File size: 3,062 Bytes
43749b7
 
 
 
 
 
 
 
 
341a985
43749b7
 
 
 
 
 
be05e33
43749b7
5cc5144
43749b7
 
 
 
 
 
 
 
 
 
 
 
 
 
4ad6cda
 
 
 
 
 
 
 
 
 
 
43749b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
---
library_name: transformers
license: apache-2.0
base_model: google-t5/t5-small
tags:
- generated_from_trainer
metrics:
- rouge
model-index:
- name: t5-small-billsum
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# t5-small-billsum

This model is a fine-tuned version of [google-t5/t5-small](https://huggingface.co/google-t5/t5-small) on Billsum dataset.
It achieves the following results on the evaluation set:
- Loss: 2.3195
- Rouge1: 0.205
- Rouge2: 0.0994
- Rougel: 0.1708
- Rougelsum: 0.1704
- Gen Len: 20.0

## Model description

More information needed

## Intended uses & limitations

- Research: Explore summarization performance of small models on long-text datasets like BillSum.

- Applications: Legislative bill summarization (short drafts), document compression.

**Not intended for production without further fine-tuning and testing.**

- Short outputs: Current fine-tuning capped at 20 tokens → summaries are incomplete.

- Model size: T5-small struggles with very long inputs (BillSum often >512 tokens).

- Performance: ROUGE scores (~0.20) are well below state-of-the-art.

## 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: 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
- num_epochs: 10
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|:-------------:|:-----:|:----:|:---------------:|:------:|:------:|:------:|:---------:|:-------:|
| No log        | 1.0   | 62   | 2.7672          | 0.139  | 0.0457 | 0.114  | 0.1141    | 20.0    |
| No log        | 2.0   | 124  | 2.5470          | 0.148  | 0.0535 | 0.121  | 0.1209    | 20.0    |
| No log        | 3.0   | 186  | 2.4587          | 0.16   | 0.0645 | 0.1316 | 0.1311    | 20.0    |
| No log        | 4.0   | 248  | 2.4068          | 0.1849 | 0.0833 | 0.1541 | 0.1541    | 20.0    |
| No log        | 5.0   | 310  | 2.3747          | 0.1994 | 0.0945 | 0.1666 | 0.1663    | 20.0    |
| No log        | 6.0   | 372  | 2.3537          | 0.2032 | 0.0976 | 0.1686 | 0.1683    | 20.0    |
| No log        | 7.0   | 434  | 2.3364          | 0.203  | 0.0973 | 0.1694 | 0.169     | 20.0    |
| No log        | 8.0   | 496  | 2.3259          | 0.2045 | 0.0984 | 0.1704 | 0.1701    | 20.0    |
| 2.7564        | 9.0   | 558  | 2.3217          | 0.2052 | 0.0997 | 0.1709 | 0.1704    | 20.0    |
| 2.7564        | 10.0  | 620  | 2.3195          | 0.205  | 0.0994 | 0.1708 | 0.1704    | 20.0    |


### Framework versions

- Transformers 4.55.4
- Pytorch 2.8.0+cu126
- Datasets 4.0.0
- Tokenizers 0.21.4