Model save
Browse files
README.md
ADDED
|
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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-dpo-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-dpo-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: 0.6862
|
| 22 |
+
- Rewards/chosen: -0.0539
|
| 23 |
+
- Rewards/rejected: -0.0744
|
| 24 |
+
- Rewards/accuracies: 0.5870
|
| 25 |
+
- Rewards/margins: 0.0206
|
| 26 |
+
- Rewards/margins Max: 0.1479
|
| 27 |
+
- Rewards/margins Min: -0.0932
|
| 28 |
+
- Rewards/margins Std: 0.0795
|
| 29 |
+
- Logps/rejected: -266.0218
|
| 30 |
+
- Logps/chosen: -289.9789
|
| 31 |
+
- Logits/rejected: -2.7191
|
| 32 |
+
- Logits/chosen: -2.7573
|
| 33 |
+
|
| 34 |
+
## Model description
|
| 35 |
+
|
| 36 |
+
More information needed
|
| 37 |
+
|
| 38 |
+
## Intended uses & limitations
|
| 39 |
+
|
| 40 |
+
More information needed
|
| 41 |
+
|
| 42 |
+
## Training and evaluation data
|
| 43 |
+
|
| 44 |
+
More information needed
|
| 45 |
+
|
| 46 |
+
## Training procedure
|
| 47 |
+
|
| 48 |
+
### Training hyperparameters
|
| 49 |
+
|
| 50 |
+
The following hyperparameters were used during training:
|
| 51 |
+
- learning_rate: 5e-07
|
| 52 |
+
- train_batch_size: 4
|
| 53 |
+
- eval_batch_size: 8
|
| 54 |
+
- seed: 42
|
| 55 |
+
- distributed_type: multi-GPU
|
| 56 |
+
- num_devices: 2
|
| 57 |
+
- gradient_accumulation_steps: 2
|
| 58 |
+
- total_train_batch_size: 16
|
| 59 |
+
- total_eval_batch_size: 16
|
| 60 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 61 |
+
- lr_scheduler_type: cosine
|
| 62 |
+
- lr_scheduler_warmup_ratio: 0.1
|
| 63 |
+
- num_epochs: 1
|
| 64 |
+
|
| 65 |
+
### Training results
|
| 66 |
+
|
| 67 |
+
| Training Loss | Epoch | Step | Validation Loss | 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 |
|
| 68 |
+
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:-------------------:|:-------------------:|:-------------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
| 69 |
+
| 0.6666 | 0.28 | 100 | 0.6907 | -0.0061 | -0.0128 | 0.5780 | 0.0067 | 0.0489 | -0.0295 | 0.0259 | -259.8633 | -285.2067 | -2.7596 | -2.7978 |
|
| 70 |
+
| 0.6251 | 0.56 | 200 | 0.6876 | -0.0360 | -0.0522 | 0.5870 | 0.0162 | 0.1195 | -0.0752 | 0.0641 | -263.7989 | -288.1948 | -2.7319 | -2.7700 |
|
| 71 |
+
| 0.6029 | 0.85 | 300 | 0.6862 | -0.0539 | -0.0744 | 0.5870 | 0.0206 | 0.1479 | -0.0932 | 0.0795 | -266.0218 | -289.9789 | -2.7191 | -2.7573 |
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
### Framework versions
|
| 75 |
+
|
| 76 |
+
- PEFT 0.7.1
|
| 77 |
+
- Transformers 4.39.0.dev0
|
| 78 |
+
- Pytorch 2.1.2+cu121
|
| 79 |
+
- Datasets 2.14.6
|
| 80 |
+
- 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 sha256:7a4a037c90b4c405b59a5974e15735c71b54ccc9029b823605aad60e65a95c10
|
| 3 |
size 671150064
|
all_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.6394854995566355,
|
| 4 |
+
"train_runtime": 4022.9516,
|
| 5 |
+
"train_samples": 5678,
|
| 6 |
+
"train_samples_per_second": 1.411,
|
| 7 |
+
"train_steps_per_second": 0.088
|
| 8 |
+
}
|
runs/Jul28_07-09-30_notebook-deployment-48-7d9b6c99-khd85/events.out.tfevents.1722150666.notebook-deployment-48-7d9b6c99-khd85.1186379.0
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 sha256:49f4a432632994ade570fdbc2be07835bdd7de53be862412ca172ffb3c6072e8
|
| 3 |
+
size 39947
|
train_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.6394854995566355,
|
| 4 |
+
"train_runtime": 4022.9516,
|
| 5 |
+
"train_samples": 5678,
|
| 6 |
+
"train_samples_per_second": 1.411,
|
| 7 |
+
"train_steps_per_second": 0.088
|
| 8 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,735 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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 |
+
"epoch": 0.0,
|
| 13 |
+
"grad_norm": 2.158629831791895,
|
| 14 |
+
"learning_rate": 1.3888888888888887e-08,
|
| 15 |
+
"logits/chosen": -2.804708957672119,
|
| 16 |
+
"logits/rejected": -2.8150453567504883,
|
| 17 |
+
"logps/chosen": -217.97438049316406,
|
| 18 |
+
"logps/rejected": -216.58865356445312,
|
| 19 |
+
"loss": 0.6931,
|
| 20 |
+
"rewards/accuracies": 0.0,
|
| 21 |
+
"rewards/chosen": 0.0,
|
| 22 |
+
"rewards/margins": 0.0,
|
| 23 |
+
"rewards/margins_max": 0.0,
|
| 24 |
+
"rewards/margins_min": 0.0,
|
| 25 |
+
"rewards/margins_std": 0.0,
|
| 26 |
+
"rewards/rejected": 0.0,
|
| 27 |
+
"step": 1
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"epoch": 0.03,
|
| 31 |
+
"grad_norm": 8.21578821692664,
|
| 32 |
+
"learning_rate": 1.3888888888888888e-07,
|
| 33 |
+
"logits/chosen": -2.8844423294067383,
|
| 34 |
+
"logits/rejected": -2.799159526824951,
|
| 35 |
+
"logps/chosen": -366.7507629394531,
|
| 36 |
+
"logps/rejected": -275.4356384277344,
|
| 37 |
+
"loss": 0.6932,
|
| 38 |
+
"rewards/accuracies": 0.4027777910232544,
|
| 39 |
+
"rewards/chosen": -0.00016000178584363312,
|
| 40 |
+
"rewards/margins": -0.0003212409501429647,
|
| 41 |
+
"rewards/margins_max": 0.002417487557977438,
|
| 42 |
+
"rewards/margins_min": -0.004127700813114643,
|
| 43 |
+
"rewards/margins_std": 0.0029805246740579605,
|
| 44 |
+
"rewards/rejected": 0.00016123917885124683,
|
| 45 |
+
"step": 10
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"epoch": 0.06,
|
| 49 |
+
"grad_norm": 2.0515925664507124,
|
| 50 |
+
"learning_rate": 2.7777777777777776e-07,
|
| 51 |
+
"logits/chosen": -2.74739670753479,
|
| 52 |
+
"logits/rejected": -2.6935513019561768,
|
| 53 |
+
"logps/chosen": -329.11138916015625,
|
| 54 |
+
"logps/rejected": -216.7494659423828,
|
| 55 |
+
"loss": 0.6929,
|
| 56 |
+
"rewards/accuracies": 0.625,
|
| 57 |
+
"rewards/chosen": 0.00035773837589658797,
|
| 58 |
+
"rewards/margins": 0.00041456689359620214,
|
| 59 |
+
"rewards/margins_max": 0.0031081512570381165,
|
| 60 |
+
"rewards/margins_min": -0.002314414829015732,
|
| 61 |
+
"rewards/margins_std": 0.0023837233893573284,
|
| 62 |
+
"rewards/rejected": -5.682848859578371e-05,
|
| 63 |
+
"step": 20
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"epoch": 0.08,
|
| 67 |
+
"grad_norm": 2.310838256933408,
|
| 68 |
+
"learning_rate": 4.1666666666666667e-07,
|
| 69 |
+
"logits/chosen": -2.835951328277588,
|
| 70 |
+
"logits/rejected": -2.7546262741088867,
|
| 71 |
+
"logps/chosen": -329.10211181640625,
|
| 72 |
+
"logps/rejected": -233.05508422851562,
|
| 73 |
+
"loss": 0.6917,
|
| 74 |
+
"rewards/accuracies": 0.75,
|
| 75 |
+
"rewards/chosen": 0.0018310332670807838,
|
| 76 |
+
"rewards/margins": 0.002559047192335129,
|
| 77 |
+
"rewards/margins_max": 0.006475468166172504,
|
| 78 |
+
"rewards/margins_min": -0.0006590075790882111,
|
| 79 |
+
"rewards/margins_std": 0.0032629654742777348,
|
| 80 |
+
"rewards/rejected": -0.0007280135178007185,
|
| 81 |
+
"step": 30
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"epoch": 0.11,
|
| 85 |
+
"grad_norm": 1.9718465521886885,
|
| 86 |
+
"learning_rate": 4.998060489154965e-07,
|
| 87 |
+
"logits/chosen": -2.8140110969543457,
|
| 88 |
+
"logits/rejected": -2.76088285446167,
|
| 89 |
+
"logps/chosen": -285.4794006347656,
|
| 90 |
+
"logps/rejected": -227.7167205810547,
|
| 91 |
+
"loss": 0.6905,
|
| 92 |
+
"rewards/accuracies": 0.800000011920929,
|
| 93 |
+
"rewards/chosen": 0.0035794698633253574,
|
| 94 |
+
"rewards/margins": 0.0046408590860664845,
|
| 95 |
+
"rewards/margins_max": 0.011065036058425903,
|
| 96 |
+
"rewards/margins_min": -0.0008336328901350498,
|
| 97 |
+
"rewards/margins_std": 0.005423419643193483,
|
| 98 |
+
"rewards/rejected": -0.0010613898048177361,
|
| 99 |
+
"step": 40
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"epoch": 0.14,
|
| 103 |
+
"grad_norm": 2.3429157043740894,
|
| 104 |
+
"learning_rate": 4.976275538042932e-07,
|
| 105 |
+
"logits/chosen": -2.813694477081299,
|
| 106 |
+
"logits/rejected": -2.7310705184936523,
|
| 107 |
+
"logps/chosen": -317.00640869140625,
|
| 108 |
+
"logps/rejected": -234.43209838867188,
|
| 109 |
+
"loss": 0.688,
|
| 110 |
+
"rewards/accuracies": 0.887499988079071,
|
| 111 |
+
"rewards/chosen": 0.007789776660501957,
|
| 112 |
+
"rewards/margins": 0.010368594899773598,
|
| 113 |
+
"rewards/margins_max": 0.021328028291463852,
|
| 114 |
+
"rewards/margins_min": 0.0012870692880824208,
|
| 115 |
+
"rewards/margins_std": 0.009302936494350433,
|
| 116 |
+
"rewards/rejected": -0.00257881754077971,
|
| 117 |
+
"step": 50
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"epoch": 0.17,
|
| 121 |
+
"grad_norm": 2.326595034093221,
|
| 122 |
+
"learning_rate": 4.930493069997119e-07,
|
| 123 |
+
"logits/chosen": -2.7512717247009277,
|
| 124 |
+
"logits/rejected": -2.7030184268951416,
|
| 125 |
+
"logps/chosen": -343.24273681640625,
|
| 126 |
+
"logps/rejected": -264.2438049316406,
|
| 127 |
+
"loss": 0.6845,
|
| 128 |
+
"rewards/accuracies": 0.925000011920929,
|
| 129 |
+
"rewards/chosen": 0.015377024188637733,
|
| 130 |
+
"rewards/margins": 0.01811446249485016,
|
| 131 |
+
"rewards/margins_max": 0.03716810420155525,
|
| 132 |
+
"rewards/margins_min": 0.003869078354910016,
|
| 133 |
+
"rewards/margins_std": 0.01510803122073412,
|
| 134 |
+
"rewards/rejected": -0.0027374387718737125,
|
| 135 |
+
"step": 60
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 0.2,
|
| 139 |
+
"grad_norm": 1.802371962995753,
|
| 140 |
+
"learning_rate": 4.861156761634013e-07,
|
| 141 |
+
"logits/chosen": -2.8008124828338623,
|
| 142 |
+
"logits/rejected": -2.7141239643096924,
|
| 143 |
+
"logps/chosen": -360.14227294921875,
|
| 144 |
+
"logps/rejected": -237.1912841796875,
|
| 145 |
+
"loss": 0.6809,
|
| 146 |
+
"rewards/accuracies": 0.9375,
|
| 147 |
+
"rewards/chosen": 0.02196549065411091,
|
| 148 |
+
"rewards/margins": 0.026430394500494003,
|
| 149 |
+
"rewards/margins_max": 0.05226575583219528,
|
| 150 |
+
"rewards/margins_min": 0.005380354821681976,
|
| 151 |
+
"rewards/margins_std": 0.02180148847401142,
|
| 152 |
+
"rewards/rejected": -0.0044649080373346806,
|
| 153 |
+
"step": 70
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"epoch": 0.23,
|
| 157 |
+
"grad_norm": 2.0226092318731266,
|
| 158 |
+
"learning_rate": 4.768938549177392e-07,
|
| 159 |
+
"logits/chosen": -2.842362403869629,
|
| 160 |
+
"logits/rejected": -2.778277635574341,
|
| 161 |
+
"logps/chosen": -329.4476318359375,
|
| 162 |
+
"logps/rejected": -288.3177795410156,
|
| 163 |
+
"loss": 0.6774,
|
| 164 |
+
"rewards/accuracies": 0.9624999761581421,
|
| 165 |
+
"rewards/chosen": 0.02400829829275608,
|
| 166 |
+
"rewards/margins": 0.03225512057542801,
|
| 167 |
+
"rewards/margins_max": 0.06589716672897339,
|
| 168 |
+
"rewards/margins_min": 0.006357196718454361,
|
| 169 |
+
"rewards/margins_std": 0.027716059237718582,
|
| 170 |
+
"rewards/rejected": -0.008246822282671928,
|
| 171 |
+
"step": 80
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"epoch": 0.25,
|
| 175 |
+
"grad_norm": 2.439721586727848,
|
| 176 |
+
"learning_rate": 4.654732116743193e-07,
|
| 177 |
+
"logits/chosen": -2.7840921878814697,
|
| 178 |
+
"logits/rejected": -2.700878620147705,
|
| 179 |
+
"logps/chosen": -336.05194091796875,
|
| 180 |
+
"logps/rejected": -200.1630096435547,
|
| 181 |
+
"loss": 0.672,
|
| 182 |
+
"rewards/accuracies": 0.949999988079071,
|
| 183 |
+
"rewards/chosen": 0.029071927070617676,
|
| 184 |
+
"rewards/margins": 0.04093674570322037,
|
| 185 |
+
"rewards/margins_max": 0.08617201447486877,
|
| 186 |
+
"rewards/margins_min": 0.006826425436884165,
|
| 187 |
+
"rewards/margins_std": 0.0362737737596035,
|
| 188 |
+
"rewards/rejected": -0.011864816769957542,
|
| 189 |
+
"step": 90
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 0.28,
|
| 193 |
+
"grad_norm": 2.2225465127306605,
|
| 194 |
+
"learning_rate": 4.519644235671752e-07,
|
| 195 |
+
"logits/chosen": -2.8582470417022705,
|
| 196 |
+
"logits/rejected": -2.7655489444732666,
|
| 197 |
+
"logps/chosen": -342.58416748046875,
|
| 198 |
+
"logps/rejected": -265.08441162109375,
|
| 199 |
+
"loss": 0.6666,
|
| 200 |
+
"rewards/accuracies": 0.862500011920929,
|
| 201 |
+
"rewards/chosen": 0.037609733641147614,
|
| 202 |
+
"rewards/margins": 0.050220172852277756,
|
| 203 |
+
"rewards/margins_max": 0.10150803625583649,
|
| 204 |
+
"rewards/margins_min": 0.007549063768237829,
|
| 205 |
+
"rewards/margins_std": 0.0440022274851799,
|
| 206 |
+
"rewards/rejected": -0.01261043269187212,
|
| 207 |
+
"step": 100
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"epoch": 0.28,
|
| 211 |
+
"eval_logits/chosen": -2.7978174686431885,
|
| 212 |
+
"eval_logits/rejected": -2.7595677375793457,
|
| 213 |
+
"eval_logps/chosen": -285.2066650390625,
|
| 214 |
+
"eval_logps/rejected": -259.86334228515625,
|
| 215 |
+
"eval_loss": 0.6906961798667908,
|
| 216 |
+
"eval_rewards/accuracies": 0.578000009059906,
|
| 217 |
+
"eval_rewards/chosen": -0.00613220501691103,
|
| 218 |
+
"eval_rewards/margins": 0.006711836438626051,
|
| 219 |
+
"eval_rewards/margins_max": 0.04891812801361084,
|
| 220 |
+
"eval_rewards/margins_min": -0.02950645610690117,
|
| 221 |
+
"eval_rewards/margins_std": 0.025911005213856697,
|
| 222 |
+
"eval_rewards/rejected": -0.012844040989875793,
|
| 223 |
+
"eval_runtime": 428.4446,
|
| 224 |
+
"eval_samples_per_second": 4.668,
|
| 225 |
+
"eval_steps_per_second": 0.292,
|
| 226 |
+
"step": 100
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 0.31,
|
| 230 |
+
"grad_norm": 2.4760394100510057,
|
| 231 |
+
"learning_rate": 4.364984038837727e-07,
|
| 232 |
+
"logits/chosen": -2.8690743446350098,
|
| 233 |
+
"logits/rejected": -2.7577908039093018,
|
| 234 |
+
"logps/chosen": -385.70233154296875,
|
| 235 |
+
"logps/rejected": -288.461669921875,
|
| 236 |
+
"loss": 0.6591,
|
| 237 |
+
"rewards/accuracies": 0.9375,
|
| 238 |
+
"rewards/chosen": 0.0566016249358654,
|
| 239 |
+
"rewards/margins": 0.06779664754867554,
|
| 240 |
+
"rewards/margins_max": 0.13572999835014343,
|
| 241 |
+
"rewards/margins_min": 0.010938728228211403,
|
| 242 |
+
"rewards/margins_std": 0.05764765292406082,
|
| 243 |
+
"rewards/rejected": -0.011195014230906963,
|
| 244 |
+
"step": 110
|
| 245 |
+
},
|
| 246 |
+
{
|
| 247 |
+
"epoch": 0.34,
|
| 248 |
+
"grad_norm": 2.035920276115207,
|
| 249 |
+
"learning_rate": 4.1922503338800447e-07,
|
| 250 |
+
"logits/chosen": -2.8610854148864746,
|
| 251 |
+
"logits/rejected": -2.7858219146728516,
|
| 252 |
+
"logps/chosen": -387.9818115234375,
|
| 253 |
+
"logps/rejected": -267.68585205078125,
|
| 254 |
+
"loss": 0.657,
|
| 255 |
+
"rewards/accuracies": 0.925000011920929,
|
| 256 |
+
"rewards/chosen": 0.06763629615306854,
|
| 257 |
+
"rewards/margins": 0.07911892235279083,
|
| 258 |
+
"rewards/margins_max": 0.16764040291309357,
|
| 259 |
+
"rewards/margins_min": 0.013401249423623085,
|
| 260 |
+
"rewards/margins_std": 0.07113669812679291,
|
| 261 |
+
"rewards/rejected": -0.01148262806236744,
|
| 262 |
+
"step": 120
|
| 263 |
+
},
|
| 264 |
+
{
|
| 265 |
+
"epoch": 0.37,
|
| 266 |
+
"grad_norm": 2.010676971608138,
|
| 267 |
+
"learning_rate": 4.003117078299021e-07,
|
| 268 |
+
"logits/chosen": -2.818753957748413,
|
| 269 |
+
"logits/rejected": -2.741856098175049,
|
| 270 |
+
"logps/chosen": -396.28985595703125,
|
| 271 |
+
"logps/rejected": -302.45050048828125,
|
| 272 |
+
"loss": 0.6454,
|
| 273 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 274 |
+
"rewards/chosen": 0.08936750888824463,
|
| 275 |
+
"rewards/margins": 0.10461701452732086,
|
| 276 |
+
"rewards/margins_max": 0.20179173350334167,
|
| 277 |
+
"rewards/margins_min": 0.02413741685450077,
|
| 278 |
+
"rewards/margins_std": 0.08073713630437851,
|
| 279 |
+
"rewards/rejected": -0.015249502845108509,
|
| 280 |
+
"step": 130
|
| 281 |
+
},
|
| 282 |
+
{
|
| 283 |
+
"epoch": 0.39,
|
| 284 |
+
"grad_norm": 1.7425219980216828,
|
| 285 |
+
"learning_rate": 3.799417157181075e-07,
|
| 286 |
+
"logits/chosen": -2.7920029163360596,
|
| 287 |
+
"logits/rejected": -2.7359843254089355,
|
| 288 |
+
"logps/chosen": -364.29058837890625,
|
| 289 |
+
"logps/rejected": -272.58355712890625,
|
| 290 |
+
"loss": 0.6467,
|
| 291 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 292 |
+
"rewards/chosen": 0.08406248688697815,
|
| 293 |
+
"rewards/margins": 0.10730169713497162,
|
| 294 |
+
"rewards/margins_max": 0.22186696529388428,
|
| 295 |
+
"rewards/margins_min": 0.012349050492048264,
|
| 296 |
+
"rewards/margins_std": 0.09653683751821518,
|
| 297 |
+
"rewards/rejected": -0.02323923259973526,
|
| 298 |
+
"step": 140
|
| 299 |
+
},
|
| 300 |
+
{
|
| 301 |
+
"epoch": 0.42,
|
| 302 |
+
"grad_norm": 2.0933384277297957,
|
| 303 |
+
"learning_rate": 3.583124620760659e-07,
|
| 304 |
+
"logits/chosen": -2.825629711151123,
|
| 305 |
+
"logits/rejected": -2.7282826900482178,
|
| 306 |
+
"logps/chosen": -315.4014892578125,
|
| 307 |
+
"logps/rejected": -216.2842254638672,
|
| 308 |
+
"loss": 0.6435,
|
| 309 |
+
"rewards/accuracies": 0.9375,
|
| 310 |
+
"rewards/chosen": 0.07449642568826675,
|
| 311 |
+
"rewards/margins": 0.09953755140304565,
|
| 312 |
+
"rewards/margins_max": 0.21898682415485382,
|
| 313 |
+
"rewards/margins_min": 0.014027351513504982,
|
| 314 |
+
"rewards/margins_std": 0.09459034353494644,
|
| 315 |
+
"rewards/rejected": -0.0250411219894886,
|
| 316 |
+
"step": 150
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"epoch": 0.45,
|
| 320 |
+
"grad_norm": 1.769669348441601,
|
| 321 |
+
"learning_rate": 3.356335553954679e-07,
|
| 322 |
+
"logits/chosen": -2.74135684967041,
|
| 323 |
+
"logits/rejected": -2.6822197437286377,
|
| 324 |
+
"logps/chosen": -335.69464111328125,
|
| 325 |
+
"logps/rejected": -237.88046264648438,
|
| 326 |
+
"loss": 0.6336,
|
| 327 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 328 |
+
"rewards/chosen": 0.09816019237041473,
|
| 329 |
+
"rewards/margins": 0.1330377608537674,
|
| 330 |
+
"rewards/margins_max": 0.2625694274902344,
|
| 331 |
+
"rewards/margins_min": 0.02169904112815857,
|
| 332 |
+
"rewards/margins_std": 0.1116378903388977,
|
| 333 |
+
"rewards/rejected": -0.03487757220864296,
|
| 334 |
+
"step": 160
|
| 335 |
+
},
|
| 336 |
+
{
|
| 337 |
+
"epoch": 0.48,
|
| 338 |
+
"grad_norm": 1.8260362870579057,
|
| 339 |
+
"learning_rate": 3.121247763262235e-07,
|
| 340 |
+
"logits/chosen": -2.8216443061828613,
|
| 341 |
+
"logits/rejected": -2.7401599884033203,
|
| 342 |
+
"logps/chosen": -364.33587646484375,
|
| 343 |
+
"logps/rejected": -299.15887451171875,
|
| 344 |
+
"loss": 0.635,
|
| 345 |
+
"rewards/accuracies": 0.887499988079071,
|
| 346 |
+
"rewards/chosen": 0.11069444566965103,
|
| 347 |
+
"rewards/margins": 0.1395573914051056,
|
| 348 |
+
"rewards/margins_max": 0.2983313202857971,
|
| 349 |
+
"rewards/margins_min": 0.007289635483175516,
|
| 350 |
+
"rewards/margins_std": 0.13639435172080994,
|
| 351 |
+
"rewards/rejected": -0.02886294387280941,
|
| 352 |
+
"step": 170
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 0.51,
|
| 356 |
+
"grad_norm": 2.082827875491237,
|
| 357 |
+
"learning_rate": 2.880139477883347e-07,
|
| 358 |
+
"logits/chosen": -2.789100408554077,
|
| 359 |
+
"logits/rejected": -2.700629949569702,
|
| 360 |
+
"logps/chosen": -339.28125,
|
| 361 |
+
"logps/rejected": -296.9674377441406,
|
| 362 |
+
"loss": 0.6302,
|
| 363 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 364 |
+
"rewards/chosen": 0.08692200481891632,
|
| 365 |
+
"rewards/margins": 0.11842750012874603,
|
| 366 |
+
"rewards/margins_max": 0.23567883670330048,
|
| 367 |
+
"rewards/margins_min": 0.011810391210019588,
|
| 368 |
+
"rewards/margins_std": 0.10012297332286835,
|
| 369 |
+
"rewards/rejected": -0.03150549530982971,
|
| 370 |
+
"step": 180
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"epoch": 0.54,
|
| 374 |
+
"grad_norm": 2.575609586836664,
|
| 375 |
+
"learning_rate": 2.635347271463544e-07,
|
| 376 |
+
"logits/chosen": -2.787972927093506,
|
| 377 |
+
"logits/rejected": -2.6533846855163574,
|
| 378 |
+
"logps/chosen": -349.08880615234375,
|
| 379 |
+
"logps/rejected": -242.5450897216797,
|
| 380 |
+
"loss": 0.6257,
|
| 381 |
+
"rewards/accuracies": 0.987500011920929,
|
| 382 |
+
"rewards/chosen": 0.10282168537378311,
|
| 383 |
+
"rewards/margins": 0.14908696711063385,
|
| 384 |
+
"rewards/margins_max": 0.28802552819252014,
|
| 385 |
+
"rewards/margins_min": 0.025781046599149704,
|
| 386 |
+
"rewards/margins_std": 0.1190432757139206,
|
| 387 |
+
"rewards/rejected": -0.04626528546214104,
|
| 388 |
+
"step": 190
|
| 389 |
+
},
|
| 390 |
+
{
|
| 391 |
+
"epoch": 0.56,
|
| 392 |
+
"grad_norm": 2.049618345880406,
|
| 393 |
+
"learning_rate": 2.3892434184240534e-07,
|
| 394 |
+
"logits/chosen": -2.857001543045044,
|
| 395 |
+
"logits/rejected": -2.7506966590881348,
|
| 396 |
+
"logps/chosen": -387.255126953125,
|
| 397 |
+
"logps/rejected": -270.194091796875,
|
| 398 |
+
"loss": 0.6251,
|
| 399 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 400 |
+
"rewards/chosen": 0.11739673465490341,
|
| 401 |
+
"rewards/margins": 0.1590125858783722,
|
| 402 |
+
"rewards/margins_max": 0.32226094603538513,
|
| 403 |
+
"rewards/margins_min": 0.03523118048906326,
|
| 404 |
+
"rewards/margins_std": 0.12896928191184998,
|
| 405 |
+
"rewards/rejected": -0.04161586984992027,
|
| 406 |
+
"step": 200
|
| 407 |
+
},
|
| 408 |
+
{
|
| 409 |
+
"epoch": 0.56,
|
| 410 |
+
"eval_logits/chosen": -2.769979953765869,
|
| 411 |
+
"eval_logits/rejected": -2.7318813800811768,
|
| 412 |
+
"eval_logps/chosen": -288.19482421875,
|
| 413 |
+
"eval_logps/rejected": -263.79888916015625,
|
| 414 |
+
"eval_loss": 0.6876310110092163,
|
| 415 |
+
"eval_rewards/accuracies": 0.5870000123977661,
|
| 416 |
+
"eval_rewards/chosen": -0.03601397946476936,
|
| 417 |
+
"eval_rewards/margins": 0.016185704618692398,
|
| 418 |
+
"eval_rewards/margins_max": 0.11952462792396545,
|
| 419 |
+
"eval_rewards/margins_min": -0.07521206140518188,
|
| 420 |
+
"eval_rewards/margins_std": 0.0641048476099968,
|
| 421 |
+
"eval_rewards/rejected": -0.05219968408346176,
|
| 422 |
+
"eval_runtime": 427.8872,
|
| 423 |
+
"eval_samples_per_second": 4.674,
|
| 424 |
+
"eval_steps_per_second": 0.292,
|
| 425 |
+
"step": 200
|
| 426 |
+
},
|
| 427 |
+
{
|
| 428 |
+
"epoch": 0.59,
|
| 429 |
+
"grad_norm": 1.9845466839870691,
|
| 430 |
+
"learning_rate": 2.1442129043167873e-07,
|
| 431 |
+
"logits/chosen": -2.751984119415283,
|
| 432 |
+
"logits/rejected": -2.6815638542175293,
|
| 433 |
+
"logps/chosen": -344.3485412597656,
|
| 434 |
+
"logps/rejected": -262.97393798828125,
|
| 435 |
+
"loss": 0.6188,
|
| 436 |
+
"rewards/accuracies": 0.9750000238418579,
|
| 437 |
+
"rewards/chosen": 0.10976312309503555,
|
| 438 |
+
"rewards/margins": 0.16779468953609467,
|
| 439 |
+
"rewards/margins_max": 0.34073972702026367,
|
| 440 |
+
"rewards/margins_min": 0.03692127764225006,
|
| 441 |
+
"rewards/margins_std": 0.14643600583076477,
|
| 442 |
+
"rewards/rejected": -0.05803157761693001,
|
| 443 |
+
"step": 210
|
| 444 |
+
},
|
| 445 |
+
{
|
| 446 |
+
"epoch": 0.62,
|
| 447 |
+
"grad_norm": 2.0373527988549145,
|
| 448 |
+
"learning_rate": 1.9026303129961048e-07,
|
| 449 |
+
"logits/chosen": -2.8502397537231445,
|
| 450 |
+
"logits/rejected": -2.7268834114074707,
|
| 451 |
+
"logps/chosen": -393.9187927246094,
|
| 452 |
+
"logps/rejected": -280.2196960449219,
|
| 453 |
+
"loss": 0.6142,
|
| 454 |
+
"rewards/accuracies": 0.9375,
|
| 455 |
+
"rewards/chosen": 0.13092514872550964,
|
| 456 |
+
"rewards/margins": 0.18590331077575684,
|
| 457 |
+
"rewards/margins_max": 0.3502216637134552,
|
| 458 |
+
"rewards/margins_min": 0.03089449368417263,
|
| 459 |
+
"rewards/margins_std": 0.14906269311904907,
|
| 460 |
+
"rewards/rejected": -0.0549781434237957,
|
| 461 |
+
"step": 220
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.65,
|
| 465 |
+
"grad_norm": 2.1828985591845402,
|
| 466 |
+
"learning_rate": 1.6668368145931396e-07,
|
| 467 |
+
"logits/chosen": -2.875049114227295,
|
| 468 |
+
"logits/rejected": -2.744711399078369,
|
| 469 |
+
"logps/chosen": -390.4495849609375,
|
| 470 |
+
"logps/rejected": -268.98565673828125,
|
| 471 |
+
"loss": 0.6067,
|
| 472 |
+
"rewards/accuracies": 0.949999988079071,
|
| 473 |
+
"rewards/chosen": 0.12363851070404053,
|
| 474 |
+
"rewards/margins": 0.17946995794773102,
|
| 475 |
+
"rewards/margins_max": 0.34318283200263977,
|
| 476 |
+
"rewards/margins_min": 0.036444298923015594,
|
| 477 |
+
"rewards/margins_std": 0.13844837248325348,
|
| 478 |
+
"rewards/rejected": -0.05583144351840019,
|
| 479 |
+
"step": 230
|
| 480 |
+
},
|
| 481 |
+
{
|
| 482 |
+
"epoch": 0.68,
|
| 483 |
+
"grad_norm": 1.8193698783653034,
|
| 484 |
+
"learning_rate": 1.4391174773015834e-07,
|
| 485 |
+
"logits/chosen": -2.802640199661255,
|
| 486 |
+
"logits/rejected": -2.71109938621521,
|
| 487 |
+
"logps/chosen": -333.38397216796875,
|
| 488 |
+
"logps/rejected": -289.92462158203125,
|
| 489 |
+
"loss": 0.6224,
|
| 490 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 491 |
+
"rewards/chosen": 0.09798085689544678,
|
| 492 |
+
"rewards/margins": 0.14519774913787842,
|
| 493 |
+
"rewards/margins_max": 0.293338418006897,
|
| 494 |
+
"rewards/margins_min": 0.01530275959521532,
|
| 495 |
+
"rewards/margins_std": 0.12239019572734833,
|
| 496 |
+
"rewards/rejected": -0.047216884791851044,
|
| 497 |
+
"step": 240
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"epoch": 0.7,
|
| 501 |
+
"grad_norm": 1.9709337022102749,
|
| 502 |
+
"learning_rate": 1.2216791228457775e-07,
|
| 503 |
+
"logits/chosen": -2.7975411415100098,
|
| 504 |
+
"logits/rejected": -2.6804046630859375,
|
| 505 |
+
"logps/chosen": -351.70257568359375,
|
| 506 |
+
"logps/rejected": -260.0617370605469,
|
| 507 |
+
"loss": 0.6084,
|
| 508 |
+
"rewards/accuracies": 0.9624999761581421,
|
| 509 |
+
"rewards/chosen": 0.12191393226385117,
|
| 510 |
+
"rewards/margins": 0.1914171278476715,
|
| 511 |
+
"rewards/margins_max": 0.36713144183158875,
|
| 512 |
+
"rewards/margins_min": 0.05136305093765259,
|
| 513 |
+
"rewards/margins_std": 0.142560213804245,
|
| 514 |
+
"rewards/rejected": -0.06950321048498154,
|
| 515 |
+
"step": 250
|
| 516 |
+
},
|
| 517 |
+
{
|
| 518 |
+
"epoch": 0.73,
|
| 519 |
+
"grad_norm": 1.752772587188972,
|
| 520 |
+
"learning_rate": 1.0166289402331391e-07,
|
| 521 |
+
"logits/chosen": -2.8487606048583984,
|
| 522 |
+
"logits/rejected": -2.737738847732544,
|
| 523 |
+
"logps/chosen": -345.0237731933594,
|
| 524 |
+
"logps/rejected": -265.47198486328125,
|
| 525 |
+
"loss": 0.6074,
|
| 526 |
+
"rewards/accuracies": 0.9375,
|
| 527 |
+
"rewards/chosen": 0.11421672999858856,
|
| 528 |
+
"rewards/margins": 0.17133933305740356,
|
| 529 |
+
"rewards/margins_max": 0.3663348853588104,
|
| 530 |
+
"rewards/margins_min": 0.02202555350959301,
|
| 531 |
+
"rewards/margins_std": 0.15875253081321716,
|
| 532 |
+
"rewards/rejected": -0.057122599333524704,
|
| 533 |
+
"step": 260
|
| 534 |
+
},
|
| 535 |
+
{
|
| 536 |
+
"epoch": 0.76,
|
| 537 |
+
"grad_norm": 2.056956615608033,
|
| 538 |
+
"learning_rate": 8.259540650444734e-08,
|
| 539 |
+
"logits/chosen": -2.8006067276000977,
|
| 540 |
+
"logits/rejected": -2.7100348472595215,
|
| 541 |
+
"logps/chosen": -365.325927734375,
|
| 542 |
+
"logps/rejected": -270.2814636230469,
|
| 543 |
+
"loss": 0.6098,
|
| 544 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 545 |
+
"rewards/chosen": 0.1257423460483551,
|
| 546 |
+
"rewards/margins": 0.20250901579856873,
|
| 547 |
+
"rewards/margins_max": 0.3718946874141693,
|
| 548 |
+
"rewards/margins_min": 0.04439568892121315,
|
| 549 |
+
"rewards/margins_std": 0.1491011530160904,
|
| 550 |
+
"rewards/rejected": -0.07676666229963303,
|
| 551 |
+
"step": 270
|
| 552 |
+
},
|
| 553 |
+
{
|
| 554 |
+
"epoch": 0.79,
|
| 555 |
+
"grad_norm": 1.9417182779069821,
|
| 556 |
+
"learning_rate": 6.515023221586721e-08,
|
| 557 |
+
"logits/chosen": -2.7494287490844727,
|
| 558 |
+
"logits/rejected": -2.7017343044281006,
|
| 559 |
+
"logps/chosen": -320.38360595703125,
|
| 560 |
+
"logps/rejected": -279.5456848144531,
|
| 561 |
+
"loss": 0.6125,
|
| 562 |
+
"rewards/accuracies": 0.8999999761581421,
|
| 563 |
+
"rewards/chosen": 0.10271165519952774,
|
| 564 |
+
"rewards/margins": 0.16274654865264893,
|
| 565 |
+
"rewards/margins_max": 0.3613172173500061,
|
| 566 |
+
"rewards/margins_min": 0.03712720423936844,
|
| 567 |
+
"rewards/margins_std": 0.14939478039741516,
|
| 568 |
+
"rewards/rejected": -0.060034893453121185,
|
| 569 |
+
"step": 280
|
| 570 |
+
},
|
| 571 |
+
{
|
| 572 |
+
"epoch": 0.82,
|
| 573 |
+
"grad_norm": 2.159880856830845,
|
| 574 |
+
"learning_rate": 4.949643185335287e-08,
|
| 575 |
+
"logits/chosen": -2.7616562843322754,
|
| 576 |
+
"logits/rejected": -2.6814732551574707,
|
| 577 |
+
"logps/chosen": -331.0811462402344,
|
| 578 |
+
"logps/rejected": -272.906982421875,
|
| 579 |
+
"loss": 0.6168,
|
| 580 |
+
"rewards/accuracies": 0.9375,
|
| 581 |
+
"rewards/chosen": 0.0991094559431076,
|
| 582 |
+
"rewards/margins": 0.16621682047843933,
|
| 583 |
+
"rewards/margins_max": 0.34492212533950806,
|
| 584 |
+
"rewards/margins_min": 0.02063518390059471,
|
| 585 |
+
"rewards/margins_std": 0.14825591444969177,
|
| 586 |
+
"rewards/rejected": -0.06710737198591232,
|
| 587 |
+
"step": 290
|
| 588 |
+
},
|
| 589 |
+
{
|
| 590 |
+
"epoch": 0.85,
|
| 591 |
+
"grad_norm": 2.247267229121733,
|
| 592 |
+
"learning_rate": 3.578570595810274e-08,
|
| 593 |
+
"logits/chosen": -2.805422306060791,
|
| 594 |
+
"logits/rejected": -2.7308857440948486,
|
| 595 |
+
"logps/chosen": -351.537109375,
|
| 596 |
+
"logps/rejected": -296.57861328125,
|
| 597 |
+
"loss": 0.6029,
|
| 598 |
+
"rewards/accuracies": 0.9375,
|
| 599 |
+
"rewards/chosen": 0.12097059190273285,
|
| 600 |
+
"rewards/margins": 0.19219207763671875,
|
| 601 |
+
"rewards/margins_max": 0.3688820004463196,
|
| 602 |
+
"rewards/margins_min": 0.05189325660467148,
|
| 603 |
+
"rewards/margins_std": 0.14433594048023224,
|
| 604 |
+
"rewards/rejected": -0.07122147083282471,
|
| 605 |
+
"step": 300
|
| 606 |
+
},
|
| 607 |
+
{
|
| 608 |
+
"epoch": 0.85,
|
| 609 |
+
"eval_logits/chosen": -2.757275342941284,
|
| 610 |
+
"eval_logits/rejected": -2.7190775871276855,
|
| 611 |
+
"eval_logps/chosen": -289.97894287109375,
|
| 612 |
+
"eval_logps/rejected": -266.02178955078125,
|
| 613 |
+
"eval_loss": 0.6861926913261414,
|
| 614 |
+
"eval_rewards/accuracies": 0.5870000123977661,
|
| 615 |
+
"eval_rewards/chosen": -0.05385516211390495,
|
| 616 |
+
"eval_rewards/margins": 0.020573224872350693,
|
| 617 |
+
"eval_rewards/margins_max": 0.14790384471416473,
|
| 618 |
+
"eval_rewards/margins_min": -0.09322728216648102,
|
| 619 |
+
"eval_rewards/margins_std": 0.079450324177742,
|
| 620 |
+
"eval_rewards/rejected": -0.07442838698625565,
|
| 621 |
+
"eval_runtime": 427.9454,
|
| 622 |
+
"eval_samples_per_second": 4.673,
|
| 623 |
+
"eval_steps_per_second": 0.292,
|
| 624 |
+
"step": 300
|
| 625 |
+
},
|
| 626 |
+
{
|
| 627 |
+
"epoch": 0.87,
|
| 628 |
+
"grad_norm": 1.8974226134430692,
|
| 629 |
+
"learning_rate": 2.415092479103503e-08,
|
| 630 |
+
"logits/chosen": -2.840935230255127,
|
| 631 |
+
"logits/rejected": -2.709672212600708,
|
| 632 |
+
"logps/chosen": -345.2643737792969,
|
| 633 |
+
"logps/rejected": -222.6641082763672,
|
| 634 |
+
"loss": 0.6093,
|
| 635 |
+
"rewards/accuracies": 0.9624999761581421,
|
| 636 |
+
"rewards/chosen": 0.1101798266172409,
|
| 637 |
+
"rewards/margins": 0.1844477355480194,
|
| 638 |
+
"rewards/margins_max": 0.3727789521217346,
|
| 639 |
+
"rewards/margins_min": 0.046926215291023254,
|
| 640 |
+
"rewards/margins_std": 0.1535920351743698,
|
| 641 |
+
"rewards/rejected": -0.0742679089307785,
|
| 642 |
+
"step": 310
|
| 643 |
+
},
|
| 644 |
+
{
|
| 645 |
+
"epoch": 0.9,
|
| 646 |
+
"grad_norm": 1.7695491697233519,
|
| 647 |
+
"learning_rate": 1.4704840690808656e-08,
|
| 648 |
+
"logits/chosen": -2.796245813369751,
|
| 649 |
+
"logits/rejected": -2.7119815349578857,
|
| 650 |
+
"logps/chosen": -339.24664306640625,
|
| 651 |
+
"logps/rejected": -268.58984375,
|
| 652 |
+
"loss": 0.6037,
|
| 653 |
+
"rewards/accuracies": 0.949999988079071,
|
| 654 |
+
"rewards/chosen": 0.1139553040266037,
|
| 655 |
+
"rewards/margins": 0.17954252660274506,
|
| 656 |
+
"rewards/margins_max": 0.37754225730895996,
|
| 657 |
+
"rewards/margins_min": 0.0253077894449234,
|
| 658 |
+
"rewards/margins_std": 0.16321782767772675,
|
| 659 |
+
"rewards/rejected": -0.06558724492788315,
|
| 660 |
+
"step": 320
|
| 661 |
+
},
|
| 662 |
+
{
|
| 663 |
+
"epoch": 0.93,
|
| 664 |
+
"grad_norm": 2.0645664151522136,
|
| 665 |
+
"learning_rate": 7.538995394063995e-09,
|
| 666 |
+
"logits/chosen": -2.8658013343811035,
|
| 667 |
+
"logits/rejected": -2.760768175125122,
|
| 668 |
+
"logps/chosen": -386.96258544921875,
|
| 669 |
+
"logps/rejected": -275.399658203125,
|
| 670 |
+
"loss": 0.6073,
|
| 671 |
+
"rewards/accuracies": 0.887499988079071,
|
| 672 |
+
"rewards/chosen": 0.13314954936504364,
|
| 673 |
+
"rewards/margins": 0.20348164439201355,
|
| 674 |
+
"rewards/margins_max": 0.3888625502586365,
|
| 675 |
+
"rewards/margins_min": 0.05053550750017166,
|
| 676 |
+
"rewards/margins_std": 0.15889115631580353,
|
| 677 |
+
"rewards/rejected": -0.0703321173787117,
|
| 678 |
+
"step": 330
|
| 679 |
+
},
|
| 680 |
+
{
|
| 681 |
+
"epoch": 0.96,
|
| 682 |
+
"grad_norm": 2.0081843762801617,
|
| 683 |
+
"learning_rate": 2.7228329070159705e-09,
|
| 684 |
+
"logits/chosen": -2.7621803283691406,
|
| 685 |
+
"logits/rejected": -2.6747400760650635,
|
| 686 |
+
"logps/chosen": -334.4164123535156,
|
| 687 |
+
"logps/rejected": -258.71417236328125,
|
| 688 |
+
"loss": 0.607,
|
| 689 |
+
"rewards/accuracies": 0.8999999761581421,
|
| 690 |
+
"rewards/chosen": 0.10603541135787964,
|
| 691 |
+
"rewards/margins": 0.17292913794517517,
|
| 692 |
+
"rewards/margins_max": 0.366424024105072,
|
| 693 |
+
"rewards/margins_min": 0.027338892221450806,
|
| 694 |
+
"rewards/margins_std": 0.15261869132518768,
|
| 695 |
+
"rewards/rejected": -0.06689374148845673,
|
| 696 |
+
"step": 340
|
| 697 |
+
},
|
| 698 |
+
{
|
| 699 |
+
"epoch": 0.99,
|
| 700 |
+
"grad_norm": 3.396556816184061,
|
| 701 |
+
"learning_rate": 3.0302652553296226e-10,
|
| 702 |
+
"logits/chosen": -2.754178285598755,
|
| 703 |
+
"logits/rejected": -2.6804542541503906,
|
| 704 |
+
"logps/chosen": -348.5409851074219,
|
| 705 |
+
"logps/rejected": -294.7231750488281,
|
| 706 |
+
"loss": 0.6046,
|
| 707 |
+
"rewards/accuracies": 0.9624999761581421,
|
| 708 |
+
"rewards/chosen": 0.10722661018371582,
|
| 709 |
+
"rewards/margins": 0.18621695041656494,
|
| 710 |
+
"rewards/margins_max": 0.36825960874557495,
|
| 711 |
+
"rewards/margins_min": 0.045198000967502594,
|
| 712 |
+
"rewards/margins_std": 0.14424237608909607,
|
| 713 |
+
"rewards/rejected": -0.07899035513401031,
|
| 714 |
+
"step": 350
|
| 715 |
+
},
|
| 716 |
+
{
|
| 717 |
+
"epoch": 1.0,
|
| 718 |
+
"step": 355,
|
| 719 |
+
"total_flos": 0.0,
|
| 720 |
+
"train_loss": 0.6394854995566355,
|
| 721 |
+
"train_runtime": 4022.9516,
|
| 722 |
+
"train_samples_per_second": 1.411,
|
| 723 |
+
"train_steps_per_second": 0.088
|
| 724 |
+
}
|
| 725 |
+
],
|
| 726 |
+
"logging_steps": 10,
|
| 727 |
+
"max_steps": 355,
|
| 728 |
+
"num_input_tokens_seen": 0,
|
| 729 |
+
"num_train_epochs": 1,
|
| 730 |
+
"save_steps": 100,
|
| 731 |
+
"total_flos": 0.0,
|
| 732 |
+
"train_batch_size": 4,
|
| 733 |
+
"trial_name": null,
|
| 734 |
+
"trial_params": null
|
| 735 |
+
}
|