Alant2000/ruadapt_qwen2.5_1.5B_test

Модель обучена в рамках практического задания по курсу "Практические аспекты обучения LLM": https://github.com/RefalMachine/rcc_msu_llm_course_autumn_2024/tree/main

This model is a fine-tuned version of ruadapt_qwen2.5_1.5B_ext_u48_mean_init on the Alant2000/taiga_cleared dataset. It achieves the following results on the evaluation set:

• Loss: 4.0484
• Accuracy: 0.3071

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 16
• gradient_accumulation_steps: 4
• total_train_batch_size: 1024
• total_eval_batch_size: 256
• optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.0008	1	7.8632	0.1493
4.0503	0.7593	1000	4.0488	0.3070

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.0a0+6ddf5cf85e.nv24.04
• Datasets 2.18.0
• Tokenizers 0.20.3