|
--- |
|
license: apache-2.0 |
|
base_model: hZzy/qwen2.5-0.5b-sft-news-IFT |
|
tags: |
|
- trl |
|
- expo |
|
- alignment-handbook |
|
- ndcg |
|
- generated_from_trainer |
|
model-index: |
|
- name: qwen2.5-0.5b-expo-DPO-ES-0.01 |
|
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. --> |
|
|
|
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/zhiyuzha-university-of-florida/huggingface/runs/r1jti62c) |
|
# qwen2.5-0.5b-expo-DPO-ES-0.01 |
|
|
|
This model is a fine-tuned version of [hZzy/qwen2.5-0.5b-sft-news-IFT](https://huggingface.co/hZzy/qwen2.5-0.5b-sft-news-IFT) on an unknown dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.6901 |
|
- Logps: -349.5348 |
|
- Logits: -5.2492 |
|
- Objective: 0.6972 |
|
- Dpo Loss: 0.6972 |
|
- Regularize: 0.6972 |
|
- Ranking Simple: 0.5719 |
|
- Ranking Idealized: 0.8732 |
|
- Ranking Idealized Expo: 0.5321 |
|
- Wo Beta: 31.1765 |
|
|
|
## 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-06 |
|
- train_batch_size: 4 |
|
- eval_batch_size: 4 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 3 |
|
- gradient_accumulation_steps: 12 |
|
- total_train_batch_size: 144 |
|
- total_eval_batch_size: 12 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: cosine |
|
- lr_scheduler_warmup_ratio: 0.1 |
|
- num_epochs: 5 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Dpo Loss | Logits | Logps | Validation Loss | Objective | Ranking Idealized | Ranking Idealized Expo | Ranking Simple | Regularize | Wo Beta | |
|
|:-------------:|:------:|:----:|:--------:|:-------:|:---------:|:---------------:|:---------:|:-----------------:|:----------------------:|:--------------:|:----------:|:-------:| |
|
| 0.6907 | 0.1417 | 50 | 0.6902 | -1.7611 | -99.3821 | 0.6894 | 0.6902 | 0.8732 | 0.5321 | 0.5305 | 0.6902 | 9.3575 | |
|
| 0.6701 | 0.2834 | 100 | 0.6896 | -1.8467 | -153.5716 | 0.6837 | 0.6896 | 0.8732 | 0.5321 | 0.5518 | 0.6896 | 14.1741 | |
|
| 0.637 | 0.4251 | 150 | 0.6765 | -2.4614 | -198.9269 | 0.6723 | 0.6765 | 0.8732 | 0.5321 | 0.5823 | 0.6765 | 17.2098 | |
|
| 0.5833 | 0.5668 | 200 | 0.6780 | -3.3478 | -256.0312 | 0.6729 | 0.6780 | 0.8732 | 0.5321 | 0.5797 | 0.6780 | 21.7109 | |
|
| 0.5439 | 0.7085 | 250 | 0.6858 | -3.6269 | -257.3546 | 0.6781 | 0.6858 | 0.8732 | 0.5321 | 0.5683 | 0.6858 | 22.9139 | |
|
| 0.5077 | 0.8503 | 300 | 0.6685 | -4.6100 | -319.3935 | 0.6640 | 0.6685 | 0.8732 | 0.5321 | 0.5828 | 0.6685 | 23.6506 | |
|
| 0.4786 | 0.9920 | 350 | 0.6867 | -368.7014| -4.9192 | 0.6897 | 0.6897 | 0.6897 | 0.5751 | 0.8732 | 0.5321 | 27.7113 | |
|
| 0.3619 | 1.1337 | 400 | 0.6961 | -392.7018| -5.5801 | 0.6990 | 0.6990 | 0.6990 | 0.5849 | 0.8732 | 0.5321 | 32.0730 | |
|
| 0.3679 | 1.2754 | 450 | 0.6843 | -349.2029| -5.2450 | 0.6953 | 0.6953 | 0.6953 | 0.5885 | 0.8732 | 0.5321 | 31.3199 | |
|
| 0.3662 | 1.4171 | 500 | 0.6858 | -350.7137| -5.8233 | 0.6903 | 0.6903 | 0.6903 | 0.5890 | 0.8732 | 0.5321 | 30.2726 | |
|
| 0.3485 | 1.5588 | 550 | 0.6901 | -349.5348| -5.2492 | 0.6972 | 0.6972 | 0.6972 | 0.5719 | 0.8732 | 0.5321 | 31.1765 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.42.0 |
|
- Pytorch 2.3.0+cu121 |
|
- Datasets 2.19.1 |
|
- Tokenizers 0.19.1 |
|
|
