layoutlmv2-base-uncased_finetuned_docvqa

该模型是 microsoft/layoutlmv2-base-uncased 在一个未知数据集上的微调版本。它在评估集上取得了以下结果：

损失：4.5469

模型描述

需要更多信息

预期用途与局限性

需要更多信息

训练和评估数据

需要更多信息

训练过程

训练超参数

训练过程中使用了以下超参数：

learning_rate：5e-05
train_batch_size：4
eval_batch_size：8
seed：42
optimizer：使用 OptimizerNames.ADAMW_TORCH_FUSED，betas=(0.9,0.999)，epsilon=1e-08，optimizer_args=无额外优化器参数
lr_scheduler_type：linear
num_epochs：20

训练结果

训练损失	轮次	步数	验证损失
5.2971	0.2212	50	4.5199
4.4509	0.4425	100	4.0939
4.0522	0.6637	150	3.8095
3.8104	0.8850	200	3.6099
3.5676	1.1062	250	3.6083
3.1991	1.3274	300	3.2266
3.1147	1.5487	350	3.0062
2.8610	1.7699	400	2.8407
2.6843	1.9912	450	2.7080
2.2057	2.2124	500	2.7440
1.9496	2.4336	550	2.4666
1.9274	2.6549	600	2.4632
1.8665	2.8761	650	2.4528
1.6049	3.0973	700	2.9239
1.7440	3.3186	750	2.6191
1.5232	3.5398	800	2.6962
1.3731	3.7611	850	2.4688
1.4415	3.9823	900	2.7062
1.2257	4.2035	950	2.7565
1.3680	4.4248	1000	2.6935
1.1731	4.6460	1050	2.5318
1.1767	4.8673	1100	2.6353
1.0538	5.0885	1150	2.8096
1.1712	5.3097	1200	2.8875
1.2517	5.5310	1250	3.6131
0.6705	5.7522	1300	3.6724
0.8024	5.9735	1350	3.4343
0.6279	6.1947	1400	2.9707
0.4980	6.4159	1450	3.6763
0.6252	6.6372	1500	3.2491
0.6904	6.8584	1550	3.7241
0.8179	7.0796	1600	3.6211
0.4764	7.3009	1650	3.9078
0.5007	7.5221	1700	4.4816
0.6872	7.7434	1750	3.6041
0.4735	7.9646	1800	3.4764
0.3815	8.1858	1850	3.7402
0.3994	8.4071	1900	3.8179
0.4076	8.6283	1950	3.7463
0.4628	8.8496	2000	3.5482
0.3891	9.0708	2050	3.7790
0.2142	9.2920	2100	3.9935
0.2324	9.5133	2150	4.1999
0.5996	9.7345	2200	3.7067
0.1822	9.9558	2250	4.0189
0.3799	10.1770	2300	4.0172
0.1988	10.3982	2350	4.2575
0.2392	10.6195	2400	4.1520
0.3086	10.8407	2450	3.9827
0.1570	11.0619	2500	4.2772
0.1668	11.2832	2550	4.2690
0.2935	11.5044	2600	4.0455
0.1665	11.7257	2650	4.0969
0.0886	11.9469	2700	4.2723
0.0997	12.1681	2750	4.4508
0.0355	12.3894	2800	4.3501
0.1220	12.6106	2850	4.0862
0.2402	12.8319	2900	4.2115
0.1178	13.0531	2950	4.0085
0.0881	13.2743	3000	4.0700
0.0867	13.4956	3050	4.3007
0.0950	13.7168	3100	4.3927
0.1399	13.9381	3150	4.4827
0.1882	14.1593	3200	4.3956
0.1268	14.3805	3250	4.2252
0.0761	14.6018	3300	4.2256
0.0797	14.8230	3350	4.2576
0.0662	15.0442	3400	4.3513
0.1340	15.2655	3450	4.3753
0.0337	15.4867	3500	4.4571
0.1954	15.7080	3550	4.4622
0.0318	15.9292	3600	4.7108
0.0762	16.1504	3650	4.5958
0.0305	16.3717	3700	4.7203
0.0632	16.5929	3750	4.6967
0.1050	16.8142	3800	4.5247
0.0168	17.0354	3850	4.5173
0.0444	17.2566	3900	4.5160
0.0995	17.4779	3950	4.4270
0.0319	17.6991	4000	4.4033
0.0184	17.9204	4050	4.4754
0.0292	18.1416	4100	4.4763
0.0040	18.3628	4150	4.4920
0.0313	18.5841	4200	4.5430
0.0035	18.8053	4250	4.5429
0.0599	19.0265	4300	4.5338
0.0114	19.2478	4350	4.5285
0.0349	19.4690	4400	4.5413
0.0429	19.6903	4450	4.5425
0.0267	19.9115	4500	4.5469

框架版本

Transformers 5.0.0
Pytorch 2.10.0+cu128
Datasets 4.0.0
Tokenizers 0.22.2

训练过程

训练超参数

训练过程中使用了以下超参数：

learning_rate：5e-05

train_batch_size：4

eval_batch_size：8

seed：42

optimizer：使用 OptimizerNames.ADAMW_TORCH_FUSED，betas=(0.9,0.999)，epsilon=1e-08，optimizer_args=无额外优化器参数

lr_scheduler_type：linear

num_epochs：20

训练结果

训练损失	轮次	步数	验证损失
5.2971	0.2212	50	4.5199
4.4509	0.4425	100	4.0939
4.0522	0.6637	150	3.8095
3.8104	0.8850	200	3.6099
3.5676	1.1062	250	3.6083
3.1991	1.3274	300	3.2266
3.1147	1.5487	350	3.0062
2.8610	1.7699	400	2.8407
2.6843	1.9912	450	2.7080
2.2057	2.2124	500	2.7440
1.9496	2.4336	550	2.4666
1.9274	2.6549	600	2.4632
1.8665	2.8761	650	2.4528
1.6049	3.0973	700	2.9239
1.7440	3.3186	750	2.6191
1.5232	3.5398	800	2.6962
1.3731	3.7611	850	2.4688
1.4415	3.9823	900	2.7062
1.2257	4.2035	950	2.7565
1.3680	4.4248	1000	2.6935
1.1731	4.6460	1050	2.5318
1.1767	4.8673	1100	2.6353
1.0538	5.0885	1150	2.8096
1.1712	5.3097	1200	2.8875
1.2517	5.5310	1250	3.6131
0.6705	5.7522	1300	3.6724
0.8024	5.9735	1350	3.4343
0.6279	6.1947	1400	2.9707
0.4980	6.4159	1450	3.6763
0.6252	6.6372	1500	3.2491
0.6904	6.8584	1550	3.7241
0.8179	7.0796	1600	3.6211
0.4764	7.3009	1650	3.9078
0.5007	7.5221	1700	4.4816
0.6872	7.7434	1750	3.6041
0.4735	7.9646	1800	3.4764
0.3815	8.1858	1850	3.7402
0.3994	8.4071	1900	3.8179
0.4076	8.6283	1950	3.7463
0.4628	8.8496	2000	3.5482
0.3891	9.0708	2050	3.7790
0.2142	9.2920	2100	3.9935
0.2324	9.5133	2150	4.1999
0.5996	9.7345	2200	3.7067
0.1822	9.9558	2250	4.0189
0.3799	10.1770	2300	4.0172
0.1988	10.3982	2350	4.2575
0.2392	10.6195	2400	4.1520
0.3086	10.8407	2450	3.9827
0.1570	11.0619	2500	4.2772
0.1668	11.2832	2550	4.2690
0.2935	11.5044	2600	4.0455
0.1665	11.7257	2650	4.0969
0.0886	11.9469	2700	4.2723
0.0997	12.1681	2750	4.4508
0.0355	12.3894	2800	4.3501
0.1220	12.6106	2850	4.0862
0.2402	12.8319	2900	4.2115
0.1178	13.0531	2950	4.0085
0.0881	13.2743	3000	4.0700
0.0867	13.4956	3050	4.3007
0.0950	13.7168	3100	4.3927
0.1399	13.9381	3150	4.4827
0.1882	14.1593	3200	4.3956
0.1268	14.3805	3250	4.2252
0.0761	14.6018	3300	4.2256
0.0797	14.8230	3350	4.2576
0.0662	15.0442	3400	4.3513
0.1340	15.2655	3450	4.3753
0.0337	15.4867	3500	4.4571
0.1954	15.7080	3550	4.4622
0.0318	15.9292	3600	4.7108
0.0762	16.1504	3650	4.5958
0.0305	16.3717	3700	4.7203
0.0632	16.5929	3750	4.6967
0.1050	16.8142	3800	4.5247
0.0168	17.0354	3850	4.5173
0.0444	17.2566	3900	4.5160
0.0995	17.4779	3950	4.4270
0.0319	17.6991	4000	4.4033
0.0184	17.9204	4050	4.4754
0.0292	18.1416	4100	4.4763
0.0040	18.3628	4150	4.4920
0.0313	18.5841	4200	4.5430
0.0035	18.8053	4250	4.5429
0.0599	19.0265	4300	4.5338
0.0114	19.2478	4350	4.5285
0.0349	19.4690	4400	4.5413
0.0429	19.6903	4450	4.5425
0.0267	19.9115	4500	4.5469

框架版本

Transformers 5.0.0

Pytorch 2.10.0+cu128

Datasets 4.0.0

Tokenizers 0.22.2