Qwen2.5-1.5B-Instruct-finetune-ru-news-lora

This model is a fine-tuned version of Qwen/Qwen2.5-1.5B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.5295
• Perplexity: 4.6645

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: 2e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 111
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Perplexity
No log	0	0	1.7086	5.5805
1.5638	1.0	75	1.6235	5.1242
1.6127	2.0	150	1.5856	4.9323
1.6656	3.0	225	1.5689	4.8497
1.6207	4.0	300	1.5578	4.7967
1.5559	5.0	375	1.5510	4.7642
1.5766	6.0	450	1.5463	4.7420
1.5744	7.0	525	1.5428	4.7257
1.5892	8.0	600	1.5401	4.7129
1.4133	9.0	675	1.5378	4.7022
1.6007	10.0	750	1.5360	4.6939
1.6776	11.0	825	1.5345	4.6872
1.4363	12.0	900	1.5332	4.6814
1.3633	13.0	975	1.5323	4.6771
1.4944	14.0	1050	1.5314	4.6730
1.4514	15.0	1125	1.5308	4.6703
1.4892	16.0	1200	1.5303	4.6681
1.3994	17.0	1275	1.5299	4.6664
1.507	18.0	1350	1.5296	4.6651
1.4906	19.0	1425	1.5295	4.6645
1.4982	19.7383	1480	1.5295	4.6645

Framework versions

• PEFT 0.14.0
• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0