b8848b309daf9cbd7251f04ac0a77b0d

This model is a fine-tuned version of distilbert/distilbert-base-german-cased on the nyu-mll/glue [qnli] dataset. It achieves the following results on the evaluation set:

• Loss: 0.5541
• Data Size: 1.0
• Epoch Runtime: 86.5958
• Accuracy: 0.8046
• F1 Macro: 0.8039
• Rouge1: 0.8046
• Rouge2: 0.0
• Rougel: 0.8046
• Rougelsum: 0.8042

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Data Size	Epoch Runtime	Accuracy	F1 Macro	Rouge1	Rouge2	Rougel	Rougelsum
No log	0	0	0.7029	0	2.0318	0.4943	0.3308	0.4947	0.0	0.4943	0.4944
No log	1	3273	0.6051	0.0078	4.0144	0.7233	0.7233	0.7233	0.0	0.7232	0.7230
0.0102	2	6546	0.5370	0.0156	3.6080	0.7546	0.7535	0.7548	0.0	0.7544	0.7542
0.5417	3	9819	0.4824	0.0312	5.0655	0.7704	0.7700	0.7705	0.0	0.7702	0.7702
0.5088	4	13092	0.4726	0.0625	7.9030	0.7774	0.7756	0.7774	0.0	0.7776	0.7772
0.4589	5	16365	0.4344	0.125	13.2498	0.8053	0.8051	0.8055	0.0	0.8054	0.8053
0.4912	6	19638	0.4588	0.25	23.4593	0.7853	0.7831	0.7853	0.0	0.7855	0.7853
0.4348	7	22911	0.4302	0.5	44.0955	0.8050	0.8042	0.8050	0.0	0.8051	0.8048
0.4331	8.0	26184	0.4341	1.0	85.0020	0.8187	0.8185	0.8187	0.0	0.8189	0.8186
0.3468	9.0	29457	0.4354	1.0	90.1068	0.8123	0.8120	0.8121	0.0	0.8121	0.8121
0.2824	10.0	32730	0.4595	1.0	86.3276	0.8173	0.8173	0.8176	0.0	0.8175	0.8173
0.254	11.0	36003	0.5541	1.0	86.5958	0.8046	0.8039	0.8046	0.0	0.8046	0.8042

Framework versions

• Transformers 4.57.0
• Pytorch 2.8.0+cu128
• Datasets 4.3.0
• Tokenizers 0.22.1