h0-1

This model is a fine-tuned version of microsoft/CodeGPT-small-py on hearthstone dataset. GitHub repo. It achieves the following results on the evaluation set:

• Loss: 0.3622
• Exact Match: 0.1970
• Bleu: 0.9193
• Codebleu: 0.7686
• Chrf: 93.5686

Model description

CodeGPT-small-py fine-tuned on HearthStone dataset for 200 epochs

Intended uses & limitations

HearthStone card code synthesis.

Training and evaluation data

See split of hearthstone dataset

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 17
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Exact Match	Bleu	Codebleu	Chrf
0.2482	11.94	1600	0.2828	0.1364	0.9012	0.7012	92.2247
0.0203	23.88	3200	0.2968	0.1970	0.9114	0.7298	93.0236
0.0082	35.82	4800	0.3049	0.1970	0.9125	0.7480	93.1997
0.0049	47.76	6400	0.3190	0.1818	0.9125	0.7526	93.0967
0.0038	59.7	8000	0.3289	0.1818	0.9117	0.7348	93.1293
0.0024	71.64	9600	0.3358	0.1970	0.9142	0.7555	93.0747
0.0022	83.58	11200	0.3379	0.1970	0.9164	0.7642	93.2931
0.0013	95.52	12800	0.3444	0.2121	0.9189	0.7700	93.4456
0.0009	107.46	14400	0.3408	0.1970	0.9188	0.7655	93.4808
0.0006	119.4	16000	0.3522	0.1970	0.9177	0.7510	93.4061
0.0003	131.34	17600	0.3589	0.2121	0.9178	0.7614	93.3980
0.0002	143.28	19200	0.3562	0.2121	0.9179	0.7634	93.5130
0.0002	155.22	20800	0.3624	0.1970	0.9208	0.7699	93.6707
0.0001	167.16	22400	0.3608	0.1970	0.9193	0.7703	93.6082
0.0001	179.1	24000	0.3620	0.1970	0.9190	0.7667	93.5154
0.0001	191.04	25600	0.3622	0.1970	0.9193	0.7686	93.5686

Framework versions

• Transformers 4.24.0
• Pytorch 1.13.0
• Datasets 2.6.1
• Tokenizers 0.13.1