|
--- |
|
license: apache-2.0 |
|
base_model: nnheui/pythia-1.4b-sft-full |
|
tags: |
|
- alignment-handbook |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
- trl |
|
- dpo |
|
- generated_from_trainer |
|
datasets: |
|
- HuggingFaceH4/ultrafeedback_binarized |
|
model-index: |
|
- name: pythia-1.4b-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. --> |
|
|
|
# pythia-1.4b-dpo-full |
|
|
|
This model is a fine-tuned version of [nnheui/pythia-1.4b-sft-full](https://huggingface.co/nnheui/pythia-1.4b-sft-full) on the HuggingFaceH4/ultrafeedback_binarized dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 0.6257 |
|
- Rewards/chosen: -0.5234 |
|
- Rewards/rejected: -0.7812 |
|
- Rewards/accuracies: 0.6597 |
|
- Rewards/margins: 0.2578 |
|
- Logps/rejected: -416.0 |
|
- Logps/chosen: -446.0 |
|
- Logits/rejected: -1.2422 |
|
- Logits/chosen: -1.1953 |
|
- Logps/chosen Top Tokens: -0.0007 |
|
- Logps/rejected Top Tokens: -0.0007 |
|
- Logps/chosen Bottom Tokens: -14.375 |
|
- Logps/rejected Bottom Tokens: -14.3125 |
|
|
|
## 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: 5 |
|
- eval_batch_size: 5 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 6 |
|
- gradient_accumulation_steps: 4 |
|
- total_train_batch_size: 120 |
|
- total_eval_batch_size: 30 |
|
- 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 | Logps/chosen Top Tokens | Logps/rejected Top Tokens | Logps/chosen Bottom Tokens | Logps/rejected Bottom Tokens | |
|
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:-----------------------:|:-------------------------:|:--------------------------:|:----------------------------:| |
|
| 0.678 | 0.1963 | 100 | 0.6789 | -0.0275 | -0.0608 | 0.5881 | 0.0332 | -344.0 | -396.0 | -1.1562 | -1.0938 | -0.0009 | -0.0009 | -14.0625 | -14.0 | |
|
| 0.645 | 0.3925 | 200 | 0.6489 | -0.2871 | -0.4238 | 0.6448 | 0.1367 | -380.0 | -422.0 | -1.2031 | -1.1562 | -0.0009 | -0.0009 | -14.375 | -14.3125 | |
|
| 0.6396 | 0.5888 | 300 | 0.6304 | -0.4512 | -0.6797 | 0.6627 | 0.2275 | -406.0 | -438.0 | -1.2344 | -1.1875 | -0.0007 | -0.0008 | -14.375 | -14.3125 | |
|
| 0.6102 | 0.7851 | 400 | 0.6268 | -0.5039 | -0.7617 | 0.6567 | 0.2578 | -414.0 | -444.0 | -1.2344 | -1.1875 | -0.0007 | -0.0007 | -14.3125 | -14.25 | |
|
| 0.6084 | 0.9814 | 500 | 0.6259 | -0.5234 | -0.7852 | 0.6567 | 0.2617 | -416.0 | -446.0 | -1.2422 | -1.1953 | -0.0007 | -0.0007 | -14.375 | -14.3125 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.40.0 |
|
- Pytorch 2.2.2+cu121 |
|
- Datasets 2.19.0 |
|
- Tokenizers 0.19.1 |
|
|