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,
|