santyzenith/Qwen-1-5-4B-UDA-Rules-SFT

temps

This model is a fine-tuned version of Qwen/Qwen1.5-4B-Chat on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2647

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: 0.0002
• train_batch_size: 4
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• lr_scheduler_warmup_steps: 2
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss
2.4933	0.95	9	1.3903
1.1607	2.0	19	0.9786
0.9208	2.95	28	0.6901
0.5404	4.0	38	0.4664
0.3362	4.95	47	0.3509
0.1628	6.0	57	0.2848
0.117	6.95	66	0.2630
0.074	8.0	76	0.2563
0.0727	8.95	85	0.2402
0.0613	10.0	95	0.2536
0.0597	10.95	104	0.2642
0.0508	12.0	114	0.2567
0.0558	12.95	123	0.2476
0.0477	14.0	133	0.2648
0.0537	14.95	142	0.2691
0.0447	16.0	152	0.2629
0.0473	16.95	161	0.2812
0.0421	18.0	171	0.2566
0.0428	18.95	180	0.2647

Framework versions

• PEFT 0.8.2
• Transformers 4.39.3
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1