peft-starcoder-lora-a100

This model is a fine-tuned version of bigcode/starcoderbase-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.1085

• 使用ubuntu22+4090 训练步骤2000，训练时长28906秒

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: 0.0005
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 30
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss
1.0146	0.05	100	0.9460
0.9847	0.1	200	0.9616
0.608	0.15	300	0.9858
0.9076	0.2	400	0.9852
0.8904	0.25	500	1.0052
0.5104	0.3	600	1.0349
0.7794	0.35	700	1.0406
0.744	0.4	800	1.0338
0.569	0.45	900	1.0795
0.6755	0.5	1000	1.0739
0.5978	0.55	1100	1.0655
0.749	0.6	1200	1.0841
0.5411	0.65	1300	1.0907
0.7742	0.7	1400	1.0923
0.8629	0.75	1500	1.1010
0.469	0.8	1600	1.1080
0.7893	0.85	1700	1.1040
0.7894	0.9	1800	1.1055
0.4221	0.95	1900	1.1085
0.7565	1.0	2000	1.1085

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0