mistral-alpaca-qlora

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the mhenrichsen/alpaca_2k_test dataset. It achieves the following results on the evaluation set:

• Loss: 1.3095

Model description

Standard mistral 7B fine tuned with alpaca format.

Intended uses & limitations

More information needed

Training and evaluation data

mhenrichsen/alpaca_2k_test

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
5.5317	0.07	1	5.2182
5.438	0.2	3	4.7897
4.1476	0.4	6	3.4313
3.2037	0.6	9	2.8663
2.7895	0.8	12	2.5112
2.3139	1.0	15	2.1467
2.1672	1.2	18	1.8620
1.9095	1.4	21	1.6519
1.5397	1.6	24	1.5429
1.6327	1.8	27	1.4518
1.3676	2.0	30	1.3892
1.3906	2.2	33	1.3531
1.4096	2.4	36	1.3314
1.3278	2.6	39	1.3165
1.3007	2.8	42	1.3107
1.2848	3.0	45	1.3095

Framework versions

• Transformers 4.35.2
• Pytorch 2.0.1+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0