Howard881010/combined_sft_10000_mcq_2epoch

combined_sft_10000_mcq

This model is a fine-tuned version of mistralai/Mistral-Nemo-Instruct-2407 on the combined_10000_mcq dataset. It achieves the following results on the evaluation set:

• Loss: 0.0016

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: 0.0001
• train_batch_size: 12
• eval_batch_size: 12
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 48
• total_eval_batch_size: 48
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
0.0053	0.08	30	0.0054
0.0055	0.16	60	0.0052
0.005	0.24	90	0.0050
0.0049	0.32	120	0.0050
0.0052	0.4	150	0.0052
0.0045	0.48	180	0.0044
0.0036	0.56	210	0.0034
0.0032	0.64	240	0.0030
0.0033	0.72	270	0.0031
0.0031	0.8	300	0.0031
0.0029	0.88	330	0.0026
0.0029	0.96	360	0.0025
0.0023	1.04	390	0.0022
0.0021	1.12	420	0.0022
0.002	1.2	450	0.0020
0.0017	1.28	480	0.0020
0.0017	1.3600	510	0.0020
0.0016	1.44	540	0.0019
0.0017	1.52	570	0.0020
0.0017	1.6	600	0.0018
0.0017	1.6800	630	0.0018
0.0017	1.76	660	0.0017
0.0017	1.8400	690	0.0017
0.0016	1.92	720	0.0016
0.0017	2.0	750	0.0016

Framework versions

• PEFT 0.12.0
• Transformers 4.46.0
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.20.1