Model save
Browse files- README.md +19 -30
- all_results.json +5 -18
- config.json +1 -1
- model-00001-of-00004.safetensors +1 -1
- model-00002-of-00004.safetensors +1 -1
- model-00003-of-00004.safetensors +1 -1
- model-00004-of-00004.safetensors +1 -1
- runs/Jul25_01-23-17_action-graph-trainer/events.out.tfevents.1721871175.action-graph-trainer.762006.0 +3 -0
- train_results.json +5 -5
- trainer_state.json +628 -1428
- training_args.bin +1 -1
README.md
CHANGED
|
@@ -2,15 +2,9 @@
|
|
| 2 |
license: gemma
|
| 3 |
base_model: tanliboy/zephyr-gemma-2-9b-sft
|
| 4 |
tags:
|
| 5 |
-
- alignment-handbook
|
| 6 |
- trl
|
| 7 |
- dpo
|
| 8 |
- generated_from_trainer
|
| 9 |
-
- trl
|
| 10 |
-
- dpo
|
| 11 |
-
- generated_from_trainer
|
| 12 |
-
datasets:
|
| 13 |
-
- HuggingFaceH4/ultrafeedback_binarized
|
| 14 |
model-index:
|
| 15 |
- name: zephyr-gemma-2-9b-dpo-2
|
| 16 |
results: []
|
|
@@ -19,20 +13,20 @@ model-index:
|
|
| 19 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 20 |
should probably proofread and complete it, then remove this comment. -->
|
| 21 |
|
| 22 |
-
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/tanliboy/huggingface/runs/
|
| 23 |
# zephyr-gemma-2-9b-dpo-2
|
| 24 |
|
| 25 |
-
This model is a fine-tuned version of [tanliboy/zephyr-gemma-2-9b-sft](https://huggingface.co/tanliboy/zephyr-gemma-2-9b-sft) on
|
| 26 |
It achieves the following results on the evaluation set:
|
| 27 |
-
- Loss: 0.
|
| 28 |
-
- Rewards/chosen: -0.
|
| 29 |
-
- Rewards/rejected: -1.
|
| 30 |
-
- Rewards/accuracies: 0.
|
| 31 |
-
- Rewards/margins: 0.
|
| 32 |
-
- Logps/rejected: -
|
| 33 |
-
- Logps/chosen: -
|
| 34 |
-
- Logits/rejected: -14.
|
| 35 |
-
- Logits/chosen: -14.
|
| 36 |
|
| 37 |
## Model description
|
| 38 |
|
|
@@ -51,33 +45,28 @@ More information needed
|
|
| 51 |
### Training hyperparameters
|
| 52 |
|
| 53 |
The following hyperparameters were used during training:
|
| 54 |
-
- learning_rate:
|
| 55 |
- train_batch_size: 2
|
| 56 |
- eval_batch_size: 2
|
| 57 |
- seed: 42
|
| 58 |
- distributed_type: multi-GPU
|
| 59 |
- num_devices: 8
|
| 60 |
-
- gradient_accumulation_steps:
|
| 61 |
-
- total_train_batch_size:
|
| 62 |
- total_eval_batch_size: 16
|
| 63 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 64 |
- lr_scheduler_type: cosine
|
| 65 |
-
- lr_scheduler_warmup_ratio: 0.
|
| 66 |
- num_epochs: 1
|
| 67 |
|
| 68 |
### Training results
|
| 69 |
|
| 70 |
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
| 71 |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
| 72 |
-
| 0.
|
| 73 |
-
| 0.
|
| 74 |
-
| 0.
|
| 75 |
-
| 0.
|
| 76 |
-
| 0.5441 | 0.5234 | 500 | 0.5363 | -0.6274 | -1.2091 | 0.6680 | 0.5817 | -405.3189 | -377.0607 | -14.3976 | -14.1308 |
|
| 77 |
-
| 0.5125 | 0.6281 | 600 | 0.5344 | -0.5757 | -1.1705 | 0.6840 | 0.5948 | -401.4605 | -371.8937 | -14.3713 | -14.1120 |
|
| 78 |
-
| 0.5158 | 0.7328 | 700 | 0.5316 | -0.6220 | -1.2328 | 0.6760 | 0.6108 | -407.6867 | -376.5182 | -14.2832 | -14.1010 |
|
| 79 |
-
| 0.5133 | 0.8375 | 800 | 0.5278 | -0.6258 | -1.2452 | 0.6800 | 0.6193 | -408.9254 | -376.9043 | -14.2747 | -14.0908 |
|
| 80 |
-
| 0.5098 | 0.9422 | 900 | 0.5276 | -0.6043 | -1.2270 | 0.6960 | 0.6227 | -407.1073 | -374.7531 | -14.2849 | -14.1010 |
|
| 81 |
|
| 82 |
|
| 83 |
### Framework versions
|
|
|
|
| 2 |
license: gemma
|
| 3 |
base_model: tanliboy/zephyr-gemma-2-9b-sft
|
| 4 |
tags:
|
|
|
|
| 5 |
- trl
|
| 6 |
- dpo
|
| 7 |
- generated_from_trainer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
model-index:
|
| 9 |
- name: zephyr-gemma-2-9b-dpo-2
|
| 10 |
results: []
|
|
|
|
| 13 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 14 |
should probably proofread and complete it, then remove this comment. -->
|
| 15 |
|
| 16 |
+
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/tanliboy/huggingface/runs/dikk0994)
|
| 17 |
# zephyr-gemma-2-9b-dpo-2
|
| 18 |
|
| 19 |
+
This model is a fine-tuned version of [tanliboy/zephyr-gemma-2-9b-sft](https://huggingface.co/tanliboy/zephyr-gemma-2-9b-sft) on an unknown dataset.
|
| 20 |
It achieves the following results on the evaluation set:
|
| 21 |
+
- Loss: 0.5647
|
| 22 |
+
- Rewards/chosen: -0.6443
|
| 23 |
+
- Rewards/rejected: -1.1499
|
| 24 |
+
- Rewards/accuracies: 0.6920
|
| 25 |
+
- Rewards/margins: 0.5057
|
| 26 |
+
- Logps/rejected: -1434.3335
|
| 27 |
+
- Logps/chosen: -958.5825
|
| 28 |
+
- Logits/rejected: -14.1861
|
| 29 |
+
- Logits/chosen: -14.6684
|
| 30 |
|
| 31 |
## Model description
|
| 32 |
|
|
|
|
| 45 |
### Training hyperparameters
|
| 46 |
|
| 47 |
The following hyperparameters were used during training:
|
| 48 |
+
- learning_rate: 5e-07
|
| 49 |
- train_batch_size: 2
|
| 50 |
- eval_batch_size: 2
|
| 51 |
- seed: 42
|
| 52 |
- distributed_type: multi-GPU
|
| 53 |
- num_devices: 8
|
| 54 |
+
- gradient_accumulation_steps: 16
|
| 55 |
+
- total_train_batch_size: 256
|
| 56 |
- total_eval_batch_size: 16
|
| 57 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 58 |
- lr_scheduler_type: cosine
|
| 59 |
+
- lr_scheduler_warmup_ratio: 0.2
|
| 60 |
- num_epochs: 1
|
| 61 |
|
| 62 |
### Training results
|
| 63 |
|
| 64 |
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
| 65 |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
| 66 |
+
| 0.6835 | 0.2094 | 50 | 0.6815 | -0.0218 | -0.0436 | 0.6560 | 0.0218 | -328.0053 | -336.0947 | -11.6381 | -11.3403 |
|
| 67 |
+
| 0.6243 | 0.4187 | 100 | 0.6229 | -0.5238 | -0.7528 | 0.6600 | 0.2290 | -1037.2136 | -838.1255 | -15.5098 | -15.6787 |
|
| 68 |
+
| 0.5625 | 0.6281 | 150 | 0.5793 | -0.7186 | -1.1873 | 0.6880 | 0.4688 | -1471.7362 | -1032.8834 | -14.7746 | -15.1797 |
|
| 69 |
+
| 0.5699 | 0.8375 | 200 | 0.5647 | -0.6443 | -1.1499 | 0.6920 | 0.5057 | -1434.3335 | -958.5825 | -14.1861 | -14.6684 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
|
| 71 |
|
| 72 |
### Framework versions
|
all_results.json
CHANGED
|
@@ -1,22 +1,9 @@
|
|
| 1 |
{
|
| 2 |
-
"epoch": 0.
|
| 3 |
-
"eval_logits/chosen": -14.105589866638184,
|
| 4 |
-
"eval_logits/rejected": -14.292818069458008,
|
| 5 |
-
"eval_logps/chosen": -375.1571960449219,
|
| 6 |
-
"eval_logps/rejected": -407.44989013671875,
|
| 7 |
-
"eval_loss": 0.5276930332183838,
|
| 8 |
-
"eval_rewards/accuracies": 0.6880000233650208,
|
| 9 |
-
"eval_rewards/chosen": -0.6083627939224243,
|
| 10 |
-
"eval_rewards/margins": 0.6220458149909973,
|
| 11 |
-
"eval_rewards/rejected": -1.2304086685180664,
|
| 12 |
-
"eval_runtime": 148.4787,
|
| 13 |
-
"eval_samples": 2000,
|
| 14 |
-
"eval_samples_per_second": 13.47,
|
| 15 |
-
"eval_steps_per_second": 0.842,
|
| 16 |
"total_flos": 0.0,
|
| 17 |
-
"train_loss": 0.
|
| 18 |
-
"train_runtime":
|
| 19 |
"train_samples": 61134,
|
| 20 |
-
"train_samples_per_second":
|
| 21 |
-
"train_steps_per_second": 0.
|
| 22 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"epoch": 0.9965977492802931,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
"total_flos": 0.0,
|
| 4 |
+
"train_loss": 0.6157421479706003,
|
| 5 |
+
"train_runtime": 11733.4361,
|
| 6 |
"train_samples": 61134,
|
| 7 |
+
"train_samples_per_second": 5.21,
|
| 8 |
+
"train_steps_per_second": 0.02
|
| 9 |
}
|
config.json
CHANGED
|
@@ -29,6 +29,6 @@
|
|
| 29 |
"sliding_window_size": 4096,
|
| 30 |
"torch_dtype": "bfloat16",
|
| 31 |
"transformers_version": "4.43.1",
|
| 32 |
-
"use_cache":
|
| 33 |
"vocab_size": 256000
|
| 34 |
}
|
|
|
|
| 29 |
"sliding_window_size": 4096,
|
| 30 |
"torch_dtype": "bfloat16",
|
| 31 |
"transformers_version": "4.43.1",
|
| 32 |
+
"use_cache": false,
|
| 33 |
"vocab_size": 256000
|
| 34 |
}
|
model-00001-of-00004.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4903351912
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:233036a8075e3f8bec5c65ee1e3f12c4a2bc723a17f775a61b5623c9c39d2552
|
| 3 |
size 4903351912
|
model-00002-of-00004.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4947570872
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:75daea75c054634bd81b8427bd48c44d89567ad0e7eebabb4a3645c63db63bb6
|
| 3 |
size 4947570872
|
model-00003-of-00004.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4962221464
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d7c2ebf9602671d9dc02fe11665e1c700e053d62b1df61cdb44df2fa1bd3a475
|
| 3 |
size 4962221464
|
model-00004-of-00004.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 3670322200
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e9060c192a77500562639f82aaec40aa8938d90459195bfac950a0999b1655f9
|
| 3 |
size 3670322200
|
runs/Jul25_01-23-17_action-graph-trainer/events.out.tfevents.1721871175.action-graph-trainer.762006.0
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2171071bdb920780b0e1dbe0e1266805d4d430111886287382a69a4d9d6480be
|
| 3 |
+
size 42413
|
train_results.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
{
|
| 2 |
-
"epoch": 0.
|
| 3 |
"total_flos": 0.0,
|
| 4 |
-
"train_loss": 0.
|
| 5 |
-
"train_runtime":
|
| 6 |
"train_samples": 61134,
|
| 7 |
-
"train_samples_per_second":
|
| 8 |
-
"train_steps_per_second": 0.
|
| 9 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"epoch": 0.9965977492802931,
|
| 3 |
"total_flos": 0.0,
|
| 4 |
+
"train_loss": 0.6157421479706003,
|
| 5 |
+
"train_runtime": 11733.4361,
|
| 6 |
"train_samples": 61134,
|
| 7 |
+
"train_samples_per_second": 5.21,
|
| 8 |
+
"train_steps_per_second": 0.02
|
| 9 |
}
|
trainer_state.json
CHANGED
|
@@ -1,21 +1,21 @@
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
-
"epoch": 0.
|
| 5 |
-
"eval_steps":
|
| 6 |
-
"global_step":
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
-
"epoch": 0.
|
| 13 |
-
"grad_norm":
|
| 14 |
-
"learning_rate":
|
| 15 |
-
"logits/chosen": -
|
| 16 |
-
"logits/rejected": -8.
|
| 17 |
-
"logps/chosen": -
|
| 18 |
-
"logps/rejected": -
|
| 19 |
"loss": 0.6931,
|
| 20 |
"rewards/accuracies": 0.0,
|
| 21 |
"rewards/chosen": 0.0,
|
|
@@ -23,1587 +23,787 @@
|
|
| 23 |
"rewards/rejected": 0.0,
|
| 24 |
"step": 1
|
| 25 |
},
|
| 26 |
-
{
|
| 27 |
-
"epoch": 0.010468463752944255,
|
| 28 |
-
"grad_norm": 6.230445884319703,
|
| 29 |
-
"learning_rate": 2.0833333333333335e-08,
|
| 30 |
-
"logits/chosen": -7.981190204620361,
|
| 31 |
-
"logits/rejected": -8.130318641662598,
|
| 32 |
-
"logps/chosen": -306.2852783203125,
|
| 33 |
-
"logps/rejected": -309.1938781738281,
|
| 34 |
-
"loss": 0.6928,
|
| 35 |
-
"rewards/accuracies": 0.4722222089767456,
|
| 36 |
-
"rewards/chosen": 0.0020241288002580404,
|
| 37 |
-
"rewards/margins": 0.0018345804419368505,
|
| 38 |
-
"rewards/rejected": 0.000189548620255664,
|
| 39 |
-
"step": 10
|
| 40 |
-
},
|
| 41 |
{
|
| 42 |
"epoch": 0.02093692750588851,
|
| 43 |
-
"grad_norm":
|
| 44 |
-
"learning_rate":
|
| 45 |
-
"logits/chosen": -8.
|
| 46 |
-
"logits/rejected": -8.
|
| 47 |
-
"logps/chosen": -
|
| 48 |
-
"logps/rejected": -
|
| 49 |
-
"loss": 0.6934,
|
| 50 |
-
"rewards/accuracies": 0.42500001192092896,
|
| 51 |
-
"rewards/chosen": 0.0012120783794671297,
|
| 52 |
-
"rewards/margins": -0.0002803016686812043,
|
| 53 |
-
"rewards/rejected": 0.0014923801645636559,
|
| 54 |
-
"step": 20
|
| 55 |
-
},
|
| 56 |
-
{
|
| 57 |
-
"epoch": 0.031405391258832765,
|
| 58 |
-
"grad_norm": 8.288905108120925,
|
| 59 |
-
"learning_rate": 6.25e-08,
|
| 60 |
-
"logits/chosen": -7.823553562164307,
|
| 61 |
-
"logits/rejected": -8.030963897705078,
|
| 62 |
-
"logps/chosen": -294.57769775390625,
|
| 63 |
-
"logps/rejected": -290.71209716796875,
|
| 64 |
"loss": 0.6931,
|
| 65 |
-
"rewards/accuracies": 0.
|
| 66 |
-
"rewards/chosen":
|
| 67 |
-
"rewards/margins":
|
| 68 |
-
"rewards/rejected":
|
| 69 |
-
"step":
|
| 70 |
},
|
| 71 |
{
|
| 72 |
"epoch": 0.04187385501177702,
|
| 73 |
-
"grad_norm":
|
| 74 |
-
"learning_rate": 8.333333333333334e-08,
|
| 75 |
-
"logits/chosen": -8.137880325317383,
|
| 76 |
-
"logits/rejected": -8.182608604431152,
|
| 77 |
-
"logps/chosen": -302.7641906738281,
|
| 78 |
-
"logps/rejected": -278.7587890625,
|
| 79 |
-
"loss": 0.6925,
|
| 80 |
-
"rewards/accuracies": 0.5249999761581421,
|
| 81 |
-
"rewards/chosen": 0.004604184068739414,
|
| 82 |
-
"rewards/margins": 0.0005916848895139992,
|
| 83 |
-
"rewards/rejected": 0.004012499004602432,
|
| 84 |
-
"step": 40
|
| 85 |
-
},
|
| 86 |
-
{
|
| 87 |
-
"epoch": 0.05234231876472128,
|
| 88 |
-
"grad_norm": 11.714084906492422,
|
| 89 |
"learning_rate": 1.0416666666666667e-07,
|
| 90 |
-
"logits/chosen": -
|
| 91 |
-
"logits/rejected": -
|
| 92 |
-
"logps/chosen": -
|
| 93 |
-
"logps/rejected": -
|
| 94 |
-
"loss": 0.
|
| 95 |
-
"rewards/accuracies": 0.
|
| 96 |
-
"rewards/chosen":
|
| 97 |
-
"rewards/margins":
|
| 98 |
-
"rewards/rejected":
|
| 99 |
-
"step":
|
| 100 |
},
|
| 101 |
{
|
| 102 |
"epoch": 0.06281078251766553,
|
| 103 |
-
"grad_norm":
|
| 104 |
-
"learning_rate": 1.
|
| 105 |
-
"logits/chosen": -8.
|
| 106 |
-
"logits/rejected": -8.
|
| 107 |
-
"logps/chosen": -
|
| 108 |
-
"logps/rejected": -
|
| 109 |
-
"loss": 0.
|
| 110 |
-
"rewards/accuracies": 0.
|
| 111 |
-
"rewards/chosen": 0.
|
| 112 |
-
"rewards/margins":
|
| 113 |
-
"rewards/rejected": 0.
|
| 114 |
-
"step":
|
| 115 |
-
},
|
| 116 |
-
{
|
| 117 |
-
"epoch": 0.07327924627060979,
|
| 118 |
-
"grad_norm": 6.439781455829167,
|
| 119 |
-
"learning_rate": 1.4583333333333332e-07,
|
| 120 |
-
"logits/chosen": -8.384071350097656,
|
| 121 |
-
"logits/rejected": -8.226759910583496,
|
| 122 |
-
"logps/chosen": -240.31106567382812,
|
| 123 |
-
"logps/rejected": -258.47332763671875,
|
| 124 |
-
"loss": 0.6867,
|
| 125 |
-
"rewards/accuracies": 0.6625000238418579,
|
| 126 |
-
"rewards/chosen": 0.0298004187643528,
|
| 127 |
-
"rewards/margins": 0.011175071820616722,
|
| 128 |
-
"rewards/rejected": 0.018625345081090927,
|
| 129 |
-
"step": 70
|
| 130 |
},
|
| 131 |
{
|
| 132 |
"epoch": 0.08374771002355404,
|
| 133 |
-
"grad_norm":
|
| 134 |
-
"learning_rate":
|
| 135 |
-
"logits/chosen": -8.
|
| 136 |
-
"logits/rejected": -
|
| 137 |
-
"logps/chosen": -
|
| 138 |
-
"logps/rejected": -
|
| 139 |
-
"loss": 0.
|
| 140 |
-
"rewards/accuracies": 0.
|
| 141 |
-
"rewards/chosen": 0.
|
| 142 |
-
"rewards/margins": 0.
|
| 143 |
-
"rewards/rejected": 0.
|
| 144 |
-
"step":
|
| 145 |
-
},
|
| 146 |
-
{
|
| 147 |
-
"epoch": 0.0942161737764983,
|
| 148 |
-
"grad_norm": 5.920042917298396,
|
| 149 |
-
"learning_rate": 1.875e-07,
|
| 150 |
-
"logits/chosen": -8.7146635055542,
|
| 151 |
-
"logits/rejected": -8.876623153686523,
|
| 152 |
-
"logps/chosen": -272.12420654296875,
|
| 153 |
-
"logps/rejected": -255.40225219726562,
|
| 154 |
-
"loss": 0.6782,
|
| 155 |
-
"rewards/accuracies": 0.7875000238418579,
|
| 156 |
-
"rewards/chosen": 0.06685945391654968,
|
| 157 |
-
"rewards/margins": 0.04631630331277847,
|
| 158 |
-
"rewards/rejected": 0.02054314874112606,
|
| 159 |
-
"step": 90
|
| 160 |
-
},
|
| 161 |
-
{
|
| 162 |
-
"epoch": 0.10468463752944256,
|
| 163 |
-
"grad_norm": 7.4287436912349065,
|
| 164 |
-
"learning_rate": 1.9998929970725745e-07,
|
| 165 |
-
"logits/chosen": -8.944499969482422,
|
| 166 |
-
"logits/rejected": -8.88210678100586,
|
| 167 |
-
"logps/chosen": -293.67266845703125,
|
| 168 |
-
"logps/rejected": -300.55419921875,
|
| 169 |
-
"loss": 0.6677,
|
| 170 |
-
"rewards/accuracies": 0.7124999761581421,
|
| 171 |
-
"rewards/chosen": 0.049799270927906036,
|
| 172 |
-
"rewards/margins": 0.04667241498827934,
|
| 173 |
-
"rewards/rejected": 0.003126861061900854,
|
| 174 |
-
"step": 100
|
| 175 |
},
|
| 176 |
{
|
| 177 |
"epoch": 0.10468463752944256,
|
| 178 |
-
"
|
| 179 |
-
"
|
| 180 |
-
"
|
| 181 |
-
"
|
| 182 |
-
"
|
| 183 |
-
"
|
| 184 |
-
"
|
| 185 |
-
"eval_rewards/margins": 0.057010453194379807,
|
| 186 |
-
"eval_rewards/rejected": -0.030102219432592392,
|
| 187 |
-
"eval_runtime": 151.8713,
|
| 188 |
-
"eval_samples_per_second": 13.169,
|
| 189 |
-
"eval_steps_per_second": 0.823,
|
| 190 |
-
"step": 100
|
| 191 |
-
},
|
| 192 |
-
{
|
| 193 |
-
"epoch": 0.11515310128238682,
|
| 194 |
-
"grad_norm": 7.059170038910516,
|
| 195 |
-
"learning_rate": 1.9986894771071702e-07,
|
| 196 |
-
"logits/chosen": -9.338754653930664,
|
| 197 |
-
"logits/rejected": -9.311285972595215,
|
| 198 |
-
"logps/chosen": -288.4460144042969,
|
| 199 |
-
"logps/rejected": -247.97964477539062,
|
| 200 |
-
"loss": 0.6636,
|
| 201 |
-
"rewards/accuracies": 0.675000011920929,
|
| 202 |
-
"rewards/chosen": 0.021794503554701805,
|
| 203 |
-
"rewards/margins": 0.10265706479549408,
|
| 204 |
-
"rewards/rejected": -0.08086254447698593,
|
| 205 |
-
"step": 110
|
| 206 |
-
},
|
| 207 |
-
{
|
| 208 |
-
"epoch": 0.12562156503533106,
|
| 209 |
-
"grad_norm": 8.852936631402919,
|
| 210 |
-
"learning_rate": 1.996150298485439e-07,
|
| 211 |
-
"logits/chosen": -10.604690551757812,
|
| 212 |
-
"logits/rejected": -10.366097450256348,
|
| 213 |
-
"logps/chosen": -265.26751708984375,
|
| 214 |
-
"logps/rejected": -300.37530517578125,
|
| 215 |
-
"loss": 0.6409,
|
| 216 |
"rewards/accuracies": 0.737500011920929,
|
| 217 |
-
"rewards/chosen":
|
| 218 |
-
"rewards/margins": 0.
|
| 219 |
-
"rewards/rejected":
|
| 220 |
-
"step":
|
| 221 |
},
|
| 222 |
{
|
| 223 |
-
"epoch": 0.
|
| 224 |
-
"grad_norm":
|
| 225 |
-
"learning_rate":
|
| 226 |
-
"logits/chosen": -
|
| 227 |
-
"logits/rejected": -
|
| 228 |
-
"logps/chosen": -
|
| 229 |
-
"logps/rejected": -
|
| 230 |
-
"loss": 0.
|
| 231 |
-
"rewards/accuracies": 0.
|
| 232 |
-
"rewards/chosen":
|
| 233 |
-
"rewards/margins": 0.
|
| 234 |
-
"rewards/rejected":
|
| 235 |
-
"step":
|
| 236 |
},
|
| 237 |
{
|
| 238 |
"epoch": 0.14655849254121958,
|
| 239 |
-
"grad_norm":
|
| 240 |
-
"learning_rate":
|
| 241 |
-
"logits/chosen": -
|
| 242 |
-
"logits/rejected": -
|
| 243 |
-
"logps/chosen": -
|
| 244 |
-
"logps/rejected": -
|
| 245 |
-
"loss": 0.
|
| 246 |
-
"rewards/accuracies": 0.
|
| 247 |
-
"rewards/chosen":
|
| 248 |
-
"rewards/margins": 0.
|
| 249 |
-
"rewards/rejected":
|
| 250 |
-
"step":
|
| 251 |
-
},
|
| 252 |
-
{
|
| 253 |
-
"epoch": 0.15702695629416383,
|
| 254 |
-
"grad_norm": 7.638915011188106,
|
| 255 |
-
"learning_rate": 1.9805616719375848e-07,
|
| 256 |
-
"logits/chosen": -10.7134370803833,
|
| 257 |
-
"logits/rejected": -11.019803047180176,
|
| 258 |
-
"logps/chosen": -330.085205078125,
|
| 259 |
-
"logps/rejected": -305.45208740234375,
|
| 260 |
-
"loss": 0.6196,
|
| 261 |
-
"rewards/accuracies": 0.737500011920929,
|
| 262 |
-
"rewards/chosen": -0.15285547077655792,
|
| 263 |
-
"rewards/margins": 0.1812134087085724,
|
| 264 |
-
"rewards/rejected": -0.3340689241886139,
|
| 265 |
-
"step": 150
|
| 266 |
},
|
| 267 |
{
|
| 268 |
"epoch": 0.16749542004710807,
|
| 269 |
-
"grad_norm":
|
| 270 |
-
"learning_rate":
|
| 271 |
-
"logits/chosen": -
|
| 272 |
-
"logits/rejected": -
|
| 273 |
-
"logps/chosen": -
|
| 274 |
-
"logps/rejected": -
|
| 275 |
-
"loss": 0.
|
| 276 |
-
"rewards/accuracies": 0.
|
| 277 |
-
"rewards/chosen":
|
| 278 |
-
"rewards/margins": 0.
|
| 279 |
-
"rewards/rejected": -0.
|
| 280 |
-
"step":
|
| 281 |
-
},
|
| 282 |
-
{
|
| 283 |
-
"epoch": 0.17796388380005235,
|
| 284 |
-
"grad_norm": 32.698955102660705,
|
| 285 |
-
"learning_rate": 1.9636005833471467e-07,
|
| 286 |
-
"logits/chosen": -11.027162551879883,
|
| 287 |
-
"logits/rejected": -11.326080322265625,
|
| 288 |
-
"logps/chosen": -288.31817626953125,
|
| 289 |
-
"logps/rejected": -291.74908447265625,
|
| 290 |
-
"loss": 0.5927,
|
| 291 |
-
"rewards/accuracies": 0.6000000238418579,
|
| 292 |
-
"rewards/chosen": -0.3270031809806824,
|
| 293 |
-
"rewards/margins": 0.15106120705604553,
|
| 294 |
-
"rewards/rejected": -0.4780643582344055,
|
| 295 |
-
"step": 170
|
| 296 |
},
|
| 297 |
{
|
| 298 |
"epoch": 0.1884323475529966,
|
| 299 |
-
"grad_norm":
|
| 300 |
-
"learning_rate":
|
| 301 |
-
"logits/chosen": -10.
|
| 302 |
-
"logits/rejected": -
|
| 303 |
-
"logps/chosen": -
|
| 304 |
-
"logps/rejected": -
|
| 305 |
-
"loss": 0.
|
| 306 |
-
"rewards/accuracies": 0.
|
| 307 |
-
"rewards/chosen": -0.
|
| 308 |
-
"rewards/margins": 0.
|
| 309 |
-
"rewards/rejected": -0.
|
| 310 |
-
"step":
|
| 311 |
-
},
|
| 312 |
-
{
|
| 313 |
-
"epoch": 0.19890081130594087,
|
| 314 |
-
"grad_norm": 10.803006572264644,
|
| 315 |
-
"learning_rate": 1.9414862968351785e-07,
|
| 316 |
-
"logits/chosen": -11.358689308166504,
|
| 317 |
-
"logits/rejected": -11.146299362182617,
|
| 318 |
-
"logps/chosen": -284.0877380371094,
|
| 319 |
-
"logps/rejected": -376.7961120605469,
|
| 320 |
-
"loss": 0.5878,
|
| 321 |
-
"rewards/accuracies": 0.8500000238418579,
|
| 322 |
-
"rewards/chosen": -0.08324204385280609,
|
| 323 |
-
"rewards/margins": 0.37117457389831543,
|
| 324 |
-
"rewards/rejected": -0.4544166624546051,
|
| 325 |
-
"step": 190
|
| 326 |
},
|
| 327 |
{
|
| 328 |
"epoch": 0.2093692750588851,
|
| 329 |
-
"grad_norm":
|
| 330 |
-
"learning_rate":
|
| 331 |
-
"logits/chosen": -11.
|
| 332 |
-
"logits/rejected": -11.
|
| 333 |
-
"logps/chosen": -
|
| 334 |
-
"logps/rejected": -
|
| 335 |
-
"loss": 0.
|
| 336 |
-
"rewards/accuracies": 0.
|
| 337 |
-
"rewards/chosen": -0.
|
| 338 |
-
"rewards/margins": 0.
|
| 339 |
-
"rewards/rejected": -0.
|
| 340 |
-
"step":
|
| 341 |
},
|
| 342 |
{
|
| 343 |
"epoch": 0.2093692750588851,
|
| 344 |
-
"eval_logits/chosen": -11.
|
| 345 |
-
"eval_logits/rejected": -11.
|
| 346 |
-
"eval_logps/chosen": -
|
| 347 |
-
"eval_logps/rejected": -
|
| 348 |
-
"eval_loss": 0.
|
| 349 |
-
"eval_rewards/accuracies": 0.
|
| 350 |
-
"eval_rewards/chosen": -0.
|
| 351 |
-
"eval_rewards/margins": 0.
|
| 352 |
-
"eval_rewards/rejected": -0.
|
| 353 |
-
"eval_runtime": 150.
|
| 354 |
-
"eval_samples_per_second": 13.
|
| 355 |
-
"eval_steps_per_second": 0.
|
| 356 |
-
"step":
|
| 357 |
-
},
|
| 358 |
-
{
|
| 359 |
-
"epoch": 0.21983773881182936,
|
| 360 |
-
"grad_norm": 25.282722057255278,
|
| 361 |
-
"learning_rate": 1.914337076438937e-07,
|
| 362 |
-
"logits/chosen": -11.216215133666992,
|
| 363 |
-
"logits/rejected": -11.796686172485352,
|
| 364 |
-
"logps/chosen": -339.0605163574219,
|
| 365 |
-
"logps/rejected": -345.29901123046875,
|
| 366 |
-
"loss": 0.6089,
|
| 367 |
-
"rewards/accuracies": 0.6875,
|
| 368 |
-
"rewards/chosen": -0.44959211349487305,
|
| 369 |
-
"rewards/margins": 0.2086714208126068,
|
| 370 |
-
"rewards/rejected": -0.6582635641098022,
|
| 371 |
-
"step": 210
|
| 372 |
},
|
| 373 |
{
|
| 374 |
"epoch": 0.23030620256477363,
|
| 375 |
-
"grad_norm":
|
| 376 |
-
"learning_rate":
|
| 377 |
-
"logits/chosen": -11.
|
| 378 |
-
"logits/rejected": -12.
|
| 379 |
-
"logps/chosen": -
|
| 380 |
-
"logps/rejected": -
|
| 381 |
-
"loss": 0.
|
| 382 |
-
"rewards/accuracies": 0.
|
| 383 |
-
"rewards/chosen": -0.
|
| 384 |
-
"rewards/margins": 0.
|
| 385 |
-
"rewards/rejected": -0.
|
| 386 |
-
"step":
|
| 387 |
-
},
|
| 388 |
-
{
|
| 389 |
-
"epoch": 0.24077466631771788,
|
| 390 |
-
"grad_norm": 12.45188359067346,
|
| 391 |
-
"learning_rate": 1.882298112301034e-07,
|
| 392 |
-
"logits/chosen": -11.14279556274414,
|
| 393 |
-
"logits/rejected": -11.60063362121582,
|
| 394 |
-
"logps/chosen": -281.5401916503906,
|
| 395 |
-
"logps/rejected": -305.64166259765625,
|
| 396 |
-
"loss": 0.5863,
|
| 397 |
-
"rewards/accuracies": 0.6499999761581421,
|
| 398 |
-
"rewards/chosen": -0.23669323325157166,
|
| 399 |
-
"rewards/margins": 0.28197115659713745,
|
| 400 |
-
"rewards/rejected": -0.5186644196510315,
|
| 401 |
-
"step": 230
|
| 402 |
},
|
| 403 |
{
|
| 404 |
"epoch": 0.2512431300706621,
|
| 405 |
-
"grad_norm":
|
| 406 |
-
"learning_rate":
|
| 407 |
-
"logits/chosen": -
|
| 408 |
-
"logits/rejected": -
|
| 409 |
-
"logps/chosen": -
|
| 410 |
-
"logps/rejected": -
|
| 411 |
-
"loss": 0.
|
| 412 |
-
"rewards/accuracies": 0.
|
| 413 |
-
"rewards/chosen": -0.
|
| 414 |
-
"rewards/margins": 0.
|
| 415 |
-
"rewards/rejected": -0.
|
| 416 |
-
"step":
|
| 417 |
-
},
|
| 418 |
-
{
|
| 419 |
-
"epoch": 0.26171159382360637,
|
| 420 |
-
"grad_norm": 14.325794995366559,
|
| 421 |
-
"learning_rate": 1.8455407442133465e-07,
|
| 422 |
-
"logits/chosen": -11.799114227294922,
|
| 423 |
-
"logits/rejected": -12.152410507202148,
|
| 424 |
-
"logps/chosen": -329.87677001953125,
|
| 425 |
-
"logps/rejected": -324.18115234375,
|
| 426 |
-
"loss": 0.5655,
|
| 427 |
-
"rewards/accuracies": 0.7875000238418579,
|
| 428 |
-
"rewards/chosen": -0.24216821789741516,
|
| 429 |
-
"rewards/margins": 0.43084821105003357,
|
| 430 |
-
"rewards/rejected": -0.6730164289474487,
|
| 431 |
-
"step": 250
|
| 432 |
},
|
| 433 |
{
|
| 434 |
"epoch": 0.2721800575765506,
|
| 435 |
-
"grad_norm":
|
| 436 |
-
"learning_rate":
|
| 437 |
-
"logits/chosen": -12.
|
| 438 |
-
"logits/rejected": -
|
| 439 |
-
"logps/chosen": -
|
| 440 |
-
"logps/rejected": -
|
| 441 |
-
"loss": 0.
|
| 442 |
-
"rewards/accuracies": 0.
|
| 443 |
-
"rewards/chosen": -0.
|
| 444 |
-
"rewards/margins": 0.
|
| 445 |
-
"rewards/rejected": -0.
|
| 446 |
-
"step":
|
| 447 |
-
},
|
| 448 |
-
{
|
| 449 |
-
"epoch": 0.2826485213294949,
|
| 450 |
-
"grad_norm": 17.931356253932698,
|
| 451 |
-
"learning_rate": 1.8042615453163484e-07,
|
| 452 |
-
"logits/chosen": -12.003366470336914,
|
| 453 |
-
"logits/rejected": -12.557150840759277,
|
| 454 |
-
"logps/chosen": -372.60821533203125,
|
| 455 |
-
"logps/rejected": -349.703125,
|
| 456 |
-
"loss": 0.5514,
|
| 457 |
-
"rewards/accuracies": 0.6875,
|
| 458 |
-
"rewards/chosen": -0.45285406708717346,
|
| 459 |
-
"rewards/margins": 0.3556319773197174,
|
| 460 |
-
"rewards/rejected": -0.8084859848022461,
|
| 461 |
-
"step": 270
|
| 462 |
},
|
| 463 |
{
|
| 464 |
"epoch": 0.29311698508243916,
|
| 465 |
-
"grad_norm":
|
| 466 |
-
"learning_rate":
|
| 467 |
-
"logits/chosen": -
|
| 468 |
-
"logits/rejected": -
|
| 469 |
-
"logps/chosen": -
|
| 470 |
-
"logps/rejected": -
|
| 471 |
-
"loss": 0.
|
| 472 |
-
"rewards/accuracies": 0.
|
| 473 |
-
"rewards/chosen": -0.
|
| 474 |
-
"rewards/margins": 0.
|
| 475 |
-
"rewards/rejected": -0.
|
| 476 |
-
"step":
|
| 477 |
-
},
|
| 478 |
-
{
|
| 479 |
-
"epoch": 0.3035854488353834,
|
| 480 |
-
"grad_norm": 16.797223533505605,
|
| 481 |
-
"learning_rate": 1.7586812708541044e-07,
|
| 482 |
-
"logits/chosen": -12.28768253326416,
|
| 483 |
-
"logits/rejected": -13.17219066619873,
|
| 484 |
-
"logps/chosen": -382.7824401855469,
|
| 485 |
-
"logps/rejected": -368.29449462890625,
|
| 486 |
-
"loss": 0.5672,
|
| 487 |
-
"rewards/accuracies": 0.800000011920929,
|
| 488 |
-
"rewards/chosen": -0.40373754501342773,
|
| 489 |
-
"rewards/margins": 0.4520534873008728,
|
| 490 |
-
"rewards/rejected": -0.8557910919189453,
|
| 491 |
-
"step": 290
|
| 492 |
-
},
|
| 493 |
-
{
|
| 494 |
-
"epoch": 0.31405391258832765,
|
| 495 |
-
"grad_norm": 17.74775759360231,
|
| 496 |
-
"learning_rate": 1.7343535406158772e-07,
|
| 497 |
-
"logits/chosen": -12.226592063903809,
|
| 498 |
-
"logits/rejected": -12.483770370483398,
|
| 499 |
-
"logps/chosen": -300.1335144042969,
|
| 500 |
-
"logps/rejected": -350.2237854003906,
|
| 501 |
-
"loss": 0.5723,
|
| 502 |
-
"rewards/accuracies": 0.75,
|
| 503 |
-
"rewards/chosen": -0.401310533285141,
|
| 504 |
-
"rewards/margins": 0.4268767833709717,
|
| 505 |
-
"rewards/rejected": -0.8281872868537903,
|
| 506 |
-
"step": 300
|
| 507 |
},
|
| 508 |
{
|
| 509 |
"epoch": 0.31405391258832765,
|
| 510 |
-
"
|
| 511 |
-
"
|
| 512 |
-
"
|
| 513 |
-
"
|
| 514 |
-
"
|
| 515 |
-
"
|
| 516 |
-
"
|
| 517 |
-
"
|
| 518 |
-
"
|
| 519 |
-
"
|
| 520 |
-
"
|
| 521 |
-
"
|
| 522 |
-
"step": 300
|
| 523 |
-
},
|
| 524 |
-
{
|
| 525 |
-
"epoch": 0.3245223763412719,
|
| 526 |
-
"grad_norm": 21.123290325111288,
|
| 527 |
-
"learning_rate": 1.709043677606842e-07,
|
| 528 |
-
"logits/chosen": -13.11596393585205,
|
| 529 |
-
"logits/rejected": -13.485511779785156,
|
| 530 |
-
"logps/chosen": -372.96563720703125,
|
| 531 |
-
"logps/rejected": -374.05731201171875,
|
| 532 |
-
"loss": 0.5502,
|
| 533 |
-
"rewards/accuracies": 0.737500011920929,
|
| 534 |
-
"rewards/chosen": -0.4754568040370941,
|
| 535 |
-
"rewards/margins": 0.5059635043144226,
|
| 536 |
-
"rewards/rejected": -0.9814203381538391,
|
| 537 |
-
"step": 310
|
| 538 |
},
|
| 539 |
{
|
| 540 |
"epoch": 0.33499084009421615,
|
| 541 |
-
"grad_norm":
|
| 542 |
-
"learning_rate":
|
| 543 |
-
"logits/chosen": -
|
| 544 |
-
"logits/rejected": -
|
| 545 |
-
"logps/chosen": -
|
| 546 |
-
"logps/rejected": -
|
| 547 |
-
"loss": 0.
|
| 548 |
-
"rewards/accuracies": 0.
|
| 549 |
-
"rewards/chosen": -0.
|
| 550 |
-
"rewards/margins": 0.
|
| 551 |
-
"rewards/rejected": -
|
| 552 |
-
"step":
|
| 553 |
-
},
|
| 554 |
-
{
|
| 555 |
-
"epoch": 0.34545930384716045,
|
| 556 |
-
"grad_norm": 18.33219470868773,
|
| 557 |
-
"learning_rate": 1.6556142203145976e-07,
|
| 558 |
-
"logits/chosen": -12.740918159484863,
|
| 559 |
-
"logits/rejected": -13.098161697387695,
|
| 560 |
-
"logps/chosen": -339.1952819824219,
|
| 561 |
-
"logps/rejected": -363.0919189453125,
|
| 562 |
-
"loss": 0.5684,
|
| 563 |
-
"rewards/accuracies": 0.637499988079071,
|
| 564 |
-
"rewards/chosen": -0.5905572175979614,
|
| 565 |
-
"rewards/margins": 0.25887611508369446,
|
| 566 |
-
"rewards/rejected": -0.8494332432746887,
|
| 567 |
-
"step": 330
|
| 568 |
},
|
| 569 |
{
|
| 570 |
"epoch": 0.3559277676001047,
|
| 571 |
-
"grad_norm":
|
| 572 |
-
"learning_rate":
|
| 573 |
-
"logits/chosen": -
|
| 574 |
-
"logits/rejected": -
|
| 575 |
-
"logps/chosen": -
|
| 576 |
-
"logps/rejected": -
|
| 577 |
-
"loss": 0.
|
| 578 |
-
"rewards/accuracies": 0.
|
| 579 |
-
"rewards/chosen": -0.
|
| 580 |
-
"rewards/margins": 0.
|
| 581 |
-
"rewards/rejected": -0.
|
| 582 |
-
"step":
|
| 583 |
-
},
|
| 584 |
-
{
|
| 585 |
-
"epoch": 0.36639623135304894,
|
| 586 |
-
"grad_norm": 20.453314348602326,
|
| 587 |
-
"learning_rate": 1.598678632063284e-07,
|
| 588 |
-
"logits/chosen": -12.832880973815918,
|
| 589 |
-
"logits/rejected": -13.420519828796387,
|
| 590 |
-
"logps/chosen": -368.6810302734375,
|
| 591 |
-
"logps/rejected": -389.3878479003906,
|
| 592 |
-
"loss": 0.5505,
|
| 593 |
-
"rewards/accuracies": 0.737500011920929,
|
| 594 |
-
"rewards/chosen": -0.4756461977958679,
|
| 595 |
-
"rewards/margins": 0.39826783537864685,
|
| 596 |
-
"rewards/rejected": -0.8739139437675476,
|
| 597 |
-
"step": 350
|
| 598 |
},
|
| 599 |
{
|
| 600 |
"epoch": 0.3768646951059932,
|
| 601 |
-
"grad_norm":
|
| 602 |
-
"learning_rate":
|
| 603 |
-
"logits/chosen": -
|
| 604 |
-
"logits/rejected": -
|
| 605 |
-
"logps/chosen": -
|
| 606 |
-
"logps/rejected": -
|
| 607 |
-
"loss": 0.
|
| 608 |
-
"rewards/accuracies": 0.7250000238418579,
|
| 609 |
-
"rewards/chosen": -0.4747592806816101,
|
| 610 |
-
"rewards/margins": 0.43383827805519104,
|
| 611 |
-
"rewards/rejected": -0.9085975885391235,
|
| 612 |
-
"step": 360
|
| 613 |
-
},
|
| 614 |
-
{
|
| 615 |
-
"epoch": 0.38733315885893743,
|
| 616 |
-
"grad_norm": 12.256605456435558,
|
| 617 |
-
"learning_rate": 1.5385413962250656e-07,
|
| 618 |
-
"logits/chosen": -13.217790603637695,
|
| 619 |
-
"logits/rejected": -13.344598770141602,
|
| 620 |
-
"logps/chosen": -338.6850280761719,
|
| 621 |
-
"logps/rejected": -376.1455993652344,
|
| 622 |
-
"loss": 0.545,
|
| 623 |
-
"rewards/accuracies": 0.7749999761581421,
|
| 624 |
-
"rewards/chosen": -0.41851702332496643,
|
| 625 |
-
"rewards/margins": 0.6487592458724976,
|
| 626 |
-
"rewards/rejected": -1.0672763586044312,
|
| 627 |
-
"step": 370
|
| 628 |
-
},
|
| 629 |
-
{
|
| 630 |
-
"epoch": 0.39780162261188173,
|
| 631 |
-
"grad_norm": 23.457846993021622,
|
| 632 |
-
"learning_rate": 1.507372039740978e-07,
|
| 633 |
-
"logits/chosen": -13.952138900756836,
|
| 634 |
-
"logits/rejected": -13.335546493530273,
|
| 635 |
-
"logps/chosen": -337.4014892578125,
|
| 636 |
-
"logps/rejected": -424.3997497558594,
|
| 637 |
-
"loss": 0.5639,
|
| 638 |
"rewards/accuracies": 0.6499999761581421,
|
| 639 |
-
"rewards/chosen": -0.
|
| 640 |
-
"rewards/margins": 0.
|
| 641 |
-
"rewards/rejected": -0.
|
| 642 |
-
"step":
|
| 643 |
},
|
| 644 |
{
|
| 645 |
-
"epoch": 0.
|
| 646 |
-
"grad_norm":
|
| 647 |
-
"learning_rate":
|
| 648 |
-
"logits/chosen": -
|
| 649 |
-
"logits/rejected": -
|
| 650 |
-
"logps/chosen": -
|
| 651 |
-
"logps/rejected": -
|
| 652 |
-
"loss": 0.
|
| 653 |
-
"rewards/accuracies": 0.
|
| 654 |
-
"rewards/chosen": -0.
|
| 655 |
-
"rewards/margins": 0.
|
| 656 |
-
"rewards/rejected": -0.
|
| 657 |
-
"step":
|
| 658 |
},
|
| 659 |
{
|
| 660 |
"epoch": 0.4187385501177702,
|
| 661 |
-
"grad_norm":
|
| 662 |
-
"learning_rate":
|
| 663 |
-
"logits/chosen": -
|
| 664 |
-
"logits/rejected": -
|
| 665 |
-
"logps/chosen": -
|
| 666 |
-
"logps/rejected": -
|
| 667 |
-
"loss": 0.
|
| 668 |
-
"rewards/accuracies": 0.
|
| 669 |
-
"rewards/chosen": -0.
|
| 670 |
-
"rewards/margins": 0.
|
| 671 |
-
"rewards/rejected": -0.
|
| 672 |
-
"step":
|
| 673 |
},
|
| 674 |
{
|
| 675 |
"epoch": 0.4187385501177702,
|
| 676 |
-
"eval_logits/chosen": -
|
| 677 |
-
"eval_logits/rejected": -
|
| 678 |
-
"eval_logps/chosen": -
|
| 679 |
-
"eval_logps/rejected": -
|
| 680 |
-
"eval_loss": 0.
|
| 681 |
-
"eval_rewards/accuracies": 0.
|
| 682 |
-
"eval_rewards/chosen": -0.
|
| 683 |
-
"eval_rewards/margins": 0.
|
| 684 |
-
"eval_rewards/rejected": -
|
| 685 |
-
"eval_runtime":
|
| 686 |
-
"eval_samples_per_second": 13.
|
| 687 |
-
"eval_steps_per_second": 0.
|
| 688 |
-
"step":
|
| 689 |
-
},
|
| 690 |
-
{
|
| 691 |
-
"epoch": 0.42920701387071447,
|
| 692 |
-
"grad_norm": 16.540681614535696,
|
| 693 |
-
"learning_rate": 1.4099638051412743e-07,
|
| 694 |
-
"logits/chosen": -13.835235595703125,
|
| 695 |
-
"logits/rejected": -13.860295295715332,
|
| 696 |
-
"logps/chosen": -377.9153747558594,
|
| 697 |
-
"logps/rejected": -418.71014404296875,
|
| 698 |
-
"loss": 0.565,
|
| 699 |
-
"rewards/accuracies": 0.75,
|
| 700 |
-
"rewards/chosen": -0.5656043291091919,
|
| 701 |
-
"rewards/margins": 0.5640857815742493,
|
| 702 |
-
"rewards/rejected": -1.129690170288086,
|
| 703 |
-
"step": 410
|
| 704 |
},
|
| 705 |
{
|
| 706 |
"epoch": 0.4396754776236587,
|
| 707 |
-
"grad_norm":
|
| 708 |
-
"learning_rate":
|
| 709 |
-
"logits/chosen": -
|
| 710 |
-
"logits/rejected": -
|
| 711 |
-
"logps/chosen": -
|
| 712 |
-
"logps/rejected": -
|
| 713 |
-
"loss": 0.
|
| 714 |
-
"rewards/accuracies": 0.
|
| 715 |
-
"rewards/chosen": -0.
|
| 716 |
-
"rewards/margins": 0.
|
| 717 |
-
"rewards/rejected": -
|
| 718 |
-
"step":
|
| 719 |
-
},
|
| 720 |
-
{
|
| 721 |
-
"epoch": 0.45014394137660296,
|
| 722 |
-
"grad_norm": 18.49857147965794,
|
| 723 |
-
"learning_rate": 1.342211064439091e-07,
|
| 724 |
-
"logits/chosen": -13.539401054382324,
|
| 725 |
-
"logits/rejected": -13.536432266235352,
|
| 726 |
-
"logps/chosen": -348.41888427734375,
|
| 727 |
-
"logps/rejected": -399.24371337890625,
|
| 728 |
-
"loss": 0.5585,
|
| 729 |
-
"rewards/accuracies": 0.637499988079071,
|
| 730 |
-
"rewards/chosen": -0.8608742952346802,
|
| 731 |
-
"rewards/margins": 0.3582659065723419,
|
| 732 |
-
"rewards/rejected": -1.2191402912139893,
|
| 733 |
-
"step": 430
|
| 734 |
},
|
| 735 |
{
|
| 736 |
"epoch": 0.46061240512954726,
|
| 737 |
-
"grad_norm":
|
| 738 |
-
"learning_rate":
|
| 739 |
-
"logits/chosen": -
|
| 740 |
-
"logits/rejected": -
|
| 741 |
-
"logps/chosen": -
|
| 742 |
-
"logps/rejected": -
|
| 743 |
-
"loss": 0.
|
| 744 |
-
"rewards/accuracies": 0.
|
| 745 |
-
"rewards/chosen": -0.
|
| 746 |
-
"rewards/margins": 0.
|
| 747 |
-
"rewards/rejected": -
|
| 748 |
-
"step":
|
| 749 |
-
},
|
| 750 |
-
{
|
| 751 |
-
"epoch": 0.4710808688824915,
|
| 752 |
-
"grad_norm": 18.20248083655786,
|
| 753 |
-
"learning_rate": 1.2726282279726786e-07,
|
| 754 |
-
"logits/chosen": -14.505528450012207,
|
| 755 |
-
"logits/rejected": -14.693391799926758,
|
| 756 |
-
"logps/chosen": -330.843017578125,
|
| 757 |
-
"logps/rejected": -396.43011474609375,
|
| 758 |
-
"loss": 0.5424,
|
| 759 |
-
"rewards/accuracies": 0.8125,
|
| 760 |
-
"rewards/chosen": -0.5095499157905579,
|
| 761 |
-
"rewards/margins": 0.6803628206253052,
|
| 762 |
-
"rewards/rejected": -1.1899127960205078,
|
| 763 |
-
"step": 450
|
| 764 |
},
|
| 765 |
{
|
| 766 |
"epoch": 0.48154933263543576,
|
| 767 |
-
"grad_norm":
|
| 768 |
-
"learning_rate":
|
| 769 |
-
"logits/chosen": -
|
| 770 |
-
"logits/rejected": -
|
| 771 |
-
"logps/chosen": -
|
| 772 |
-
"logps/rejected": -
|
| 773 |
-
"loss": 0.
|
| 774 |
-
"rewards/accuracies": 0.
|
| 775 |
-
"rewards/chosen": -0.
|
| 776 |
-
"rewards/margins": 0.
|
| 777 |
-
"rewards/rejected": -1.
|
| 778 |
-
"step":
|
| 779 |
-
},
|
| 780 |
-
{
|
| 781 |
-
"epoch": 0.49201779638838,
|
| 782 |
-
"grad_norm": 19.430025911116154,
|
| 783 |
-
"learning_rate": 1.2015874147864312e-07,
|
| 784 |
-
"logits/chosen": -14.247779846191406,
|
| 785 |
-
"logits/rejected": -14.570295333862305,
|
| 786 |
-
"logps/chosen": -338.3775939941406,
|
| 787 |
-
"logps/rejected": -354.2191467285156,
|
| 788 |
-
"loss": 0.5134,
|
| 789 |
-
"rewards/accuracies": 0.7124999761581421,
|
| 790 |
-
"rewards/chosen": -0.4138456881046295,
|
| 791 |
-
"rewards/margins": 0.44109171628952026,
|
| 792 |
-
"rewards/rejected": -0.8549374341964722,
|
| 793 |
-
"step": 470
|
| 794 |
},
|
| 795 |
{
|
| 796 |
"epoch": 0.5024862601413242,
|
| 797 |
-
"grad_norm":
|
| 798 |
-
"learning_rate":
|
| 799 |
-
"logits/chosen": -
|
| 800 |
-
"logits/rejected": -
|
| 801 |
-
"logps/chosen": -
|
| 802 |
-
"logps/rejected": -
|
| 803 |
-
"loss": 0.
|
| 804 |
-
"rewards/accuracies": 0.
|
| 805 |
-
"rewards/chosen": -0.
|
| 806 |
-
"rewards/margins": 0.
|
| 807 |
-
"rewards/rejected": -
|
| 808 |
-
"step":
|
| 809 |
-
},
|
| 810 |
-
{
|
| 811 |
-
"epoch": 0.5129547238942685,
|
| 812 |
-
"grad_norm": 28.874755648357077,
|
| 813 |
-
"learning_rate": 1.1294685409754433e-07,
|
| 814 |
-
"logits/chosen": -12.982122421264648,
|
| 815 |
-
"logits/rejected": -14.110940933227539,
|
| 816 |
-
"logps/chosen": -390.028076171875,
|
| 817 |
-
"logps/rejected": -422.0865173339844,
|
| 818 |
-
"loss": 0.5217,
|
| 819 |
-
"rewards/accuracies": 0.824999988079071,
|
| 820 |
-
"rewards/chosen": -0.47721004486083984,
|
| 821 |
-
"rewards/margins": 0.8017258644104004,
|
| 822 |
-
"rewards/rejected": -1.2789360284805298,
|
| 823 |
-
"step": 490
|
| 824 |
-
},
|
| 825 |
-
{
|
| 826 |
-
"epoch": 0.5234231876472127,
|
| 827 |
-
"grad_norm": 14.310282443089868,
|
| 828 |
-
"learning_rate": 1.093125187804288e-07,
|
| 829 |
-
"logits/chosen": -13.757654190063477,
|
| 830 |
-
"logits/rejected": -13.917083740234375,
|
| 831 |
-
"logps/chosen": -368.4084777832031,
|
| 832 |
-
"logps/rejected": -402.80621337890625,
|
| 833 |
-
"loss": 0.5441,
|
| 834 |
-
"rewards/accuracies": 0.6875,
|
| 835 |
-
"rewards/chosen": -0.6358953714370728,
|
| 836 |
-
"rewards/margins": 0.5089365243911743,
|
| 837 |
-
"rewards/rejected": -1.144831895828247,
|
| 838 |
-
"step": 500
|
| 839 |
},
|
| 840 |
{
|
| 841 |
"epoch": 0.5234231876472127,
|
| 842 |
-
"
|
| 843 |
-
"
|
| 844 |
-
"
|
| 845 |
-
"
|
| 846 |
-
"
|
| 847 |
-
"
|
| 848 |
-
"
|
| 849 |
-
"
|
| 850 |
-
"
|
| 851 |
-
"
|
| 852 |
-
"
|
| 853 |
-
"
|
| 854 |
-
"step": 500
|
| 855 |
-
},
|
| 856 |
-
{
|
| 857 |
-
"epoch": 0.533891651400157,
|
| 858 |
-
"grad_norm": 20.201758093145344,
|
| 859 |
-
"learning_rate": 1.0566572879486386e-07,
|
| 860 |
-
"logits/chosen": -13.861791610717773,
|
| 861 |
-
"logits/rejected": -14.239435195922852,
|
| 862 |
-
"logps/chosen": -357.29888916015625,
|
| 863 |
-
"logps/rejected": -406.57366943359375,
|
| 864 |
-
"loss": 0.5252,
|
| 865 |
-
"rewards/accuracies": 0.762499988079071,
|
| 866 |
-
"rewards/chosen": -0.498563289642334,
|
| 867 |
-
"rewards/margins": 0.7506011724472046,
|
| 868 |
-
"rewards/rejected": -1.249164342880249,
|
| 869 |
-
"step": 510
|
| 870 |
},
|
| 871 |
{
|
| 872 |
"epoch": 0.5443601151531012,
|
| 873 |
-
"grad_norm":
|
| 874 |
-
"learning_rate":
|
| 875 |
-
"logits/chosen": -14.
|
| 876 |
-
"logits/rejected": -14.
|
| 877 |
-
"logps/chosen": -
|
| 878 |
-
"logps/rejected": -
|
| 879 |
-
"loss": 0.
|
| 880 |
-
"rewards/accuracies": 0.
|
| 881 |
-
"rewards/chosen": -0.
|
| 882 |
-
"rewards/margins": 0.
|
| 883 |
-
"rewards/rejected": -1.
|
| 884 |
-
"step":
|
| 885 |
-
},
|
| 886 |
-
{
|
| 887 |
-
"epoch": 0.5548285789060455,
|
| 888 |
-
"grad_norm": 20.995271815473735,
|
| 889 |
-
"learning_rate": 9.835430398598318e-08,
|
| 890 |
-
"logits/chosen": -14.159637451171875,
|
| 891 |
-
"logits/rejected": -14.701879501342773,
|
| 892 |
-
"logps/chosen": -377.4815979003906,
|
| 893 |
-
"logps/rejected": -436.66632080078125,
|
| 894 |
-
"loss": 0.5273,
|
| 895 |
-
"rewards/accuracies": 0.75,
|
| 896 |
-
"rewards/chosen": -0.3921489119529724,
|
| 897 |
-
"rewards/margins": 0.8064279556274414,
|
| 898 |
-
"rewards/rejected": -1.1985770463943481,
|
| 899 |
-
"step": 530
|
| 900 |
},
|
| 901 |
{
|
| 902 |
"epoch": 0.5652970426589898,
|
| 903 |
-
"grad_norm":
|
| 904 |
-
"learning_rate":
|
| 905 |
-
"logits/chosen": -13.
|
| 906 |
-
"logits/rejected": -
|
| 907 |
-
"logps/chosen": -
|
| 908 |
-
"logps/rejected": -
|
| 909 |
-
"loss": 0.
|
| 910 |
-
"rewards/accuracies": 0.
|
| 911 |
-
"rewards/chosen": -0.
|
| 912 |
-
"rewards/margins": 0.
|
| 913 |
-
"rewards/rejected": -
|
| 914 |
-
"step":
|
| 915 |
-
},
|
| 916 |
-
{
|
| 917 |
-
"epoch": 0.575765506411934,
|
| 918 |
-
"grad_norm": 21.357543363893058,
|
| 919 |
-
"learning_rate": 9.105168012370371e-08,
|
| 920 |
-
"logits/chosen": -13.735044479370117,
|
| 921 |
-
"logits/rejected": -14.701571464538574,
|
| 922 |
-
"logps/chosen": -379.26495361328125,
|
| 923 |
-
"logps/rejected": -371.4606628417969,
|
| 924 |
-
"loss": 0.5351,
|
| 925 |
-
"rewards/accuracies": 0.7250000238418579,
|
| 926 |
-
"rewards/chosen": -0.69088214635849,
|
| 927 |
-
"rewards/margins": 0.4814137816429138,
|
| 928 |
-
"rewards/rejected": -1.1722959280014038,
|
| 929 |
-
"step": 550
|
| 930 |
},
|
| 931 |
{
|
| 932 |
"epoch": 0.5862339701648783,
|
| 933 |
-
"grad_norm":
|
| 934 |
-
"learning_rate":
|
| 935 |
-
"logits/chosen": -14.
|
| 936 |
-
"logits/rejected": -
|
| 937 |
-
"logps/chosen": -
|
| 938 |
-
"logps/rejected": -
|
| 939 |
-
"loss": 0.
|
| 940 |
-
"rewards/accuracies": 0.
|
| 941 |
-
"rewards/chosen": -0.
|
| 942 |
-
"rewards/margins": 0.
|
| 943 |
-
"rewards/rejected": -
|
| 944 |
-
"step":
|
| 945 |
-
},
|
| 946 |
-
{
|
| 947 |
-
"epoch": 0.5967024339178225,
|
| 948 |
-
"grad_norm": 16.735647281024168,
|
| 949 |
-
"learning_rate": 8.379691059462476e-08,
|
| 950 |
-
"logits/chosen": -13.99199104309082,
|
| 951 |
-
"logits/rejected": -14.157026290893555,
|
| 952 |
-
"logps/chosen": -374.80938720703125,
|
| 953 |
-
"logps/rejected": -430.4781799316406,
|
| 954 |
-
"loss": 0.5302,
|
| 955 |
-
"rewards/accuracies": 0.699999988079071,
|
| 956 |
-
"rewards/chosen": -0.6314277648925781,
|
| 957 |
-
"rewards/margins": 0.5574957132339478,
|
| 958 |
-
"rewards/rejected": -1.1889234781265259,
|
| 959 |
-
"step": 570
|
| 960 |
},
|
| 961 |
{
|
| 962 |
"epoch": 0.6071708976707668,
|
| 963 |
-
"grad_norm":
|
| 964 |
-
"learning_rate":
|
| 965 |
-
"logits/chosen": -14.
|
| 966 |
-
"logits/rejected": -14.
|
| 967 |
-
"logps/chosen": -
|
| 968 |
-
"logps/rejected": -
|
| 969 |
-
"loss": 0.
|
| 970 |
-
"rewards/accuracies": 0.
|
| 971 |
-
"rewards/chosen": -0.
|
| 972 |
-
"rewards/margins": 0.
|
| 973 |
-
"rewards/rejected": -1.
|
| 974 |
-
"step":
|
| 975 |
-
},
|
| 976 |
-
{
|
| 977 |
-
"epoch": 0.6176393614237111,
|
| 978 |
-
"grad_norm": 26.479615574042146,
|
| 979 |
-
"learning_rate": 7.662879286722496e-08,
|
| 980 |
-
"logits/chosen": -13.325056076049805,
|
| 981 |
-
"logits/rejected": -13.995033264160156,
|
| 982 |
-
"logps/chosen": -348.37322998046875,
|
| 983 |
-
"logps/rejected": -410.867431640625,
|
| 984 |
-
"loss": 0.5221,
|
| 985 |
-
"rewards/accuracies": 0.7875000238418579,
|
| 986 |
-
"rewards/chosen": -0.5271872282028198,
|
| 987 |
-
"rewards/margins": 0.7677302360534668,
|
| 988 |
-
"rewards/rejected": -1.2949175834655762,
|
| 989 |
-
"step": 590
|
| 990 |
},
|
| 991 |
{
|
| 992 |
"epoch": 0.6281078251766553,
|
| 993 |
-
"grad_norm":
|
| 994 |
-
"learning_rate":
|
| 995 |
-
"logits/chosen": -
|
| 996 |
-
"logits/rejected": -14.
|
| 997 |
-
"logps/chosen": -
|
| 998 |
-
"logps/rejected": -
|
| 999 |
-
"loss": 0.
|
| 1000 |
-
"rewards/accuracies": 0.
|
| 1001 |
-
"rewards/chosen": -0.
|
| 1002 |
-
"rewards/margins": 0.
|
| 1003 |
-
"rewards/rejected": -1.
|
| 1004 |
-
"step":
|
| 1005 |
},
|
| 1006 |
{
|
| 1007 |
"epoch": 0.6281078251766553,
|
| 1008 |
-
"eval_logits/chosen": -
|
| 1009 |
-
"eval_logits/rejected": -14.
|
| 1010 |
-
"eval_logps/chosen": -
|
| 1011 |
-
"eval_logps/rejected": -
|
| 1012 |
-
"eval_loss": 0.
|
| 1013 |
-
"eval_rewards/accuracies": 0.
|
| 1014 |
-
"eval_rewards/chosen": -0.
|
| 1015 |
-
"eval_rewards/margins": 0.
|
| 1016 |
-
"eval_rewards/rejected": -1.
|
| 1017 |
-
"eval_runtime": 150.
|
| 1018 |
-
"eval_samples_per_second": 13.
|
| 1019 |
"eval_steps_per_second": 0.832,
|
| 1020 |
-
"step":
|
| 1021 |
-
},
|
| 1022 |
-
{
|
| 1023 |
-
"epoch": 0.6385762889295996,
|
| 1024 |
-
"grad_norm": 15.068671602779133,
|
| 1025 |
-
"learning_rate": 6.958566100855715e-08,
|
| 1026 |
-
"logits/chosen": -13.8591890335083,
|
| 1027 |
-
"logits/rejected": -14.538678169250488,
|
| 1028 |
-
"logps/chosen": -354.1308288574219,
|
| 1029 |
-
"logps/rejected": -361.6885070800781,
|
| 1030 |
-
"loss": 0.535,
|
| 1031 |
-
"rewards/accuracies": 0.800000011920929,
|
| 1032 |
-
"rewards/chosen": -0.481017529964447,
|
| 1033 |
-
"rewards/margins": 0.7814317941665649,
|
| 1034 |
-
"rewards/rejected": -1.2624493837356567,
|
| 1035 |
-
"step": 610
|
| 1036 |
},
|
| 1037 |
{
|
| 1038 |
"epoch": 0.6490447526825438,
|
| 1039 |
-
"grad_norm":
|
| 1040 |
-
"learning_rate":
|
| 1041 |
-
"logits/chosen": -
|
| 1042 |
-
"logits/rejected": -
|
| 1043 |
-
"logps/chosen": -
|
| 1044 |
-
"logps/rejected": -
|
| 1045 |
-
"loss": 0.
|
| 1046 |
-
"rewards/accuracies": 0.
|
| 1047 |
-
"rewards/chosen": -0.
|
| 1048 |
-
"rewards/margins": 0.
|
| 1049 |
-
"rewards/rejected": -1.
|
| 1050 |
-
"step":
|
| 1051 |
-
},
|
| 1052 |
-
{
|
| 1053 |
-
"epoch": 0.6595132164354881,
|
| 1054 |
-
"grad_norm": 19.103702131187973,
|
| 1055 |
-
"learning_rate": 6.270518067914745e-08,
|
| 1056 |
-
"logits/chosen": -13.852685928344727,
|
| 1057 |
-
"logits/rejected": -14.353567123413086,
|
| 1058 |
-
"logps/chosen": -374.9162902832031,
|
| 1059 |
-
"logps/rejected": -387.75860595703125,
|
| 1060 |
-
"loss": 0.5089,
|
| 1061 |
-
"rewards/accuracies": 0.625,
|
| 1062 |
-
"rewards/chosen": -0.5707219243049622,
|
| 1063 |
-
"rewards/margins": 0.6046653389930725,
|
| 1064 |
-
"rewards/rejected": -1.1753873825073242,
|
| 1065 |
-
"step": 630
|
| 1066 |
},
|
| 1067 |
{
|
| 1068 |
"epoch": 0.6699816801884323,
|
| 1069 |
-
"grad_norm":
|
| 1070 |
-
"learning_rate":
|
| 1071 |
-
"logits/chosen": -
|
| 1072 |
-
"logits/rejected": -
|
| 1073 |
-
"logps/chosen": -
|
| 1074 |
-
"logps/rejected": -
|
| 1075 |
-
"loss": 0.
|
| 1076 |
-
"rewards/accuracies": 0.
|
| 1077 |
-
"rewards/chosen": -0.
|
| 1078 |
-
"rewards/margins": 0.
|
| 1079 |
-
"rewards/rejected": -1.
|
| 1080 |
-
"step":
|
| 1081 |
-
},
|
| 1082 |
-
{
|
| 1083 |
-
"epoch": 0.6804501439413766,
|
| 1084 |
-
"grad_norm": 26.384619321462868,
|
| 1085 |
-
"learning_rate": 5.6024147702436975e-08,
|
| 1086 |
-
"logits/chosen": -13.914807319641113,
|
| 1087 |
-
"logits/rejected": -14.317163467407227,
|
| 1088 |
-
"logps/chosen": -386.0091247558594,
|
| 1089 |
-
"logps/rejected": -435.4970703125,
|
| 1090 |
-
"loss": 0.5178,
|
| 1091 |
-
"rewards/accuracies": 0.699999988079071,
|
| 1092 |
-
"rewards/chosen": -0.7489283680915833,
|
| 1093 |
-
"rewards/margins": 0.6266080141067505,
|
| 1094 |
-
"rewards/rejected": -1.3755362033843994,
|
| 1095 |
-
"step": 650
|
| 1096 |
},
|
| 1097 |
{
|
| 1098 |
"epoch": 0.6909186076943209,
|
| 1099 |
-
"grad_norm":
|
| 1100 |
-
"learning_rate":
|
| 1101 |
-
"logits/chosen": -
|
| 1102 |
-
"logits/rejected": -
|
| 1103 |
-
"logps/chosen": -
|
| 1104 |
-
"logps/rejected": -
|
| 1105 |
-
"loss": 0.
|
| 1106 |
-
"rewards/accuracies": 0.
|
| 1107 |
-
"rewards/chosen": -0.
|
| 1108 |
-
"rewards/margins": 0.
|
| 1109 |
-
"rewards/rejected": -1.
|
| 1110 |
-
"step":
|
| 1111 |
-
},
|
| 1112 |
-
{
|
| 1113 |
-
"epoch": 0.7013870714472651,
|
| 1114 |
-
"grad_norm": 24.949872599196897,
|
| 1115 |
-
"learning_rate": 4.95782912859878e-08,
|
| 1116 |
-
"logits/chosen": -13.573728561401367,
|
| 1117 |
-
"logits/rejected": -13.890779495239258,
|
| 1118 |
-
"logps/chosen": -366.00323486328125,
|
| 1119 |
-
"logps/rejected": -416.9151916503906,
|
| 1120 |
-
"loss": 0.5086,
|
| 1121 |
-
"rewards/accuracies": 0.800000011920929,
|
| 1122 |
-
"rewards/chosen": -0.6050105094909668,
|
| 1123 |
-
"rewards/margins": 0.8032246828079224,
|
| 1124 |
-
"rewards/rejected": -1.4082351922988892,
|
| 1125 |
-
"step": 670
|
| 1126 |
},
|
| 1127 |
{
|
| 1128 |
"epoch": 0.7118555352002094,
|
| 1129 |
-
"grad_norm":
|
| 1130 |
-
"learning_rate":
|
| 1131 |
-
"logits/chosen": -
|
| 1132 |
-
"logits/rejected": -14.
|
| 1133 |
-
"logps/chosen": -
|
| 1134 |
-
"logps/rejected": -
|
| 1135 |
-
"loss": 0.
|
| 1136 |
-
"rewards/accuracies": 0.6499999761581421,
|
| 1137 |
-
"rewards/chosen": -0.6707770824432373,
|
| 1138 |
-
"rewards/margins": 0.42906999588012695,
|
| 1139 |
-
"rewards/rejected": -1.0998470783233643,
|
| 1140 |
-
"step": 680
|
| 1141 |
-
},
|
| 1142 |
-
{
|
| 1143 |
-
"epoch": 0.7223239989531536,
|
| 1144 |
-
"grad_norm": 24.422507206522518,
|
| 1145 |
-
"learning_rate": 4.340208294679745e-08,
|
| 1146 |
-
"logits/chosen": -13.968500137329102,
|
| 1147 |
-
"logits/rejected": -14.3223876953125,
|
| 1148 |
-
"logps/chosen": -341.5804748535156,
|
| 1149 |
-
"logps/rejected": -367.90045166015625,
|
| 1150 |
-
"loss": 0.5023,
|
| 1151 |
-
"rewards/accuracies": 0.6875,
|
| 1152 |
-
"rewards/chosen": -0.6214663982391357,
|
| 1153 |
-
"rewards/margins": 0.579698383808136,
|
| 1154 |
-
"rewards/rejected": -1.201164722442627,
|
| 1155 |
-
"step": 690
|
| 1156 |
-
},
|
| 1157 |
-
{
|
| 1158 |
-
"epoch": 0.7327924627060979,
|
| 1159 |
-
"grad_norm": 16.46994924986635,
|
| 1160 |
-
"learning_rate": 4.042547973446017e-08,
|
| 1161 |
-
"logits/chosen": -13.902259826660156,
|
| 1162 |
-
"logits/rejected": -14.259056091308594,
|
| 1163 |
-
"logps/chosen": -358.53607177734375,
|
| 1164 |
-
"logps/rejected": -394.62664794921875,
|
| 1165 |
-
"loss": 0.5158,
|
| 1166 |
"rewards/accuracies": 0.699999988079071,
|
| 1167 |
-
"rewards/chosen": -0.
|
| 1168 |
-
"rewards/margins": 0.
|
| 1169 |
-
"rewards/rejected": -1.
|
| 1170 |
-
"step":
|
| 1171 |
},
|
| 1172 |
{
|
| 1173 |
"epoch": 0.7327924627060979,
|
| 1174 |
-
"
|
| 1175 |
-
"
|
| 1176 |
-
"
|
| 1177 |
-
"
|
| 1178 |
-
"
|
| 1179 |
-
"
|
| 1180 |
-
"
|
| 1181 |
-
"
|
| 1182 |
-
"
|
| 1183 |
-
"
|
| 1184 |
-
"
|
| 1185 |
-
"
|
| 1186 |
-
"step": 700
|
| 1187 |
-
},
|
| 1188 |
-
{
|
| 1189 |
-
"epoch": 0.7432609264590422,
|
| 1190 |
-
"grad_norm": 21.302516106563797,
|
| 1191 |
-
"learning_rate": 3.7528552162562855e-08,
|
| 1192 |
-
"logits/chosen": -14.20964527130127,
|
| 1193 |
-
"logits/rejected": -14.762395858764648,
|
| 1194 |
-
"logps/chosen": -352.42864990234375,
|
| 1195 |
-
"logps/rejected": -357.80755615234375,
|
| 1196 |
-
"loss": 0.5624,
|
| 1197 |
-
"rewards/accuracies": 0.762499988079071,
|
| 1198 |
-
"rewards/chosen": -0.6552818417549133,
|
| 1199 |
-
"rewards/margins": 0.563475489616394,
|
| 1200 |
-
"rewards/rejected": -1.2187573909759521,
|
| 1201 |
-
"step": 710
|
| 1202 |
},
|
| 1203 |
{
|
| 1204 |
"epoch": 0.7537293902119864,
|
| 1205 |
-
"grad_norm":
|
| 1206 |
-
"learning_rate":
|
| 1207 |
-
"logits/chosen": -
|
| 1208 |
-
"logits/rejected": -14.
|
| 1209 |
-
"logps/chosen": -
|
| 1210 |
-
"logps/rejected": -
|
| 1211 |
-
"loss": 0.
|
| 1212 |
-
"rewards/accuracies": 0.
|
| 1213 |
-
"rewards/chosen": -0.
|
| 1214 |
-
"rewards/margins": 0.
|
| 1215 |
-
"rewards/rejected": -1.
|
| 1216 |
-
"step":
|
| 1217 |
-
},
|
| 1218 |
-
{
|
| 1219 |
-
"epoch": 0.7641978539649307,
|
| 1220 |
-
"grad_norm": 14.730020567879517,
|
| 1221 |
-
"learning_rate": 3.198910973476393e-08,
|
| 1222 |
-
"logits/chosen": -13.825230598449707,
|
| 1223 |
-
"logits/rejected": -14.049738883972168,
|
| 1224 |
-
"logps/chosen": -334.89404296875,
|
| 1225 |
-
"logps/rejected": -387.7596740722656,
|
| 1226 |
-
"loss": 0.5295,
|
| 1227 |
-
"rewards/accuracies": 0.7875000238418579,
|
| 1228 |
-
"rewards/chosen": -0.5604512691497803,
|
| 1229 |
-
"rewards/margins": 0.7153105735778809,
|
| 1230 |
-
"rewards/rejected": -1.2757618427276611,
|
| 1231 |
-
"step": 730
|
| 1232 |
},
|
| 1233 |
{
|
| 1234 |
"epoch": 0.7746663177178749,
|
| 1235 |
-
"grad_norm":
|
| 1236 |
-
"learning_rate":
|
| 1237 |
-
"logits/chosen": -
|
| 1238 |
-
"logits/rejected": -
|
| 1239 |
-
"logps/chosen": -
|
| 1240 |
-
"logps/rejected": -
|
| 1241 |
-
"loss": 0.
|
| 1242 |
-
"rewards/accuracies": 0.
|
| 1243 |
-
"rewards/chosen": -0.
|
| 1244 |
-
"rewards/margins": 0.
|
| 1245 |
-
"rewards/rejected": -1.
|
| 1246 |
-
"step":
|
| 1247 |
-
},
|
| 1248 |
-
{
|
| 1249 |
-
"epoch": 0.7851347814708192,
|
| 1250 |
-
"grad_norm": 13.767843648079221,
|
| 1251 |
-
"learning_rate": 2.6813379808195357e-08,
|
| 1252 |
-
"logits/chosen": -14.288251876831055,
|
| 1253 |
-
"logits/rejected": -14.359420776367188,
|
| 1254 |
-
"logps/chosen": -339.8260498046875,
|
| 1255 |
-
"logps/rejected": -450.6610412597656,
|
| 1256 |
-
"loss": 0.509,
|
| 1257 |
-
"rewards/accuracies": 0.800000011920929,
|
| 1258 |
-
"rewards/chosen": -0.5569700598716736,
|
| 1259 |
-
"rewards/margins": 0.9691677093505859,
|
| 1260 |
-
"rewards/rejected": -1.5261377096176147,
|
| 1261 |
-
"step": 750
|
| 1262 |
},
|
| 1263 |
{
|
| 1264 |
"epoch": 0.7956032452237635,
|
| 1265 |
-
"grad_norm":
|
| 1266 |
-
"learning_rate":
|
| 1267 |
-
"logits/chosen": -
|
| 1268 |
-
"logits/rejected": -
|
| 1269 |
-
"logps/chosen": -
|
| 1270 |
-
"logps/rejected": -
|
| 1271 |
-
"loss": 0.
|
| 1272 |
-
"rewards/accuracies": 0.
|
| 1273 |
-
"rewards/chosen": -0.
|
| 1274 |
-
"rewards/margins": 0.
|
| 1275 |
-
"rewards/rejected": -1.
|
| 1276 |
-
"step":
|
| 1277 |
-
},
|
| 1278 |
-
{
|
| 1279 |
-
"epoch": 0.8060717089767077,
|
| 1280 |
-
"grad_norm": 23.918260657300134,
|
| 1281 |
-
"learning_rate": 2.202904144528295e-08,
|
| 1282 |
-
"logits/chosen": -14.572360038757324,
|
| 1283 |
-
"logits/rejected": -14.208137512207031,
|
| 1284 |
-
"logps/chosen": -367.49468994140625,
|
| 1285 |
-
"logps/rejected": -460.3101501464844,
|
| 1286 |
-
"loss": 0.5074,
|
| 1287 |
-
"rewards/accuracies": 0.675000011920929,
|
| 1288 |
-
"rewards/chosen": -0.5758759379386902,
|
| 1289 |
-
"rewards/margins": 0.6287983655929565,
|
| 1290 |
-
"rewards/rejected": -1.204674243927002,
|
| 1291 |
-
"step": 770
|
| 1292 |
},
|
| 1293 |
{
|
| 1294 |
"epoch": 0.816540172729652,
|
| 1295 |
-
"grad_norm":
|
| 1296 |
-
"learning_rate":
|
| 1297 |
-
"logits/chosen": -
|
| 1298 |
-
"logits/rejected": -14.
|
| 1299 |
-
"logps/chosen": -
|
| 1300 |
-
"logps/rejected": -
|
| 1301 |
-
"loss": 0.
|
| 1302 |
-
"rewards/accuracies": 0.
|
| 1303 |
-
"rewards/chosen": -0.
|
| 1304 |
-
"rewards/margins": 0.
|
| 1305 |
-
"rewards/rejected": -1.
|
| 1306 |
-
"step":
|
| 1307 |
-
},
|
| 1308 |
-
{
|
| 1309 |
-
"epoch": 0.8270086364825961,
|
| 1310 |
-
"grad_norm": 25.443458513615624,
|
| 1311 |
-
"learning_rate": 1.766168060242159e-08,
|
| 1312 |
-
"logits/chosen": -14.38599681854248,
|
| 1313 |
-
"logits/rejected": -14.689620971679688,
|
| 1314 |
-
"logps/chosen": -313.2379455566406,
|
| 1315 |
-
"logps/rejected": -354.0423889160156,
|
| 1316 |
-
"loss": 0.5395,
|
| 1317 |
-
"rewards/accuracies": 0.699999988079071,
|
| 1318 |
-
"rewards/chosen": -0.5758354663848877,
|
| 1319 |
-
"rewards/margins": 0.5628177523612976,
|
| 1320 |
-
"rewards/rejected": -1.1386531591415405,
|
| 1321 |
-
"step": 790
|
| 1322 |
},
|
| 1323 |
{
|
| 1324 |
"epoch": 0.8374771002355405,
|
| 1325 |
-
"grad_norm":
|
| 1326 |
-
"learning_rate":
|
| 1327 |
-
"logits/chosen": -
|
| 1328 |
-
"logits/rejected": -
|
| 1329 |
-
"logps/chosen": -
|
| 1330 |
-
"logps/rejected": -
|
| 1331 |
-
"loss": 0.
|
| 1332 |
-
"rewards/accuracies": 0.
|
| 1333 |
-
"rewards/chosen": -0.
|
| 1334 |
-
"rewards/margins": 0.
|
| 1335 |
-
"rewards/rejected": -1.
|
| 1336 |
-
"step":
|
| 1337 |
},
|
| 1338 |
{
|
| 1339 |
"epoch": 0.8374771002355405,
|
| 1340 |
-
"eval_logits/chosen": -14.
|
| 1341 |
-
"eval_logits/rejected": -14.
|
| 1342 |
-
"eval_logps/chosen": -
|
| 1343 |
-
"eval_logps/rejected": -
|
| 1344 |
-
"eval_loss": 0.
|
| 1345 |
-
"eval_rewards/accuracies": 0.
|
| 1346 |
-
"eval_rewards/chosen": -0.
|
| 1347 |
-
"eval_rewards/margins": 0.
|
| 1348 |
-
"eval_rewards/rejected": -1.
|
| 1349 |
-
"eval_runtime": 149.
|
| 1350 |
-
"eval_samples_per_second": 13.
|
| 1351 |
-
"eval_steps_per_second": 0.
|
| 1352 |
-
"step":
|
| 1353 |
-
},
|
| 1354 |
-
{
|
| 1355 |
-
"epoch": 0.8479455639884846,
|
| 1356 |
-
"grad_norm": 22.038018685540855,
|
| 1357 |
-
"learning_rate": 1.3734653299944831e-08,
|
| 1358 |
-
"logits/chosen": -13.653346061706543,
|
| 1359 |
-
"logits/rejected": -14.236944198608398,
|
| 1360 |
-
"logps/chosen": -420.22412109375,
|
| 1361 |
-
"logps/rejected": -421.231201171875,
|
| 1362 |
-
"loss": 0.532,
|
| 1363 |
-
"rewards/accuracies": 0.7749999761581421,
|
| 1364 |
-
"rewards/chosen": -0.6482435464859009,
|
| 1365 |
-
"rewards/margins": 0.6211605072021484,
|
| 1366 |
-
"rewards/rejected": -1.2694040536880493,
|
| 1367 |
-
"step": 810
|
| 1368 |
},
|
| 1369 |
{
|
| 1370 |
"epoch": 0.8584140277414289,
|
| 1371 |
-
"grad_norm":
|
| 1372 |
-
"learning_rate":
|
| 1373 |
-
"logits/chosen": -
|
| 1374 |
-
"logits/rejected": -
|
| 1375 |
-
"logps/chosen": -
|
| 1376 |
-
"logps/rejected": -
|
| 1377 |
-
"loss": 0.
|
| 1378 |
-
"rewards/accuracies": 0.
|
| 1379 |
-
"rewards/chosen": -0.
|
| 1380 |
-
"rewards/margins": 0.
|
| 1381 |
-
"rewards/rejected": -1.
|
| 1382 |
-
"step":
|
| 1383 |
-
},
|
| 1384 |
-
{
|
| 1385 |
-
"epoch": 0.8688824914943732,
|
| 1386 |
-
"grad_norm": 16.83996659028827,
|
| 1387 |
-
"learning_rate": 1.0268960717472741e-08,
|
| 1388 |
-
"logits/chosen": -14.19860553741455,
|
| 1389 |
-
"logits/rejected": -14.166885375976562,
|
| 1390 |
-
"logps/chosen": -347.3482971191406,
|
| 1391 |
-
"logps/rejected": -415.2900390625,
|
| 1392 |
-
"loss": 0.5355,
|
| 1393 |
-
"rewards/accuracies": 0.7749999761581421,
|
| 1394 |
-
"rewards/chosen": -0.5952852368354797,
|
| 1395 |
-
"rewards/margins": 0.6116087436676025,
|
| 1396 |
-
"rewards/rejected": -1.2068939208984375,
|
| 1397 |
-
"step": 830
|
| 1398 |
},
|
| 1399 |
{
|
| 1400 |
"epoch": 0.8793509552473174,
|
| 1401 |
-
"grad_norm":
|
| 1402 |
-
"learning_rate":
|
| 1403 |
-
"logits/chosen": -
|
| 1404 |
-
"logits/rejected": -14.
|
| 1405 |
-
"logps/chosen": -
|
| 1406 |
-
"logps/rejected": -
|
| 1407 |
-
"loss": 0.
|
| 1408 |
-
"rewards/accuracies": 0.
|
| 1409 |
-
"rewards/chosen": -0.
|
| 1410 |
-
"rewards/margins": 0.
|
| 1411 |
-
"rewards/rejected": -1.
|
| 1412 |
-
"step":
|
| 1413 |
-
},
|
| 1414 |
-
{
|
| 1415 |
-
"epoch": 0.8898194190002617,
|
| 1416 |
-
"grad_norm": 39.033655096711556,
|
| 1417 |
-
"learning_rate": 7.2831368826110625e-09,
|
| 1418 |
-
"logits/chosen": -14.004640579223633,
|
| 1419 |
-
"logits/rejected": -14.758051872253418,
|
| 1420 |
-
"logps/chosen": -382.56427001953125,
|
| 1421 |
-
"logps/rejected": -397.9862365722656,
|
| 1422 |
-
"loss": 0.5084,
|
| 1423 |
-
"rewards/accuracies": 0.800000011920929,
|
| 1424 |
-
"rewards/chosen": -0.5773628354072571,
|
| 1425 |
-
"rewards/margins": 0.7263933420181274,
|
| 1426 |
-
"rewards/rejected": -1.3037563562393188,
|
| 1427 |
-
"step": 850
|
| 1428 |
},
|
| 1429 |
{
|
| 1430 |
"epoch": 0.9002878827532059,
|
| 1431 |
-
"grad_norm":
|
| 1432 |
-
"learning_rate":
|
| 1433 |
-
"logits/chosen": -
|
| 1434 |
-
"logits/rejected": -
|
| 1435 |
-
"logps/chosen": -
|
| 1436 |
-
"logps/rejected": -
|
| 1437 |
-
"loss": 0.
|
| 1438 |
-
"rewards/accuracies": 0.
|
| 1439 |
-
"rewards/chosen": -0.
|
| 1440 |
-
"rewards/margins": 0.
|
| 1441 |
-
"rewards/rejected": -1.
|
| 1442 |
-
"step":
|
| 1443 |
-
},
|
| 1444 |
-
{
|
| 1445 |
-
"epoch": 0.9107563465061502,
|
| 1446 |
-
"grad_norm": 59.31572753609887,
|
| 1447 |
-
"learning_rate": 4.793149553625786e-09,
|
| 1448 |
-
"logits/chosen": -14.279424667358398,
|
| 1449 |
-
"logits/rejected": -14.221160888671875,
|
| 1450 |
-
"logps/chosen": -333.5687255859375,
|
| 1451 |
-
"logps/rejected": -372.06304931640625,
|
| 1452 |
-
"loss": 0.5295,
|
| 1453 |
-
"rewards/accuracies": 0.699999988079071,
|
| 1454 |
-
"rewards/chosen": -0.5818564295768738,
|
| 1455 |
-
"rewards/margins": 0.6132394075393677,
|
| 1456 |
-
"rewards/rejected": -1.1950958967208862,
|
| 1457 |
-
"step": 870
|
| 1458 |
},
|
| 1459 |
{
|
| 1460 |
"epoch": 0.9212248102590945,
|
| 1461 |
-
"grad_norm":
|
| 1462 |
-
"learning_rate":
|
| 1463 |
-
"logits/chosen": -14.
|
| 1464 |
-
"logits/rejected": -14.
|
| 1465 |
-
"logps/chosen": -
|
| 1466 |
-
"logps/rejected": -
|
| 1467 |
-
"loss": 0.
|
| 1468 |
-
"rewards/accuracies": 0.
|
| 1469 |
-
"rewards/chosen": -0.
|
| 1470 |
-
"rewards/margins": 0.
|
| 1471 |
-
"rewards/rejected": -1.
|
| 1472 |
-
"step":
|
| 1473 |
-
},
|
| 1474 |
-
{
|
| 1475 |
-
"epoch": 0.9316932740120387,
|
| 1476 |
-
"grad_norm": 23.713770091143388,
|
| 1477 |
-
"learning_rate": 2.812314826158746e-09,
|
| 1478 |
-
"logits/chosen": -13.538187026977539,
|
| 1479 |
-
"logits/rejected": -13.821019172668457,
|
| 1480 |
-
"logps/chosen": -342.670654296875,
|
| 1481 |
-
"logps/rejected": -417.42608642578125,
|
| 1482 |
-
"loss": 0.5146,
|
| 1483 |
-
"rewards/accuracies": 0.699999988079071,
|
| 1484 |
-
"rewards/chosen": -0.5183156728744507,
|
| 1485 |
-
"rewards/margins": 0.8244325518608093,
|
| 1486 |
-
"rewards/rejected": -1.3427482843399048,
|
| 1487 |
-
"step": 890
|
| 1488 |
-
},
|
| 1489 |
-
{
|
| 1490 |
-
"epoch": 0.942161737764983,
|
| 1491 |
-
"grad_norm": 17.247650815291333,
|
| 1492 |
-
"learning_rate": 2.016248081729144e-09,
|
| 1493 |
-
"logits/chosen": -14.117114067077637,
|
| 1494 |
-
"logits/rejected": -14.199200630187988,
|
| 1495 |
-
"logps/chosen": -360.4630126953125,
|
| 1496 |
-
"logps/rejected": -437.9790954589844,
|
| 1497 |
-
"loss": 0.5098,
|
| 1498 |
-
"rewards/accuracies": 0.75,
|
| 1499 |
-
"rewards/chosen": -0.5845374464988708,
|
| 1500 |
-
"rewards/margins": 0.8339082598686218,
|
| 1501 |
-
"rewards/rejected": -1.4184458255767822,
|
| 1502 |
-
"step": 900
|
| 1503 |
},
|
| 1504 |
{
|
| 1505 |
"epoch": 0.942161737764983,
|
| 1506 |
-
"
|
| 1507 |
-
"
|
| 1508 |
-
"
|
| 1509 |
-
"
|
| 1510 |
-
"
|
| 1511 |
-
"
|
| 1512 |
-
"
|
| 1513 |
-
"
|
| 1514 |
-
"
|
| 1515 |
-
"
|
| 1516 |
-
"
|
| 1517 |
-
"
|
| 1518 |
-
"step": 900
|
| 1519 |
-
},
|
| 1520 |
-
{
|
| 1521 |
-
"epoch": 0.9526302015179272,
|
| 1522 |
-
"grad_norm": 38.823289395104936,
|
| 1523 |
-
"learning_rate": 1.3512259206550746e-09,
|
| 1524 |
-
"logits/chosen": -13.173808097839355,
|
| 1525 |
-
"logits/rejected": -14.06005573272705,
|
| 1526 |
-
"logps/chosen": -376.40521240234375,
|
| 1527 |
-
"logps/rejected": -369.52972412109375,
|
| 1528 |
-
"loss": 0.5286,
|
| 1529 |
-
"rewards/accuracies": 0.6875,
|
| 1530 |
-
"rewards/chosen": -0.6154804229736328,
|
| 1531 |
-
"rewards/margins": 0.589421272277832,
|
| 1532 |
-
"rewards/rejected": -1.2049016952514648,
|
| 1533 |
-
"step": 910
|
| 1534 |
},
|
| 1535 |
{
|
| 1536 |
"epoch": 0.9630986652708715,
|
| 1537 |
-
"grad_norm":
|
| 1538 |
-
"learning_rate":
|
| 1539 |
-
"logits/chosen": -14.
|
| 1540 |
-
"logits/rejected": -
|
| 1541 |
-
"logps/chosen": -
|
| 1542 |
-
"logps/rejected": -
|
| 1543 |
-
"loss": 0.
|
| 1544 |
-
"rewards/accuracies": 0.7749999761581421,
|
| 1545 |
-
"rewards/chosen": -0.569662868976593,
|
| 1546 |
-
"rewards/margins": 0.7953134179115295,
|
| 1547 |
-
"rewards/rejected": -1.3649762868881226,
|
| 1548 |
-
"step": 920
|
| 1549 |
-
},
|
| 1550 |
-
{
|
| 1551 |
-
"epoch": 0.9735671290238157,
|
| 1552 |
-
"grad_norm": 24.37374985565681,
|
| 1553 |
-
"learning_rate": 4.1769653133743035e-10,
|
| 1554 |
-
"logits/chosen": -14.06200885772705,
|
| 1555 |
-
"logits/rejected": -14.65931510925293,
|
| 1556 |
-
"logps/chosen": -348.43487548828125,
|
| 1557 |
-
"logps/rejected": -376.7773132324219,
|
| 1558 |
-
"loss": 0.4989,
|
| 1559 |
"rewards/accuracies": 0.7124999761581421,
|
| 1560 |
-
"rewards/chosen": -0.
|
| 1561 |
-
"rewards/margins": 0.
|
| 1562 |
-
"rewards/rejected": -1.
|
| 1563 |
-
"step":
|
| 1564 |
},
|
| 1565 |
{
|
| 1566 |
"epoch": 0.98403559277676,
|
| 1567 |
-
"grad_norm":
|
| 1568 |
-
"learning_rate":
|
| 1569 |
-
"logits/chosen": -14.
|
| 1570 |
-
"logits/rejected": -14.
|
| 1571 |
-
"logps/chosen": -
|
| 1572 |
-
"logps/rejected": -
|
| 1573 |
-
"loss": 0.
|
| 1574 |
"rewards/accuracies": 0.675000011920929,
|
| 1575 |
-
"rewards/chosen": -0.
|
| 1576 |
-
"rewards/margins": 0.
|
| 1577 |
-
"rewards/rejected": -1.
|
| 1578 |
-
"step":
|
| 1579 |
-
},
|
| 1580 |
-
{
|
| 1581 |
-
"epoch": 0.9945040565297043,
|
| 1582 |
-
"grad_norm": 22.704537546218425,
|
| 1583 |
-
"learning_rate": 1.671903968816224e-11,
|
| 1584 |
-
"logits/chosen": -13.458340644836426,
|
| 1585 |
-
"logits/rejected": -13.864636421203613,
|
| 1586 |
-
"logps/chosen": -367.2121276855469,
|
| 1587 |
-
"logps/rejected": -387.31488037109375,
|
| 1588 |
-
"loss": 0.5191,
|
| 1589 |
-
"rewards/accuracies": 0.75,
|
| 1590 |
-
"rewards/chosen": -0.6486170887947083,
|
| 1591 |
-
"rewards/margins": 0.6029497385025024,
|
| 1592 |
-
"rewards/rejected": -1.2515666484832764,
|
| 1593 |
-
"step": 950
|
| 1594 |
},
|
| 1595 |
{
|
| 1596 |
-
"epoch": 0.
|
| 1597 |
-
"step":
|
| 1598 |
"total_flos": 0.0,
|
| 1599 |
-
"train_loss": 0.
|
| 1600 |
-
"train_runtime":
|
| 1601 |
-
"train_samples_per_second":
|
| 1602 |
-
"train_steps_per_second": 0.
|
| 1603 |
}
|
| 1604 |
],
|
| 1605 |
-
"logging_steps":
|
| 1606 |
-
"max_steps":
|
| 1607 |
"num_input_tokens_seen": 0,
|
| 1608 |
"num_train_epochs": 1,
|
| 1609 |
"save_steps": 500,
|
|
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 0.9965977492802931,
|
| 5 |
+
"eval_steps": 50,
|
| 6 |
+
"global_step": 238,
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
+
"epoch": 0.004187385501177702,
|
| 13 |
+
"grad_norm": 0.41322922094683573,
|
| 14 |
+
"learning_rate": 1.0416666666666666e-08,
|
| 15 |
+
"logits/chosen": -8.027767181396484,
|
| 16 |
+
"logits/rejected": -8.085662841796875,
|
| 17 |
+
"logps/chosen": -315.0633544921875,
|
| 18 |
+
"logps/rejected": -339.5360412597656,
|
| 19 |
"loss": 0.6931,
|
| 20 |
"rewards/accuracies": 0.0,
|
| 21 |
"rewards/chosen": 0.0,
|
|
|
|
| 23 |
"rewards/rejected": 0.0,
|
| 24 |
"step": 1
|
| 25 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
{
|
| 27 |
"epoch": 0.02093692750588851,
|
| 28 |
+
"grad_norm": 0.3749268511204142,
|
| 29 |
+
"learning_rate": 5.208333333333333e-08,
|
| 30 |
+
"logits/chosen": -8.285957336425781,
|
| 31 |
+
"logits/rejected": -8.363251686096191,
|
| 32 |
+
"logps/chosen": -328.44573974609375,
|
| 33 |
+
"logps/rejected": -294.34893798828125,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
"loss": 0.6931,
|
| 35 |
+
"rewards/accuracies": 0.3984375,
|
| 36 |
+
"rewards/chosen": 5.56520426471252e-05,
|
| 37 |
+
"rewards/margins": -1.4014758562552743e-05,
|
| 38 |
+
"rewards/rejected": 6.966680666664615e-05,
|
| 39 |
+
"step": 5
|
| 40 |
},
|
| 41 |
{
|
| 42 |
"epoch": 0.04187385501177702,
|
| 43 |
+
"grad_norm": 0.3783566632626705,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
"learning_rate": 1.0416666666666667e-07,
|
| 45 |
+
"logits/chosen": -7.978540897369385,
|
| 46 |
+
"logits/rejected": -8.10208797454834,
|
| 47 |
+
"logps/chosen": -298.9517822265625,
|
| 48 |
+
"logps/rejected": -284.9400329589844,
|
| 49 |
+
"loss": 0.6931,
|
| 50 |
+
"rewards/accuracies": 0.512499988079071,
|
| 51 |
+
"rewards/chosen": 4.78029360237997e-05,
|
| 52 |
+
"rewards/margins": -1.2775728464475833e-05,
|
| 53 |
+
"rewards/rejected": 6.0578669945243746e-05,
|
| 54 |
+
"step": 10
|
| 55 |
},
|
| 56 |
{
|
| 57 |
"epoch": 0.06281078251766553,
|
| 58 |
+
"grad_norm": 0.365811115381345,
|
| 59 |
+
"learning_rate": 1.5624999999999999e-07,
|
| 60 |
+
"logits/chosen": -8.248689651489258,
|
| 61 |
+
"logits/rejected": -8.808431625366211,
|
| 62 |
+
"logps/chosen": -327.92962646484375,
|
| 63 |
+
"logps/rejected": -271.6505432128906,
|
| 64 |
+
"loss": 0.6931,
|
| 65 |
+
"rewards/accuracies": 0.4937500059604645,
|
| 66 |
+
"rewards/chosen": 0.00027168082306161523,
|
| 67 |
+
"rewards/margins": -3.7765556044178084e-05,
|
| 68 |
+
"rewards/rejected": 0.0003094463318120688,
|
| 69 |
+
"step": 15
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
},
|
| 71 |
{
|
| 72 |
"epoch": 0.08374771002355404,
|
| 73 |
+
"grad_norm": 0.38119222418916227,
|
| 74 |
+
"learning_rate": 2.0833333333333333e-07,
|
| 75 |
+
"logits/chosen": -8.283721923828125,
|
| 76 |
+
"logits/rejected": -8.563437461853027,
|
| 77 |
+
"logps/chosen": -270.73004150390625,
|
| 78 |
+
"logps/rejected": -264.69805908203125,
|
| 79 |
+
"loss": 0.6929,
|
| 80 |
+
"rewards/accuracies": 0.6312500238418579,
|
| 81 |
+
"rewards/chosen": 0.0011856909841299057,
|
| 82 |
+
"rewards/margins": 0.00045584110193885863,
|
| 83 |
+
"rewards/rejected": 0.0007298499112948775,
|
| 84 |
+
"step": 20
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 85 |
},
|
| 86 |
{
|
| 87 |
"epoch": 0.10468463752944256,
|
| 88 |
+
"grad_norm": 0.3717131307404656,
|
| 89 |
+
"learning_rate": 2.604166666666667e-07,
|
| 90 |
+
"logits/chosen": -8.32921028137207,
|
| 91 |
+
"logits/rejected": -8.343989372253418,
|
| 92 |
+
"logps/chosen": -285.73162841796875,
|
| 93 |
+
"logps/rejected": -277.33428955078125,
|
| 94 |
+
"loss": 0.6926,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 95 |
"rewards/accuracies": 0.737500011920929,
|
| 96 |
+
"rewards/chosen": 0.0029997099190950394,
|
| 97 |
+
"rewards/margins": 0.001172252232208848,
|
| 98 |
+
"rewards/rejected": 0.0018274573376402259,
|
| 99 |
+
"step": 25
|
| 100 |
},
|
| 101 |
{
|
| 102 |
+
"epoch": 0.12562156503533106,
|
| 103 |
+
"grad_norm": 0.41761867743254133,
|
| 104 |
+
"learning_rate": 3.1249999999999997e-07,
|
| 105 |
+
"logits/chosen": -8.842605590820312,
|
| 106 |
+
"logits/rejected": -8.617189407348633,
|
| 107 |
+
"logps/chosen": -271.97125244140625,
|
| 108 |
+
"logps/rejected": -261.83111572265625,
|
| 109 |
+
"loss": 0.6921,
|
| 110 |
+
"rewards/accuracies": 0.706250011920929,
|
| 111 |
+
"rewards/chosen": 0.005933535750955343,
|
| 112 |
+
"rewards/margins": 0.002372791524976492,
|
| 113 |
+
"rewards/rejected": 0.0035607446916401386,
|
| 114 |
+
"step": 30
|
| 115 |
},
|
| 116 |
{
|
| 117 |
"epoch": 0.14655849254121958,
|
| 118 |
+
"grad_norm": 0.3733417083873318,
|
| 119 |
+
"learning_rate": 3.645833333333333e-07,
|
| 120 |
+
"logits/chosen": -8.783945083618164,
|
| 121 |
+
"logits/rejected": -8.695141792297363,
|
| 122 |
+
"logps/chosen": -318.70452880859375,
|
| 123 |
+
"logps/rejected": -294.575927734375,
|
| 124 |
+
"loss": 0.6911,
|
| 125 |
+
"rewards/accuracies": 0.6625000238418579,
|
| 126 |
+
"rewards/chosen": 0.007537335157394409,
|
| 127 |
+
"rewards/margins": 0.0031830158550292253,
|
| 128 |
+
"rewards/rejected": 0.0043543195351958275,
|
| 129 |
+
"step": 35
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 130 |
},
|
| 131 |
{
|
| 132 |
"epoch": 0.16749542004710807,
|
| 133 |
+
"grad_norm": 0.4685368279017966,
|
| 134 |
+
"learning_rate": 4.1666666666666667e-07,
|
| 135 |
+
"logits/chosen": -8.875042915344238,
|
| 136 |
+
"logits/rejected": -9.194344520568848,
|
| 137 |
+
"logps/chosen": -313.56195068359375,
|
| 138 |
+
"logps/rejected": -266.4806823730469,
|
| 139 |
+
"loss": 0.6898,
|
| 140 |
+
"rewards/accuracies": 0.731249988079071,
|
| 141 |
+
"rewards/chosen": 0.005330189131200314,
|
| 142 |
+
"rewards/margins": 0.00792426336556673,
|
| 143 |
+
"rewards/rejected": -0.0025940726045519114,
|
| 144 |
+
"step": 40
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
},
|
| 146 |
{
|
| 147 |
"epoch": 0.1884323475529966,
|
| 148 |
+
"grad_norm": 0.5132211674073757,
|
| 149 |
+
"learning_rate": 4.6874999999999996e-07,
|
| 150 |
+
"logits/chosen": -10.095111846923828,
|
| 151 |
+
"logits/rejected": -10.34981632232666,
|
| 152 |
+
"logps/chosen": -288.3968200683594,
|
| 153 |
+
"logps/rejected": -279.54132080078125,
|
| 154 |
+
"loss": 0.6871,
|
| 155 |
+
"rewards/accuracies": 0.668749988079071,
|
| 156 |
+
"rewards/chosen": -0.006939026527106762,
|
| 157 |
+
"rewards/margins": 0.010522229596972466,
|
| 158 |
+
"rewards/rejected": -0.01746125891804695,
|
| 159 |
+
"step": 45
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 160 |
},
|
| 161 |
{
|
| 162 |
"epoch": 0.2093692750588851,
|
| 163 |
+
"grad_norm": 0.670419675961168,
|
| 164 |
+
"learning_rate": 4.998633143352315e-07,
|
| 165 |
+
"logits/chosen": -11.052873611450195,
|
| 166 |
+
"logits/rejected": -11.20246696472168,
|
| 167 |
+
"logps/chosen": -314.8858337402344,
|
| 168 |
+
"logps/rejected": -331.8305969238281,
|
| 169 |
+
"loss": 0.6835,
|
| 170 |
+
"rewards/accuracies": 0.8062499761581421,
|
| 171 |
+
"rewards/chosen": -0.007041523698717356,
|
| 172 |
+
"rewards/margins": 0.02735614776611328,
|
| 173 |
+
"rewards/rejected": -0.0343976691365242,
|
| 174 |
+
"step": 50
|
| 175 |
},
|
| 176 |
{
|
| 177 |
"epoch": 0.2093692750588851,
|
| 178 |
+
"eval_logits/chosen": -11.340270042419434,
|
| 179 |
+
"eval_logits/rejected": -11.638143539428711,
|
| 180 |
+
"eval_logps/chosen": -336.09466552734375,
|
| 181 |
+
"eval_logps/rejected": -328.0053405761719,
|
| 182 |
+
"eval_loss": 0.6815094351768494,
|
| 183 |
+
"eval_rewards/accuracies": 0.656000018119812,
|
| 184 |
+
"eval_rewards/chosen": -0.021773764863610268,
|
| 185 |
+
"eval_rewards/margins": 0.02182256057858467,
|
| 186 |
+
"eval_rewards/rejected": -0.04359632730484009,
|
| 187 |
+
"eval_runtime": 150.9435,
|
| 188 |
+
"eval_samples_per_second": 13.25,
|
| 189 |
+
"eval_steps_per_second": 0.828,
|
| 190 |
+
"step": 50
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 191 |
},
|
| 192 |
{
|
| 193 |
"epoch": 0.23030620256477363,
|
| 194 |
+
"grad_norm": 0.7268891633247492,
|
| 195 |
+
"learning_rate": 4.983273165884096e-07,
|
| 196 |
+
"logits/chosen": -11.379460334777832,
|
| 197 |
+
"logits/rejected": -12.12627124786377,
|
| 198 |
+
"logps/chosen": -355.1887512207031,
|
| 199 |
+
"logps/rejected": -343.373046875,
|
| 200 |
+
"loss": 0.6809,
|
| 201 |
+
"rewards/accuracies": 0.6875,
|
| 202 |
+
"rewards/chosen": -0.04076331481337547,
|
| 203 |
+
"rewards/margins": 0.02745387889444828,
|
| 204 |
+
"rewards/rejected": -0.0682171955704689,
|
| 205 |
+
"step": 55
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
},
|
| 207 |
{
|
| 208 |
"epoch": 0.2512431300706621,
|
| 209 |
+
"grad_norm": 1.1651426700711027,
|
| 210 |
+
"learning_rate": 4.950949914687024e-07,
|
| 211 |
+
"logits/chosen": -12.13168716430664,
|
| 212 |
+
"logits/rejected": -12.4894380569458,
|
| 213 |
+
"logps/chosen": -346.7480163574219,
|
| 214 |
+
"logps/rejected": -366.7667236328125,
|
| 215 |
+
"loss": 0.674,
|
| 216 |
+
"rewards/accuracies": 0.6499999761581421,
|
| 217 |
+
"rewards/chosen": -0.06704720854759216,
|
| 218 |
+
"rewards/margins": 0.03310731425881386,
|
| 219 |
+
"rewards/rejected": -0.10015451908111572,
|
| 220 |
+
"step": 60
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 221 |
},
|
| 222 |
{
|
| 223 |
"epoch": 0.2721800575765506,
|
| 224 |
+
"grad_norm": 1.3796192618478396,
|
| 225 |
+
"learning_rate": 4.901884190342121e-07,
|
| 226 |
+
"logits/chosen": -12.9432373046875,
|
| 227 |
+
"logits/rejected": -13.217244148254395,
|
| 228 |
+
"logps/chosen": -414.9393615722656,
|
| 229 |
+
"logps/rejected": -431.466796875,
|
| 230 |
+
"loss": 0.6675,
|
| 231 |
+
"rewards/accuracies": 0.731249988079071,
|
| 232 |
+
"rewards/chosen": -0.10360779613256454,
|
| 233 |
+
"rewards/margins": 0.05821988731622696,
|
| 234 |
+
"rewards/rejected": -0.16182765364646912,
|
| 235 |
+
"step": 65
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 236 |
},
|
| 237 |
{
|
| 238 |
"epoch": 0.29311698508243916,
|
| 239 |
+
"grad_norm": 1.3517064281641342,
|
| 240 |
+
"learning_rate": 4.836411161498652e-07,
|
| 241 |
+
"logits/chosen": -13.74645709991455,
|
| 242 |
+
"logits/rejected": -14.12658977508545,
|
| 243 |
+
"logps/chosen": -480.12762451171875,
|
| 244 |
+
"logps/rejected": -511.24481201171875,
|
| 245 |
+
"loss": 0.6615,
|
| 246 |
+
"rewards/accuracies": 0.6187499761581421,
|
| 247 |
+
"rewards/chosen": -0.1500854343175888,
|
| 248 |
+
"rewards/margins": 0.07933008670806885,
|
| 249 |
+
"rewards/rejected": -0.22941550612449646,
|
| 250 |
+
"step": 70
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 251 |
},
|
| 252 |
{
|
| 253 |
"epoch": 0.31405391258832765,
|
| 254 |
+
"grad_norm": 2.2837508840523095,
|
| 255 |
+
"learning_rate": 4.754978075332398e-07,
|
| 256 |
+
"logits/chosen": -15.218640327453613,
|
| 257 |
+
"logits/rejected": -15.693450927734375,
|
| 258 |
+
"logps/chosen": -540.690185546875,
|
| 259 |
+
"logps/rejected": -628.9052124023438,
|
| 260 |
+
"loss": 0.6596,
|
| 261 |
+
"rewards/accuracies": 0.6875,
|
| 262 |
+
"rewards/chosen": -0.23948459327220917,
|
| 263 |
+
"rewards/margins": 0.11436040699481964,
|
| 264 |
+
"rewards/rejected": -0.3538450300693512,
|
| 265 |
+
"step": 75
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 266 |
},
|
| 267 |
{
|
| 268 |
"epoch": 0.33499084009421615,
|
| 269 |
+
"grad_norm": 1.7535426886006145,
|
| 270 |
+
"learning_rate": 4.6581412023939346e-07,
|
| 271 |
+
"logits/chosen": -16.318897247314453,
|
| 272 |
+
"logits/rejected": -16.302444458007812,
|
| 273 |
+
"logps/chosen": -605.7731323242188,
|
| 274 |
+
"logps/rejected": -758.9752807617188,
|
| 275 |
+
"loss": 0.6472,
|
| 276 |
+
"rewards/accuracies": 0.6875,
|
| 277 |
+
"rewards/chosen": -0.3184322416782379,
|
| 278 |
+
"rewards/margins": 0.16458377242088318,
|
| 279 |
+
"rewards/rejected": -0.4830159544944763,
|
| 280 |
+
"step": 80
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 281 |
},
|
| 282 |
{
|
| 283 |
"epoch": 0.3559277676001047,
|
| 284 |
+
"grad_norm": 2.6529186768831545,
|
| 285 |
+
"learning_rate": 4.546562036716731e-07,
|
| 286 |
+
"logits/chosen": -16.481460571289062,
|
| 287 |
+
"logits/rejected": -16.788829803466797,
|
| 288 |
+
"logps/chosen": -638.745849609375,
|
| 289 |
+
"logps/rejected": -767.808837890625,
|
| 290 |
+
"loss": 0.6476,
|
| 291 |
+
"rewards/accuracies": 0.606249988079071,
|
| 292 |
+
"rewards/chosen": -0.3655020594596863,
|
| 293 |
+
"rewards/margins": 0.1500101387500763,
|
| 294 |
+
"rewards/rejected": -0.5155122876167297,
|
| 295 |
+
"step": 85
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 296 |
},
|
| 297 |
{
|
| 298 |
"epoch": 0.3768646951059932,
|
| 299 |
+
"grad_norm": 2.72153138496113,
|
| 300 |
+
"learning_rate": 4.4210027771421476e-07,
|
| 301 |
+
"logits/chosen": -17.021060943603516,
|
| 302 |
+
"logits/rejected": -17.08652114868164,
|
| 303 |
+
"logps/chosen": -752.3135375976562,
|
| 304 |
+
"logps/rejected": -926.8680419921875,
|
| 305 |
+
"loss": 0.6236,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 306 |
"rewards/accuracies": 0.6499999761581421,
|
| 307 |
+
"rewards/chosen": -0.4733971953392029,
|
| 308 |
+
"rewards/margins": 0.16153177618980408,
|
| 309 |
+
"rewards/rejected": -0.6349289417266846,
|
| 310 |
+
"step": 90
|
| 311 |
},
|
| 312 |
{
|
| 313 |
+
"epoch": 0.39780162261188173,
|
| 314 |
+
"grad_norm": 2.6592008129120215,
|
| 315 |
+
"learning_rate": 4.282321120728493e-07,
|
| 316 |
+
"logits/chosen": -16.358983993530273,
|
| 317 |
+
"logits/rejected": -16.262645721435547,
|
| 318 |
+
"logps/chosen": -745.8214111328125,
|
| 319 |
+
"logps/rejected": -967.6781005859375,
|
| 320 |
+
"loss": 0.635,
|
| 321 |
+
"rewards/accuracies": 0.706250011920929,
|
| 322 |
+
"rewards/chosen": -0.457266241312027,
|
| 323 |
+
"rewards/margins": 0.20888647437095642,
|
| 324 |
+
"rewards/rejected": -0.6661526560783386,
|
| 325 |
+
"step": 95
|
| 326 |
},
|
| 327 |
{
|
| 328 |
"epoch": 0.4187385501177702,
|
| 329 |
+
"grad_norm": 3.1808401845251506,
|
| 330 |
+
"learning_rate": 4.1314644038104213e-07,
|
| 331 |
+
"logits/chosen": -15.009060859680176,
|
| 332 |
+
"logits/rejected": -15.336482048034668,
|
| 333 |
+
"logps/chosen": -740.966064453125,
|
| 334 |
+
"logps/rejected": -862.05712890625,
|
| 335 |
+
"loss": 0.6243,
|
| 336 |
+
"rewards/accuracies": 0.59375,
|
| 337 |
+
"rewards/chosen": -0.45071443915367126,
|
| 338 |
+
"rewards/margins": 0.12646019458770752,
|
| 339 |
+
"rewards/rejected": -0.5771746635437012,
|
| 340 |
+
"step": 100
|
| 341 |
},
|
| 342 |
{
|
| 343 |
"epoch": 0.4187385501177702,
|
| 344 |
+
"eval_logits/chosen": -15.678736686706543,
|
| 345 |
+
"eval_logits/rejected": -15.50975227355957,
|
| 346 |
+
"eval_logps/chosen": -838.1255493164062,
|
| 347 |
+
"eval_logps/rejected": -1037.213623046875,
|
| 348 |
+
"eval_loss": 0.6229372024536133,
|
| 349 |
+
"eval_rewards/accuracies": 0.6600000262260437,
|
| 350 |
+
"eval_rewards/chosen": -0.5238046050071716,
|
| 351 |
+
"eval_rewards/margins": 0.229000061750412,
|
| 352 |
+
"eval_rewards/rejected": -0.752804696559906,
|
| 353 |
+
"eval_runtime": 150.6688,
|
| 354 |
+
"eval_samples_per_second": 13.274,
|
| 355 |
+
"eval_steps_per_second": 0.83,
|
| 356 |
+
"step": 100
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 357 |
},
|
| 358 |
{
|
| 359 |
"epoch": 0.4396754776236587,
|
| 360 |
+
"grad_norm": 3.6393194741175043,
|
| 361 |
+
"learning_rate": 3.9694631307311825e-07,
|
| 362 |
+
"logits/chosen": -15.765314102172852,
|
| 363 |
+
"logits/rejected": -15.19013500213623,
|
| 364 |
+
"logps/chosen": -911.4410400390625,
|
| 365 |
+
"logps/rejected": -1072.3721923828125,
|
| 366 |
+
"loss": 0.6298,
|
| 367 |
+
"rewards/accuracies": 0.6625000238418579,
|
| 368 |
+
"rewards/chosen": -0.5906545519828796,
|
| 369 |
+
"rewards/margins": 0.2006884515285492,
|
| 370 |
+
"rewards/rejected": -0.7913430333137512,
|
| 371 |
+
"step": 105
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 372 |
},
|
| 373 |
{
|
| 374 |
"epoch": 0.46061240512954726,
|
| 375 |
+
"grad_norm": 4.619726408628701,
|
| 376 |
+
"learning_rate": 3.797423934453038e-07,
|
| 377 |
+
"logits/chosen": -15.227154731750488,
|
| 378 |
+
"logits/rejected": -15.2573881149292,
|
| 379 |
+
"logps/chosen": -897.2864379882812,
|
| 380 |
+
"logps/rejected": -1104.17578125,
|
| 381 |
+
"loss": 0.6083,
|
| 382 |
+
"rewards/accuracies": 0.675000011920929,
|
| 383 |
+
"rewards/chosen": -0.6274054050445557,
|
| 384 |
+
"rewards/margins": 0.2075110673904419,
|
| 385 |
+
"rewards/rejected": -0.8349164724349976,
|
| 386 |
+
"step": 110
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 387 |
},
|
| 388 |
{
|
| 389 |
"epoch": 0.48154933263543576,
|
| 390 |
+
"grad_norm": 4.936193430576564,
|
| 391 |
+
"learning_rate": 3.6165220171320164e-07,
|
| 392 |
+
"logits/chosen": -15.75273609161377,
|
| 393 |
+
"logits/rejected": -15.582818984985352,
|
| 394 |
+
"logps/chosen": -940.6126098632812,
|
| 395 |
+
"logps/rejected": -1408.8499755859375,
|
| 396 |
+
"loss": 0.6105,
|
| 397 |
+
"rewards/accuracies": 0.768750011920929,
|
| 398 |
+
"rewards/chosen": -0.6616717576980591,
|
| 399 |
+
"rewards/margins": 0.451225221157074,
|
| 400 |
+
"rewards/rejected": -1.1128969192504883,
|
| 401 |
+
"step": 115
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 402 |
},
|
| 403 |
{
|
| 404 |
"epoch": 0.5024862601413242,
|
| 405 |
+
"grad_norm": 4.0220824325322155,
|
| 406 |
+
"learning_rate": 3.4279931222955516e-07,
|
| 407 |
+
"logits/chosen": -14.798166275024414,
|
| 408 |
+
"logits/rejected": -14.821731567382812,
|
| 409 |
+
"logps/chosen": -762.6677856445312,
|
| 410 |
+
"logps/rejected": -1019.94677734375,
|
| 411 |
+
"loss": 0.5999,
|
| 412 |
+
"rewards/accuracies": 0.7437499761581421,
|
| 413 |
+
"rewards/chosen": -0.48045119643211365,
|
| 414 |
+
"rewards/margins": 0.26533186435699463,
|
| 415 |
+
"rewards/rejected": -0.7457829713821411,
|
| 416 |
+
"step": 120
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 417 |
},
|
| 418 |
{
|
| 419 |
"epoch": 0.5234231876472127,
|
| 420 |
+
"grad_norm": 5.124454609764516,
|
| 421 |
+
"learning_rate": 3.233125093461162e-07,
|
| 422 |
+
"logits/chosen": -14.470125198364258,
|
| 423 |
+
"logits/rejected": -14.480206489562988,
|
| 424 |
+
"logps/chosen": -922.26708984375,
|
| 425 |
+
"logps/rejected": -1210.3658447265625,
|
| 426 |
+
"loss": 0.6,
|
| 427 |
+
"rewards/accuracies": 0.699999988079071,
|
| 428 |
+
"rewards/chosen": -0.5987042188644409,
|
| 429 |
+
"rewards/margins": 0.3204038441181183,
|
| 430 |
+
"rewards/rejected": -0.9191079139709473,
|
| 431 |
+
"step": 125
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 432 |
},
|
| 433 |
{
|
| 434 |
"epoch": 0.5443601151531012,
|
| 435 |
+
"grad_norm": 13.38213845011268,
|
| 436 |
+
"learning_rate": 3.033249076859367e-07,
|
| 437 |
+
"logits/chosen": -14.465197563171387,
|
| 438 |
+
"logits/rejected": -14.281991958618164,
|
| 439 |
+
"logps/chosen": -1182.00732421875,
|
| 440 |
+
"logps/rejected": -1630.92333984375,
|
| 441 |
+
"loss": 0.5923,
|
| 442 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 443 |
+
"rewards/chosen": -0.8785597085952759,
|
| 444 |
+
"rewards/margins": 0.4840659201145172,
|
| 445 |
+
"rewards/rejected": -1.3626257181167603,
|
| 446 |
+
"step": 130
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 447 |
},
|
| 448 |
{
|
| 449 |
"epoch": 0.5652970426589898,
|
| 450 |
+
"grad_norm": 4.505485837442961,
|
| 451 |
+
"learning_rate": 2.8297304283551724e-07,
|
| 452 |
+
"logits/chosen": -13.903594970703125,
|
| 453 |
+
"logits/rejected": -13.29607105255127,
|
| 454 |
+
"logps/chosen": -865.8912963867188,
|
| 455 |
+
"logps/rejected": -1264.981201171875,
|
| 456 |
+
"loss": 0.6025,
|
| 457 |
+
"rewards/accuracies": 0.6812499761581421,
|
| 458 |
+
"rewards/chosen": -0.5570029020309448,
|
| 459 |
+
"rewards/margins": 0.4205314517021179,
|
| 460 |
+
"rewards/rejected": -0.9775344729423523,
|
| 461 |
+
"step": 135
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 462 |
},
|
| 463 |
{
|
| 464 |
"epoch": 0.5862339701648783,
|
| 465 |
+
"grad_norm": 7.955445077360379,
|
| 466 |
+
"learning_rate": 2.6239593866830556e-07,
|
| 467 |
+
"logits/chosen": -14.410888671875,
|
| 468 |
+
"logits/rejected": -13.651044845581055,
|
| 469 |
+
"logps/chosen": -876.46044921875,
|
| 470 |
+
"logps/rejected": -1102.4383544921875,
|
| 471 |
+
"loss": 0.5961,
|
| 472 |
+
"rewards/accuracies": 0.6625000238418579,
|
| 473 |
+
"rewards/chosen": -0.5745535492897034,
|
| 474 |
+
"rewards/margins": 0.26768410205841064,
|
| 475 |
+
"rewards/rejected": -0.842237651348114,
|
| 476 |
+
"step": 140
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 477 |
},
|
| 478 |
{
|
| 479 |
"epoch": 0.6071708976707668,
|
| 480 |
+
"grad_norm": 6.562592794957003,
|
| 481 |
+
"learning_rate": 2.4173415767067293e-07,
|
| 482 |
+
"logits/chosen": -14.869283676147461,
|
| 483 |
+
"logits/rejected": -14.766156196594238,
|
| 484 |
+
"logps/chosen": -1157.8944091796875,
|
| 485 |
+
"logps/rejected": -1581.955810546875,
|
| 486 |
+
"loss": 0.5867,
|
| 487 |
+
"rewards/accuracies": 0.699999988079071,
|
| 488 |
+
"rewards/chosen": -0.866470992565155,
|
| 489 |
+
"rewards/margins": 0.4261693060398102,
|
| 490 |
+
"rewards/rejected": -1.2926403284072876,
|
| 491 |
+
"step": 145
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 492 |
},
|
| 493 |
{
|
| 494 |
"epoch": 0.6281078251766553,
|
| 495 |
+
"grad_norm": 5.969306865423692,
|
| 496 |
+
"learning_rate": 2.2112884075760347e-07,
|
| 497 |
+
"logits/chosen": -15.083928108215332,
|
| 498 |
+
"logits/rejected": -14.58642578125,
|
| 499 |
+
"logps/chosen": -949.1544189453125,
|
| 500 |
+
"logps/rejected": -1470.0855712890625,
|
| 501 |
+
"loss": 0.5625,
|
| 502 |
+
"rewards/accuracies": 0.78125,
|
| 503 |
+
"rewards/chosen": -0.628490686416626,
|
| 504 |
+
"rewards/margins": 0.5490113496780396,
|
| 505 |
+
"rewards/rejected": -1.177502155303955,
|
| 506 |
+
"step": 150
|
| 507 |
},
|
| 508 |
{
|
| 509 |
"epoch": 0.6281078251766553,
|
| 510 |
+
"eval_logits/chosen": -15.179731369018555,
|
| 511 |
+
"eval_logits/rejected": -14.774624824523926,
|
| 512 |
+
"eval_logps/chosen": -1032.8834228515625,
|
| 513 |
+
"eval_logps/rejected": -1471.7362060546875,
|
| 514 |
+
"eval_loss": 0.5793285965919495,
|
| 515 |
+
"eval_rewards/accuracies": 0.6880000233650208,
|
| 516 |
+
"eval_rewards/chosen": -0.7185624241828918,
|
| 517 |
+
"eval_rewards/margins": 0.4687648415565491,
|
| 518 |
+
"eval_rewards/rejected": -1.187327265739441,
|
| 519 |
+
"eval_runtime": 150.2684,
|
| 520 |
+
"eval_samples_per_second": 13.31,
|
| 521 |
"eval_steps_per_second": 0.832,
|
| 522 |
+
"step": 150
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 523 |
},
|
| 524 |
{
|
| 525 |
"epoch": 0.6490447526825438,
|
| 526 |
+
"grad_norm": 11.88961718187316,
|
| 527 |
+
"learning_rate": 2.0072074313712993e-07,
|
| 528 |
+
"logits/chosen": -15.340092658996582,
|
| 529 |
+
"logits/rejected": -15.312405586242676,
|
| 530 |
+
"logps/chosen": -1098.570068359375,
|
| 531 |
+
"logps/rejected": -1679.193603515625,
|
| 532 |
+
"loss": 0.5753,
|
| 533 |
+
"rewards/accuracies": 0.768750011920929,
|
| 534 |
+
"rewards/chosen": -0.7703229188919067,
|
| 535 |
+
"rewards/margins": 0.6483911275863647,
|
| 536 |
+
"rewards/rejected": -1.418714165687561,
|
| 537 |
+
"step": 155
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 538 |
},
|
| 539 |
{
|
| 540 |
"epoch": 0.6699816801884323,
|
| 541 |
+
"grad_norm": 8.601486869561926,
|
| 542 |
+
"learning_rate": 1.806492728095389e-07,
|
| 543 |
+
"logits/chosen": -15.2809476852417,
|
| 544 |
+
"logits/rejected": -14.993128776550293,
|
| 545 |
+
"logps/chosen": -1034.3973388671875,
|
| 546 |
+
"logps/rejected": -1639.748779296875,
|
| 547 |
+
"loss": 0.5623,
|
| 548 |
+
"rewards/accuracies": 0.7437499761581421,
|
| 549 |
+
"rewards/chosen": -0.7416967153549194,
|
| 550 |
+
"rewards/margins": 0.634353518486023,
|
| 551 |
+
"rewards/rejected": -1.376050353050232,
|
| 552 |
+
"step": 160
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 553 |
},
|
| 554 |
{
|
| 555 |
"epoch": 0.6909186076943209,
|
| 556 |
+
"grad_norm": 9.897007035455301,
|
| 557 |
+
"learning_rate": 1.6105153826937085e-07,
|
| 558 |
+
"logits/chosen": -14.975976943969727,
|
| 559 |
+
"logits/rejected": -14.773465156555176,
|
| 560 |
+
"logps/chosen": -1030.3551025390625,
|
| 561 |
+
"logps/rejected": -1616.5765380859375,
|
| 562 |
+
"loss": 0.5668,
|
| 563 |
+
"rewards/accuracies": 0.731249988079071,
|
| 564 |
+
"rewards/chosen": -0.7377808690071106,
|
| 565 |
+
"rewards/margins": 0.6023792028427124,
|
| 566 |
+
"rewards/rejected": -1.3401601314544678,
|
| 567 |
+
"step": 165
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 568 |
},
|
| 569 |
{
|
| 570 |
"epoch": 0.7118555352002094,
|
| 571 |
+
"grad_norm": 10.944418149122821,
|
| 572 |
+
"learning_rate": 1.420614119153768e-07,
|
| 573 |
+
"logits/chosen": -15.038368225097656,
|
| 574 |
+
"logits/rejected": -14.671664237976074,
|
| 575 |
+
"logps/chosen": -1127.2037353515625,
|
| 576 |
+
"logps/rejected": -1648.9619140625,
|
| 577 |
+
"loss": 0.5612,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 578 |
"rewards/accuracies": 0.699999988079071,
|
| 579 |
+
"rewards/chosen": -0.8304765820503235,
|
| 580 |
+
"rewards/margins": 0.5383009314537048,
|
| 581 |
+
"rewards/rejected": -1.3687775135040283,
|
| 582 |
+
"step": 170
|
| 583 |
},
|
| 584 |
{
|
| 585 |
"epoch": 0.7327924627060979,
|
| 586 |
+
"grad_norm": 7.272876700396834,
|
| 587 |
+
"learning_rate": 1.2380861556628914e-07,
|
| 588 |
+
"logits/chosen": -14.817059516906738,
|
| 589 |
+
"logits/rejected": -14.438855171203613,
|
| 590 |
+
"logps/chosen": -1002.3743896484375,
|
| 591 |
+
"logps/rejected": -1483.3427734375,
|
| 592 |
+
"loss": 0.545,
|
| 593 |
+
"rewards/accuracies": 0.7124999761581421,
|
| 594 |
+
"rewards/chosen": -0.7164356112480164,
|
| 595 |
+
"rewards/margins": 0.5084205865859985,
|
| 596 |
+
"rewards/rejected": -1.2248561382293701,
|
| 597 |
+
"step": 175
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 598 |
},
|
| 599 |
{
|
| 600 |
"epoch": 0.7537293902119864,
|
| 601 |
+
"grad_norm": 7.940966891497204,
|
| 602 |
+
"learning_rate": 1.064178343292641e-07,
|
| 603 |
+
"logits/chosen": -14.983675956726074,
|
| 604 |
+
"logits/rejected": -14.386784553527832,
|
| 605 |
+
"logps/chosen": -991.7278442382812,
|
| 606 |
+
"logps/rejected": -1572.2987060546875,
|
| 607 |
+
"loss": 0.5762,
|
| 608 |
+
"rewards/accuracies": 0.800000011920929,
|
| 609 |
+
"rewards/chosen": -0.6495895981788635,
|
| 610 |
+
"rewards/margins": 0.6349653601646423,
|
| 611 |
+
"rewards/rejected": -1.2845550775527954,
|
| 612 |
+
"step": 180
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 613 |
},
|
| 614 |
{
|
| 615 |
"epoch": 0.7746663177178749,
|
| 616 |
+
"grad_norm": 8.710562600121326,
|
| 617 |
+
"learning_rate": 9.000786487417084e-08,
|
| 618 |
+
"logits/chosen": -14.231167793273926,
|
| 619 |
+
"logits/rejected": -13.895710945129395,
|
| 620 |
+
"logps/chosen": -935.4019775390625,
|
| 621 |
+
"logps/rejected": -1424.656005859375,
|
| 622 |
+
"loss": 0.5656,
|
| 623 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 624 |
+
"rewards/chosen": -0.6355662941932678,
|
| 625 |
+
"rewards/margins": 0.5334969758987427,
|
| 626 |
+
"rewards/rejected": -1.1690632104873657,
|
| 627 |
+
"step": 185
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 628 |
},
|
| 629 |
{
|
| 630 |
"epoch": 0.7956032452237635,
|
| 631 |
+
"grad_norm": 6.761392853792158,
|
| 632 |
+
"learning_rate": 7.469080393187785e-08,
|
| 633 |
+
"logits/chosen": -14.6929292678833,
|
| 634 |
+
"logits/rejected": -13.966386795043945,
|
| 635 |
+
"logps/chosen": -1072.4215087890625,
|
| 636 |
+
"logps/rejected": -1634.0445556640625,
|
| 637 |
+
"loss": 0.568,
|
| 638 |
+
"rewards/accuracies": 0.71875,
|
| 639 |
+
"rewards/chosen": -0.7651950120925903,
|
| 640 |
+
"rewards/margins": 0.58966064453125,
|
| 641 |
+
"rewards/rejected": -1.3548556566238403,
|
| 642 |
+
"step": 190
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 643 |
},
|
| 644 |
{
|
| 645 |
"epoch": 0.816540172729652,
|
| 646 |
+
"grad_norm": 9.876582631533086,
|
| 647 |
+
"learning_rate": 6.057128255991637e-08,
|
| 648 |
+
"logits/chosen": -14.620927810668945,
|
| 649 |
+
"logits/rejected": -14.377177238464355,
|
| 650 |
+
"logps/chosen": -1004.9943237304688,
|
| 651 |
+
"logps/rejected": -1455.802001953125,
|
| 652 |
+
"loss": 0.5626,
|
| 653 |
+
"rewards/accuracies": 0.6875,
|
| 654 |
+
"rewards/chosen": -0.6936627626419067,
|
| 655 |
+
"rewards/margins": 0.4617386758327484,
|
| 656 |
+
"rewards/rejected": -1.1554014682769775,
|
| 657 |
+
"step": 195
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 658 |
},
|
| 659 |
{
|
| 660 |
"epoch": 0.8374771002355405,
|
| 661 |
+
"grad_norm": 8.039290948138406,
|
| 662 |
+
"learning_rate": 4.774575140626316e-08,
|
| 663 |
+
"logits/chosen": -15.10753345489502,
|
| 664 |
+
"logits/rejected": -14.191171646118164,
|
| 665 |
+
"logps/chosen": -837.1350708007812,
|
| 666 |
+
"logps/rejected": -1408.885498046875,
|
| 667 |
+
"loss": 0.5699,
|
| 668 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 669 |
+
"rewards/chosen": -0.5800159573554993,
|
| 670 |
+
"rewards/margins": 0.580711841583252,
|
| 671 |
+
"rewards/rejected": -1.1607277393341064,
|
| 672 |
+
"step": 200
|
| 673 |
},
|
| 674 |
{
|
| 675 |
"epoch": 0.8374771002355405,
|
| 676 |
+
"eval_logits/chosen": -14.668434143066406,
|
| 677 |
+
"eval_logits/rejected": -14.186065673828125,
|
| 678 |
+
"eval_logps/chosen": -958.5824584960938,
|
| 679 |
+
"eval_logps/rejected": -1434.33349609375,
|
| 680 |
+
"eval_loss": 0.5646860599517822,
|
| 681 |
+
"eval_rewards/accuracies": 0.6919999718666077,
|
| 682 |
+
"eval_rewards/chosen": -0.6442615985870361,
|
| 683 |
+
"eval_rewards/margins": 0.5056628584861755,
|
| 684 |
+
"eval_rewards/rejected": -1.149924397468567,
|
| 685 |
+
"eval_runtime": 149.7174,
|
| 686 |
+
"eval_samples_per_second": 13.358,
|
| 687 |
+
"eval_steps_per_second": 0.835,
|
| 688 |
+
"step": 200
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 689 |
},
|
| 690 |
{
|
| 691 |
"epoch": 0.8584140277414289,
|
| 692 |
+
"grad_norm": 7.7753080838732265,
|
| 693 |
+
"learning_rate": 3.6301821853615216e-08,
|
| 694 |
+
"logits/chosen": -14.531840324401855,
|
| 695 |
+
"logits/rejected": -13.954656600952148,
|
| 696 |
+
"logps/chosen": -979.337890625,
|
| 697 |
+
"logps/rejected": -1347.329833984375,
|
| 698 |
+
"loss": 0.5735,
|
| 699 |
+
"rewards/accuracies": 0.6937500238418579,
|
| 700 |
+
"rewards/chosen": -0.6474758982658386,
|
| 701 |
+
"rewards/margins": 0.4113900065422058,
|
| 702 |
+
"rewards/rejected": -1.0588659048080444,
|
| 703 |
+
"step": 205
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 704 |
},
|
| 705 |
{
|
| 706 |
"epoch": 0.8793509552473174,
|
| 707 |
+
"grad_norm": 8.900926616124899,
|
| 708 |
+
"learning_rate": 2.631766754480913e-08,
|
| 709 |
+
"logits/chosen": -15.085081100463867,
|
| 710 |
+
"logits/rejected": -14.495912551879883,
|
| 711 |
+
"logps/chosen": -993.9776611328125,
|
| 712 |
+
"logps/rejected": -1629.8128662109375,
|
| 713 |
+
"loss": 0.554,
|
| 714 |
+
"rewards/accuracies": 0.731249988079071,
|
| 715 |
+
"rewards/chosen": -0.7016497850418091,
|
| 716 |
+
"rewards/margins": 0.627029299736023,
|
| 717 |
+
"rewards/rejected": -1.328679084777832,
|
| 718 |
+
"step": 210
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 719 |
},
|
| 720 |
{
|
| 721 |
"epoch": 0.9002878827532059,
|
| 722 |
+
"grad_norm": 7.108989309187613,
|
| 723 |
+
"learning_rate": 1.786149037757326e-08,
|
| 724 |
+
"logits/chosen": -14.463391304016113,
|
| 725 |
+
"logits/rejected": -14.160985946655273,
|
| 726 |
+
"logps/chosen": -1008.32080078125,
|
| 727 |
+
"logps/rejected": -1559.0267333984375,
|
| 728 |
+
"loss": 0.5445,
|
| 729 |
+
"rewards/accuracies": 0.6812499761581421,
|
| 730 |
+
"rewards/chosen": -0.678644061088562,
|
| 731 |
+
"rewards/margins": 0.6104253530502319,
|
| 732 |
+
"rewards/rejected": -1.289069414138794,
|
| 733 |
+
"step": 215
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 734 |
},
|
| 735 |
{
|
| 736 |
"epoch": 0.9212248102590945,
|
| 737 |
+
"grad_norm": 9.175335683582457,
|
| 738 |
+
"learning_rate": 1.0991054616410588e-08,
|
| 739 |
+
"logits/chosen": -14.525983810424805,
|
| 740 |
+
"logits/rejected": -14.118609428405762,
|
| 741 |
+
"logps/chosen": -938.26904296875,
|
| 742 |
+
"logps/rejected": -1491.09814453125,
|
| 743 |
+
"loss": 0.5515,
|
| 744 |
+
"rewards/accuracies": 0.706250011920929,
|
| 745 |
+
"rewards/chosen": -0.6404693126678467,
|
| 746 |
+
"rewards/margins": 0.5899735689163208,
|
| 747 |
+
"rewards/rejected": -1.230442762374878,
|
| 748 |
+
"step": 220
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 749 |
},
|
| 750 |
{
|
| 751 |
"epoch": 0.942161737764983,
|
| 752 |
+
"grad_norm": 9.106722734484554,
|
| 753 |
+
"learning_rate": 5.753292304100182e-09,
|
| 754 |
+
"logits/chosen": -14.361761093139648,
|
| 755 |
+
"logits/rejected": -14.124635696411133,
|
| 756 |
+
"logps/chosen": -928.7554931640625,
|
| 757 |
+
"logps/rejected": -1761.9857177734375,
|
| 758 |
+
"loss": 0.5522,
|
| 759 |
+
"rewards/accuracies": 0.824999988079071,
|
| 760 |
+
"rewards/chosen": -0.632331371307373,
|
| 761 |
+
"rewards/margins": 0.840011477470398,
|
| 762 |
+
"rewards/rejected": -1.472342848777771,
|
| 763 |
+
"step": 225
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 764 |
},
|
| 765 |
{
|
| 766 |
"epoch": 0.9630986652708715,
|
| 767 |
+
"grad_norm": 9.134170824690674,
|
| 768 |
+
"learning_rate": 2.1839826682562014e-09,
|
| 769 |
+
"logits/chosen": -14.802205085754395,
|
| 770 |
+
"logits/rejected": -13.968228340148926,
|
| 771 |
+
"logps/chosen": -993.044921875,
|
| 772 |
+
"logps/rejected": -1571.5948486328125,
|
| 773 |
+
"loss": 0.5697,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 774 |
"rewards/accuracies": 0.7124999761581421,
|
| 775 |
+
"rewards/chosen": -0.6945916414260864,
|
| 776 |
+
"rewards/margins": 0.6186043620109558,
|
| 777 |
+
"rewards/rejected": -1.3131959438323975,
|
| 778 |
+
"step": 230
|
| 779 |
},
|
| 780 |
{
|
| 781 |
"epoch": 0.98403559277676,
|
| 782 |
+
"grad_norm": 7.825653316640832,
|
| 783 |
+
"learning_rate": 3.075077129238157e-10,
|
| 784 |
+
"logits/chosen": -14.510324478149414,
|
| 785 |
+
"logits/rejected": -14.37182903289795,
|
| 786 |
+
"logps/chosen": -1027.906982421875,
|
| 787 |
+
"logps/rejected": -1584.23828125,
|
| 788 |
+
"loss": 0.5535,
|
| 789 |
"rewards/accuracies": 0.675000011920929,
|
| 790 |
+
"rewards/chosen": -0.7289544343948364,
|
| 791 |
+
"rewards/margins": 0.562315821647644,
|
| 792 |
+
"rewards/rejected": -1.29127037525177,
|
| 793 |
+
"step": 235
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 794 |
},
|
| 795 |
{
|
| 796 |
+
"epoch": 0.9965977492802931,
|
| 797 |
+
"step": 238,
|
| 798 |
"total_flos": 0.0,
|
| 799 |
+
"train_loss": 0.6157421479706003,
|
| 800 |
+
"train_runtime": 11733.4361,
|
| 801 |
+
"train_samples_per_second": 5.21,
|
| 802 |
+
"train_steps_per_second": 0.02
|
| 803 |
}
|
| 804 |
],
|
| 805 |
+
"logging_steps": 5,
|
| 806 |
+
"max_steps": 238,
|
| 807 |
"num_input_tokens_seen": 0,
|
| 808 |
"num_train_epochs": 1,
|
| 809 |
"save_steps": 500,
|
training_args.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 7544
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:640f2657d58cabae6bc26036066fcbcd91efca82106edb6249b200527a89c437
|
| 3 |
size 7544
|