Breyten/mistral-instruct-dutch-syntax-10000

Mistral-7B-Instruct-v0.1-syntax2023-12-16-21-24

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.1 on a Lassy_small dataset curated for dutch syntax. 10000 samples where used, batch-size 2, runtime 2 epochs. It achieves the following results on the evaluation set:

• Loss: 0.2522

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: 2.5e-05
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.03
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss
0.7075	0.11	500	0.6710
0.3569	0.21	1000	0.4348
0.3458	0.32	1500	0.3517
0.3325	0.42	2000	0.3151
0.3014	0.53	2500	0.2928
0.2304	0.63	3000	0.2817
0.2984	0.74	3500	0.2736
0.2283	0.84	4000	0.2680
0.2399	0.95	4500	0.2640
0.24	1.05	5000	0.2609
0.2039	1.16	5500	0.2588
0.2447	1.26	6000	0.2558
0.2377	1.37	6500	0.2544
0.2399	1.47	7000	0.2544
0.2424	1.58	7500	0.2532
0.2626	1.68	8000	0.2527
0.2346	1.79	8500	0.2524
0.2194	1.89	9000	0.2522
0.2123	2.0	9500	0.2522
0.2618	2.11	10000	0.2522

Framework versions

• PEFT 0.7.1
• Transformers 4.36.1
• Pytorch 2.1.2+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0