|
--- |
|
license: apache-2.0 |
|
base_model: hZzy/qwen2.5-0.5b-sft-news-IFT |
|
tags: |
|
- alignment-handbook |
|
- ndcg |
|
- trl |
|
- expo |
|
- generated_from_trainer |
|
- trl |
|
- expo |
|
- generated_from_trainer |
|
datasets: |
|
- hZzy/train_pairwise |
|
model-index: |
|
- name: qwen2.5-0.5b-expo-DPO-ES-TRY |
|
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/gcfd4lf7) |
|
# qwen2.5-0.5b-expo-DPO-ES-TRY |
|
|
|
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 the hZzy/train_pairwise dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.6811 |
|
- Logps: -89.5089 |
|
- Logits: -2.2697 |
|
- Objective: 0.6619 |
|
- Dpo Loss: 0.6619 |
|
- Regularize: 0.6619 |
|
- Ranking Simple: 0.5735 |
|
- Ranking Idealized: 0.6046 |
|
- Ranking Idealized Expo: 0.5280 |
|
- Dpo Wo Beta: -2.3796 |
|
|
|
## 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: 2 |
|
- eval_batch_size: 2 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 6 |
|
- gradient_accumulation_steps: 6 |
|
- total_train_batch_size: 72 |
|
- 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: 3 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Dpo Loss | Dpo Wo Beta | Logits | Logps | Validation Loss | Objective | Ranking Idealized | Ranking Idealized Expo | Ranking Simple | Regularize | |
|
|:-------------:|:------:|:----:|:--------:|:-----------:|:-------:|:---------:|:---------------:|:---------:|:-----------------:|:----------------------:|:--------------:|:----------:| |
|
| 0.6857 | 0.0709 | 50 | 0.6927 | -1.2807 | -1.9606 | -88.9841 | 0.6914 | 0.6927 | 0.6046 | 0.5280 | 0.5362 | 0.6927 | |
|
| 0.6524 | 0.1417 | 100 | 0.7010 | -1.8911 | -2.0579 | -98.6358 | 0.6922 | 0.7010 | 0.6046 | 0.5280 | 0.5269 | 0.7010 | |
|
| 0.6123 | 0.2126 | 150 | 0.7015 | -2.1166 | -1.9033 | -102.8927 | 0.6967 | 0.7015 | 0.6046 | 0.5280 | 0.5280 | 0.7015 | |
|
| 0.5779 | 0.2834 | 200 | 0.6816 | -2.1417 | -2.0716 | -106.4944 | 0.6794 | 0.6816 | 0.6046 | 0.5280 | 0.5507 | 0.6816 | |
|
| 0.5709 | 0.3543 | 250 | 0.6817 | -2.2676 | -2.2470 | -87.7326 | 0.6883 | 0.6817 | 0.6046 | 0.5280 | 0.5424 | 0.6817 | |
|
| 0.5563 | 0.4251 | 300 | 0.6619 | -2.3796 | -2.2697 | -89.5089 | 0.6811 | 0.6619 | 0.6046 | 0.5280 | 0.5735 | 0.6619 | |
|
| 0.5321 | 0.4960 | 350 | 0.6773 | -2.6295 | -2.3683 | -99.0927 | 0.6926 | 0.6773 | 0.6046 | 0.5280 | 0.5735 | 0.6773 | |
|
| 0.4963 | 0.5668 | 400 | 0.6836 | -2.6913 | -2.2508 | -106.7073 | 0.6914 | 0.6836 | 0.6046 | 0.5280 | 0.5673 | 0.6836 | |
|
| 0.4745 | 0.6377 | 450 | 0.6938 | -105.8669 | -2.2347 | 0.6815 | 0.6815 | 0.6815 | 0.5631 | 0.6046 | 0.5280 | -2.6738 | |
|
| 0.4867 | 0.7085 | 500 | 0.7040 | -105.1848 | -2.2182 | 0.6995 | 0.6995 | 0.6995 | 0.5507 | 0.6046 | 0.5280 | -2.7257 | |
|
| 0.4582 | 0.7794 | 550 | 0.6995 | -102.6643 | -2.3855 | 0.7027 | 0.7027 | 0.7027 | 0.5683 | 0.6046 | 0.5280 | -3.1023 | |
|
| 0.4339 | 0.8503 | 600 | 0.6965 | -103.5456 | -2.4456 | 0.7050 | 0.7050 | 0.7050 | 0.5735 | 0.6046 | 0.5280 | -3.2166 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.42.0 |
|
- Pytorch 2.3.0+cu121 |
|
- Datasets 2.19.1 |
|
- Tokenizers 0.19.1 |
|
|
