metadata
base_model: JunxiongWang/llama3_0_50_mamba2_sft
tags:
- alignment-handbook
- generated_from_trainer
datasets:
- HuggingFaceH4/ultrafeedback_binarized
- HuggingFaceH4/orca_dpo_pairs
- JunxiongWang/llama3-ultrafeedback-armorm
model-index:
- name: JunxiongWang/Mamba2InLlama_0_50
results: []
Please check here for details.
JunxiongWang/Mamba2InLlama_0_50
This model is a fine-tuned version of [JunxiongWang/llama3_0_50_mamba2_sft] on the HuggingFaceH4/ultrafeedback_binarized, the HuggingFaceH4/orca_dpo_pairs and the JunxiongWang/llama3-ultrafeedback-armorm datasets. It achieves the following results on the evaluation set:
- Loss: 0.4340
- Rewards/chosen: -2.3310
- Rewards/rejected: -4.2908
- Rewards/accuracies: 0.8214
- Rewards/margins: 1.9598
- Logps/rejected: -707.1605
- Logps/chosen: -505.8361
- Logits/rejected: 1.0544
- Logits/chosen: 1.1061
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: 4
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 32
- total_eval_batch_size: 64
- 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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.4605 | 0.4798 | 2000 | 0.4675 | -1.7509 | -3.3086 | 0.8107 | 1.5578 | -608.9371 | -447.8168 | 0.6185 | 0.6654 |
| 0.4475 | 0.9597 | 4000 | 0.4340 | -2.3310 | -4.2908 | 0.8214 | 1.9598 | -707.1605 | -505.8361 | 1.0544 | 1.1061 |
Framework versions
- Transformers 4.43.1
- Pytorch 2.1.1+cu118
- Datasets 2.20.0
- Tokenizers 0.19.1