metadata

license: apache-2.0
tags:
  - generated_from_trainer
datasets:
  - winograd_wsc
metrics:
  - rouge
model-index:
  - name: flan-t5-large-coref
    results:
      - task:
          name: Sequence-to-sequence Language Modeling
          type: text2text-generation
        dataset:
          name: winograd_wsc
          type: winograd_wsc
          config: wsc285
          split: test
          args: wsc285
        metrics:
          - name: Rouge1
            type: rouge
            value: 0.9495
widget:
  - text: Sam has a Parker pen. He loves writing with it.
    example_title: Example 1
  - text: >-
      Coronavirus quickly spread worldwide in 2020. The virus mostly affects
      elderly people. They can easily catch it.
    example_title: Example 2
  - text: >-
      First, the manager evaluates the candidates. Afterwards, he notifies the
      candidates regarding the evaluation.
    example_title: Example 3

flan-t5-large-coref

This model is a fine-tuned version of google/flan-t5-large on the winograd_wsc dataset.

The model was trained on the task of coreference resolution.

It achieves the following results on the evaluation set:

Loss: 0.2404
Rouge1: 0.9495
Rouge2: 0.9107
Rougel: 0.9494
Rougelsum: 0.9494
Gen Len: 23.4828

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: 2e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
1.0169	1.0	16	0.6742	0.7918	0.6875	0.7836	0.7847	18.2414
0.6275	2.0	32	0.5093	0.8776	0.7947	0.8734	0.8732	21.5517
0.596	3.0	48	0.4246	0.9104	0.8486	0.9085	0.9091	22.5172
0.743	4.0	64	0.3632	0.9247	0.8661	0.9235	0.9231	22.8621
0.5007	5.0	80	0.3301	0.9353	0.8845	0.9357	0.9353	22.8621
0.2567	6.0	96	0.3093	0.9388	0.8962	0.9392	0.9388	22.9655
0.4146	7.0	112	0.2978	0.9449	0.907	0.9455	0.9458	23.1034
0.1991	8.0	128	0.2853	0.9454	0.9064	0.946	0.9462	23.069
0.1786	9.0	144	0.2794	0.9475	0.9097	0.9475	0.9477	23.069
0.3559	10.0	160	0.2701	0.9424	0.9013	0.9428	0.9426	23.0345
0.2059	11.0	176	0.2636	0.9472	0.9069	0.9472	0.9472	23.0345
0.199	12.0	192	0.2592	0.9523	0.9141	0.9521	0.9524	23.4483
0.1634	13.0	208	0.2553	0.9523	0.9141	0.9521	0.9524	23.4483
0.2006	14.0	224	0.2518	0.9523	0.9141	0.9521	0.9524	23.4483
0.1419	15.0	240	0.2487	0.9523	0.9141	0.9521	0.9524	23.4483
0.2089	16.0	256	0.2456	0.9523	0.9141	0.9521	0.9524	23.4483
0.1007	17.0	272	0.2431	0.9523	0.9141	0.9521	0.9524	23.4483
0.1598	18.0	288	0.2415	0.9495	0.9107	0.9494	0.9494	23.4828
0.3088	19.0	304	0.2407	0.9495	0.9107	0.9494	0.9494	23.4828
0.2003	20.0	320	0.2404	0.9495	0.9107	0.9494	0.9494	23.4828

Framework versions

Transformers 4.25.1
Pytorch 1.13.0+cu116
Datasets 2.7.1
Tokenizers 0.13.2