--- license: mit --- # MirrorAPI This model is a fine-tuned version of [Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) ### Training and evaluation data The training data of MirrorAPI consists of: - [`train_sft.json`](https://huggingface.co/datasets/stabletoolbench/MirrorAPI/blob/main/train/train_sft.json) - [`train_cot.json`](https://huggingface.co/datasets/stabletoolbench/MirrorAPI/blob/main/train/train_cot.json) - [`train_augment.json`](https://huggingface.co/datasets/stabletoolbench/MirrorAPI/blob/main/train/train_augment.json) The testing data are under [`test_sft`](https://huggingface.co/datasets/stabletoolbench/MirrorAPI/tree/main/test/test_sft) and [`test_cot`](https://huggingface.co/datasets/stabletoolbench/MirrorAPI/tree/main/test/test_cot) for SFT and CoT modes, respectively. ### Training hyperparameters The following hyperparameters were used during training: - learning_rate: 2e-05 - train_batch_size: 2 - eval_batch_size: 2 - seed: 42 - distributed_type: multi-GPU - num_devices: 8 - gradient_accumulation_steps: 8 - 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: 0.04 - lr_scheduler_warmup_steps: 100 - num_epochs: 5.0 ### Framework versions - Transformers 4.44.2 - Pytorch 2.4.1+cu118 - Datasets 2.21.0 - Tokenizers 0.19.1