
"eval_logits/chosen": -2.142347574234009,
|
1974 |
+
"eval_logits/rejected": -2.137584686279297,
|
1975 |
+
"eval_logps/chosen": -34.35411071777344,
|
1976 |
+
"eval_logps/rejected": -38.025672912597656,
|
1977 |
+
"eval_loss": 0.6907772421836853,
|
1978 |
+
"eval_rewards/accuracies": 0.5747508406639099,
|
1979 |
+
"eval_rewards/chosen": -0.2876059412956238,
|
1980 |
+
"eval_rewards/margins": 0.1705409586429596,
|
1981 |
+
"eval_rewards/rejected": -0.45814695954322815,
|
1982 |
+
"eval_runtime": 145.3519,
|
1983 |
+
"eval_samples_per_second": 2.36,
|
1984 |
+
"eval_steps_per_second": 0.296,
|
1985 |
+
"step": 1200
|
1986 |
+
},
|
1987 |
+
{
|
1988 |
+
"epoch": 3.14,
|
1989 |
+
"grad_norm": 9.9375,
|
1990 |
+
"learning_rate": 6.673703204254348e-07,
|
1991 |
+
"logits/chosen": -1.8635101318359375,
|
1992 |
+
"logits/rejected": -1.858994722366333,
|
1993 |
+
"logps/chosen": -29.87615966796875,
|
1994 |
+
"logps/rejected": -32.12238311767578,
|
1995 |
+
"loss": 0.1548,
|
1996 |
+
"rewards/accuracies": 0.949999988079071,
|
1997 |
+
"rewards/chosen": 1.2577589750289917,
|
1998 |
+
"rewards/margins": 2.350618362426758,
|
1999 |
+
"rewards/rejected": -1.0928595066070557,
|
2000 |
+
"step": 1210
|
2001 |
+
},
|
2002 |
+
{
|
2003 |
+
"epoch": 3.17,
|
2004 |
+
"grad_norm": 5.375,
|
2005 |
+
"learning_rate": 6.293013489185315e-07,
|
2006 |
+
"logits/chosen": -1.8775825500488281,
|
2007 |
+
"logits/rejected": -1.8721100091934204,
|
2008 |
+
"logps/chosen": -33.751792907714844,
|
2009 |
+
"logps/rejected": -35.85100555419922,
|
2010 |
+
"loss": 0.1601,
|
2011 |
+
"rewards/accuracies": 0.9375,
|
2012 |
+
"rewards/chosen": 1.3699350357055664,
|
2013 |
+
"rewards/margins": 2.522946834564209,
|
2014 |
+
"rewards/rejected": -1.1530119180679321,
|
2015 |
+
"step": 1220
|
2016 |
+
},
|
2017 |
+
{
|
2018 |
+
"epoch": 3.19,
|
2019 |
+
"grad_norm": 4.25,
|
2020 |
+
"learning_rate": 5.921934551632086e-07,
|
2021 |
+
"logits/chosen": -1.879861831665039,
|
2022 |
+
"logits/rejected": -1.8666155338287354,
|
2023 |
+
"logps/chosen": -31.769283294677734,
|
2024 |
+
"logps/rejected": -35.38819122314453,
|
2025 |
+
"loss": 0.1889,
|
2026 |
+
"rewards/accuracies": 0.875,
|
2027 |
+
"rewards/chosen": 1.1922564506530762,
|
2028 |
+
"rewards/margins": 2.333768606185913,
|
2029 |
+
"rewards/rejected": -1.141512393951416,
|
2030 |
+
"step": 1230
|
2031 |
+
},
|
2032 |
+
{
|
2033 |
+
"epoch": 3.22,
|
2034 |
+
"grad_norm": 3.765625,
|
2035 |
+
"learning_rate": 5.560657034652405e-07,
|
2036 |
+
"logits/chosen": -1.9246861934661865,
|
2037 |
+
"logits/rejected": -1.922136664390564,
|
2038 |
+
"logps/chosen": -33.49988555908203,
|
2039 |
+
"logps/rejected": -32.82632064819336,
|
2040 |
+
"loss": 0.1608,
|
2041 |
+
"rewards/accuracies": 0.949999988079071,
|
2042 |
+
"rewards/chosen": 1.3797134160995483,
|
2043 |
+
"rewards/margins": 2.2675235271453857,
|
2044 |
+
"rewards/rejected": -0.8878101110458374,
|
2045 |
+
"step": 1240
|
2046 |
+
},
|
2047 |
+
{
|
2048 |
+
"epoch": 3.25,
|
2049 |
+
"grad_norm": 5.96875,
|
2050 |
+
"learning_rate": 5.2093665457911e-07,
|
2051 |
+
"logits/chosen": -1.851680040359497,
|
2052 |
+
"logits/rejected": -1.8488292694091797,
|
2053 |
+
"logps/chosen": -32.87358856201172,
|
2054 |
+
"logps/rejected": -36.248497009277344,
|
2055 |
+
"loss": 0.1342,
|
2056 |
+
"rewards/accuracies": 0.9624999761581421,
|
2057 |
+
"rewards/chosen": 1.3703018426895142,
|
2058 |
+
"rewards/margins": 2.6765310764312744,
|
2059 |
+
"rewards/rejected": -1.3062288761138916,
|
2060 |
+
"step": 1250
|
2061 |
+
},
|
2062 |
+
{
|
2063 |
+
"epoch": 3.27,
|
2064 |
+
"grad_norm": 4.5625,
|
2065 |
+
"learning_rate": 4.868243561723535e-07,
|
2066 |
+
"logits/chosen": -1.9186599254608154,
|
2067 |
+
"logits/rejected": -1.913433313369751,
|
2068 |
+
"logps/chosen": -29.251794815063477,
|
2069 |
+
"logps/rejected": -33.522377014160156,
|
2070 |
+
"loss": 0.1506,
|
2071 |
+
"rewards/accuracies": 0.949999988079071,
|
2072 |
+
"rewards/chosen": 1.3581030368804932,
|
2073 |
+
"rewards/margins": 2.383944511413574,
|
2074 |
+
"rewards/rejected": -1.025841474533081,
|
2075 |
+
"step": 1260
|
2076 |
+
},
|
2077 |
+
{
|
2078 |
+
"epoch": 3.3,
|
2079 |
+
"grad_norm": 4.15625,
|
2080 |
+
"learning_rate": 4.537463335535161e-07,
|
2081 |
+
"logits/chosen": -1.9449501037597656,
|
2082 |
+
"logits/rejected": -1.9497982263565063,
|
2083 |
+
"logps/chosen": -31.12078285217285,
|
2084 |
+
"logps/rejected": -32.769309997558594,
|
2085 |
+
"loss": 0.1426,
|
2086 |
+
"rewards/accuracies": 0.9624999761581421,
|
2087 |
+
"rewards/chosen": 1.4311946630477905,
|
2088 |
+
"rewards/margins": 2.410348892211914,
|
2089 |
+
"rewards/rejected": -0.9791544079780579,
|
2090 |
+
"step": 1270
|
2091 |
+
},
|
2092 |
+
{
|
2093 |
+
"epoch": 3.32,
|
2094 |
+
"grad_norm": 7.09375,
|
2095 |
+
"learning_rate": 4.217195806684629e-07,
|
2096 |
+
"logits/chosen": -1.8380733728408813,
|
2097 |
+
"logits/rejected": -1.8454326391220093,
|
2098 |
+
"logps/chosen": -33.449485778808594,
|
2099 |
+
"logps/rejected": -33.86452102661133,
|
2100 |
+
"loss": 0.1535,
|
2101 |
+
"rewards/accuracies": 0.9624999761581421,
|
2102 |
+
"rewards/chosen": 1.3119791746139526,
|
2103 |
+
"rewards/margins": 2.4516043663024902,
|
2104 |
+
"rewards/rejected": -1.1396249532699585,
|
2105 |
+
"step": 1280
|
2106 |
+
},
|
2107 |
+
{
|
2108 |
+
"epoch": 3.35,
|
2109 |
+
"grad_norm": 4.625,
|
2110 |
+
"learning_rate": 3.907605513696808e-07,
|
2111 |
+
"logits/chosen": -1.7485191822052002,
|
2112 |
+
"logits/rejected": -1.7507747411727905,
|
2113 |
+
"logps/chosen": -31.557483673095703,
|
2114 |
+
"logps/rejected": -37.77085494995117,
|
2115 |
+
"loss": 0.1711,
|
2116 |
+
"rewards/accuracies": 0.9624999761581421,
|
2117 |
+
"rewards/chosen": 1.416973352432251,
|
2118 |
+
"rewards/margins": 2.511898994445801,
|
2119 |
+
"rewards/rejected": -1.0949256420135498,
|
2120 |
+
"step": 1290
|
2121 |
+
},
|
2122 |
+
{
|
2123 |
+
"epoch": 3.38,
|
2124 |
+
"grad_norm": 4.65625,
|
2125 |
+
"learning_rate": 3.6088515096305675e-07,
|
2126 |
+
"logits/chosen": -1.7980448007583618,
|
2127 |
+
"logits/rejected": -1.8015098571777344,
|
2128 |
+
"logps/chosen": -31.520572662353516,
|
2129 |
+
"logps/rejected": -33.67357635498047,
|
2130 |
+
"loss": 0.1264,
|
2131 |
+
"rewards/accuracies": 0.987500011920929,
|
2132 |
+
"rewards/chosen": 1.5264867544174194,
|
2133 |
+
"rewards/margins": 2.614957332611084,
|
2134 |
+
"rewards/rejected": -1.088470458984375,
|
2135 |
+
"step": 1300
|
2136 |
+
},
|
2137 |
+
{
|
2138 |
+
"epoch": 3.38,
|
2139 |
+
"eval_logits/chosen": -2.1421782970428467,
|
2140 |
+
"eval_logits/rejected": -2.1374199390411377,
|
2141 |
+
"eval_logps/chosen": -34.34251022338867,
|
2142 |
+
"eval_logps/rejected": -38.01955795288086,
|
2143 |
+
"eval_loss": 0.6911265850067139,
|
2144 |
+
"eval_rewards/accuracies": 0.5892857313156128,
|
2145 |
+
"eval_rewards/chosen": -0.27716198563575745,
|
2146 |
+
"eval_rewards/margins": 0.17548424005508423,
|
2147 |
+
"eval_rewards/rejected": -0.4526461660861969,
|
2148 |
+
"eval_runtime": 145.3336,
|
2149 |
+
"eval_samples_per_second": 2.36,
|
2150 |
+
"eval_steps_per_second": 0.296,
|
2151 |
+
"step": 1300
|
2152 |
+
},
|
2153 |
+
{
|
2154 |
+
"epoch": 3.4,
|
2155 |
+
"grad_norm": 3.578125,
|
2156 |
+
"learning_rate": 3.321087280364757e-07,
|
2157 |
+
"logits/chosen": -1.9077885150909424,
|
2158 |
+
"logits/rejected": -1.8874790668487549,
|
2159 |
+
"logps/chosen": -29.57822036743164,
|
2160 |
+
"logps/rejected": -36.54710388183594,
|
2161 |
+
"loss": 0.1528,
|
2162 |
+
"rewards/accuracies": 0.9375,
|
2163 |
+
"rewards/chosen": 1.3816629648208618,
|
2164 |
+
"rewards/margins": 2.624289035797119,
|
2165 |
+
"rewards/rejected": -1.242626428604126,
|
2166 |
+
"step": 1310
|
2167 |
+
},
|
2168 |
+
{
|
2169 |
+
"epoch": 3.43,
|
2170 |
+
"grad_norm": 5.71875,
|
2171 |
+
"learning_rate": 3.044460665744284e-07,
|
2172 |
+
"logits/chosen": -1.8845760822296143,
|
2173 |
+
"logits/rejected": -1.8903357982635498,
|
2174 |
+
"logps/chosen": -31.33782958984375,
|
2175 |
+
"logps/rejected": -33.613182067871094,
|
2176 |
+
"loss": 0.1368,
|
2177 |
+
"rewards/accuracies": 0.9624999761581421,
|
2178 |
+
"rewards/chosen": 1.5263617038726807,
|
2179 |
+
"rewards/margins": 2.5365498065948486,
|
2180 |
+
"rewards/rejected": -1.0101878643035889,
|
2181 |
+
"step": 1320
|
2182 |
+
},
|
2183 |
+
{
|
2184 |
+
"epoch": 3.45,
|
2185 |
+
"grad_norm": 3.9375,
|
2186 |
+
"learning_rate": 2.779113783626916e-07,
|
2187 |
+
"logits/chosen": -1.8593124151229858,
|
2188 |
+
"logits/rejected": -1.8543720245361328,
|
2189 |
+
"logps/chosen": -31.642696380615234,
|
2190 |
+
"logps/rejected": -36.003231048583984,
|
2191 |
+
"loss": 0.1361,
|
2192 |
+
"rewards/accuracies": 0.9624999761581421,
|
2193 |
+
"rewards/chosen": 1.2418328523635864,
|
2194 |
+
"rewards/margins": 2.529675245285034,
|
2195 |
+
"rewards/rejected": -1.2878425121307373,
|
2196 |
+
"step": 1330
|
2197 |
+
},
|
2198 |
+
{
|
2199 |
+
"epoch": 3.48,
|
2200 |
+
"grad_norm": 5.03125,
|
2201 |
+
"learning_rate": 2.5251829568697204e-07,
|
2202 |
+
"logits/chosen": -1.7216873168945312,
|
2203 |
+
"logits/rejected": -1.731299638748169,
|
2204 |
+
"logps/chosen": -32.13956069946289,
|
2205 |
+
"logps/rejected": -32.022666931152344,
|
2206 |
+
"loss": 0.153,
|
2207 |
+
"rewards/accuracies": 0.987500011920929,
|
2208 |
+
"rewards/chosen": 1.4940283298492432,
|
2209 |
+
"rewards/margins": 2.383723735809326,
|
2210 |
+
"rewards/rejected": -0.8896951675415039,
|
2211 |
+
"step": 1340
|
2212 |
+
},
|
2213 |
+
{
|
2214 |
+
"epoch": 3.51,
|
2215 |
+
"grad_norm": 4.46875,
|
2216 |
+
"learning_rate": 2.2827986432927774e-07,
|
2217 |
+
"logits/chosen": -1.779524803161621,
|
2218 |
+
"logits/rejected": -1.784233808517456,
|
2219 |
+
"logps/chosen": -32.241004943847656,
|
2220 |
+
"logps/rejected": -34.2801513671875,
|
2221 |
+
"loss": 0.1178,
|
2222 |
+
"rewards/accuracies": 0.987500011920929,
|
2223 |
+
"rewards/chosen": 1.5389398336410522,
|
2224 |
+
"rewards/margins": 2.5067484378814697,
|
2225 |
+
"rewards/rejected": -0.9678082466125488,
|
2226 |
+
"step": 1350
|
2227 |
+
},
|
2228 |
+
{
|
2229 |
+
"epoch": 3.53,
|
2230 |
+
"grad_norm": 22.75,
|
2231 |
+
"learning_rate": 2.0520853686560177e-07,
|
2232 |
+
"logits/chosen": -1.808258056640625,
|
2233 |
+
"logits/rejected": -1.813680648803711,
|
2234 |
+
"logps/chosen": -32.30830001831055,
|
2235 |
+
"logps/rejected": -35.810401916503906,
|
2236 |
+
"loss": 0.1575,
|
2237 |
+
"rewards/accuracies": 0.9624999761581421,
|
2238 |
+
"rewards/chosen": 1.371025800704956,
|
2239 |
+
"rewards/margins": 2.493631601333618,
|
2240 |
+
"rewards/rejected": -1.1226056814193726,
|
2241 |
+
"step": 1360
|
2242 |
+
},
|
2243 |
+
{
|
2244 |
+
"epoch": 3.56,
|
2245 |
+
"grad_norm": 3.953125,
|
2246 |
+
"learning_rate": 1.833161662683672e-07,
|
2247 |
+
"logits/chosen": -1.900948166847229,
|
2248 |
+
"logits/rejected": -1.8964240550994873,
|
2249 |
+
"logps/chosen": -31.239208221435547,
|
2250 |
+
"logps/rejected": -32.06488800048828,
|
2251 |
+
"loss": 0.1478,
|
2252 |
+
"rewards/accuracies": 0.9750000238418579,
|
2253 |
+
"rewards/chosen": 1.2568284273147583,
|
2254 |
+
"rewards/margins": 2.2945187091827393,
|
2255 |
+
"rewards/rejected": -1.0376904010772705,
|
2256 |
+
"step": 1370
|
2257 |
+
},
|
2258 |
+
{
|
2259 |
+
"epoch": 3.58,
|
2260 |
+
"grad_norm": 6.21875,
|
2261 |
+
"learning_rate": 1.626139998169246e-07,
|
2262 |
+
"logits/chosen": -1.8157840967178345,
|
2263 |
+
"logits/rejected": -1.817800760269165,
|
2264 |
+
"logps/chosen": -28.267253875732422,
|
2265 |
+
"logps/rejected": -31.886409759521484,
|
2266 |
+
"loss": 0.1623,
|
2267 |
+
"rewards/accuracies": 0.9624999761581421,
|
2268 |
+
"rewards/chosen": 1.2528432607650757,
|
2269 |
+
"rewards/margins": 2.2808501720428467,
|
2270 |
+
"rewards/rejected": -1.0280072689056396,
|
2271 |
+
"step": 1380
|
2272 |
+
},
|
2273 |
+
{
|
2274 |
+
"epoch": 3.61,
|
2275 |
+
"grad_norm": 9.25,
|
2276 |
+
"learning_rate": 1.4311267331922535e-07,
|
2277 |
+
"logits/chosen": -1.8119876384735107,
|
2278 |
+
"logits/rejected": -1.8166316747665405,
|
2279 |
+
"logps/chosen": -30.596643447875977,
|
2280 |
+
"logps/rejected": -33.92557907104492,
|
2281 |
+
"loss": 0.1638,
|
2282 |
+
"rewards/accuracies": 0.9750000238418579,
|
2283 |
+
"rewards/chosen": 1.280565619468689,
|
2284 |
+
"rewards/margins": 2.124329090118408,
|
2285 |
+
"rewards/rejected": -0.8437638282775879,
|
2286 |
+
"step": 1390
|
2287 |
+
},
|
2288 |
+
{
|
2289 |
+
"epoch": 3.64,
|
2290 |
+
"grad_norm": 3.5,
|
2291 |
+
"learning_rate": 1.2482220564763669e-07,
|
2292 |
+
"logits/chosen": -1.8785665035247803,
|
2293 |
+
"logits/rejected": -1.881792664527893,
|
2294 |
+
"logps/chosen": -33.31572723388672,
|
2295 |
+
"logps/rejected": -34.651222229003906,
|
2296 |
+
"loss": 0.1206,
|
2297 |
+
"rewards/accuracies": 0.987500011920929,
|
2298 |
+
"rewards/chosen": 1.4533611536026,
|
2299 |
+
"rewards/margins": 2.5582594871520996,
|
2300 |
+
"rewards/rejected": -1.10489821434021,
|
2301 |
+
"step": 1400
|
2302 |
+
},
|
2303 |
+
{
|
2304 |
+
"epoch": 3.64,
|
2305 |
+
"eval_logits/chosen": -2.1418943405151367,
|
2306 |
+
"eval_logits/rejected": -2.137127637863159,
|
2307 |
+
"eval_logps/chosen": -34.35324478149414,
|
2308 |
+
"eval_logps/rejected": -38.02573013305664,
|
2309 |
+
"eval_loss": 0.692409336566925,
|
2310 |
+
"eval_rewards/accuracies": 0.5917773842811584,
|
2311 |
+
"eval_rewards/chosen": -0.28682059049606323,
|
2312 |
+
"eval_rewards/margins": 0.17138195037841797,
|
2313 |
+
"eval_rewards/rejected": -0.4582025408744812,
|
2314 |
+
"eval_runtime": 145.3556,
|
2315 |
+
"eval_samples_per_second": 2.36,
|
2316 |
+
"eval_steps_per_second": 0.296,
|
2317 |
+
"step": 1400
|
2318 |
+
},
|
2319 |
+
{
|
2320 |
+
"epoch": 3.66,
|
2321 |
+
"grad_norm": 6.125,
|
2322 |
+
"learning_rate": 1.0775199359171346e-07,
|
2323 |
+
"logits/chosen": -1.946838140487671,
|
2324 |
+
"logits/rejected": -1.939971923828125,
|
2325 |
+
"logps/chosen": -32.17081832885742,
|
2326 |
+
"logps/rejected": -35.340518951416016,
|
2327 |
+
"loss": 0.1438,
|
2328 |
+
"rewards/accuracies": 0.9750000238418579,
|
2329 |
+
"rewards/chosen": 1.4685560464859009,
|
2330 |
+
"rewards/margins": 2.454911470413208,
|
2331 |
+
"rewards/rejected": -0.9863556027412415,
|
2332 |
+
"step": 1410
|
2333 |
+
},
|
2334 |
+
{
|
2335 |
+
"epoch": 3.69,
|
2336 |
+
"grad_norm": 11.0,
|
2337 |
+
"learning_rate": 9.191080703056604e-08,
|
2338 |
+
"logits/chosen": -1.8355038166046143,
|
2339 |
+
"logits/rejected": -1.8465068340301514,
|
2340 |
+
"logps/chosen": -32.56521987915039,
|
2341 |
+
"logps/rejected": -34.721900939941406,
|
2342 |
+
"loss": 0.1399,
|
2343 |
+
"rewards/accuracies": 0.9750000238418579,
|
2344 |
+
"rewards/chosen": 1.4256752729415894,
|
2345 |
+
"rewards/margins": 2.4434168338775635,
|
2346 |
+
"rewards/rejected": -1.0177414417266846,
|
2347 |
+
"step": 1420
|
2348 |
+
},
|
2349 |
+
{
|
2350 |
+
"epoch": 3.71,
|
2351 |
+
"grad_norm": 7.5625,
|
2352 |
+
"learning_rate": 7.730678442730539e-08,
|
2353 |
+
"logits/chosen": -1.9164823293685913,
|
2354 |
+
"logits/rejected": -1.928348183631897,
|
2355 |
+
"logps/chosen": -33.428531646728516,
|
2356 |
+
"logps/rejected": -34.8869514465332,
|
2357 |
+
"loss": 0.1342,
|
2358 |
+
"rewards/accuracies": 0.9624999761581421,
|
2359 |
+
"rewards/chosen": 1.4690487384796143,
|
2360 |
+
"rewards/margins": 2.5928056240081787,
|
2361 |
+
"rewards/rejected": -1.123757004737854,
|
2362 |
+
"step": 1430
|
2363 |
+
},
|
2364 |
+
{
|
2365 |
+
"epoch": 3.74,
|
2366 |
+
"grad_norm": 9.5625,
|
2367 |
+
"learning_rate": 6.394742864787806e-08,
|
2368 |
+
"logits/chosen": -1.9064290523529053,
|
2369 |
+
"logits/rejected": -1.9086523056030273,
|
2370 |
+
"logps/chosen": -31.696964263916016,
|
2371 |
+
"logps/rejected": -35.06704330444336,
|
2372 |
+
"loss": 0.1635,
|
2373 |
+
"rewards/accuracies": 0.949999988079071,
|
2374 |
+
"rewards/chosen": 1.3602821826934814,
|
2375 |
+
"rewards/margins": 2.4466512203216553,
|
2376 |
+
"rewards/rejected": -1.086369276046753,
|
2377 |
+
"step": 1440
|
2378 |
+
},
|
2379 |
+
{
|
2380 |
+
"epoch": 3.77,
|
2381 |
+
"grad_norm": 5.46875,
|
2382 |
+
"learning_rate": 5.183960310644748e-08,
|
2383 |
+
"logits/chosen": -1.8891935348510742,
|
2384 |
+
"logits/rejected": -1.8821351528167725,
|
2385 |
+
"logps/chosen": -33.91851043701172,
|
2386 |
+
"logps/rejected": -34.860595703125,
|
2387 |
+
"loss": 0.1585,
|
2388 |
+
"rewards/accuracies": 0.949999988079071,
|
2389 |
+
"rewards/chosen": 1.4848333597183228,
|
2390 |
+
"rewards/margins": 2.4660050868988037,
|
2391 |
+
"rewards/rejected": -0.981171727180481,
|
2392 |
+
"step": 1450
|
2393 |
+
},
|
2394 |
+
{
|
2395 |
+
"epoch": 3.79,
|
2396 |
+
"grad_norm": 4.96875,
|
2397 |
+
"learning_rate": 4.098952823928693e-08,
|
2398 |
+
"logits/chosen": -1.8560640811920166,
|
2399 |
+
"logits/rejected": -1.8620649576187134,
|
2400 |
+
"logps/chosen": -29.741008758544922,
|
2401 |
+
"logps/rejected": -34.997169494628906,
|
2402 |
+
"loss": 0.1378,
|
2403 |
+
"rewards/accuracies": 0.9750000238418579,
|
2404 |
+
"rewards/chosen": 1.3249523639678955,
|
2405 |
+
"rewards/margins": 2.519402027130127,
|
2406 |
+
"rewards/rejected": -1.194449782371521,
|
2407 |
+
"step": 1460
|
2408 |
+
},
|
2409 |
+
{
|
2410 |
+
"epoch": 3.82,
|
2411 |
+
"grad_norm": 5.40625,
|
2412 |
+
"learning_rate": 3.1402778309014284e-08,
|
2413 |
+
"logits/chosen": -1.803776502609253,
|
2414 |
+
"logits/rejected": -1.801746129989624,
|
2415 |
+
"logps/chosen": -29.282577514648438,
|
2416 |
+
"logps/rejected": -31.621936798095703,
|
2417 |
+
"loss": 0.1565,
|
2418 |
+
"rewards/accuracies": 0.949999988079071,
|
2419 |
+
"rewards/chosen": 1.4066790342330933,
|
2420 |
+
"rewards/margins": 2.331610918045044,
|
2421 |
+
"rewards/rejected": -0.9249318242073059,
|
2422 |
+
"step": 1470
|
2423 |
+
},
|
2424 |
+
{
|
2425 |
+
"epoch": 3.84,
|
2426 |
+
"grad_norm": 6.1875,
|
2427 |
+
"learning_rate": 2.3084278540791427e-08,
|
2428 |
+
"logits/chosen": -1.9790493249893188,
|
2429 |
+
"logits/rejected": -1.9734687805175781,
|
2430 |
+
"logps/chosen": -33.892913818359375,
|
2431 |
+
"logps/rejected": -33.2518310546875,
|
2432 |
+
"loss": 0.1448,
|
2433 |
+
"rewards/accuracies": 0.987500011920929,
|
2434 |
+
"rewards/chosen": 1.253114104270935,
|
2435 |
+
"rewards/margins": 2.3287227153778076,
|
2436 |
+
"rewards/rejected": -1.0756088495254517,
|
2437 |
+
"step": 1480
|
2438 |
+
},
|
2439 |
+
{
|
2440 |
+
"epoch": 3.87,
|
2441 |
+
"grad_norm": 7.28125,
|
2442 |
+
"learning_rate": 1.6038302591975807e-08,
|
2443 |
+
"logits/chosen": -1.8717892169952393,
|
2444 |
+
"logits/rejected": -1.8740745782852173,
|
2445 |
+
"logps/chosen": -27.078582763671875,
|
2446 |
+
"logps/rejected": -29.020471572875977,
|
2447 |
+
"loss": 0.1879,
|
2448 |
+
"rewards/accuracies": 0.9125000238418579,
|
2449 |
+
"rewards/chosen": 1.2580173015594482,
|
2450 |
+
"rewards/margins": 2.156388759613037,
|
2451 |
+
"rewards/rejected": -0.8983713984489441,
|
2452 |
+
"step": 1490
|
2453 |
+
},
|
2454 |
+
{
|
2455 |
+
"epoch": 3.9,
|
2456 |
+
"grad_norm": 5.75,
|
2457 |
+
"learning_rate": 1.0268470356514237e-08,
|
2458 |
+
"logits/chosen": -1.8792669773101807,
|
2459 |
+
"logits/rejected": -1.8736785650253296,
|
2460 |
+
"logps/chosen": -31.454416275024414,
|
2461 |
+
"logps/rejected": -33.20934295654297,
|
2462 |
+
"loss": 0.1645,
|
2463 |
+
"rewards/accuracies": 0.9624999761581421,
|
2464 |
+
"rewards/chosen": 1.2117339372634888,
|
2465 |
+
"rewards/margins": 2.328080892562866,
|
2466 |
+
"rewards/rejected": -1.116347074508667,
|
2467 |
+
"step": 1500
|
2468 |
+
},
|
2469 |
+
{
|
2470 |
+
"epoch": 3.9,
|
2471 |
+
"eval_logits/chosen": -2.1418604850769043,
|
2472 |
+
"eval_logits/rejected": -2.1370973587036133,
|
2473 |
+
"eval_logps/chosen": -34.35454177856445,
|
2474 |
+
"eval_logps/rejected": -38.024742126464844,
|
2475 |
+
"eval_loss": 0.6943246126174927,
|
2476 |
+
"eval_rewards/accuracies": 0.5718438625335693,
|
2477 |
+
"eval_rewards/chosen": -0.28799253702163696,
|
2478 |
+
"eval_rewards/margins": 0.16931606829166412,
|
2479 |
+
"eval_rewards/rejected": -0.4573085606098175,
|
2480 |
+
"eval_runtime": 145.3262,
|
2481 |
+
"eval_samples_per_second": 2.36,
|
2482 |
+
"eval_steps_per_second": 0.296,
|
2483 |
+
"step": 1500
|
2484 |
+
},
|
2485 |
+
{
|
2486 |
+
"epoch": 3.92,
|
2487 |
+
"grad_norm": 9.4375,
|
2488 |
+
"learning_rate": 5.777746105209147e-09,
|
2489 |
+
"logits/chosen": -1.8055070638656616,
|
2490 |
+
"logits/rejected": -1.8097158670425415,
|
2491 |
+
"logps/chosen": -32.785335540771484,
|
2492 |
+
"logps/rejected": -35.42612075805664,
|
2493 |
+
"loss": 0.1841,
|
2494 |
+
"rewards/accuracies": 0.9125000238418579,
|
2495 |
+
"rewards/chosen": 1.2470948696136475,
|
2496 |
+
"rewards/margins": 2.1612586975097656,
|
2497 |
+
"rewards/rejected": -0.9141640663146973,
|
2498 |
+
"step": 1510
|
2499 |
+
},
|
2500 |
+
{
|
2501 |
+
"epoch": 3.95,
|
2502 |
+
"grad_norm": 5.40625,
|
2503 |
+
"learning_rate": 2.5684369628148352e-09,
|
2504 |
+
"logits/chosen": -1.8615461587905884,
|
2505 |
+
"logits/rejected": -1.8599445819854736,
|
2506 |
+
"logps/chosen": -29.254053115844727,
|
2507 |
+
"logps/rejected": -33.70402908325195,
|
2508 |
+
"loss": 0.1759,
|
2509 |
+
"rewards/accuracies": 0.949999988079071,
|
2510 |
+
"rewards/chosen": 1.3321115970611572,
|
2511 |
+
"rewards/margins": 2.334989547729492,
|
2512 |
+
"rewards/rejected": -1.002877950668335,
|
2513 |
+
"step": 1520
|
2514 |
+
},
|
2515 |
+
{
|
2516 |
+
"epoch": 3.97,
|
2517 |
+
"grad_norm": 8.8125,
|
2518 |
+
"learning_rate": 6.421917227455999e-10,
|
2519 |
+
"logits/chosen": -1.960646390914917,
|
2520 |
+
"logits/rejected": -1.9528770446777344,
|
2521 |
+
"logps/chosen": -26.7325439453125,
|
2522 |
+
"logps/rejected": -29.855510711669922,
|
2523 |
+
"loss": 0.1584,
|
2524 |
+
"rewards/accuracies": 0.9750000238418579,
|
2525 |
+
"rewards/chosen": 1.1973587274551392,
|
2526 |
+
"rewards/margins": 2.2628719806671143,
|
2527 |
+
"rewards/rejected": -1.065513253211975,
|
2528 |
+
"step": 1530
|
2529 |
+
},
|
2530 |
+
{
|
2531 |
+
"epoch": 4.0,
|
2532 |
+
"grad_norm": 7.0625,
|
2533 |
+
"learning_rate": 0.0,
|
2534 |
+
"logits/chosen": -1.863152265548706,
|
2535 |
+
"logits/rejected": -1.85297429561615,
|
2536 |
+
"logps/chosen": -31.610912322998047,
|
2537 |
+
"logps/rejected": -36.648475646972656,
|
2538 |
+
"loss": 0.1361,
|
2539 |
+
"rewards/accuracies": 0.987500011920929,
|
2540 |
+
"rewards/chosen": 1.3097574710845947,
|
2541 |
+
"rewards/margins": 2.4233245849609375,
|
2542 |
+
"rewards/rejected": -1.1135669946670532,
|
2543 |
+
"step": 1540
|
2544 |
+
},
|
2545 |
+
{
|
2546 |
+
"epoch": 4.0,
|
2547 |
+
"step": 1540,
|
2548 |
"total_flos": 0.0,
|
2549 |
+
"train_loss": 0.19153660760297403,
|
2550 |
+
"train_runtime": 10767.2812,
|
2551 |
+
"train_samples_per_second": 1.144,
|
2552 |
+
"train_steps_per_second": 0.143
|
2553 |
}
|
2554 |
],
|
2555 |
"logging_steps": 10,
|
2556 |
+
"max_steps": 1540,
|
2557 |
"num_input_tokens_seen": 0,
|
2558 |
+
"num_train_epochs": 4,
|
2559 |
"save_steps": 100,
|
2560 |
"total_flos": 0.0,
|
2561 |
"train_batch_size": 4,
|