---
license: apache-2.0
base_model: studio-ousia/luke-japanese-base-lite
tags:
- generated_from_trainer
metrics:
- accuracy
- precision
- recall
- f1
model-index:
- name: out
  results: []
---

## Fine-tuning
- this model was trained to classify whether input text comes from "chosen sentence" or "rejected sentence"
- the probability (logits after passing softmax function) in last layer of this model can be used to quantify the preference from user input
- fine-tuned [studio-ousia/mluke-large-lite](https://huggingface.co/studio-ousia/mluke-large-lite) via full parameter tuning using [open-preference-v0.3](https://huggingface.co/datasets/ryota39/open_preference-v0.3)
- trained on bf16 format
- Label 0 stands for rejected sentence
- Label 1 stands for chosen sentence
- **Note that this model can handle only 512 tokens in maximum**
  - **The limitation arises from Luke-based pre-trained model**

## Metric

- train and validation split

|train loss|eval loss|accuracy|recall|precision|f1-score|
|:---|:---|:---|:---|:---|:---|
|0.1427|0.2009|9282|0.9383|0.9198|0.9290|

- test split

|accuracy|recall|precision|f1-score|
|:---|:---|:---|:---|
|0.9310|0.9199|0.9408|0.9302|

- confusion matrix when test split

![image/png](https://cdn-uploads.huggingface.co/production/uploads/651e3f30ca333f3c8df692b8/sWbpo0Hwp24SmcpvEtMlq.png)

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall | F1     |
|:-------------:|:-----:|:----:|:---------------:|:--------:|:---------:|:------:|:------:|
| 0.316         | 1.0   | 1479 | 0.2245          | 0.9127   | 0.9027    | 0.9251 | 0.9138 |
| 0.1696        | 2.0   | 2958 | 0.1869          | 0.9308   | 0.9234    | 0.9395 | 0.9314 |
| 0.1427        | 3.0   | 4437 | 0.2009          | 0.9283   | 0.9198    | 0.9384 | 0.9290 |


### Framework versions

- Transformers 4.42.3
- Pytorch 2.1.0+cu118
- Datasets 2.20.0
- Tokenizers 0.19.1