library_name, base_model, tags, datasets, model-index
| library_name | base_model | tags | datasets | model-index | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| transformers | W-61/llama-3-8b-base-sft-ultrachat-8xh200 |
|
|
|
llama-3-8b-base-simpo-8xh200
This model is a fine-tuned version of W-61/llama-3-8b-base-sft-ultrachat-8xh200 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:
- Loss: 1.0269
- Rewards/chosen: -2.9880
- Rewards/rejected: -4.0573
- Rewards/accuracies: 0.7379
- Rewards/margins: 1.0692
- Logps/rejected: -2.0286
- Logps/chosen: -1.4940
- Logits/rejected: -0.7413
- Logits/chosen: -0.7546
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: 6e-07
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- total_eval_batch_size: 64
- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 2.1958 | 0.4188 | 200 | 1.0818 | -2.4141 | -3.2133 | 0.6935 | 0.7991 | -1.6066 | -1.2071 | -0.6654 | -0.6648 |
| 2.1186 | 0.8377 | 400 | 1.0269 | -2.9880 | -4.0573 | 0.7379 | 1.0692 | -2.0286 | -1.4940 | -0.7413 | -0.7546 |
Framework versions
- Transformers 4.51.0
- Pytorch 2.3.1+cu121
- Datasets 2.21.0
- Tokenizers 0.21.4