mathpaper

This model is a fine-tuned version of bigcode/starcoderbase-1b on a dataset of 1000 arxiv category theory publications.

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: 16
• eval_batch_size: 16
• seed: 42
• 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
1.3831	0.05	100	1.4326
1.2475	0.1	200	1.4149
1.2937	0.15	300	1.3903
1.3187	0.2	400	1.3723
1.4185	0.25	500	1.3577
1.3816	0.3	600	1.3475
1.324	0.35	700	1.3467
1.3456	0.4	800	1.3347
1.2906	0.45	900	1.3360
1.2916	0.5	1000	1.3315
1.3851	0.55	1100	1.3232
1.1827	0.6	1200	1.3193
1.2704	0.65	1300	1.3180
1.2495	0.7	1400	1.3104
1.2986	0.75	1500	1.3059
1.3759	0.8	1600	1.3005
1.2775	0.85	1700	1.2983
1.2648	0.9	1800	1.2969
1.2247	0.95	1900	1.2961
1.2152	1.0	2000	1.2959

Framework versions

• PEFT 0.13.2
• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1