mbien/gpt-neo-pl-125m

gpt_neo_pl_125M_v2

This model was trained from scratch on the wikipedia 20220720.pl dataset. It achieves the following results on the evaluation set:

• Loss: 3.3862
• Accuracy: 0.4313

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: 1
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
5.9469	0.02	1000	6.5843	0.1435
4.9953	0.05	2000	5.7709	0.1911
4.3754	0.07	3000	5.2624	0.2331
3.9795	0.1	4000	4.8752	0.2731
3.7099	0.12	5000	4.5927	0.3039
3.4747	0.15	6000	4.3942	0.3230
3.343	0.17	7000	4.2879	0.3349
3.2767	0.2	8000	4.1698	0.3459
3.1852	0.22	9000	4.0925	0.3534
3.0871	0.25	10000	4.0239	0.3608
3.0746	0.27	11000	3.9646	0.3664
2.9473	0.3	12000	3.9245	0.3706
2.9737	0.32	13000	3.8742	0.3754
2.9193	0.35	14000	3.8285	0.3796
2.8833	0.37	15000	3.7952	0.3837
2.8533	0.4	16000	3.7616	0.3873
2.8654	0.42	17000	3.7296	0.3907
2.8196	0.44	18000	3.7049	0.3936
2.7883	0.47	19000	3.6786	0.3966
2.747	0.49	20000	3.6488	0.3990
2.7355	0.52	21000	3.6243	0.4021
2.7355	0.54	22000	3.5982	0.4053
2.6999	0.57	23000	3.5765	0.4075
2.7243	0.59	24000	3.5558	0.4101
2.6526	0.62	25000	3.5371	0.4125
2.641	0.64	26000	3.5150	0.4146
2.6602	0.67	27000	3.4971	0.4168
2.644	0.69	28000	3.4812	0.4192
2.6558	0.72	29000	3.4622	0.4215
2.5664	0.74	30000	3.4504	0.4229
2.5669	0.77	31000	3.4376	0.4245
2.5498	0.79	32000	3.4263	0.4263
2.5874	0.82	33000	3.4169	0.4274
2.5555	0.84	34000	3.4067	0.4286
2.5502	0.86	35000	3.3997	0.4298
2.5232	0.89	36000	3.3946	0.4302
2.5369	0.91	37000	3.3898	0.4309
2.5335	0.94	38000	3.3869	0.4313
2.6032	0.96	39000	3.3853	0.4315
2.5244	0.99	40000	3.3850	0.4314

Framework versions

• Transformers 4.22.0.dev0
• Pytorch 1.12.0
• Datasets 2.4.0
• Tokenizers 0.12.1