mistral-7b-sft-25-1

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.3 on the hZzy/SFT_new_mix_full2 dataset. It achieves the following results on the evaluation set:

• Loss: 1.5649

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: 6
• gradient_accumulation_steps: 8
• total_train_batch_size: 384
• total_eval_batch_size: 48
• 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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.4019	0.0299	5	2.4203
2.3288	0.0597	10	2.1392
2.0023	0.0896	15	1.8836
1.8366	0.1194	20	1.7753
1.7521	0.1493	25	1.7129
1.6941	0.1791	30	1.6754
1.6516	0.2090	35	1.6462
1.6219	0.2388	40	1.6239
1.6066	0.2687	45	1.6098
1.5845	0.2985	50	1.5998
1.5851	0.3284	55	1.5923
1.571	0.3582	60	1.5869
1.5676	0.3881	65	1.5824
1.5602	0.4179	70	1.5788
1.5427	0.4478	75	1.5761
1.5378	0.4776	80	1.5741
1.5366	0.5075	85	1.5723
1.534	0.5373	90	1.5706
1.525	0.5672	95	1.5695
1.5238	0.5970	100	1.5685
1.5254	0.6269	105	1.5680
1.5209	0.6567	110	1.5672
1.5102	0.6866	115	1.5666
1.5199	0.7164	120	1.5660
1.4967	0.7463	125	1.5660
1.5206	0.7761	130	1.5656
1.5152	0.8060	135	1.5656
1.5008	0.8358	140	1.5651
1.5092	0.8657	145	1.5652
1.5064	0.8955	150	1.5652
1.5011	0.9254	155	1.5650
1.5089	0.9552	160	1.5649
1.5137	0.9851	165	1.5649

Framework versions

• Transformers 4.42.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.19.1