Model save
Browse files
README.md
ADDED
|
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: alignment-handbook/zephyr-7b-sft-full
|
| 3 |
+
library_name: peft
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
tags:
|
| 6 |
+
- trl
|
| 7 |
+
- dpo
|
| 8 |
+
- generated_from_trainer
|
| 9 |
+
model-index:
|
| 10 |
+
- name: zephyr-dpop-qlora-gpt4-5e-7
|
| 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 |
+
# zephyr-dpop-qlora-gpt4-5e-7
|
| 18 |
+
|
| 19 |
+
This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the None dataset.
|
| 20 |
+
It achieves the following results on the evaluation set:
|
| 21 |
+
- Loss: 0.7727
|
| 22 |
+
- Positive Losses: 0.8307
|
| 23 |
+
- Dpo Losses: 0.6847
|
| 24 |
+
- Rewards/chosen: 0.0252
|
| 25 |
+
- Rewards/rejected: 0.0072
|
| 26 |
+
- Rewards/accuracies: 0.5920
|
| 27 |
+
- Rewards/margins: 0.0180
|
| 28 |
+
- Rewards/margins Max: 0.1227
|
| 29 |
+
- Rewards/margins Min: -0.0701
|
| 30 |
+
- Rewards/margins Std: 0.0638
|
| 31 |
+
- Logps/rejected: -257.8545
|
| 32 |
+
- Logps/chosen: -282.0691
|
| 33 |
+
- Logits/rejected: -2.7315
|
| 34 |
+
- Logits/chosen: -2.7691
|
| 35 |
+
|
| 36 |
+
## Model description
|
| 37 |
+
|
| 38 |
+
More information needed
|
| 39 |
+
|
| 40 |
+
## Intended uses & limitations
|
| 41 |
+
|
| 42 |
+
More information needed
|
| 43 |
+
|
| 44 |
+
## Training and evaluation data
|
| 45 |
+
|
| 46 |
+
More information needed
|
| 47 |
+
|
| 48 |
+
## Training procedure
|
| 49 |
+
|
| 50 |
+
### Training hyperparameters
|
| 51 |
+
|
| 52 |
+
The following hyperparameters were used during training:
|
| 53 |
+
- learning_rate: 5e-07
|
| 54 |
+
- train_batch_size: 4
|
| 55 |
+
- eval_batch_size: 8
|
| 56 |
+
- seed: 42
|
| 57 |
+
- distributed_type: multi-GPU
|
| 58 |
+
- num_devices: 2
|
| 59 |
+
- gradient_accumulation_steps: 2
|
| 60 |
+
- total_train_batch_size: 16
|
| 61 |
+
- total_eval_batch_size: 16
|
| 62 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 63 |
+
- lr_scheduler_type: cosine
|
| 64 |
+
- lr_scheduler_warmup_ratio: 0.1
|
| 65 |
+
- num_epochs: 1
|
| 66 |
+
|
| 67 |
+
### Training results
|
| 68 |
+
|
| 69 |
+
| Training Loss | Epoch | Step | Validation Loss | Positive Losses | Dpo Losses | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Rewards/margins Max | Rewards/margins Min | Rewards/margins Std | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
| 70 |
+
|:-------------:|:-----:|:----:|:---------------:|:---------------:|:----------:|:--------------:|:----------------:|:------------------:|:---------------:|:-------------------:|:-------------------:|:-------------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
| 71 |
+
| 0.6709 | 0.28 | 100 | 0.7044 | 0.1334 | 0.6900 | 0.0150 | 0.0085 | 0.5830 | 0.0065 | 0.0434 | -0.0241 | 0.0224 | -257.7293 | -283.0980 | -2.7607 | -2.7991 |
|
| 72 |
+
| 0.6375 | 0.56 | 200 | 0.7539 | 0.6398 | 0.6862 | 0.0226 | 0.0079 | 0.5880 | 0.0147 | 0.1022 | -0.0596 | 0.0535 | -257.7878 | -282.3363 | -2.7335 | -2.7718 |
|
| 73 |
+
| 0.6203 | 0.85 | 300 | 0.7727 | 0.8307 | 0.6847 | 0.0252 | 0.0072 | 0.5920 | 0.0180 | 0.1227 | -0.0701 | 0.0638 | -257.8545 | -282.0691 | -2.7315 | -2.7691 |
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
### Framework versions
|
| 77 |
+
|
| 78 |
+
- PEFT 0.7.1
|
| 79 |
+
- Transformers 4.39.0.dev0
|
| 80 |
+
- Pytorch 2.1.2+cu121
|
| 81 |
+
- Datasets 2.14.6
|
| 82 |
+
- Tokenizers 0.15.2
|
adapter_model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 671150064
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bc7a9417d9327aa92903b352ccd4daa0abc6b92419268b6ba050d34235083d93
|
| 3 |
size 671150064
|
all_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.6494339942932129,
|
| 4 |
+
"train_runtime": 4013.1211,
|
| 5 |
+
"train_samples": 5678,
|
| 6 |
+
"train_samples_per_second": 1.415,
|
| 7 |
+
"train_steps_per_second": 0.088
|
| 8 |
+
}
|
runs/Jul28_07-28-08_notebook-deployment-48-7d9b6c99-khd85/events.out.tfevents.1722151734.notebook-deployment-48-7d9b6c99-khd85.1188279.0
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0c536ddf2089015d8a97abb4d74d4a15b39c073163230a36bdb8508b810b1820
|
| 3 |
+
size 44402
|
train_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.6494339942932129,
|
| 4 |
+
"train_runtime": 4013.1211,
|
| 5 |
+
"train_samples": 5678,
|
| 6 |
+
"train_samples_per_second": 1.415,
|
| 7 |
+
"train_steps_per_second": 0.088
|
| 8 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,813 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 1.0,
|
| 5 |
+
"eval_steps": 100,
|
| 6 |
+
"global_step": 355,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"dpo_losses": 0.6931471824645996,
|
| 13 |
+
"epoch": 0.0,
|
| 14 |
+
"grad_norm": 2.1563079383916617,
|
| 15 |
+
"learning_rate": 1.3888888888888887e-08,
|
| 16 |
+
"logits/chosen": -2.804708957672119,
|
| 17 |
+
"logits/rejected": -2.8150453567504883,
|
| 18 |
+
"logps/chosen": -217.97438049316406,
|
| 19 |
+
"logps/rejected": -216.58865356445312,
|
| 20 |
+
"loss": 0.6931,
|
| 21 |
+
"positive_losses": 0.0,
|
| 22 |
+
"rewards/accuracies": 0.0,
|
| 23 |
+
"rewards/chosen": 0.0,
|
| 24 |
+
"rewards/margins": 0.0,
|
| 25 |
+
"rewards/margins_max": 0.0,
|
| 26 |
+
"rewards/margins_min": 0.0,
|
| 27 |
+
"rewards/margins_std": 0.0,
|
| 28 |
+
"rewards/rejected": 0.0,
|
| 29 |
+
"step": 1
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"dpo_losses": 0.6929957270622253,
|
| 33 |
+
"epoch": 0.03,
|
| 34 |
+
"grad_norm": 29.018601338328015,
|
| 35 |
+
"learning_rate": 1.3888888888888888e-07,
|
| 36 |
+
"logits/chosen": -2.884633779525757,
|
| 37 |
+
"logits/rejected": -2.7994985580444336,
|
| 38 |
+
"logps/chosen": -366.7369689941406,
|
| 39 |
+
"logps/rejected": -275.4845275878906,
|
| 40 |
+
"loss": 0.7009,
|
| 41 |
+
"positive_losses": 0.08276741951704025,
|
| 42 |
+
"rewards/accuracies": 0.4027777910232544,
|
| 43 |
+
"rewards/chosen": -2.229649544460699e-05,
|
| 44 |
+
"rewards/margins": 0.00030543722095899284,
|
| 45 |
+
"rewards/margins_max": 0.003608349245041609,
|
| 46 |
+
"rewards/margins_min": -0.0025843861512839794,
|
| 47 |
+
"rewards/margins_std": 0.002678285352885723,
|
| 48 |
+
"rewards/rejected": -0.00032773372367955744,
|
| 49 |
+
"step": 10
|
| 50 |
+
},
|
| 51 |
+
{
|
| 52 |
+
"dpo_losses": 0.6929633021354675,
|
| 53 |
+
"epoch": 0.06,
|
| 54 |
+
"grad_norm": 11.724263026129163,
|
| 55 |
+
"learning_rate": 2.7777777777777776e-07,
|
| 56 |
+
"logits/chosen": -2.7479777336120605,
|
| 57 |
+
"logits/rejected": -2.6943278312683105,
|
| 58 |
+
"logps/chosen": -329.0453186035156,
|
| 59 |
+
"logps/rejected": -216.6789093017578,
|
| 60 |
+
"loss": 0.6981,
|
| 61 |
+
"positive_losses": 0.05849536508321762,
|
| 62 |
+
"rewards/accuracies": 0.5625,
|
| 63 |
+
"rewards/chosen": 0.0010186245199292898,
|
| 64 |
+
"rewards/margins": 0.00036991172237321734,
|
| 65 |
+
"rewards/margins_max": 0.003288195002824068,
|
| 66 |
+
"rewards/margins_min": -0.002696308773010969,
|
| 67 |
+
"rewards/margins_std": 0.0026607036124914885,
|
| 68 |
+
"rewards/rejected": 0.0006487127975560725,
|
| 69 |
+
"step": 20
|
| 70 |
+
},
|
| 71 |
+
{
|
| 72 |
+
"dpo_losses": 0.6923447251319885,
|
| 73 |
+
"epoch": 0.08,
|
| 74 |
+
"grad_norm": 2.313455126668315,
|
| 75 |
+
"learning_rate": 4.1666666666666667e-07,
|
| 76 |
+
"logits/chosen": -2.835400342941284,
|
| 77 |
+
"logits/rejected": -2.7531635761260986,
|
| 78 |
+
"logps/chosen": -328.7791442871094,
|
| 79 |
+
"logps/rejected": -232.63723754882812,
|
| 80 |
+
"loss": 0.6929,
|
| 81 |
+
"positive_losses": 0.0025131225120276213,
|
| 82 |
+
"rewards/accuracies": 0.6499999761581421,
|
| 83 |
+
"rewards/chosen": 0.005060128401964903,
|
| 84 |
+
"rewards/margins": 0.0016096157487481833,
|
| 85 |
+
"rewards/margins_max": 0.006041892804205418,
|
| 86 |
+
"rewards/margins_min": -0.0021330646704882383,
|
| 87 |
+
"rewards/margins_std": 0.003625147510319948,
|
| 88 |
+
"rewards/rejected": 0.003450512420386076,
|
| 89 |
+
"step": 30
|
| 90 |
+
},
|
| 91 |
+
{
|
| 92 |
+
"dpo_losses": 0.6913660764694214,
|
| 93 |
+
"epoch": 0.11,
|
| 94 |
+
"grad_norm": 1.9684885850555545,
|
| 95 |
+
"learning_rate": 4.998060489154965e-07,
|
| 96 |
+
"logits/chosen": -2.813957929611206,
|
| 97 |
+
"logits/rejected": -2.7604265213012695,
|
| 98 |
+
"logps/chosen": -284.87249755859375,
|
| 99 |
+
"logps/rejected": -227.00289916992188,
|
| 100 |
+
"loss": 0.6915,
|
| 101 |
+
"positive_losses": 0.0,
|
| 102 |
+
"rewards/accuracies": 0.800000011920929,
|
| 103 |
+
"rewards/chosen": 0.009648630395531654,
|
| 104 |
+
"rewards/margins": 0.003571807174012065,
|
| 105 |
+
"rewards/margins_max": 0.008924934081733227,
|
| 106 |
+
"rewards/margins_min": -0.001162365428172052,
|
| 107 |
+
"rewards/margins_std": 0.004516431130468845,
|
| 108 |
+
"rewards/rejected": 0.006076823454350233,
|
| 109 |
+
"step": 40
|
| 110 |
+
},
|
| 111 |
+
{
|
| 112 |
+
"dpo_losses": 0.6890588998794556,
|
| 113 |
+
"epoch": 0.14,
|
| 114 |
+
"grad_norm": 2.3506499385956174,
|
| 115 |
+
"learning_rate": 4.976275538042932e-07,
|
| 116 |
+
"logits/chosen": -2.8134660720825195,
|
| 117 |
+
"logits/rejected": -2.73073148727417,
|
| 118 |
+
"logps/chosen": -316.21136474609375,
|
| 119 |
+
"logps/rejected": -233.4212646484375,
|
| 120 |
+
"loss": 0.689,
|
| 121 |
+
"positive_losses": 0.0,
|
| 122 |
+
"rewards/accuracies": 0.8374999761581421,
|
| 123 |
+
"rewards/chosen": 0.015740128234028816,
|
| 124 |
+
"rewards/margins": 0.008210490457713604,
|
| 125 |
+
"rewards/margins_max": 0.017676424235105515,
|
| 126 |
+
"rewards/margins_min": 0.0003599710180424154,
|
| 127 |
+
"rewards/margins_std": 0.008204594254493713,
|
| 128 |
+
"rewards/rejected": 0.00752963637933135,
|
| 129 |
+
"step": 50
|
| 130 |
+
},
|
| 131 |
+
{
|
| 132 |
+
"dpo_losses": 0.6860449910163879,
|
| 133 |
+
"epoch": 0.17,
|
| 134 |
+
"grad_norm": 2.3197802822711284,
|
| 135 |
+
"learning_rate": 4.930493069997119e-07,
|
| 136 |
+
"logits/chosen": -2.7519192695617676,
|
| 137 |
+
"logits/rejected": -2.7033615112304688,
|
| 138 |
+
"logps/chosen": -342.56121826171875,
|
| 139 |
+
"logps/rejected": -263.17974853515625,
|
| 140 |
+
"loss": 0.6864,
|
| 141 |
+
"positive_losses": 0.0,
|
| 142 |
+
"rewards/accuracies": 0.8999999761581421,
|
| 143 |
+
"rewards/chosen": 0.02219192497432232,
|
| 144 |
+
"rewards/margins": 0.01428920030593872,
|
| 145 |
+
"rewards/margins_max": 0.02703510783612728,
|
| 146 |
+
"rewards/margins_min": 0.0032699487637728453,
|
| 147 |
+
"rewards/margins_std": 0.010757345706224442,
|
| 148 |
+
"rewards/rejected": 0.007902727462351322,
|
| 149 |
+
"step": 60
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"dpo_losses": 0.6826337575912476,
|
| 153 |
+
"epoch": 0.2,
|
| 154 |
+
"grad_norm": 1.81612559785793,
|
| 155 |
+
"learning_rate": 4.861156761634013e-07,
|
| 156 |
+
"logits/chosen": -2.8005824089050293,
|
| 157 |
+
"logits/rejected": -2.7131030559539795,
|
| 158 |
+
"logps/chosen": -359.32147216796875,
|
| 159 |
+
"logps/rejected": -235.8487091064453,
|
| 160 |
+
"loss": 0.6831,
|
| 161 |
+
"positive_losses": 0.0,
|
| 162 |
+
"rewards/accuracies": 0.949999988079071,
|
| 163 |
+
"rewards/chosen": 0.030173394829034805,
|
| 164 |
+
"rewards/margins": 0.021212318912148476,
|
| 165 |
+
"rewards/margins_max": 0.04077567160129547,
|
| 166 |
+
"rewards/margins_min": 0.003378701163455844,
|
| 167 |
+
"rewards/margins_std": 0.01716596446931362,
|
| 168 |
+
"rewards/rejected": 0.008961074985563755,
|
| 169 |
+
"step": 70
|
| 170 |
+
},
|
| 171 |
+
{
|
| 172 |
+
"dpo_losses": 0.6795681715011597,
|
| 173 |
+
"epoch": 0.23,
|
| 174 |
+
"grad_norm": 2.0311863815908806,
|
| 175 |
+
"learning_rate": 4.768938549177392e-07,
|
| 176 |
+
"logits/chosen": -2.8440804481506348,
|
| 177 |
+
"logits/rejected": -2.7800514698028564,
|
| 178 |
+
"logps/chosen": -328.42352294921875,
|
| 179 |
+
"logps/rejected": -286.81768798828125,
|
| 180 |
+
"loss": 0.6796,
|
| 181 |
+
"positive_losses": 0.0,
|
| 182 |
+
"rewards/accuracies": 0.949999988079071,
|
| 183 |
+
"rewards/chosen": 0.0342494361102581,
|
| 184 |
+
"rewards/margins": 0.027495408430695534,
|
| 185 |
+
"rewards/margins_max": 0.05417575687170029,
|
| 186 |
+
"rewards/margins_min": 0.004369550384581089,
|
| 187 |
+
"rewards/margins_std": 0.022978123277425766,
|
| 188 |
+
"rewards/rejected": 0.006754028145223856,
|
| 189 |
+
"step": 80
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"dpo_losses": 0.6761659383773804,
|
| 193 |
+
"epoch": 0.25,
|
| 194 |
+
"grad_norm": 2.441277633330898,
|
| 195 |
+
"learning_rate": 4.654732116743193e-07,
|
| 196 |
+
"logits/chosen": -2.784029722213745,
|
| 197 |
+
"logits/rejected": -2.700582981109619,
|
| 198 |
+
"logps/chosen": -334.9620056152344,
|
| 199 |
+
"logps/rejected": -198.4322967529297,
|
| 200 |
+
"loss": 0.6754,
|
| 201 |
+
"positive_losses": 0.0,
|
| 202 |
+
"rewards/accuracies": 0.9375,
|
| 203 |
+
"rewards/chosen": 0.03997080400586128,
|
| 204 |
+
"rewards/margins": 0.034528493881225586,
|
| 205 |
+
"rewards/margins_max": 0.07119405269622803,
|
| 206 |
+
"rewards/margins_min": 0.006210760213434696,
|
| 207 |
+
"rewards/margins_std": 0.030152523890137672,
|
| 208 |
+
"rewards/rejected": 0.00544231291860342,
|
| 209 |
+
"step": 90
|
| 210 |
+
},
|
| 211 |
+
{
|
| 212 |
+
"dpo_losses": 0.6722386479377747,
|
| 213 |
+
"epoch": 0.28,
|
| 214 |
+
"grad_norm": 2.6661097024380993,
|
| 215 |
+
"learning_rate": 4.519644235671752e-07,
|
| 216 |
+
"logits/chosen": -2.858572244644165,
|
| 217 |
+
"logits/rejected": -2.7656137943267822,
|
| 218 |
+
"logps/chosen": -341.40325927734375,
|
| 219 |
+
"logps/rejected": -263.1445007324219,
|
| 220 |
+
"loss": 0.6709,
|
| 221 |
+
"positive_losses": 0.0,
|
| 222 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 223 |
+
"rewards/chosen": 0.04941926524043083,
|
| 224 |
+
"rewards/margins": 0.042630504816770554,
|
| 225 |
+
"rewards/margins_max": 0.08383534103631973,
|
| 226 |
+
"rewards/margins_min": 0.00657398859038949,
|
| 227 |
+
"rewards/margins_std": 0.03647482395172119,
|
| 228 |
+
"rewards/rejected": 0.00678876182064414,
|
| 229 |
+
"step": 100
|
| 230 |
+
},
|
| 231 |
+
{
|
| 232 |
+
"epoch": 0.28,
|
| 233 |
+
"eval_dpo_losses": 0.6899906396865845,
|
| 234 |
+
"eval_logits/chosen": -2.799083709716797,
|
| 235 |
+
"eval_logits/rejected": -2.7607357501983643,
|
| 236 |
+
"eval_logps/chosen": -283.0979919433594,
|
| 237 |
+
"eval_logps/rejected": -257.7293395996094,
|
| 238 |
+
"eval_loss": 0.7044196128845215,
|
| 239 |
+
"eval_positive_losses": 0.13342785835266113,
|
| 240 |
+
"eval_rewards/accuracies": 0.5830000042915344,
|
| 241 |
+
"eval_rewards/chosen": 0.014954007230699062,
|
| 242 |
+
"eval_rewards/margins": 0.0064581576734781265,
|
| 243 |
+
"eval_rewards/margins_max": 0.04341236874461174,
|
| 244 |
+
"eval_rewards/margins_min": -0.024124523624777794,
|
| 245 |
+
"eval_rewards/margins_std": 0.022434763610363007,
|
| 246 |
+
"eval_rewards/rejected": 0.008495849557220936,
|
| 247 |
+
"eval_runtime": 428.6351,
|
| 248 |
+
"eval_samples_per_second": 4.666,
|
| 249 |
+
"eval_steps_per_second": 0.292,
|
| 250 |
+
"step": 100
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"dpo_losses": 0.6648027300834656,
|
| 254 |
+
"epoch": 0.31,
|
| 255 |
+
"grad_norm": 2.4903000527490997,
|
| 256 |
+
"learning_rate": 4.364984038837727e-07,
|
| 257 |
+
"logits/chosen": -2.8712093830108643,
|
| 258 |
+
"logits/rejected": -2.7597901821136475,
|
| 259 |
+
"logps/chosen": -384.63873291015625,
|
| 260 |
+
"logps/rejected": -286.42755126953125,
|
| 261 |
+
"loss": 0.6644,
|
| 262 |
+
"positive_losses": 0.0,
|
| 263 |
+
"rewards/accuracies": 0.9375,
|
| 264 |
+
"rewards/chosen": 0.06723800301551819,
|
| 265 |
+
"rewards/margins": 0.05809164047241211,
|
| 266 |
+
"rewards/margins_max": 0.11163143068552017,
|
| 267 |
+
"rewards/margins_min": 0.012343182228505611,
|
| 268 |
+
"rewards/margins_std": 0.04587184637784958,
|
| 269 |
+
"rewards/rejected": 0.009146357886493206,
|
| 270 |
+
"step": 110
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"dpo_losses": 0.6599195599555969,
|
| 274 |
+
"epoch": 0.34,
|
| 275 |
+
"grad_norm": 2.0272270859977986,
|
| 276 |
+
"learning_rate": 4.1922503338800447e-07,
|
| 277 |
+
"logits/chosen": -2.8636832237243652,
|
| 278 |
+
"logits/rejected": -2.788536548614502,
|
| 279 |
+
"logps/chosen": -386.91357421875,
|
| 280 |
+
"logps/rejected": -265.55084228515625,
|
| 281 |
+
"loss": 0.6624,
|
| 282 |
+
"positive_losses": 0.0009745836141519248,
|
| 283 |
+
"rewards/accuracies": 0.949999988079071,
|
| 284 |
+
"rewards/chosen": 0.07831867039203644,
|
| 285 |
+
"rewards/margins": 0.0684511810541153,
|
| 286 |
+
"rewards/margins_max": 0.13835951685905457,
|
| 287 |
+
"rewards/margins_min": 0.014530710875988007,
|
| 288 |
+
"rewards/margins_std": 0.057229261845350266,
|
| 289 |
+
"rewards/rejected": 0.00986748468130827,
|
| 290 |
+
"step": 120
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"dpo_losses": 0.6510322093963623,
|
| 294 |
+
"epoch": 0.37,
|
| 295 |
+
"grad_norm": 1.9824231979290536,
|
| 296 |
+
"learning_rate": 4.003117078299021e-07,
|
| 297 |
+
"logits/chosen": -2.8207156658172607,
|
| 298 |
+
"logits/rejected": -2.7434334754943848,
|
| 299 |
+
"logps/chosen": -395.66802978515625,
|
| 300 |
+
"logps/rejected": -300.08343505859375,
|
| 301 |
+
"loss": 0.6531,
|
| 302 |
+
"positive_losses": 0.0,
|
| 303 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 304 |
+
"rewards/chosen": 0.09558561444282532,
|
| 305 |
+
"rewards/margins": 0.08716461062431335,
|
| 306 |
+
"rewards/margins_max": 0.1589736044406891,
|
| 307 |
+
"rewards/margins_min": 0.021759631112217903,
|
| 308 |
+
"rewards/margins_std": 0.06302757561206818,
|
| 309 |
+
"rewards/rejected": 0.008421008475124836,
|
| 310 |
+
"step": 130
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"dpo_losses": 0.6501553654670715,
|
| 314 |
+
"epoch": 0.39,
|
| 315 |
+
"grad_norm": 1.7119399665574373,
|
| 316 |
+
"learning_rate": 3.799417157181075e-07,
|
| 317 |
+
"logits/chosen": -2.795180082321167,
|
| 318 |
+
"logits/rejected": -2.7392399311065674,
|
| 319 |
+
"logps/chosen": -363.2869567871094,
|
| 320 |
+
"logps/rejected": -269.78350830078125,
|
| 321 |
+
"loss": 0.6548,
|
| 322 |
+
"positive_losses": 0.0016069173580035567,
|
| 323 |
+
"rewards/accuracies": 0.9624999761581421,
|
| 324 |
+
"rewards/chosen": 0.09409914910793304,
|
| 325 |
+
"rewards/margins": 0.08933846652507782,
|
| 326 |
+
"rewards/margins_max": 0.172917902469635,
|
| 327 |
+
"rewards/margins_min": 0.012987839989364147,
|
| 328 |
+
"rewards/margins_std": 0.07409818470478058,
|
| 329 |
+
"rewards/rejected": 0.004760689102113247,
|
| 330 |
+
"step": 140
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"dpo_losses": 0.6526910662651062,
|
| 334 |
+
"epoch": 0.42,
|
| 335 |
+
"grad_norm": 2.0809425241432886,
|
| 336 |
+
"learning_rate": 3.583124620760659e-07,
|
| 337 |
+
"logits/chosen": -2.8274710178375244,
|
| 338 |
+
"logits/rejected": -2.7291388511657715,
|
| 339 |
+
"logps/chosen": -313.82757568359375,
|
| 340 |
+
"logps/rejected": -213.179443359375,
|
| 341 |
+
"loss": 0.6516,
|
| 342 |
+
"positive_losses": 0.0,
|
| 343 |
+
"rewards/accuracies": 0.9624999761581421,
|
| 344 |
+
"rewards/chosen": 0.09023541212081909,
|
| 345 |
+
"rewards/margins": 0.08422856032848358,
|
| 346 |
+
"rewards/margins_max": 0.17488643527030945,
|
| 347 |
+
"rewards/margins_min": 0.01652991771697998,
|
| 348 |
+
"rewards/margins_std": 0.07335906475782394,
|
| 349 |
+
"rewards/rejected": 0.006006857845932245,
|
| 350 |
+
"step": 150
|
| 351 |
+
},
|
| 352 |
+
{
|
| 353 |
+
"dpo_losses": 0.6424547433853149,
|
| 354 |
+
"epoch": 0.45,
|
| 355 |
+
"grad_norm": 1.7536600530591888,
|
| 356 |
+
"learning_rate": 3.356335553954679e-07,
|
| 357 |
+
"logits/chosen": -2.7450592517852783,
|
| 358 |
+
"logits/rejected": -2.685469627380371,
|
| 359 |
+
"logps/chosen": -334.85797119140625,
|
| 360 |
+
"logps/rejected": -234.3365020751953,
|
| 361 |
+
"loss": 0.6447,
|
| 362 |
+
"positive_losses": 0.005527782253921032,
|
| 363 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 364 |
+
"rewards/chosen": 0.10652638971805573,
|
| 365 |
+
"rewards/margins": 0.10596431791782379,
|
| 366 |
+
"rewards/margins_max": 0.20612521469593048,
|
| 367 |
+
"rewards/margins_min": 0.020822737365961075,
|
| 368 |
+
"rewards/margins_std": 0.08536245673894882,
|
| 369 |
+
"rewards/rejected": 0.0005620642914436758,
|
| 370 |
+
"step": 160
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"dpo_losses": 0.6378639936447144,
|
| 374 |
+
"epoch": 0.48,
|
| 375 |
+
"grad_norm": 6.472916445339736,
|
| 376 |
+
"learning_rate": 3.121247763262235e-07,
|
| 377 |
+
"logits/chosen": -2.82389497756958,
|
| 378 |
+
"logits/rejected": -2.7423415184020996,
|
| 379 |
+
"logps/chosen": -363.29315185546875,
|
| 380 |
+
"logps/rejected": -295.8577575683594,
|
| 381 |
+
"loss": 0.6445,
|
| 382 |
+
"positive_losses": 0.00014963149442337453,
|
| 383 |
+
"rewards/accuracies": 0.8999999761581421,
|
| 384 |
+
"rewards/chosen": 0.12112192809581757,
|
| 385 |
+
"rewards/margins": 0.11697308719158173,
|
| 386 |
+
"rewards/margins_max": 0.24103912711143494,
|
| 387 |
+
"rewards/margins_min": 0.013107486069202423,
|
| 388 |
+
"rewards/margins_std": 0.10543470084667206,
|
| 389 |
+
"rewards/rejected": 0.00414883391931653,
|
| 390 |
+
"step": 170
|
| 391 |
+
},
|
| 392 |
+
{
|
| 393 |
+
"dpo_losses": 0.6460212469100952,
|
| 394 |
+
"epoch": 0.51,
|
| 395 |
+
"grad_norm": 2.067078450412541,
|
| 396 |
+
"learning_rate": 2.880139477883347e-07,
|
| 397 |
+
"logits/chosen": -2.793970823287964,
|
| 398 |
+
"logits/rejected": -2.705854892730713,
|
| 399 |
+
"logps/chosen": -337.114501953125,
|
| 400 |
+
"logps/rejected": -292.83624267578125,
|
| 401 |
+
"loss": 0.6403,
|
| 402 |
+
"positive_losses": 0.0,
|
| 403 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 404 |
+
"rewards/chosen": 0.10858919471502304,
|
| 405 |
+
"rewards/margins": 0.098782978951931,
|
| 406 |
+
"rewards/margins_max": 0.19024236500263214,
|
| 407 |
+
"rewards/margins_min": 0.0153144970536232,
|
| 408 |
+
"rewards/margins_std": 0.07897777855396271,
|
| 409 |
+
"rewards/rejected": 0.009806220419704914,
|
| 410 |
+
"step": 180
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"dpo_losses": 0.634928286075592,
|
| 414 |
+
"epoch": 0.54,
|
| 415 |
+
"grad_norm": 2.5683292981841586,
|
| 416 |
+
"learning_rate": 2.635347271463544e-07,
|
| 417 |
+
"logits/chosen": -2.793365001678467,
|
| 418 |
+
"logits/rejected": -2.6589195728302,
|
| 419 |
+
"logps/chosen": -346.8443908691406,
|
| 420 |
+
"logps/rejected": -237.68563842773438,
|
| 421 |
+
"loss": 0.6365,
|
| 422 |
+
"positive_losses": 0.0,
|
| 423 |
+
"rewards/accuracies": 0.987500011920929,
|
| 424 |
+
"rewards/chosen": 0.12526604533195496,
|
| 425 |
+
"rewards/margins": 0.12293746322393417,
|
| 426 |
+
"rewards/margins_max": 0.2294788807630539,
|
| 427 |
+
"rewards/margins_min": 0.023495811969041824,
|
| 428 |
+
"rewards/margins_std": 0.09407813102006912,
|
| 429 |
+
"rewards/rejected": 0.002328577684238553,
|
| 430 |
+
"step": 190
|
| 431 |
+
},
|
| 432 |
+
{
|
| 433 |
+
"dpo_losses": 0.6293594837188721,
|
| 434 |
+
"epoch": 0.56,
|
| 435 |
+
"grad_norm": 6.199913442684288,
|
| 436 |
+
"learning_rate": 2.3892434184240534e-07,
|
| 437 |
+
"logits/chosen": -2.862529993057251,
|
| 438 |
+
"logits/rejected": -2.7567837238311768,
|
| 439 |
+
"logps/chosen": -384.57025146484375,
|
| 440 |
+
"logps/rejected": -265.1017761230469,
|
| 441 |
+
"loss": 0.6375,
|
| 442 |
+
"positive_losses": 0.026654411107301712,
|
| 443 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 444 |
+
"rewards/chosen": 0.14424578845500946,
|
| 445 |
+
"rewards/margins": 0.13493841886520386,
|
| 446 |
+
"rewards/margins_max": 0.2611827552318573,
|
| 447 |
+
"rewards/margins_min": 0.03425807133316994,
|
| 448 |
+
"rewards/margins_std": 0.10205696523189545,
|
| 449 |
+
"rewards/rejected": 0.009307395666837692,
|
| 450 |
+
"step": 200
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"epoch": 0.56,
|
| 454 |
+
"eval_dpo_losses": 0.6862263083457947,
|
| 455 |
+
"eval_logits/chosen": -2.7717671394348145,
|
| 456 |
+
"eval_logits/rejected": -2.7334673404693604,
|
| 457 |
+
"eval_logps/chosen": -282.3363037109375,
|
| 458 |
+
"eval_logps/rejected": -257.78778076171875,
|
| 459 |
+
"eval_loss": 0.7538697123527527,
|
| 460 |
+
"eval_positive_losses": 0.6397833228111267,
|
| 461 |
+
"eval_rewards/accuracies": 0.5879999995231628,
|
| 462 |
+
"eval_rewards/chosen": 0.022570965811610222,
|
| 463 |
+
"eval_rewards/margins": 0.0146592753008008,
|
| 464 |
+
"eval_rewards/margins_max": 0.10217105597257614,
|
| 465 |
+
"eval_rewards/margins_min": -0.05958281084895134,
|
| 466 |
+
"eval_rewards/margins_std": 0.0534750372171402,
|
| 467 |
+
"eval_rewards/rejected": 0.007911693304777145,
|
| 468 |
+
"eval_runtime": 427.9807,
|
| 469 |
+
"eval_samples_per_second": 4.673,
|
| 470 |
+
"eval_steps_per_second": 0.292,
|
| 471 |
+
"step": 200
|
| 472 |
+
},
|
| 473 |
+
{
|
| 474 |
+
"dpo_losses": 0.6285057067871094,
|
| 475 |
+
"epoch": 0.59,
|
| 476 |
+
"grad_norm": 2.0332452218189423,
|
| 477 |
+
"learning_rate": 2.1442129043167873e-07,
|
| 478 |
+
"logits/chosen": -2.756385326385498,
|
| 479 |
+
"logits/rejected": -2.685586929321289,
|
| 480 |
+
"logps/chosen": -341.33837890625,
|
| 481 |
+
"logps/rejected": -256.91766357421875,
|
| 482 |
+
"loss": 0.6311,
|
| 483 |
+
"positive_losses": 0.0025708198081701994,
|
| 484 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 485 |
+
"rewards/chosen": 0.13986453413963318,
|
| 486 |
+
"rewards/margins": 0.13733358681201935,
|
| 487 |
+
"rewards/margins_max": 0.27369070053100586,
|
| 488 |
+
"rewards/margins_min": 0.027926146984100342,
|
| 489 |
+
"rewards/margins_std": 0.11671511083841324,
|
| 490 |
+
"rewards/rejected": 0.0025309394113719463,
|
| 491 |
+
"step": 210
|
| 492 |
+
},
|
| 493 |
+
{
|
| 494 |
+
"dpo_losses": 0.621319055557251,
|
| 495 |
+
"epoch": 0.62,
|
| 496 |
+
"grad_norm": 2.027345866825123,
|
| 497 |
+
"learning_rate": 1.9026303129961048e-07,
|
| 498 |
+
"logits/chosen": -2.856189727783203,
|
| 499 |
+
"logits/rejected": -2.73331880569458,
|
| 500 |
+
"logps/chosen": -391.1002197265625,
|
| 501 |
+
"logps/rejected": -274.0777893066406,
|
| 502 |
+
"loss": 0.627,
|
| 503 |
+
"positive_losses": 0.0,
|
| 504 |
+
"rewards/accuracies": 0.9375,
|
| 505 |
+
"rewards/chosen": 0.15911123156547546,
|
| 506 |
+
"rewards/margins": 0.1526704728603363,
|
| 507 |
+
"rewards/margins_max": 0.2787471413612366,
|
| 508 |
+
"rewards/margins_min": 0.029059389606118202,
|
| 509 |
+
"rewards/margins_std": 0.11583348363637924,
|
| 510 |
+
"rewards/rejected": 0.006440767087042332,
|
| 511 |
+
"step": 220
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"dpo_losses": 0.6225648522377014,
|
| 515 |
+
"epoch": 0.65,
|
| 516 |
+
"grad_norm": 2.1898711296381363,
|
| 517 |
+
"learning_rate": 1.6668368145931396e-07,
|
| 518 |
+
"logits/chosen": -2.8812718391418457,
|
| 519 |
+
"logits/rejected": -2.7513253688812256,
|
| 520 |
+
"logps/chosen": -387.0869140625,
|
| 521 |
+
"logps/rejected": -262.6962585449219,
|
| 522 |
+
"loss": 0.6199,
|
| 523 |
+
"positive_losses": 0.0,
|
| 524 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 525 |
+
"rewards/chosen": 0.1572650671005249,
|
| 526 |
+
"rewards/margins": 0.15020257234573364,
|
| 527 |
+
"rewards/margins_max": 0.27498555183410645,
|
| 528 |
+
"rewards/margins_min": 0.032824836671352386,
|
| 529 |
+
"rewards/margins_std": 0.11001060903072357,
|
| 530 |
+
"rewards/rejected": 0.007062485907226801,
|
| 531 |
+
"step": 230
|
| 532 |
+
},
|
| 533 |
+
{
|
| 534 |
+
"dpo_losses": 0.636019766330719,
|
| 535 |
+
"epoch": 0.68,
|
| 536 |
+
"grad_norm": 1.8191230391360291,
|
| 537 |
+
"learning_rate": 1.4391174773015834e-07,
|
| 538 |
+
"logits/chosen": -2.807938814163208,
|
| 539 |
+
"logits/rejected": -2.716116428375244,
|
| 540 |
+
"logps/chosen": -329.6391296386719,
|
| 541 |
+
"logps/rejected": -283.71624755859375,
|
| 542 |
+
"loss": 0.6338,
|
| 543 |
+
"positive_losses": 0.016074705868959427,
|
| 544 |
+
"rewards/accuracies": 0.949999988079071,
|
| 545 |
+
"rewards/chosen": 0.13542965054512024,
|
| 546 |
+
"rewards/margins": 0.12056276947259903,
|
| 547 |
+
"rewards/margins_max": 0.23423099517822266,
|
| 548 |
+
"rewards/margins_min": 0.021957222372293472,
|
| 549 |
+
"rewards/margins_std": 0.0950225442647934,
|
| 550 |
+
"rewards/rejected": 0.01486687920987606,
|
| 551 |
+
"step": 240
|
| 552 |
+
},
|
| 553 |
+
{
|
| 554 |
+
"dpo_losses": 0.6204310655593872,
|
| 555 |
+
"epoch": 0.7,
|
| 556 |
+
"grad_norm": 1.9142563702239555,
|
| 557 |
+
"learning_rate": 1.2216791228457775e-07,
|
| 558 |
+
"logits/chosen": -2.8029227256774902,
|
| 559 |
+
"logits/rejected": -2.6856589317321777,
|
| 560 |
+
"logps/chosen": -348.2162780761719,
|
| 561 |
+
"logps/rejected": -252.9592742919922,
|
| 562 |
+
"loss": 0.6225,
|
| 563 |
+
"positive_losses": 0.0,
|
| 564 |
+
"rewards/accuracies": 0.987500011920929,
|
| 565 |
+
"rewards/chosen": 0.15677711367607117,
|
| 566 |
+
"rewards/margins": 0.15525582432746887,
|
| 567 |
+
"rewards/margins_max": 0.30013835430145264,
|
| 568 |
+
"rewards/margins_min": 0.039689190685749054,
|
| 569 |
+
"rewards/margins_std": 0.11695368587970734,
|
| 570 |
+
"rewards/rejected": 0.0015213176375254989,
|
| 571 |
+
"step": 250
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"dpo_losses": 0.6270495653152466,
|
| 575 |
+
"epoch": 0.73,
|
| 576 |
+
"grad_norm": 5.620542158536291,
|
| 577 |
+
"learning_rate": 1.0166289402331391e-07,
|
| 578 |
+
"logits/chosen": -2.8551700115203857,
|
| 579 |
+
"logits/rejected": -2.744173765182495,
|
| 580 |
+
"logps/chosen": -341.2341003417969,
|
| 581 |
+
"logps/rejected": -258.6510925292969,
|
| 582 |
+
"loss": 0.6278,
|
| 583 |
+
"positive_losses": 0.016833018511533737,
|
| 584 |
+
"rewards/accuracies": 0.987500011920929,
|
| 585 |
+
"rewards/chosen": 0.15211349725723267,
|
| 586 |
+
"rewards/margins": 0.14102734625339508,
|
| 587 |
+
"rewards/margins_max": 0.29794004559516907,
|
| 588 |
+
"rewards/margins_min": 0.023187965154647827,
|
| 589 |
+
"rewards/margins_std": 0.12820349633693695,
|
| 590 |
+
"rewards/rejected": 0.011086158454418182,
|
| 591 |
+
"step": 260
|
| 592 |
+
},
|
| 593 |
+
{
|
| 594 |
+
"dpo_losses": 0.6170064210891724,
|
| 595 |
+
"epoch": 0.76,
|
| 596 |
+
"grad_norm": 2.1039206735574796,
|
| 597 |
+
"learning_rate": 8.259540650444734e-08,
|
| 598 |
+
"logits/chosen": -2.807176351547241,
|
| 599 |
+
"logits/rejected": -2.7167766094207764,
|
| 600 |
+
"logps/chosen": -361.86981201171875,
|
| 601 |
+
"logps/rejected": -262.83673095703125,
|
| 602 |
+
"loss": 0.6263,
|
| 603 |
+
"positive_losses": 0.006254339125007391,
|
| 604 |
+
"rewards/accuracies": 0.949999988079071,
|
| 605 |
+
"rewards/chosen": 0.16030336916446686,
|
| 606 |
+
"rewards/margins": 0.16262248158454895,
|
| 607 |
+
"rewards/margins_max": 0.2933579683303833,
|
| 608 |
+
"rewards/margins_min": 0.03731368109583855,
|
| 609 |
+
"rewards/margins_std": 0.11614950001239777,
|
| 610 |
+
"rewards/rejected": -0.002319112652912736,
|
| 611 |
+
"step": 270
|
| 612 |
+
},
|
| 613 |
+
{
|
| 614 |
+
"dpo_losses": 0.6323626041412354,
|
| 615 |
+
"epoch": 0.79,
|
| 616 |
+
"grad_norm": 3.6286773445196094,
|
| 617 |
+
"learning_rate": 6.515023221586721e-08,
|
| 618 |
+
"logits/chosen": -2.756715774536133,
|
| 619 |
+
"logits/rejected": -2.709307909011841,
|
| 620 |
+
"logps/chosen": -316.50714111328125,
|
| 621 |
+
"logps/rejected": -272.3856506347656,
|
| 622 |
+
"loss": 0.6292,
|
| 623 |
+
"positive_losses": 0.029051829129457474,
|
| 624 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 625 |
+
"rewards/chosen": 0.1414763480424881,
|
| 626 |
+
"rewards/margins": 0.12991073727607727,
|
| 627 |
+
"rewards/margins_max": 0.2901129424571991,
|
| 628 |
+
"rewards/margins_min": 0.026102622970938683,
|
| 629 |
+
"rewards/margins_std": 0.12171909958124161,
|
| 630 |
+
"rewards/rejected": 0.01156560517847538,
|
| 631 |
+
"step": 280
|
| 632 |
+
},
|
| 633 |
+
{
|
| 634 |
+
"dpo_losses": 0.6301154494285583,
|
| 635 |
+
"epoch": 0.82,
|
| 636 |
+
"grad_norm": 2.1273936086157175,
|
| 637 |
+
"learning_rate": 4.949643185335287e-08,
|
| 638 |
+
"logits/chosen": -2.7689366340637207,
|
| 639 |
+
"logits/rejected": -2.6886589527130127,
|
| 640 |
+
"logps/chosen": -326.63336181640625,
|
| 641 |
+
"logps/rejected": -265.2763671875,
|
| 642 |
+
"loss": 0.6315,
|
| 643 |
+
"positive_losses": 0.0,
|
| 644 |
+
"rewards/accuracies": 0.949999988079071,
|
| 645 |
+
"rewards/chosen": 0.14358708262443542,
|
| 646 |
+
"rewards/margins": 0.134388267993927,
|
| 647 |
+
"rewards/margins_max": 0.2748602032661438,
|
| 648 |
+
"rewards/margins_min": 0.020886730402708054,
|
| 649 |
+
"rewards/margins_std": 0.11660180240869522,
|
| 650 |
+
"rewards/rejected": 0.009198823012411594,
|
| 651 |
+
"step": 290
|
| 652 |
+
},
|
| 653 |
+
{
|
| 654 |
+
"dpo_losses": 0.6222444772720337,
|
| 655 |
+
"epoch": 0.85,
|
| 656 |
+
"grad_norm": 3.916824188852417,
|
| 657 |
+
"learning_rate": 3.578570595810274e-08,
|
| 658 |
+
"logits/chosen": -2.813533306121826,
|
| 659 |
+
"logits/rejected": -2.7393226623535156,
|
| 660 |
+
"logps/chosen": -347.9246520996094,
|
| 661 |
+
"logps/rejected": -288.87091064453125,
|
| 662 |
+
"loss": 0.6203,
|
| 663 |
+
"positive_losses": 0.004223608877509832,
|
| 664 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 665 |
+
"rewards/chosen": 0.15709543228149414,
|
| 666 |
+
"rewards/margins": 0.15124014019966125,
|
| 667 |
+
"rewards/margins_max": 0.2926793098449707,
|
| 668 |
+
"rewards/margins_min": 0.03863045573234558,
|
| 669 |
+
"rewards/margins_std": 0.11517532914876938,
|
| 670 |
+
"rewards/rejected": 0.005855279043316841,
|
| 671 |
+
"step": 300
|
| 672 |
+
},
|
| 673 |
+
{
|
| 674 |
+
"epoch": 0.85,
|
| 675 |
+
"eval_dpo_losses": 0.6847304105758667,
|
| 676 |
+
"eval_logits/chosen": -2.769089937210083,
|
| 677 |
+
"eval_logits/rejected": -2.7314584255218506,
|
| 678 |
+
"eval_logps/chosen": -282.0690612792969,
|
| 679 |
+
"eval_logps/rejected": -257.8544921875,
|
| 680 |
+
"eval_loss": 0.7727298140525818,
|
| 681 |
+
"eval_positive_losses": 0.8306741118431091,
|
| 682 |
+
"eval_rewards/accuracies": 0.5920000076293945,
|
| 683 |
+
"eval_rewards/chosen": 0.02524358406662941,
|
| 684 |
+
"eval_rewards/margins": 0.017999326810240746,
|
| 685 |
+
"eval_rewards/margins_max": 0.12274405360221863,
|
| 686 |
+
"eval_rewards/margins_min": -0.07009908556938171,
|
| 687 |
+
"eval_rewards/margins_std": 0.06381046026945114,
|
| 688 |
+
"eval_rewards/rejected": 0.007244254928082228,
|
| 689 |
+
"eval_runtime": 428.2098,
|
| 690 |
+
"eval_samples_per_second": 4.671,
|
| 691 |
+
"eval_steps_per_second": 0.292,
|
| 692 |
+
"step": 300
|
| 693 |
+
},
|
| 694 |
+
{
|
| 695 |
+
"dpo_losses": 0.6221127510070801,
|
| 696 |
+
"epoch": 0.87,
|
| 697 |
+
"grad_norm": 1.8941985477237735,
|
| 698 |
+
"learning_rate": 2.415092479103503e-08,
|
| 699 |
+
"logits/chosen": -2.8478198051452637,
|
| 700 |
+
"logits/rejected": -2.7160096168518066,
|
| 701 |
+
"logps/chosen": -340.7911071777344,
|
| 702 |
+
"logps/rejected": -214.91964721679688,
|
| 703 |
+
"loss": 0.6237,
|
| 704 |
+
"positive_losses": 0.0,
|
| 705 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 706 |
+
"rewards/chosen": 0.154912531375885,
|
| 707 |
+
"rewards/margins": 0.15173590183258057,
|
| 708 |
+
"rewards/margins_max": 0.29650163650512695,
|
| 709 |
+
"rewards/margins_min": 0.04189513251185417,
|
| 710 |
+
"rewards/margins_std": 0.12011243402957916,
|
| 711 |
+
"rewards/rejected": 0.0031766362953931093,
|
| 712 |
+
"step": 310
|
| 713 |
+
},
|
| 714 |
+
{
|
| 715 |
+
"dpo_losses": 0.6243400573730469,
|
| 716 |
+
"epoch": 0.9,
|
| 717 |
+
"grad_norm": 1.80973934530786,
|
| 718 |
+
"learning_rate": 1.4704840690808656e-08,
|
| 719 |
+
"logits/chosen": -2.8038718700408936,
|
| 720 |
+
"logits/rejected": -2.719196319580078,
|
| 721 |
+
"logps/chosen": -334.8002014160156,
|
| 722 |
+
"logps/rejected": -260.89532470703125,
|
| 723 |
+
"loss": 0.6212,
|
| 724 |
+
"positive_losses": 0.0,
|
| 725 |
+
"rewards/accuracies": 0.9375,
|
| 726 |
+
"rewards/chosen": 0.15841911733150482,
|
| 727 |
+
"rewards/margins": 0.1470610648393631,
|
| 728 |
+
"rewards/margins_max": 0.30652207136154175,
|
| 729 |
+
"rewards/margins_min": 0.019394170492887497,
|
| 730 |
+
"rewards/margins_std": 0.13127803802490234,
|
| 731 |
+
"rewards/rejected": 0.011358057148754597,
|
| 732 |
+
"step": 320
|
| 733 |
+
},
|
| 734 |
+
{
|
| 735 |
+
"dpo_losses": 0.6155394315719604,
|
| 736 |
+
"epoch": 0.93,
|
| 737 |
+
"grad_norm": 1.9914736197963372,
|
| 738 |
+
"learning_rate": 7.538995394063995e-09,
|
| 739 |
+
"logits/chosen": -2.8728480339050293,
|
| 740 |
+
"logits/rejected": -2.768115282058716,
|
| 741 |
+
"logps/chosen": -383.182373046875,
|
| 742 |
+
"logps/rejected": -267.9715881347656,
|
| 743 |
+
"loss": 0.6235,
|
| 744 |
+
"positive_losses": 0.017542457208037376,
|
| 745 |
+
"rewards/accuracies": 0.9375,
|
| 746 |
+
"rewards/chosen": 0.17095208168029785,
|
| 747 |
+
"rewards/margins": 0.16700315475463867,
|
| 748 |
+
"rewards/margins_max": 0.31099098920822144,
|
| 749 |
+
"rewards/margins_min": 0.040303539484739304,
|
| 750 |
+
"rewards/margins_std": 0.12561064958572388,
|
| 751 |
+
"rewards/rejected": 0.003948924597352743,
|
| 752 |
+
"step": 330
|
| 753 |
+
},
|
| 754 |
+
{
|
| 755 |
+
"dpo_losses": 0.6262079477310181,
|
| 756 |
+
"epoch": 0.96,
|
| 757 |
+
"grad_norm": 2.0098982490102606,
|
| 758 |
+
"learning_rate": 2.7228329070159705e-09,
|
| 759 |
+
"logits/chosen": -2.7688167095184326,
|
| 760 |
+
"logits/rejected": -2.681577205657959,
|
| 761 |
+
"logps/chosen": -329.7587890625,
|
| 762 |
+
"logps/rejected": -251.1042022705078,
|
| 763 |
+
"loss": 0.6217,
|
| 764 |
+
"positive_losses": 0.0,
|
| 765 |
+
"rewards/accuracies": 0.925000011920929,
|
| 766 |
+
"rewards/chosen": 0.1526114046573639,
|
| 767 |
+
"rewards/margins": 0.14340537786483765,
|
| 768 |
+
"rewards/margins_max": 0.28913384675979614,
|
| 769 |
+
"rewards/margins_min": 0.027879271656274796,
|
| 770 |
+
"rewards/margins_std": 0.1190972700715065,
|
| 771 |
+
"rewards/rejected": 0.009206023998558521,
|
| 772 |
+
"step": 340
|
| 773 |
+
},
|
| 774 |
+
{
|
| 775 |
+
"dpo_losses": 0.6229046583175659,
|
| 776 |
+
"epoch": 0.99,
|
| 777 |
+
"grad_norm": 3.4990163510778087,
|
| 778 |
+
"learning_rate": 3.0302652553296226e-10,
|
| 779 |
+
"logits/chosen": -2.761152744293213,
|
| 780 |
+
"logits/rejected": -2.6877024173736572,
|
| 781 |
+
"logps/chosen": -344.25860595703125,
|
| 782 |
+
"logps/rejected": -286.8030090332031,
|
| 783 |
+
"loss": 0.6221,
|
| 784 |
+
"positive_losses": 0.0,
|
| 785 |
+
"rewards/accuracies": 0.9624999761581421,
|
| 786 |
+
"rewards/chosen": 0.15005084872245789,
|
| 787 |
+
"rewards/margins": 0.1498393714427948,
|
| 788 |
+
"rewards/margins_max": 0.2942337691783905,
|
| 789 |
+
"rewards/margins_min": 0.04408826678991318,
|
| 790 |
+
"rewards/margins_std": 0.11335112154483795,
|
| 791 |
+
"rewards/rejected": 0.00021145511709619313,
|
| 792 |
+
"step": 350
|
| 793 |
+
},
|
| 794 |
+
{
|
| 795 |
+
"epoch": 1.0,
|
| 796 |
+
"step": 355,
|
| 797 |
+
"total_flos": 0.0,
|
| 798 |
+
"train_loss": 0.6494339942932129,
|
| 799 |
+
"train_runtime": 4013.1211,
|
| 800 |
+
"train_samples_per_second": 1.415,
|
| 801 |
+
"train_steps_per_second": 0.088
|
| 802 |
+
}
|
| 803 |
+
],
|
| 804 |
+
"logging_steps": 10,
|
| 805 |
+
"max_steps": 355,
|
| 806 |
+
"num_input_tokens_seen": 0,
|
| 807 |
+
"num_train_epochs": 1,
|
| 808 |
+
"save_steps": 100,
|
| 809 |
+
"total_flos": 0.0,
|
| 810 |
+
"train_batch_size": 4,
|
| 811 |
+
"trial_name": null,
|
| 812 |
+
"trial_params": null
|
| 813 |
+
}
|