---
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: arceeai-cpt-sft-dpo-full
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# arceeai-cpt-sft-dpo-full

This model was trained from scratch on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4290
- Rewards/chosen: -2.1400
- Rewards/rejected: -3.3730
- Rewards/accuracies: 0.7680
- Rewards/margins: 1.2330
- Logps/rejected: -667.6970
- Logps/chosen: -599.8088
- Logits/rejected: -3.8772
- Logits/chosen: -3.7995

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-07
- train_batch_size: 4
- eval_batch_size: 2
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1

### Training results

| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.6051        | 0.1   | 100  | 0.5848          | -0.2214        | -0.5948          | 0.6920             | 0.3733          | -389.8707      | -407.9477    | -3.9800         | -3.8777       |
| 0.5134        | 0.21  | 200  | 0.5025          | -1.5935        | -2.5024          | 0.7160             | 0.9089          | -580.6380      | -545.1561    | -3.9154         | -3.8176       |
| 0.4489        | 0.31  | 300  | 0.4614          | -1.5620        | -2.6097          | 0.7760             | 1.0477          | -591.3610      | -542.0072    | -3.7594         | -3.6703       |
| 0.4359        | 0.42  | 400  | 0.4467          | -2.0879        | -3.2160          | 0.7680             | 1.1281          | -651.9947      | -594.5918    | -3.7022         | -3.6221       |
| 0.4271        | 0.52  | 500  | 0.4441          | -2.0549        | -3.2181          | 0.7840             | 1.1631          | -652.2027      | -591.3006    | -3.8189         | -3.7408       |
| 0.4181        | 0.63  | 600  | 0.4366          | -1.9876        | -3.1678          | 0.7760             | 1.1802          | -647.1777      | -584.5698    | -3.7950         | -3.7170       |
| 0.4           | 0.73  | 700  | 0.4317          | -2.1647        | -3.3521          | 0.7640             | 1.1874          | -665.6046      | -602.2762    | -3.8739         | -3.7970       |
| 0.4123        | 0.84  | 800  | 0.4291          | -2.2039        | -3.4491          | 0.7680             | 1.2453          | -675.3075      | -606.1934    | -3.8606         | -3.7827       |
| 0.4394        | 0.94  | 900  | 0.4292          | -2.1325        | -3.3633          | 0.7680             | 1.2308          | -666.7250      | -599.0574    | -3.8777         | -3.8001       |


### Framework versions

- Transformers 4.36.2
- Pytorch 2.1.2+cu121
- Datasets 2.16.1
- Tokenizers 0.15.0