Model save
Browse files- README.md +80 -0
- adapter_model.safetensors +1 -1
- all_results.json +9 -0
- train_results.json +9 -0
- trainer_state.json +977 -0
README.md
ADDED
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
library_name: peft
|
4 |
+
tags:
|
5 |
+
- trl
|
6 |
+
- dpo
|
7 |
+
- generated_from_trainer
|
8 |
+
base_model: microsoft/Phi-3-mini-4k-instruct
|
9 |
+
model-index:
|
10 |
+
- name: phi3-offline-dpo-lora-noise-0.0-5e-7-thre-1.5-42
|
11 |
+
results: []
|
12 |
+
---
|
13 |
+
|
14 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
15 |
+
should probably proofread and complete it, then remove this comment. -->
|
16 |
+
|
17 |
+
[<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/causal/huggingface/runs/3pqwimtf)
|
18 |
+
# phi3-offline-dpo-lora-noise-0.0-5e-7-thre-1.5-42
|
19 |
+
|
20 |
+
This model is a fine-tuned version of [microsoft/Phi-3-mini-4k-instruct](https://huggingface.co/microsoft/Phi-3-mini-4k-instruct) on the None dataset.
|
21 |
+
It achieves the following results on the evaluation set:
|
22 |
+
- Loss: 0.6131
|
23 |
+
- Rewards/chosen: -0.4150
|
24 |
+
- Rewards/rejected: -0.5892
|
25 |
+
- Rewards/accuracies: 0.7540
|
26 |
+
- Rewards/margins: 0.1743
|
27 |
+
- Logps/rejected: -459.2376
|
28 |
+
- Logps/chosen: -447.7001
|
29 |
+
- Logits/rejected: 12.5314
|
30 |
+
- Logits/chosen: 13.8427
|
31 |
+
|
32 |
+
## Model description
|
33 |
+
|
34 |
+
More information needed
|
35 |
+
|
36 |
+
## Intended uses & limitations
|
37 |
+
|
38 |
+
More information needed
|
39 |
+
|
40 |
+
## Training and evaluation data
|
41 |
+
|
42 |
+
More information needed
|
43 |
+
|
44 |
+
## Training procedure
|
45 |
+
|
46 |
+
### Training hyperparameters
|
47 |
+
|
48 |
+
The following hyperparameters were used during training:
|
49 |
+
- learning_rate: 5e-07
|
50 |
+
- train_batch_size: 4
|
51 |
+
- eval_batch_size: 4
|
52 |
+
- seed: 42
|
53 |
+
- distributed_type: multi-GPU
|
54 |
+
- num_devices: 4
|
55 |
+
- gradient_accumulation_steps: 4
|
56 |
+
- total_train_batch_size: 64
|
57 |
+
- total_eval_batch_size: 16
|
58 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
59 |
+
- lr_scheduler_type: cosine
|
60 |
+
- lr_scheduler_warmup_ratio: 0.1
|
61 |
+
- num_epochs: 1
|
62 |
+
|
63 |
+
### Training results
|
64 |
+
|
65 |
+
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
66 |
+
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
67 |
+
| 0.6885 | 0.1778 | 100 | 0.6884 | -0.0158 | -0.0244 | 0.6190 | 0.0086 | -402.7496 | -407.7805 | 12.8305 | 14.2621 |
|
68 |
+
| 0.6712 | 0.3556 | 200 | 0.6680 | -0.0971 | -0.1464 | 0.7937 | 0.0493 | -414.9504 | -415.9148 | 12.6482 | 14.0845 |
|
69 |
+
| 0.6339 | 0.5333 | 300 | 0.6389 | -0.2593 | -0.3712 | 0.7540 | 0.1119 | -437.4307 | -432.1300 | 12.8556 | 14.1744 |
|
70 |
+
| 0.6203 | 0.7111 | 400 | 0.6203 | -0.3738 | -0.5313 | 0.7540 | 0.1575 | -453.4457 | -443.5887 | 12.6256 | 13.9444 |
|
71 |
+
| 0.6102 | 0.8889 | 500 | 0.6131 | -0.4150 | -0.5892 | 0.7540 | 0.1743 | -459.2376 | -447.7001 | 12.5314 | 13.8427 |
|
72 |
+
|
73 |
+
|
74 |
+
### Framework versions
|
75 |
+
|
76 |
+
- PEFT 0.7.1
|
77 |
+
- Transformers 4.42.3
|
78 |
+
- Pytorch 2.3.0+cu121
|
79 |
+
- Datasets 2.14.6
|
80 |
+
- Tokenizers 0.19.1
|
adapter_model.safetensors
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 71320344
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:49e16b4d3c7e80563fafc534a0cf5709c0ad40d41cd3bb70e7f03e6fe8b09a76
|
3 |
size 71320344
|
all_results.json
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 0.9991111111111111,
|
3 |
+
"total_flos": 0.0,
|
4 |
+
"train_loss": 0.6475430355801701,
|
5 |
+
"train_runtime": 7558.4306,
|
6 |
+
"train_samples": 36000,
|
7 |
+
"train_samples_per_second": 4.763,
|
8 |
+
"train_steps_per_second": 0.074
|
9 |
+
}
|
train_results.json
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"epoch": 0.9991111111111111,
|
3 |
+
"total_flos": 0.0,
|
4 |
+
"train_loss": 0.6475430355801701,
|
5 |
+
"train_runtime": 7558.4306,
|
6 |
+
"train_samples": 36000,
|
7 |
+
"train_samples_per_second": 4.763,
|
8 |
+
"train_steps_per_second": 0.074
|
9 |
+
}
|
trainer_state.json
ADDED
@@ -0,0 +1,977 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"best_metric": null,
|
3 |
+
"best_model_checkpoint": null,
|
4 |
+
"epoch": 0.9991111111111111,
|
5 |
+
"eval_steps": 100,
|
6 |
+
"global_step": 562,
|
7 |
+
"is_hyper_param_search": false,
|
8 |
+
"is_local_process_zero": true,
|
9 |
+
"is_world_process_zero": true,
|
10 |
+
"log_history": [
|
11 |
+
{
|
12 |
+
"epoch": 0.0017777777777777779,
|
13 |
+
"grad_norm": 0.34293448639816615,
|
14 |
+
"learning_rate": 8.771929824561403e-09,
|
15 |
+
"logits/chosen": 13.316938400268555,
|
16 |
+
"logits/rejected": 13.604828834533691,
|
17 |
+
"logps/chosen": -366.0712890625,
|
18 |
+
"logps/rejected": -384.89288330078125,
|
19 |
+
"loss": 0.6931,
|
20 |
+
"rewards/accuracies": 0.0,
|
21 |
+
"rewards/chosen": 0.0,
|
22 |
+
"rewards/margins": 0.0,
|
23 |
+
"rewards/rejected": 0.0,
|
24 |
+
"step": 1
|
25 |
+
},
|
26 |
+
{
|
27 |
+
"epoch": 0.017777777777777778,
|
28 |
+
"grad_norm": 0.38494137076048135,
|
29 |
+
"learning_rate": 8.771929824561403e-08,
|
30 |
+
"logits/chosen": 14.320709228515625,
|
31 |
+
"logits/rejected": 12.689621925354004,
|
32 |
+
"logps/chosen": -420.3371887207031,
|
33 |
+
"logps/rejected": -431.1959228515625,
|
34 |
+
"loss": 0.6929,
|
35 |
+
"rewards/accuracies": 0.3194444477558136,
|
36 |
+
"rewards/chosen": -0.0003511953691486269,
|
37 |
+
"rewards/margins": 0.00028524029767140746,
|
38 |
+
"rewards/rejected": -0.0006364354630932212,
|
39 |
+
"step": 10
|
40 |
+
},
|
41 |
+
{
|
42 |
+
"epoch": 0.035555555555555556,
|
43 |
+
"grad_norm": 0.37882743094316385,
|
44 |
+
"learning_rate": 1.7543859649122805e-07,
|
45 |
+
"logits/chosen": 13.643712997436523,
|
46 |
+
"logits/rejected": 11.55290412902832,
|
47 |
+
"logps/chosen": -413.57061767578125,
|
48 |
+
"logps/rejected": -409.67694091796875,
|
49 |
+
"loss": 0.6937,
|
50 |
+
"rewards/accuracies": 0.4437499940395355,
|
51 |
+
"rewards/chosen": -0.0014681232860311866,
|
52 |
+
"rewards/margins": -0.0013028818648308516,
|
53 |
+
"rewards/rejected": -0.00016524126112926751,
|
54 |
+
"step": 20
|
55 |
+
},
|
56 |
+
{
|
57 |
+
"epoch": 0.05333333333333334,
|
58 |
+
"grad_norm": 0.3955514954356053,
|
59 |
+
"learning_rate": 2.631578947368421e-07,
|
60 |
+
"logits/chosen": 13.529040336608887,
|
61 |
+
"logits/rejected": 12.290833473205566,
|
62 |
+
"logps/chosen": -371.78033447265625,
|
63 |
+
"logps/rejected": -382.180419921875,
|
64 |
+
"loss": 0.6925,
|
65 |
+
"rewards/accuracies": 0.612500011920929,
|
66 |
+
"rewards/chosen": -0.00034525356022641063,
|
67 |
+
"rewards/margins": 0.004034269601106644,
|
68 |
+
"rewards/rejected": -0.0043795229867100716,
|
69 |
+
"step": 30
|
70 |
+
},
|
71 |
+
{
|
72 |
+
"epoch": 0.07111111111111111,
|
73 |
+
"grad_norm": 0.3113111173035772,
|
74 |
+
"learning_rate": 3.508771929824561e-07,
|
75 |
+
"logits/chosen": 13.795173645019531,
|
76 |
+
"logits/rejected": 12.55676555633545,
|
77 |
+
"logps/chosen": -389.0183410644531,
|
78 |
+
"logps/rejected": -415.35247802734375,
|
79 |
+
"loss": 0.6938,
|
80 |
+
"rewards/accuracies": 0.5,
|
81 |
+
"rewards/chosen": -0.003066176315769553,
|
82 |
+
"rewards/margins": -0.0013317791745066643,
|
83 |
+
"rewards/rejected": -0.0017343973740935326,
|
84 |
+
"step": 40
|
85 |
+
},
|
86 |
+
{
|
87 |
+
"epoch": 0.08888888888888889,
|
88 |
+
"grad_norm": 0.32606516023193766,
|
89 |
+
"learning_rate": 4.3859649122807013e-07,
|
90 |
+
"logits/chosen": 14.106832504272461,
|
91 |
+
"logits/rejected": 12.1203031539917,
|
92 |
+
"logps/chosen": -422.3417053222656,
|
93 |
+
"logps/rejected": -397.51385498046875,
|
94 |
+
"loss": 0.6937,
|
95 |
+
"rewards/accuracies": 0.4625000059604645,
|
96 |
+
"rewards/chosen": -0.00344509887509048,
|
97 |
+
"rewards/margins": -0.0021442542783915997,
|
98 |
+
"rewards/rejected": -0.001300844713114202,
|
99 |
+
"step": 50
|
100 |
+
},
|
101 |
+
{
|
102 |
+
"epoch": 0.10666666666666667,
|
103 |
+
"grad_norm": 0.33670626880499194,
|
104 |
+
"learning_rate": 4.999564631597801e-07,
|
105 |
+
"logits/chosen": 13.741307258605957,
|
106 |
+
"logits/rejected": 11.687819480895996,
|
107 |
+
"logps/chosen": -413.763427734375,
|
108 |
+
"logps/rejected": -397.7774353027344,
|
109 |
+
"loss": 0.6916,
|
110 |
+
"rewards/accuracies": 0.550000011920929,
|
111 |
+
"rewards/chosen": -0.0033744447864592075,
|
112 |
+
"rewards/margins": 0.003910833969712257,
|
113 |
+
"rewards/rejected": -0.007285278290510178,
|
114 |
+
"step": 60
|
115 |
+
},
|
116 |
+
{
|
117 |
+
"epoch": 0.12444444444444444,
|
118 |
+
"grad_norm": 0.34616943203548983,
|
119 |
+
"learning_rate": 4.991828966534002e-07,
|
120 |
+
"logits/chosen": 14.954236030578613,
|
121 |
+
"logits/rejected": 12.991679191589355,
|
122 |
+
"logps/chosen": -437.3307189941406,
|
123 |
+
"logps/rejected": -431.5257263183594,
|
124 |
+
"loss": 0.6914,
|
125 |
+
"rewards/accuracies": 0.550000011920929,
|
126 |
+
"rewards/chosen": -0.0038897425401955843,
|
127 |
+
"rewards/margins": 0.0035481129307299852,
|
128 |
+
"rewards/rejected": -0.007437856402248144,
|
129 |
+
"step": 70
|
130 |
+
},
|
131 |
+
{
|
132 |
+
"epoch": 0.14222222222222222,
|
133 |
+
"grad_norm": 0.4091997164465182,
|
134 |
+
"learning_rate": 4.974452899279291e-07,
|
135 |
+
"logits/chosen": 12.901407241821289,
|
136 |
+
"logits/rejected": 11.605668067932129,
|
137 |
+
"logps/chosen": -391.2544250488281,
|
138 |
+
"logps/rejected": -403.60089111328125,
|
139 |
+
"loss": 0.692,
|
140 |
+
"rewards/accuracies": 0.5375000238418579,
|
141 |
+
"rewards/chosen": -0.009723111987113953,
|
142 |
+
"rewards/margins": 0.000866956717800349,
|
143 |
+
"rewards/rejected": -0.010590068995952606,
|
144 |
+
"step": 80
|
145 |
+
},
|
146 |
+
{
|
147 |
+
"epoch": 0.16,
|
148 |
+
"grad_norm": 0.41253119841994956,
|
149 |
+
"learning_rate": 4.947503654462276e-07,
|
150 |
+
"logits/chosen": 13.798151016235352,
|
151 |
+
"logits/rejected": 12.685941696166992,
|
152 |
+
"logps/chosen": -403.194580078125,
|
153 |
+
"logps/rejected": -421.2926330566406,
|
154 |
+
"loss": 0.6888,
|
155 |
+
"rewards/accuracies": 0.5062500238418579,
|
156 |
+
"rewards/chosen": -0.012594172731041908,
|
157 |
+
"rewards/margins": 0.004347759298980236,
|
158 |
+
"rewards/rejected": -0.01694193109869957,
|
159 |
+
"step": 90
|
160 |
+
},
|
161 |
+
{
|
162 |
+
"epoch": 0.17777777777777778,
|
163 |
+
"grad_norm": 0.3696840111639605,
|
164 |
+
"learning_rate": 4.911085493475802e-07,
|
165 |
+
"logits/chosen": 14.222944259643555,
|
166 |
+
"logits/rejected": 12.481651306152344,
|
167 |
+
"logps/chosen": -427.9515075683594,
|
168 |
+
"logps/rejected": -426.4696350097656,
|
169 |
+
"loss": 0.6885,
|
170 |
+
"rewards/accuracies": 0.606249988079071,
|
171 |
+
"rewards/chosen": -0.01204882562160492,
|
172 |
+
"rewards/margins": 0.010180080309510231,
|
173 |
+
"rewards/rejected": -0.02222890593111515,
|
174 |
+
"step": 100
|
175 |
+
},
|
176 |
+
{
|
177 |
+
"epoch": 0.17777777777777778,
|
178 |
+
"eval_logits/chosen": 14.262085914611816,
|
179 |
+
"eval_logits/rejected": 12.830548286437988,
|
180 |
+
"eval_logps/chosen": -407.780517578125,
|
181 |
+
"eval_logps/rejected": -402.74957275390625,
|
182 |
+
"eval_loss": 0.6883670687675476,
|
183 |
+
"eval_rewards/accuracies": 0.6190476417541504,
|
184 |
+
"eval_rewards/chosen": -0.015755515545606613,
|
185 |
+
"eval_rewards/margins": 0.008608067408204079,
|
186 |
+
"eval_rewards/rejected": -0.024363582953810692,
|
187 |
+
"eval_runtime": 91.0668,
|
188 |
+
"eval_samples_per_second": 10.981,
|
189 |
+
"eval_steps_per_second": 0.692,
|
190 |
+
"step": 100
|
191 |
+
},
|
192 |
+
{
|
193 |
+
"epoch": 0.19555555555555557,
|
194 |
+
"grad_norm": 0.387876184968801,
|
195 |
+
"learning_rate": 4.865339311109869e-07,
|
196 |
+
"logits/chosen": 13.772363662719727,
|
197 |
+
"logits/rejected": 12.22777271270752,
|
198 |
+
"logps/chosen": -405.01214599609375,
|
199 |
+
"logps/rejected": -403.9130859375,
|
200 |
+
"loss": 0.688,
|
201 |
+
"rewards/accuracies": 0.6875,
|
202 |
+
"rewards/chosen": -0.0224502794444561,
|
203 |
+
"rewards/margins": 0.012338267639279366,
|
204 |
+
"rewards/rejected": -0.03478854522109032,
|
205 |
+
"step": 110
|
206 |
+
},
|
207 |
+
{
|
208 |
+
"epoch": 0.21333333333333335,
|
209 |
+
"grad_norm": 0.3899509909292818,
|
210 |
+
"learning_rate": 4.810442090457072e-07,
|
211 |
+
"logits/chosen": 13.653238296508789,
|
212 |
+
"logits/rejected": 12.279303550720215,
|
213 |
+
"logps/chosen": -402.15704345703125,
|
214 |
+
"logps/rejected": -404.1790466308594,
|
215 |
+
"loss": 0.6866,
|
216 |
+
"rewards/accuracies": 0.668749988079071,
|
217 |
+
"rewards/chosen": -0.024786310270428658,
|
218 |
+
"rewards/margins": 0.014094953425228596,
|
219 |
+
"rewards/rejected": -0.03888126462697983,
|
220 |
+
"step": 120
|
221 |
+
},
|
222 |
+
{
|
223 |
+
"epoch": 0.2311111111111111,
|
224 |
+
"grad_norm": 0.3557450823856152,
|
225 |
+
"learning_rate": 4.746606218199385e-07,
|
226 |
+
"logits/chosen": 14.775594711303711,
|
227 |
+
"logits/rejected": 12.874218940734863,
|
228 |
+
"logps/chosen": -390.8421630859375,
|
229 |
+
"logps/rejected": -400.7987365722656,
|
230 |
+
"loss": 0.6854,
|
231 |
+
"rewards/accuracies": 0.6937500238418579,
|
232 |
+
"rewards/chosen": -0.03051091358065605,
|
233 |
+
"rewards/margins": 0.017188305035233498,
|
234 |
+
"rewards/rejected": -0.0476992204785347,
|
235 |
+
"step": 130
|
236 |
+
},
|
237 |
+
{
|
238 |
+
"epoch": 0.24888888888888888,
|
239 |
+
"grad_norm": 0.36125778338353265,
|
240 |
+
"learning_rate": 4.674078662925359e-07,
|
241 |
+
"logits/chosen": 13.882017135620117,
|
242 |
+
"logits/rejected": 12.225305557250977,
|
243 |
+
"logps/chosen": -420.66033935546875,
|
244 |
+
"logps/rejected": -413.9700622558594,
|
245 |
+
"loss": 0.6837,
|
246 |
+
"rewards/accuracies": 0.6937500238418579,
|
247 |
+
"rewards/chosen": -0.040414221584796906,
|
248 |
+
"rewards/margins": 0.021218815818428993,
|
249 |
+
"rewards/rejected": -0.06163303926587105,
|
250 |
+
"step": 140
|
251 |
+
},
|
252 |
+
{
|
253 |
+
"epoch": 0.26666666666666666,
|
254 |
+
"grad_norm": 0.42414032297199183,
|
255 |
+
"learning_rate": 4.593140019656625e-07,
|
256 |
+
"logits/chosen": 13.458653450012207,
|
257 |
+
"logits/rejected": 11.939371109008789,
|
258 |
+
"logps/chosen": -395.70318603515625,
|
259 |
+
"logps/rejected": -426.5689392089844,
|
260 |
+
"loss": 0.6815,
|
261 |
+
"rewards/accuracies": 0.7437499761581421,
|
262 |
+
"rewards/chosen": -0.04780956357717514,
|
263 |
+
"rewards/margins": 0.022214924916625023,
|
264 |
+
"rewards/rejected": -0.07002449035644531,
|
265 |
+
"step": 150
|
266 |
+
},
|
267 |
+
{
|
268 |
+
"epoch": 0.28444444444444444,
|
269 |
+
"grad_norm": 0.31038758853221154,
|
270 |
+
"learning_rate": 4.504103424280266e-07,
|
271 |
+
"logits/chosen": 14.602518081665039,
|
272 |
+
"logits/rejected": 12.48104476928711,
|
273 |
+
"logps/chosen": -445.41900634765625,
|
274 |
+
"logps/rejected": -427.4073791503906,
|
275 |
+
"loss": 0.6785,
|
276 |
+
"rewards/accuracies": 0.7562500238418579,
|
277 |
+
"rewards/chosen": -0.04993806034326553,
|
278 |
+
"rewards/margins": 0.03340305760502815,
|
279 |
+
"rewards/rejected": -0.08334111422300339,
|
280 |
+
"step": 160
|
281 |
+
},
|
282 |
+
{
|
283 |
+
"epoch": 0.3022222222222222,
|
284 |
+
"grad_norm": 0.5166732651791626,
|
285 |
+
"learning_rate": 4.407313342086905e-07,
|
286 |
+
"logits/chosen": 14.151707649230957,
|
287 |
+
"logits/rejected": 12.189191818237305,
|
288 |
+
"logps/chosen": -463.04193115234375,
|
289 |
+
"logps/rejected": -419.1507263183594,
|
290 |
+
"loss": 0.6757,
|
291 |
+
"rewards/accuracies": 0.7124999761581421,
|
292 |
+
"rewards/chosen": -0.06344902515411377,
|
293 |
+
"rewards/margins": 0.031504396349191666,
|
294 |
+
"rewards/rejected": -0.09495342522859573,
|
295 |
+
"step": 170
|
296 |
+
},
|
297 |
+
{
|
298 |
+
"epoch": 0.32,
|
299 |
+
"grad_norm": 0.351499350492249,
|
300 |
+
"learning_rate": 4.3031442351014115e-07,
|
301 |
+
"logits/chosen": 13.737968444824219,
|
302 |
+
"logits/rejected": 12.326078414916992,
|
303 |
+
"logps/chosen": -399.82208251953125,
|
304 |
+
"logps/rejected": -415.14105224609375,
|
305 |
+
"loss": 0.6754,
|
306 |
+
"rewards/accuracies": 0.75,
|
307 |
+
"rewards/chosen": -0.08228292316198349,
|
308 |
+
"rewards/margins": 0.03530151769518852,
|
309 |
+
"rewards/rejected": -0.11758442968130112,
|
310 |
+
"step": 180
|
311 |
+
},
|
312 |
+
{
|
313 |
+
"epoch": 0.3377777777777778,
|
314 |
+
"grad_norm": 0.4342072416786669,
|
315 |
+
"learning_rate": 4.19199911336207e-07,
|
316 |
+
"logits/chosen": 13.775731086730957,
|
317 |
+
"logits/rejected": 11.498677253723145,
|
318 |
+
"logps/chosen": -436.33734130859375,
|
319 |
+
"logps/rejected": -418.3775939941406,
|
320 |
+
"loss": 0.6717,
|
321 |
+
"rewards/accuracies": 0.7437499761581421,
|
322 |
+
"rewards/chosen": -0.08298386633396149,
|
323 |
+
"rewards/margins": 0.04400986433029175,
|
324 |
+
"rewards/rejected": -0.12699371576309204,
|
325 |
+
"step": 190
|
326 |
+
},
|
327 |
+
{
|
328 |
+
"epoch": 0.35555555555555557,
|
329 |
+
"grad_norm": 0.4572664943304227,
|
330 |
+
"learning_rate": 4.0743079757530443e-07,
|
331 |
+
"logits/chosen": 14.21821403503418,
|
332 |
+
"logits/rejected": 12.080196380615234,
|
333 |
+
"logps/chosen": -428.86083984375,
|
334 |
+
"logps/rejected": -398.98223876953125,
|
335 |
+
"loss": 0.6712,
|
336 |
+
"rewards/accuracies": 0.75,
|
337 |
+
"rewards/chosen": -0.08896859735250473,
|
338 |
+
"rewards/margins": 0.042290668934583664,
|
339 |
+
"rewards/rejected": -0.1312592625617981,
|
340 |
+
"step": 200
|
341 |
+
},
|
342 |
+
{
|
343 |
+
"epoch": 0.35555555555555557,
|
344 |
+
"eval_logits/chosen": 14.084491729736328,
|
345 |
+
"eval_logits/rejected": 12.648215293884277,
|
346 |
+
"eval_logps/chosen": -415.914794921875,
|
347 |
+
"eval_logps/rejected": -414.950439453125,
|
348 |
+
"eval_loss": 0.6679643988609314,
|
349 |
+
"eval_rewards/accuracies": 0.7936508059501648,
|
350 |
+
"eval_rewards/chosen": -0.09709871560335159,
|
351 |
+
"eval_rewards/margins": 0.04927373677492142,
|
352 |
+
"eval_rewards/rejected": -0.146372452378273,
|
353 |
+
"eval_runtime": 91.0693,
|
354 |
+
"eval_samples_per_second": 10.981,
|
355 |
+
"eval_steps_per_second": 0.692,
|
356 |
+
"step": 200
|
357 |
+
},
|
358 |
+
{
|
359 |
+
"epoch": 0.37333333333333335,
|
360 |
+
"grad_norm": 0.3933409306767686,
|
361 |
+
"learning_rate": 3.9505261464222127e-07,
|
362 |
+
"logits/chosen": 14.01300048828125,
|
363 |
+
"logits/rejected": 12.405731201171875,
|
364 |
+
"logps/chosen": -414.12432861328125,
|
365 |
+
"logps/rejected": -435.1546936035156,
|
366 |
+
"loss": 0.6667,
|
367 |
+
"rewards/accuracies": 0.78125,
|
368 |
+
"rewards/chosen": -0.10342928022146225,
|
369 |
+
"rewards/margins": 0.06037931516766548,
|
370 |
+
"rewards/rejected": -0.16380859911441803,
|
371 |
+
"step": 210
|
372 |
+
},
|
373 |
+
{
|
374 |
+
"epoch": 0.39111111111111113,
|
375 |
+
"grad_norm": 0.48372655964524586,
|
376 |
+
"learning_rate": 3.821132513220511e-07,
|
377 |
+
"logits/chosen": 13.346511840820312,
|
378 |
+
"logits/rejected": 11.28764820098877,
|
379 |
+
"logps/chosen": -409.07830810546875,
|
380 |
+
"logps/rejected": -409.73162841796875,
|
381 |
+
"loss": 0.6614,
|
382 |
+
"rewards/accuracies": 0.8125,
|
383 |
+
"rewards/chosen": -0.11821464449167252,
|
384 |
+
"rewards/margins": 0.068251833319664,
|
385 |
+
"rewards/rejected": -0.18646648526191711,
|
386 |
+
"step": 220
|
387 |
+
},
|
388 |
+
{
|
389 |
+
"epoch": 0.4088888888888889,
|
390 |
+
"grad_norm": 0.4254357046966758,
|
391 |
+
"learning_rate": 3.6866276749778575e-07,
|
392 |
+
"logits/chosen": 14.086942672729492,
|
393 |
+
"logits/rejected": 12.073091506958008,
|
394 |
+
"logps/chosen": -421.7301330566406,
|
395 |
+
"logps/rejected": -425.4147033691406,
|
396 |
+
"loss": 0.6613,
|
397 |
+
"rewards/accuracies": 0.768750011920929,
|
398 |
+
"rewards/chosen": -0.1326219141483307,
|
399 |
+
"rewards/margins": 0.06661403924226761,
|
400 |
+
"rewards/rejected": -0.1992359459400177,
|
401 |
+
"step": 230
|
402 |
+
},
|
403 |
+
{
|
404 |
+
"epoch": 0.4266666666666667,
|
405 |
+
"grad_norm": 0.5227754305907224,
|
406 |
+
"learning_rate": 3.5475320047835385e-07,
|
407 |
+
"logits/chosen": 13.673505783081055,
|
408 |
+
"logits/rejected": 11.437660217285156,
|
409 |
+
"logps/chosen": -427.7577209472656,
|
410 |
+
"logps/rejected": -431.9249572753906,
|
411 |
+
"loss": 0.6557,
|
412 |
+
"rewards/accuracies": 0.7875000238418579,
|
413 |
+
"rewards/chosen": -0.14575575292110443,
|
414 |
+
"rewards/margins": 0.08081929385662079,
|
415 |
+
"rewards/rejected": -0.22657504677772522,
|
416 |
+
"step": 240
|
417 |
+
},
|
418 |
+
{
|
419 |
+
"epoch": 0.4444444444444444,
|
420 |
+
"grad_norm": 0.5124383191463783,
|
421 |
+
"learning_rate": 3.4043836367638084e-07,
|
422 |
+
"logits/chosen": 14.250768661499023,
|
423 |
+
"logits/rejected": 12.134879112243652,
|
424 |
+
"logps/chosen": -437.49945068359375,
|
425 |
+
"logps/rejected": -460.59295654296875,
|
426 |
+
"loss": 0.656,
|
427 |
+
"rewards/accuracies": 0.8500000238418579,
|
428 |
+
"rewards/chosen": -0.15989777445793152,
|
429 |
+
"rewards/margins": 0.09294405579566956,
|
430 |
+
"rewards/rejected": -0.25284186005592346,
|
431 |
+
"step": 250
|
432 |
+
},
|
433 |
+
{
|
434 |
+
"epoch": 0.4622222222222222,
|
435 |
+
"grad_norm": 0.46067856126558626,
|
436 |
+
"learning_rate": 3.257736384145506e-07,
|
437 |
+
"logits/chosen": 14.0197172164917,
|
438 |
+
"logits/rejected": 12.032342910766602,
|
439 |
+
"logps/chosen": -459.6131286621094,
|
440 |
+
"logps/rejected": -442.3273010253906,
|
441 |
+
"loss": 0.6506,
|
442 |
+
"rewards/accuracies": 0.762499988079071,
|
443 |
+
"rewards/chosen": -0.17320798337459564,
|
444 |
+
"rewards/margins": 0.0869670957326889,
|
445 |
+
"rewards/rejected": -0.26017507910728455,
|
446 |
+
"step": 260
|
447 |
+
},
|
448 |
+
{
|
449 |
+
"epoch": 0.48,
|
450 |
+
"grad_norm": 0.5260206224619711,
|
451 |
+
"learning_rate": 3.1081575966602624e-07,
|
452 |
+
"logits/chosen": 12.930171012878418,
|
453 |
+
"logits/rejected": 11.695451736450195,
|
454 |
+
"logps/chosen": -399.03790283203125,
|
455 |
+
"logps/rejected": -436.02532958984375,
|
456 |
+
"loss": 0.6438,
|
457 |
+
"rewards/accuracies": 0.75,
|
458 |
+
"rewards/chosen": -0.21191565692424774,
|
459 |
+
"rewards/margins": 0.0779983252286911,
|
460 |
+
"rewards/rejected": -0.28991398215293884,
|
461 |
+
"step": 270
|
462 |
+
},
|
463 |
+
{
|
464 |
+
"epoch": 0.49777777777777776,
|
465 |
+
"grad_norm": 0.49006540494490103,
|
466 |
+
"learning_rate": 2.9562259655786065e-07,
|
467 |
+
"logits/chosen": 14.197479248046875,
|
468 |
+
"logits/rejected": 12.239548683166504,
|
469 |
+
"logps/chosen": -450.38623046875,
|
470 |
+
"logps/rejected": -440.9632873535156,
|
471 |
+
"loss": 0.6414,
|
472 |
+
"rewards/accuracies": 0.793749988079071,
|
473 |
+
"rewards/chosen": -0.21933193504810333,
|
474 |
+
"rewards/margins": 0.11443768441677094,
|
475 |
+
"rewards/rejected": -0.33376961946487427,
|
476 |
+
"step": 280
|
477 |
+
},
|
478 |
+
{
|
479 |
+
"epoch": 0.5155555555555555,
|
480 |
+
"grad_norm": 0.43019209624061255,
|
481 |
+
"learning_rate": 2.8025292848658625e-07,
|
482 |
+
"logits/chosen": 13.873621940612793,
|
483 |
+
"logits/rejected": 11.836385726928711,
|
484 |
+
"logps/chosen": -467.55108642578125,
|
485 |
+
"logps/rejected": -473.07818603515625,
|
486 |
+
"loss": 0.6342,
|
487 |
+
"rewards/accuracies": 0.800000011920929,
|
488 |
+
"rewards/chosen": -0.23439328372478485,
|
489 |
+
"rewards/margins": 0.13239440321922302,
|
490 |
+
"rewards/rejected": -0.3667876720428467,
|
491 |
+
"step": 290
|
492 |
+
},
|
493 |
+
{
|
494 |
+
"epoch": 0.5333333333333333,
|
495 |
+
"grad_norm": 0.46075707476643984,
|
496 |
+
"learning_rate": 2.647662177121486e-07,
|
497 |
+
"logits/chosen": 13.895523071289062,
|
498 |
+
"logits/rejected": 11.491305351257324,
|
499 |
+
"logps/chosen": -416.88592529296875,
|
500 |
+
"logps/rejected": -417.43243408203125,
|
501 |
+
"loss": 0.6339,
|
502 |
+
"rewards/accuracies": 0.7875000238418579,
|
503 |
+
"rewards/chosen": -0.2427409142255783,
|
504 |
+
"rewards/margins": 0.10592355579137802,
|
505 |
+
"rewards/rejected": -0.3486644923686981,
|
506 |
+
"step": 300
|
507 |
+
},
|
508 |
+
{
|
509 |
+
"epoch": 0.5333333333333333,
|
510 |
+
"eval_logits/chosen": 14.174385070800781,
|
511 |
+
"eval_logits/rejected": 12.855634689331055,
|
512 |
+
"eval_logps/chosen": -432.1300354003906,
|
513 |
+
"eval_logps/rejected": -437.43072509765625,
|
514 |
+
"eval_loss": 0.6389195322990417,
|
515 |
+
"eval_rewards/accuracies": 0.7539682388305664,
|
516 |
+
"eval_rewards/chosen": -0.2592509388923645,
|
517 |
+
"eval_rewards/margins": 0.11192431300878525,
|
518 |
+
"eval_rewards/rejected": -0.37117522954940796,
|
519 |
+
"eval_runtime": 91.0722,
|
520 |
+
"eval_samples_per_second": 10.98,
|
521 |
+
"eval_steps_per_second": 0.692,
|
522 |
+
"step": 300
|
523 |
+
},
|
524 |
+
{
|
525 |
+
"epoch": 0.5511111111111111,
|
526 |
+
"grad_norm": 0.5603386362578662,
|
527 |
+
"learning_rate": 2.492223793099743e-07,
|
528 |
+
"logits/chosen": 13.52906608581543,
|
529 |
+
"logits/rejected": 12.464258193969727,
|
530 |
+
"logps/chosen": -425.34600830078125,
|
531 |
+
"logps/rejected": -457.48968505859375,
|
532 |
+
"loss": 0.6345,
|
533 |
+
"rewards/accuracies": 0.8062499761581421,
|
534 |
+
"rewards/chosen": -0.26422396302223206,
|
535 |
+
"rewards/margins": 0.11599861085414886,
|
536 |
+
"rewards/rejected": -0.3802226185798645,
|
537 |
+
"step": 310
|
538 |
+
},
|
539 |
+
{
|
540 |
+
"epoch": 0.5688888888888889,
|
541 |
+
"grad_norm": 0.5899560581428063,
|
542 |
+
"learning_rate": 2.3368154937118352e-07,
|
543 |
+
"logits/chosen": 14.568766593933105,
|
544 |
+
"logits/rejected": 12.381284713745117,
|
545 |
+
"logps/chosen": -482.33648681640625,
|
546 |
+
"logps/rejected": -460.39959716796875,
|
547 |
+
"loss": 0.6352,
|
548 |
+
"rewards/accuracies": 0.737500011920929,
|
549 |
+
"rewards/chosen": -0.2616764307022095,
|
550 |
+
"rewards/margins": 0.11895668506622314,
|
551 |
+
"rewards/rejected": -0.3806331157684326,
|
552 |
+
"step": 320
|
553 |
+
},
|
554 |
+
{
|
555 |
+
"epoch": 0.5866666666666667,
|
556 |
+
"grad_norm": 0.5671140786141569,
|
557 |
+
"learning_rate": 2.1820385234773604e-07,
|
558 |
+
"logits/chosen": 13.732978820800781,
|
559 |
+
"logits/rejected": 11.89165210723877,
|
560 |
+
"logps/chosen": -432.77435302734375,
|
561 |
+
"logps/rejected": -449.07147216796875,
|
562 |
+
"loss": 0.6348,
|
563 |
+
"rewards/accuracies": 0.78125,
|
564 |
+
"rewards/chosen": -0.28699907660484314,
|
565 |
+
"rewards/margins": 0.1393791139125824,
|
566 |
+
"rewards/rejected": -0.42637819051742554,
|
567 |
+
"step": 330
|
568 |
+
},
|
569 |
+
{
|
570 |
+
"epoch": 0.6044444444444445,
|
571 |
+
"grad_norm": 0.494895870546235,
|
572 |
+
"learning_rate": 2.0284916844260697e-07,
|
573 |
+
"logits/chosen": 14.013145446777344,
|
574 |
+
"logits/rejected": 12.176668167114258,
|
575 |
+
"logps/chosen": -466.3651428222656,
|
576 |
+
"logps/rejected": -457.82196044921875,
|
577 |
+
"loss": 0.6271,
|
578 |
+
"rewards/accuracies": 0.78125,
|
579 |
+
"rewards/chosen": -0.2999596893787384,
|
580 |
+
"rewards/margins": 0.150094673037529,
|
581 |
+
"rewards/rejected": -0.4500543177127838,
|
582 |
+
"step": 340
|
583 |
+
},
|
584 |
+
{
|
585 |
+
"epoch": 0.6222222222222222,
|
586 |
+
"grad_norm": 0.44235392764613574,
|
587 |
+
"learning_rate": 1.876769019449141e-07,
|
588 |
+
"logits/chosen": 13.546051025390625,
|
589 |
+
"logits/rejected": 11.55055046081543,
|
590 |
+
"logps/chosen": -455.70892333984375,
|
591 |
+
"logps/rejected": -447.29156494140625,
|
592 |
+
"loss": 0.6205,
|
593 |
+
"rewards/accuracies": 0.768750011920929,
|
594 |
+
"rewards/chosen": -0.32695135474205017,
|
595 |
+
"rewards/margins": 0.1475459635257721,
|
596 |
+
"rewards/rejected": -0.47449731826782227,
|
597 |
+
"step": 350
|
598 |
+
},
|
599 |
+
{
|
600 |
+
"epoch": 0.64,
|
601 |
+
"grad_norm": 0.4906124921285579,
|
602 |
+
"learning_rate": 1.7274575140626315e-07,
|
603 |
+
"logits/chosen": 13.1173095703125,
|
604 |
+
"logits/rejected": 10.774048805236816,
|
605 |
+
"logps/chosen": -454.48797607421875,
|
606 |
+
"logps/rejected": -431.7547912597656,
|
607 |
+
"loss": 0.6226,
|
608 |
+
"rewards/accuracies": 0.793749988079071,
|
609 |
+
"rewards/chosen": -0.30864274501800537,
|
610 |
+
"rewards/margins": 0.155232235789299,
|
611 |
+
"rewards/rejected": -0.4638749659061432,
|
612 |
+
"step": 360
|
613 |
+
},
|
614 |
+
{
|
615 |
+
"epoch": 0.6577777777777778,
|
616 |
+
"grad_norm": 0.632117867399185,
|
617 |
+
"learning_rate": 1.5811348254745572e-07,
|
618 |
+
"logits/chosen": 13.715913772583008,
|
619 |
+
"logits/rejected": 12.367976188659668,
|
620 |
+
"logps/chosen": -395.98883056640625,
|
621 |
+
"logps/rejected": -431.3076171875,
|
622 |
+
"loss": 0.6198,
|
623 |
+
"rewards/accuracies": 0.762499988079071,
|
624 |
+
"rewards/chosen": -0.3397838771343231,
|
625 |
+
"rewards/margins": 0.15032809972763062,
|
626 |
+
"rewards/rejected": -0.49011197686195374,
|
627 |
+
"step": 370
|
628 |
+
},
|
629 |
+
{
|
630 |
+
"epoch": 0.6755555555555556,
|
631 |
+
"grad_norm": 0.49281947117085684,
|
632 |
+
"learning_rate": 1.4383670477413674e-07,
|
633 |
+
"logits/chosen": 13.90184211730957,
|
634 |
+
"logits/rejected": 11.279874801635742,
|
635 |
+
"logps/chosen": -461.05078125,
|
636 |
+
"logps/rejected": -442.17449951171875,
|
637 |
+
"loss": 0.6203,
|
638 |
+
"rewards/accuracies": 0.8187500238418579,
|
639 |
+
"rewards/chosen": -0.3398580551147461,
|
640 |
+
"rewards/margins": 0.16354627907276154,
|
641 |
+
"rewards/rejected": -0.5034043192863464,
|
642 |
+
"step": 380
|
643 |
+
},
|
644 |
+
{
|
645 |
+
"epoch": 0.6933333333333334,
|
646 |
+
"grad_norm": 0.42880409636253375,
|
647 |
+
"learning_rate": 1.2997065216600178e-07,
|
648 |
+
"logits/chosen": 13.401517868041992,
|
649 |
+
"logits/rejected": 12.245413780212402,
|
650 |
+
"logps/chosen": -430.324462890625,
|
651 |
+
"logps/rejected": -473.84893798828125,
|
652 |
+
"loss": 0.6211,
|
653 |
+
"rewards/accuracies": 0.762499988079071,
|
654 |
+
"rewards/chosen": -0.3511757552623749,
|
655 |
+
"rewards/margins": 0.1507997363805771,
|
656 |
+
"rewards/rejected": -0.5019755363464355,
|
657 |
+
"step": 390
|
658 |
+
},
|
659 |
+
{
|
660 |
+
"epoch": 0.7111111111111111,
|
661 |
+
"grad_norm": 0.40477681812860294,
|
662 |
+
"learning_rate": 1.1656896978687259e-07,
|
663 |
+
"logits/chosen": 13.779144287109375,
|
664 |
+
"logits/rejected": 11.85938835144043,
|
665 |
+
"logps/chosen": -440.189208984375,
|
666 |
+
"logps/rejected": -442.59576416015625,
|
667 |
+
"loss": 0.6203,
|
668 |
+
"rewards/accuracies": 0.800000011920929,
|
669 |
+
"rewards/chosen": -0.3478553891181946,
|
670 |
+
"rewards/margins": 0.1676493138074875,
|
671 |
+
"rewards/rejected": -0.5155047178268433,
|
672 |
+
"step": 400
|
673 |
+
},
|
674 |
+
{
|
675 |
+
"epoch": 0.7111111111111111,
|
676 |
+
"eval_logits/chosen": 13.944354057312012,
|
677 |
+
"eval_logits/rejected": 12.625584602355957,
|
678 |
+
"eval_logps/chosen": -443.5887451171875,
|
679 |
+
"eval_logps/rejected": -453.4456787109375,
|
680 |
+
"eval_loss": 0.6203290224075317,
|
681 |
+
"eval_rewards/accuracies": 0.7539682388305664,
|
682 |
+
"eval_rewards/chosen": -0.37383800745010376,
|
683 |
+
"eval_rewards/margins": 0.15748701989650726,
|
684 |
+
"eval_rewards/rejected": -0.531325101852417,
|
685 |
+
"eval_runtime": 91.0982,
|
686 |
+
"eval_samples_per_second": 10.977,
|
687 |
+
"eval_steps_per_second": 0.692,
|
688 |
+
"step": 400
|
689 |
+
},
|
690 |
+
{
|
691 |
+
"epoch": 0.7288888888888889,
|
692 |
+
"grad_norm": 0.4976401369848891,
|
693 |
+
"learning_rate": 1.0368350614236685e-07,
|
694 |
+
"logits/chosen": 14.432429313659668,
|
695 |
+
"logits/rejected": 12.734024047851562,
|
696 |
+
"logps/chosen": -437.3631286621094,
|
697 |
+
"logps/rejected": -450.3123474121094,
|
698 |
+
"loss": 0.6156,
|
699 |
+
"rewards/accuracies": 0.8062499761581421,
|
700 |
+
"rewards/chosen": -0.37959882616996765,
|
701 |
+
"rewards/margins": 0.17338675260543823,
|
702 |
+
"rewards/rejected": -0.5529855489730835,
|
703 |
+
"step": 410
|
704 |
+
},
|
705 |
+
{
|
706 |
+
"epoch": 0.7466666666666667,
|
707 |
+
"grad_norm": 0.49003467823924163,
|
708 |
+
"learning_rate": 9.136411258810229e-08,
|
709 |
+
"logits/chosen": 14.124117851257324,
|
710 |
+
"logits/rejected": 12.797445297241211,
|
711 |
+
"logps/chosen": -429.11767578125,
|
712 |
+
"logps/rejected": -457.7822265625,
|
713 |
+
"loss": 0.6164,
|
714 |
+
"rewards/accuracies": 0.824999988079071,
|
715 |
+
"rewards/chosen": -0.3667467534542084,
|
716 |
+
"rewards/margins": 0.17160849273204803,
|
717 |
+
"rewards/rejected": -0.5383552312850952,
|
718 |
+
"step": 420
|
719 |
+
},
|
720 |
+
{
|
721 |
+
"epoch": 0.7644444444444445,
|
722 |
+
"grad_norm": 0.5544257285079301,
|
723 |
+
"learning_rate": 7.965845046448657e-08,
|
724 |
+
"logits/chosen": 14.171014785766602,
|
725 |
+
"logits/rejected": 11.860380172729492,
|
726 |
+
"logps/chosen": -460.3724670410156,
|
727 |
+
"logps/rejected": -464.814208984375,
|
728 |
+
"loss": 0.6111,
|
729 |
+
"rewards/accuracies": 0.84375,
|
730 |
+
"rewards/chosen": -0.3682584762573242,
|
731 |
+
"rewards/margins": 0.21057240664958954,
|
732 |
+
"rewards/rejected": -0.578830897808075,
|
733 |
+
"step": 430
|
734 |
+
},
|
735 |
+
{
|
736 |
+
"epoch": 0.7822222222222223,
|
737 |
+
"grad_norm": 0.4893812986921009,
|
738 |
+
"learning_rate": 6.861180670424982e-08,
|
739 |
+
"logits/chosen": 14.053049087524414,
|
740 |
+
"logits/rejected": 12.361749649047852,
|
741 |
+
"logps/chosen": -458.22113037109375,
|
742 |
+
"logps/rejected": -473.56927490234375,
|
743 |
+
"loss": 0.6182,
|
744 |
+
"rewards/accuracies": 0.84375,
|
745 |
+
"rewards/chosen": -0.38496845960617065,
|
746 |
+
"rewards/margins": 0.1849747598171234,
|
747 |
+
"rewards/rejected": -0.5699432492256165,
|
748 |
+
"step": 440
|
749 |
+
},
|
750 |
+
{
|
751 |
+
"epoch": 0.8,
|
752 |
+
"grad_norm": 0.451783793255229,
|
753 |
+
"learning_rate": 5.826691862609986e-08,
|
754 |
+
"logits/chosen": 14.11127758026123,
|
755 |
+
"logits/rejected": 12.265645027160645,
|
756 |
+
"logps/chosen": -446.2030334472656,
|
757 |
+
"logps/rejected": -448.130859375,
|
758 |
+
"loss": 0.6154,
|
759 |
+
"rewards/accuracies": 0.793749988079071,
|
760 |
+
"rewards/chosen": -0.3934079110622406,
|
761 |
+
"rewards/margins": 0.20773550868034363,
|
762 |
+
"rewards/rejected": -0.601143479347229,
|
763 |
+
"step": 450
|
764 |
+
},
|
765 |
+
{
|
766 |
+
"epoch": 0.8177777777777778,
|
767 |
+
"grad_norm": 0.5757576180607112,
|
768 |
+
"learning_rate": 4.86638085923389e-08,
|
769 |
+
"logits/chosen": 13.425259590148926,
|
770 |
+
"logits/rejected": 11.878642082214355,
|
771 |
+
"logps/chosen": -452.25042724609375,
|
772 |
+
"logps/rejected": -480.77154541015625,
|
773 |
+
"loss": 0.6026,
|
774 |
+
"rewards/accuracies": 0.8125,
|
775 |
+
"rewards/chosen": -0.42305225133895874,
|
776 |
+
"rewards/margins": 0.19301000237464905,
|
777 |
+
"rewards/rejected": -0.6160622835159302,
|
778 |
+
"step": 460
|
779 |
+
},
|
780 |
+
{
|
781 |
+
"epoch": 0.8355555555555556,
|
782 |
+
"grad_norm": 0.6028404780293103,
|
783 |
+
"learning_rate": 3.983962917011829e-08,
|
784 |
+
"logits/chosen": 13.737588882446289,
|
785 |
+
"logits/rejected": 11.253530502319336,
|
786 |
+
"logps/chosen": -453.21466064453125,
|
787 |
+
"logps/rejected": -455.3412170410156,
|
788 |
+
"loss": 0.6099,
|
789 |
+
"rewards/accuracies": 0.793749988079071,
|
790 |
+
"rewards/chosen": -0.40479034185409546,
|
791 |
+
"rewards/margins": 0.23396852612495422,
|
792 |
+
"rewards/rejected": -0.6387587785720825,
|
793 |
+
"step": 470
|
794 |
+
},
|
795 |
+
{
|
796 |
+
"epoch": 0.8533333333333334,
|
797 |
+
"grad_norm": 0.4822754579818044,
|
798 |
+
"learning_rate": 3.182851939537409e-08,
|
799 |
+
"logits/chosen": 13.7351655960083,
|
800 |
+
"logits/rejected": 12.563407897949219,
|
801 |
+
"logps/chosen": -420.5502014160156,
|
802 |
+
"logps/rejected": -467.66143798828125,
|
803 |
+
"loss": 0.6121,
|
804 |
+
"rewards/accuracies": 0.768750011920929,
|
805 |
+
"rewards/chosen": -0.4305228590965271,
|
806 |
+
"rewards/margins": 0.1863301545381546,
|
807 |
+
"rewards/rejected": -0.6168529987335205,
|
808 |
+
"step": 480
|
809 |
+
},
|
810 |
+
{
|
811 |
+
"epoch": 0.8711111111111111,
|
812 |
+
"grad_norm": 0.49761272281985847,
|
813 |
+
"learning_rate": 2.466147269552893e-08,
|
814 |
+
"logits/chosen": 13.694448471069336,
|
815 |
+
"logits/rejected": 12.38302230834961,
|
816 |
+
"logps/chosen": -429.01910400390625,
|
817 |
+
"logps/rejected": -461.55718994140625,
|
818 |
+
"loss": 0.6128,
|
819 |
+
"rewards/accuracies": 0.793749988079071,
|
820 |
+
"rewards/chosen": -0.4084044396877289,
|
821 |
+
"rewards/margins": 0.16671887040138245,
|
822 |
+
"rewards/rejected": -0.5751233100891113,
|
823 |
+
"step": 490
|
824 |
+
},
|
825 |
+
{
|
826 |
+
"epoch": 0.8888888888888888,
|
827 |
+
"grad_norm": 0.6004438030491424,
|
828 |
+
"learning_rate": 1.8366216981942628e-08,
|
829 |
+
"logits/chosen": 13.4122896194458,
|
830 |
+
"logits/rejected": 11.679363250732422,
|
831 |
+
"logps/chosen": -436.2166442871094,
|
832 |
+
"logps/rejected": -459.05841064453125,
|
833 |
+
"loss": 0.6102,
|
834 |
+
"rewards/accuracies": 0.7749999761581421,
|
835 |
+
"rewards/chosen": -0.4055519104003906,
|
836 |
+
"rewards/margins": 0.21610493957996368,
|
837 |
+
"rewards/rejected": -0.6216568946838379,
|
838 |
+
"step": 500
|
839 |
+
},
|
840 |
+
{
|
841 |
+
"epoch": 0.8888888888888888,
|
842 |
+
"eval_logits/chosen": 13.842653274536133,
|
843 |
+
"eval_logits/rejected": 12.531366348266602,
|
844 |
+
"eval_logps/chosen": -447.70013427734375,
|
845 |
+
"eval_logps/rejected": -459.2376403808594,
|
846 |
+
"eval_loss": 0.6131365299224854,
|
847 |
+
"eval_rewards/accuracies": 0.7539682388305664,
|
848 |
+
"eval_rewards/chosen": -0.4149521589279175,
|
849 |
+
"eval_rewards/margins": 0.17429186403751373,
|
850 |
+
"eval_rewards/rejected": -0.5892440676689148,
|
851 |
+
"eval_runtime": 91.0747,
|
852 |
+
"eval_samples_per_second": 10.98,
|
853 |
+
"eval_steps_per_second": 0.692,
|
854 |
+
"step": 500
|
855 |
+
},
|
856 |
+
{
|
857 |
+
"epoch": 0.9066666666666666,
|
858 |
+
"grad_norm": 0.512528339625896,
|
859 |
+
"learning_rate": 1.296710737600934e-08,
|
860 |
+
"logits/chosen": 14.248278617858887,
|
861 |
+
"logits/rejected": 11.994184494018555,
|
862 |
+
"logps/chosen": -501.92333984375,
|
863 |
+
"logps/rejected": -475.35723876953125,
|
864 |
+
"loss": 0.6136,
|
865 |
+
"rewards/accuracies": 0.78125,
|
866 |
+
"rewards/chosen": -0.40441107749938965,
|
867 |
+
"rewards/margins": 0.22114601731300354,
|
868 |
+
"rewards/rejected": -0.6255571246147156,
|
869 |
+
"step": 510
|
870 |
+
},
|
871 |
+
{
|
872 |
+
"epoch": 0.9244444444444444,
|
873 |
+
"grad_norm": 0.46717306963218175,
|
874 |
+
"learning_rate": 8.485031983924557e-09,
|
875 |
+
"logits/chosen": 13.698689460754395,
|
876 |
+
"logits/rejected": 11.797239303588867,
|
877 |
+
"logps/chosen": -473.1220703125,
|
878 |
+
"logps/rejected": -475.6698303222656,
|
879 |
+
"loss": 0.609,
|
880 |
+
"rewards/accuracies": 0.824999988079071,
|
881 |
+
"rewards/chosen": -0.40674418210983276,
|
882 |
+
"rewards/margins": 0.2259480506181717,
|
883 |
+
"rewards/rejected": -0.6326922178268433,
|
884 |
+
"step": 520
|
885 |
+
},
|
886 |
+
{
|
887 |
+
"epoch": 0.9422222222222222,
|
888 |
+
"grad_norm": 0.46210848243091895,
|
889 |
+
"learning_rate": 4.937331084660129e-09,
|
890 |
+
"logits/chosen": 13.86840534210205,
|
891 |
+
"logits/rejected": 11.828570365905762,
|
892 |
+
"logps/chosen": -465.6477966308594,
|
893 |
+
"logps/rejected": -473.4181213378906,
|
894 |
+
"loss": 0.607,
|
895 |
+
"rewards/accuracies": 0.793749988079071,
|
896 |
+
"rewards/chosen": -0.4186909794807434,
|
897 |
+
"rewards/margins": 0.19602572917938232,
|
898 |
+
"rewards/rejected": -0.614716649055481,
|
899 |
+
"step": 530
|
900 |
+
},
|
901 |
+
{
|
902 |
+
"epoch": 0.96,
|
903 |
+
"grad_norm": 0.48983149292621647,
|
904 |
+
"learning_rate": 2.337730043793423e-09,
|
905 |
+
"logits/chosen": 13.412510871887207,
|
906 |
+
"logits/rejected": 11.925616264343262,
|
907 |
+
"logps/chosen": -441.61871337890625,
|
908 |
+
"logps/rejected": -453.08489990234375,
|
909 |
+
"loss": 0.6106,
|
910 |
+
"rewards/accuracies": 0.7562500238418579,
|
911 |
+
"rewards/chosen": -0.42602628469467163,
|
912 |
+
"rewards/margins": 0.19294434785842896,
|
913 |
+
"rewards/rejected": -0.6189705729484558,
|
914 |
+
"step": 540
|
915 |
+
},
|
916 |
+
{
|
917 |
+
"epoch": 0.9777777777777777,
|
918 |
+
"grad_norm": 0.46702206339372576,
|
919 |
+
"learning_rate": 6.962862127343205e-10,
|
920 |
+
"logits/chosen": 14.47253131866455,
|
921 |
+
"logits/rejected": 12.454832077026367,
|
922 |
+
"logps/chosen": -464.0704650878906,
|
923 |
+
"logps/rejected": -453.48187255859375,
|
924 |
+
"loss": 0.6073,
|
925 |
+
"rewards/accuracies": 0.800000011920929,
|
926 |
+
"rewards/chosen": -0.4236368238925934,
|
927 |
+
"rewards/margins": 0.20897309482097626,
|
928 |
+
"rewards/rejected": -0.6326099038124084,
|
929 |
+
"step": 550
|
930 |
+
},
|
931 |
+
{
|
932 |
+
"epoch": 0.9955555555555555,
|
933 |
+
"grad_norm": 0.620631975977973,
|
934 |
+
"learning_rate": 1.9350018786556956e-11,
|
935 |
+
"logits/chosen": 13.917093276977539,
|
936 |
+
"logits/rejected": 12.686841011047363,
|
937 |
+
"logps/chosen": -464.35980224609375,
|
938 |
+
"logps/rejected": -498.465576171875,
|
939 |
+
"loss": 0.6102,
|
940 |
+
"rewards/accuracies": 0.7250000238418579,
|
941 |
+
"rewards/chosen": -0.4351399838924408,
|
942 |
+
"rewards/margins": 0.1616152673959732,
|
943 |
+
"rewards/rejected": -0.5967552661895752,
|
944 |
+
"step": 560
|
945 |
+
},
|
946 |
+
{
|
947 |
+
"epoch": 0.9991111111111111,
|
948 |
+
"step": 562,
|
949 |
+
"total_flos": 0.0,
|
950 |
+
"train_loss": 0.6475430355801701,
|
951 |
+
"train_runtime": 7558.4306,
|
952 |
+
"train_samples_per_second": 4.763,
|
953 |
+
"train_steps_per_second": 0.074
|
954 |
+
}
|
955 |
+
],
|
956 |
+
"logging_steps": 10,
|
957 |
+
"max_steps": 562,
|
958 |
+
"num_input_tokens_seen": 0,
|
959 |
+
"num_train_epochs": 1,
|
960 |
+
"save_steps": 100,
|
961 |
+
"stateful_callbacks": {
|
962 |
+
"TrainerControl": {
|
963 |
+
"args": {
|
964 |
+
"should_epoch_stop": false,
|
965 |
+
"should_evaluate": false,
|
966 |
+
"should_log": false,
|
967 |
+
"should_save": true,
|
968 |
+
"should_training_stop": true
|
969 |
+
},
|
970 |
+
"attributes": {}
|
971 |
+
}
|
972 |
+
},
|
973 |
+
"total_flos": 0.0,
|
974 |
+
"train_batch_size": 4,
|
975 |
+
"trial_name": null,
|
976 |
+
"trial_params": null
|
977 |
+
}
|