AA_preference_l0_new_step10_0_40
This model is a fine-tuned version of llava-hf/llava-v1.6-mistral-7b-hf on the AA_preference_l0_new_step10_0_40 dataset. It achieves the following results on the evaluation set:
- Loss: 0.5324
- Rewards/chosen: 1.2557
- Rewards/rejected: -1.3879
- Rewards/accuracies: 0.8073
- Rewards/margins: 2.6437
- Logps/rejected: -236.0845
- Logps/chosen: -247.5159
- Logits/rejected: -2.4707
- Logits/chosen: -2.4992
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: 8
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 4
- total_train_batch_size: 256
- total_eval_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 10
- num_epochs: 3.0
Training results
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5653 | 0.9346 | 50 | 0.5525 | 0.9593 | -0.6801 | 0.7656 | 1.6394 | -229.0063 | -250.4805 | -2.5530 | -2.5697 |
| 0.2953 | 1.8692 | 100 | 0.5387 | 1.1507 | -1.3892 | 0.8229 | 2.5398 | -236.0967 | -248.5665 | -2.5532 | -2.5732 |
| 0.1521 | 2.8037 | 150 | 0.5318 | 1.2562 | -1.3891 | 0.8073 | 2.6453 | -236.0965 | -247.5115 | -2.4707 | -2.4993 |
Framework versions
- Transformers 4.45.2
- Pytorch 2.4.0+cu121
- Datasets 2.21.0
- Tokenizers 0.20.3
- Downloads last month
- 8
Inference API (serverless) does not yet support transformers models for this pipeline type.
Model tree for htlou/mm-interp-AA_preference_l0_new_step10_0_40
Base model
llava-hf/llava-v1.6-mistral-7b-hf