MetaReason-SFT
This model is a fine-tuned version of Qwen/Qwen3-VL-8B-Instruct on the MathVRTrain_multimodal_train, the ZKPG_multimodal_train and the MathCanvasInstruct_multimodal_train datasets. It achieves the following results on the evaluation set:
- Loss: 0.2481
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-05
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 4
- total_train_batch_size: 32
- total_eval_batch_size: 8
- optimizer: Use 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: 2.0
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.2922 | 0.2867 | 500 | 0.2858 |
| 0.2697 | 0.5734 | 1000 | 0.2753 |
| 0.2635 | 0.8601 | 1500 | 0.2656 |
| 0.2103 | 1.1468 | 2000 | 0.2603 |
| 0.2022 | 1.4335 | 2500 | 0.2535 |
| 0.2048 | 1.7202 | 3000 | 0.2490 |
Framework versions
- Transformers 4.57.1
- Pytorch 2.6.0+cu124
- Datasets 4.0.0
- Tokenizers 0.22.1