m2m100-ky-en

This model is a fine-tuned version of facebook/m2m100_418M on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.1425
• Chrf++: 42.9292
• Bleu: 0.1388

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10
• num_epochs: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Chrf++	Bleu
3.453	0.4639	90	1.4171	31.5406	0.0619
0.7793	0.9278	180	1.2463	36.7296	0.0912
0.5573	1.3918	270	1.2096	38.6167	0.1059
0.506	1.8557	360	1.1481	39.9577	0.1186
0.3964	2.3196	450	1.1479	40.8569	0.1258
0.3543	2.7835	540	1.1372	41.3424	0.1261
0.2886	3.2474	630	1.1387	42.2531	0.1352
0.2554	3.7113	720	1.1344	42.0227	0.1348
0.2321	4.1753	810	1.1319	42.2876	0.1360
0.1921	4.6392	900	1.1365	42.7091	0.1374
0.1856	5.1031	990	1.1437	42.5069	0.1374
0.1533	5.5670	1080	1.1425	42.9292	0.1388

Framework versions

• Transformers 4.55.4
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.21.4