gemma-7b-sft-qlora-1

This model is a fine-tuned version of google/gemma-7b on the chansung/no_robots_only_coding dataset. It achieves the following results on the evaluation set:

• Loss: 2.1615

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: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss
23.7344	0.91	5	7.9584
14.6026	2.0	11	6.8289
10.8118	2.91	16	6.4185
10.8598	4.0	22	5.1061
7.9354	4.91	27	1.7011
2.0354	6.0	33	1.4461
1.4855	6.91	38	1.3565
1.326	8.0	44	1.2935
1.1375	8.91	49	1.2696
0.9091	10.0	55	1.2716
0.8111	10.91	60	1.2861
0.689	12.0	66	1.3148
0.6341	12.91	71	1.3391
0.5359	14.0	77	1.4232
0.4664	14.91	82	1.5107
0.3951	16.0	88	1.6597
0.3593	16.91	93	1.9377
0.2802	18.0	99	1.9024
0.2613	18.91	104	2.0981
0.2262	20.0	110	2.1472
0.2169	20.91	115	2.1633
0.2232	22.0	121	2.1595
0.2096	22.73	125	2.1615

Framework versions

• PEFT 0.7.1
• Transformers 4.39.3
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2