basho_haiku_gpt2

Generate a Haiku in the style of Matsuo Bashō given an initial word prompt!

This model is a fine-tuned version of gpt2 on a Haiku dataset. It achieves the following results on the evaluation set:

• Loss: 0.7002

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss
0.9095	0.11	100	0.6605
0.6638	0.22	200	0.6303
0.6456	0.33	300	0.6163
0.6421	0.45	400	0.6139
0.6421	0.56	500	0.6069
0.6182	0.67	600	0.5944
0.6277	0.78	700	0.5910
0.6409	0.89	800	0.5878
0.6047	1.0	900	0.5807
0.4944	1.11	1000	0.5847
0.4878	1.23	1100	0.5897
0.4706	1.34	1200	0.5947
0.4829	1.45	1300	0.5866
0.4742	1.56	1400	0.5875
0.4555	1.67	1500	0.5884
0.4713	1.78	1600	0.5890
0.4669	1.9	1700	0.5848
0.475	2.01	1800	0.5838
0.3762	2.12	1900	0.6123
0.3703	2.23	2000	0.6172
0.3772	2.34	2100	0.6118
0.3731	2.45	2200	0.6090
0.3662	2.56	2300	0.6151
0.3894	2.68	2400	0.6132
0.3663	2.79	2500	0.6195
0.368	2.9	2600	0.6163
0.3735	3.01	2700	0.6191
0.3006	3.12	2800	0.6518
0.3071	3.23	2900	0.6603
0.2898	3.34	3000	0.6629
0.2986	3.46	3100	0.6648
0.3107	3.57	3200	0.6558
0.3064	3.68	3300	0.6568
0.3052	3.79	3400	0.6633
0.3069	3.9	3500	0.6626
0.2872	4.01	3600	0.6641
0.2711	4.12	3700	0.6848
0.2584	4.24	3800	0.6944
0.2606	4.35	3900	0.7007
0.2538	4.46	4000	0.7029
0.2481	4.57	4100	0.7014
0.2466	4.68	4200	0.7006
0.25	4.79	4300	0.6990
0.2568	4.91	4400	0.7002

Framework versions

• Transformers 4.38.2
• Pytorch 2.0.1+cu117
• Datasets 2.18.0
• Tokenizers 0.15.2