starcoderbase_4096_context_length

This model is a fine-tuned version of bigcode/starcoderbase-7b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8695

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: 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_steps: 30
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss
0.4754	0.05	100	0.5687
0.3079	0.1	200	0.5495
0.1744	0.15	300	0.5829
0.1132	0.2	400	0.6087
0.1177	0.25	500	0.6039
0.0716	0.3	600	0.6470
0.0653	0.35	700	0.6801
0.1156	0.4	800	0.6671
0.0536	0.45	900	0.6707
0.0582	0.5	1000	0.7050
0.0313	0.55	1100	0.7505
0.0301	0.6	1200	0.7616
0.022	0.65	1300	0.7948
0.0241	0.7	1400	0.8004
0.0197	0.75	1500	0.8240
0.0181	0.8	1600	0.8357
0.0175	0.85	1700	0.8638
0.0162	0.9	1800	0.8632
0.0172	0.95	1900	0.8710
0.0152	1.0	2000	0.8695

Framework versions

• Transformers 4.43.3
• Pytorch 2.4.0a0+07cecf4168.nv24.05
• Datasets 2.20.0
• Tokenizers 0.19.1