metadata
library_name: transformers
license: llama3
base_model: meta-llama/Meta-Llama-3-8B-Instruct
tags:
- alignment-handbook
- trl
- simpo
- generated_from_trainer
- trl
- simpo
- generated_from_trainer
datasets:
- yakazimir/llama3-ultrafeedback-armorm
model-index:
- name: llama3instruct_-l5-10-0_3-1e-6-2_best
results: []
llama3instruct_-l5-10-0_3-1e-6-2_best
This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on the yakazimir/llama3-ultrafeedback-armorm dataset. It achieves the following results on the evaluation set:
- Loss: 1.2685
- Rewards/chosen: -8.0711
- Rewards/rejected: -19.7238
- Rewards/accuracies: 0.8675
- Rewards/margins: 11.6527
- Logps/rejected: -1.9724
- Logps/chosen: -0.8071
- Logits/rejected: -1.3327
- Logits/chosen: -1.4140
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: 1e-06
- train_batch_size: 2
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 16
- total_train_batch_size: 128
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 2.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1.4299 | 0.8743 | 400 | 1.4682 | -8.3837 | -17.5861 | 0.8705 | 9.2024 | -1.7586 | -0.8384 | -1.2770 | -1.3300 |
| 0.7858 | 1.7486 | 800 | 1.2716 | -7.9331 | -19.2874 | 0.8614 | 11.3543 | -1.9287 | -0.7933 | -1.2977 | -1.3755 |
Framework versions
- Transformers 4.44.2
- Pytorch 2.2.2+cu121
- Datasets 2.18.0
- Tokenizers 0.19.1