bart-large-wikilarge

This model is a fine-tuned version of facebook/bart-large on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9051
• Sari: 37.0457
• Paper Sari: 37.1345
• Sari Add: 4.1443
• Sari Keep: 77.5177
• Sari Del: 29.475
• Paper Sari Add: 4.1443
• Paper Sari Keep: 77.7084
• Paper Sari Del: 29.5508
• Fkgl: 8.4619
• Bleu: 92.6286

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.06
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Sari	Paper Sari	Sari Add	Sari Keep	Sari Del	Paper Sari Add	Paper Sari Keep	Paper Sari Del	Fkgl	Bleu
2.0874	0.1080	1000	0.9401	31.9299	32.0978	0.5451	77.7799	17.4648	0.5456	78.1089	17.6388	8.868	96.2302
1.5465	0.2159	2000	0.8906	31.4483	31.5849	0.7191	78.6496	14.9763	0.7195	78.9791	15.056	9.0121	96.5849
1.4814	0.3239	3000	0.8681	31.2134	31.3383	0.9158	79.1129	13.6116	0.9163	79.4448	13.6539	9.1436	97.0903
1.4484	0.4318	4000	0.8677	31.4137	31.5424	0.9679	79.2607	14.0123	0.9683	79.5972	14.0619	9.133	97.3217
1.4237	0.5398	5000	0.8712	32.6532	32.7787	1.3592	78.7355	17.8648	1.3601	79.0417	17.9343	8.9947	96.5928
1.4088	0.6477	6000	0.8711	33.8505	33.9891	1.6794	77.7335	22.1385	1.6804	78.0123	22.2745	8.7814	95.2315
1.3827	0.7557	7000	0.8577	33.4554	33.5788	1.8273	78.6389	19.8999	1.8278	78.9318	19.9767	8.9175	96.2523
1.3724	0.8637	8000	0.8546	33.9822	34.1092	1.9029	78.1136	21.93	1.9038	78.3902	22.0335	8.8339	95.6744
1.3765	0.9716	9000	0.8560	34.3858	34.5199	2.0618	77.5883	23.5074	2.0627	77.8534	23.6436	8.7336	94.8992
1.3394	1.0796	10000	0.8543	33.9466	34.0591	2.2936	78.893	20.6532	2.2939	79.1763	20.7072	8.8746	95.9382
1.3215	1.1875	11000	0.8586	34.8154	34.9403	2.5226	77.7114	24.2122	2.5232	77.967	24.3307	8.6369	94.8519
1.3201	1.2955	12000	0.8635	35.0642	35.1821	2.6475	77.9966	24.5485	2.648	78.2487	24.6497	8.6923	94.8076
1.305	1.4034	13000	0.8538	34.3125	34.4191	2.5477	78.9463	21.4435	2.548	79.2199	21.4895	8.8855	95.7623
1.3007	1.5114	14000	0.8594	35.2924	35.4213	2.6975	77.1695	26.0102	2.698	77.4097	26.1563	8.5157	93.9044
1.2941	1.6193	15000	0.8670	35.1955	35.3046	2.9201	78.0836	24.5827	2.9204	78.3297	24.6637	8.6739	94.4185
1.2843	1.7273	16000	0.8618	34.7826	34.9015	2.5214	78.3308	23.4956	2.5216	78.5955	23.5875	8.7743	95.1965
1.2845	1.8353	17000	0.8661	34.4209	34.5391	2.2746	78.1474	22.8407	2.275	78.4122	22.9303	8.7781	95.3004
1.2752	1.9432	18000	0.8763	35.4414	35.5519	2.9318	78.0589	25.3335	2.9321	78.3016	25.422	8.6975	94.5009
1.2275	2.0512	19000	0.8691	35.6042	35.7092	3.2336	78.0741	25.5048	3.2338	78.3111	25.5826	8.6568	94.5273
1.212	2.1591	20000	0.8709	35.7725	35.8786	3.1674	77.5545	26.5955	3.1677	77.7781	26.6901	8.5776	93.7323
1.2132	2.2671	21000	0.8662	34.9038	35.0133	2.7852	78.5778	23.3486	2.7853	78.8389	23.4158	8.81	95.2465
1.2052	2.3750	22000	0.8707	35.4538	35.5547	3.1225	78.438	24.801	3.1226	78.6803	24.8613	8.7035	94.7578
1.2175	2.4830	23000	0.8745	35.6625	35.7636	3.2546	78.2719	25.4612	3.2548	78.5084	25.5276	8.6879	94.7659
1.2014	2.5910	24000	0.8725	35.6727	35.779	3.0206	77.7743	26.2232	3.0209	78.0033	26.313	8.5494	93.8561
1.195	2.6989	25000	0.8726	35.7564	35.8621	3.12	77.9879	26.1613	3.1202	78.2202	26.246	8.596	93.9884
1.2045	2.8069	26000	0.8702	35.579	35.6877	3.0601	77.8844	25.7926	3.0603	78.1198	25.883	8.6031	94.1975
1.1882	2.9148	27000	0.8659	35.1077	35.2157	2.9039	78.3907	24.0286	2.9041	78.6437	24.0993	8.7458	94.7959
1.1946	3.0228	28000	0.8895	36.2244	36.3277	3.4612	77.5469	27.6651	3.4613	77.7619	27.7599	8.5522	93.2346
1.1346	3.1307	29000	0.8777	36.4978	36.5959	3.6593	77.8239	28.0103	3.6594	78.0366	28.0917	8.5201	93.3092
1.1325	3.2387	30000	0.8841	35.8627	35.9653	3.3274	77.7601	26.5006	3.3275	77.9843	26.5841	8.5529	93.4478
1.1345	3.3466	31000	0.8774	36.0619	36.159	3.4726	77.8183	26.8947	3.4728	78.0361	26.9679	8.5371	93.6775
1.1456	3.4546	32000	0.8806	36.154	36.258	3.3219	77.7428	27.3972	3.3221	77.9628	27.4892	8.5411	93.3262
1.1239	3.5626	33000	0.8762	36.2594	36.3554	3.5953	77.9167	27.2661	3.5955	78.1329	27.3379	8.5613	93.631
1.1227	3.6705	34000	0.8809	35.6	35.7039	3.1605	78.0969	25.5426	3.1606	78.3333	25.6179	8.687	94.0505
1.1413	3.7785	35000	0.8794	36.6955	36.7995	3.708	77.1888	29.1897	3.7082	77.3896	29.3007	8.4227	92.7837
1.1423	3.8864	36000	0.8956	36.1318	36.2347	3.3576	77.581	27.4568	3.3578	77.7975	27.5488	8.5717	93.1018
1.1339	3.9944	37000	0.8829	36.4258	36.5211	3.6958	77.4683	28.1132	3.696	77.6717	28.1957	8.5213	92.9871
1.0747	4.1023	38000	0.9051	37.0457	37.1345	4.1443	77.5177	29.475	4.1443	77.7084	29.5508	8.4619	92.6286
1.0711	4.2103	39000	0.8930	36.2805	36.3798	3.5565	78.0056	27.2794	3.5565	78.2265	27.3563	8.611	93.7963
1.0785	4.3183	40000	0.8929	36.618	36.7162	3.6923	77.392	28.7699	3.6924	77.5932	28.8631	8.4574	92.7116
1.0818	4.4262	41000	0.8896	36.8682	36.9674	3.8285	77.4044	29.3716	3.8286	77.6034	29.4702	8.4212	92.9933
1.0811	4.5342	42000	0.9119	36.7358	36.8258	3.888	77.669	28.6503	3.8881	77.8679	28.7214	8.484	92.7557
1.0872	4.6421	43000	0.8924	36.3606	36.46	3.5563	77.7412	27.7843	3.5564	77.9549	27.8688	8.512	93.3203
1.0968	4.7501	44000	0.8921	36.4742	36.5676	3.735	77.7654	27.9221	3.7352	77.9725	27.995	8.5476	93.3622
1.0835	4.8580	45000	0.9040	36.7224	36.8112	4.053	77.4397	28.6746	4.0531	77.6335	28.747	8.5208	92.7691
1.0895	4.9660	46000	0.8906	36.706	36.7989	3.9402	77.6053	28.5724	3.9404	77.8063	28.6501	8.5282	93.1717
1.0533	5.0740	47000	0.9085	36.4749	36.5574	3.8835	78.1434	27.3978	3.8837	78.3486	27.4398	8.6253	93.4822
1.027	5.1819	48000	0.9055	36.7132	36.8084	3.7814	77.1845	29.1737	3.7815	77.3773	29.2664	8.4275	92.3575

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.22.1