Kira-Floris/TranslateGemma-4B

sft_stage2_lr

This model is a fine-tuned version of saves/translategemma3-4b/sft on the expert_en_rw_gemma__train dataset. It achieves the following results on the evaluation set:

• Loss: 0.3614

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: 1e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 8
• total_train_batch_size: 48
• total_eval_batch_size: 6
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.3937	0.0682	500	0.3737
0.3516	0.1363	1000	0.3694
0.3791	0.2045	1500	0.3690
0.3863	0.2727	2000	0.3664
0.342	0.3409	2500	0.3654
0.3528	0.4090	3000	0.3647
0.3563	0.4772	3500	0.3631
0.3877	0.5454	4000	0.3630
0.3611	0.6136	4500	0.3621
0.4012	0.6817	5000	0.3622
0.3738	0.7499	5500	0.3614
0.4021	0.8181	6000	0.3611
0.378	0.8863	6500	0.3613
0.3636	0.9544	7000	0.3613

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.1+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2