mistral-7b-sql

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on the Hyper dataset. It achieves the following results on the evaluation set:

• Loss: 1.0316

Intended uses & limitations

This model is intended to be used for Text-to-SQL tasks, such as an interface for querying a database in natural language.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.3669	0.9231	3	2.0030
1.9273	1.8462	6	1.6669
1.5048	2.7692	9	1.4241
0.915	4.0	13	1.2316
0.9884	4.9231	16	1.1354
0.8136	5.8462	19	1.0745
0.684	6.7692	22	1.0422
0.4341	8.0	26	1.0330
0.526	8.9231	29	1.0319
0.3763	9.2308	30	1.0316

Framework versions

• PEFT 0.11.1
• Transformers 4.41.2
• Pytorch 2.1.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1