Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: google/long-t5-tglobal-base
|
| 4 |
+
tags:
|
| 5 |
+
- generated_from_trainer
|
| 6 |
+
- synthsumm
|
| 7 |
+
metrics:
|
| 8 |
+
- rouge
|
| 9 |
+
datasets:
|
| 10 |
+
- pszemraj/synthsumm
|
| 11 |
+
language:
|
| 12 |
+
- en
|
| 13 |
+
pipeline_tag: summarization
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
| 17 |
+
should probably proofread and complete it, then remove this comment. -->
|
| 18 |
+
|
| 19 |
+
# long-t5-tglobal-base-synthsumm_direct
|
| 20 |
+
|
| 21 |
+
Fine-tuned on a synthetic dataset of curated long-context text and `GPT-3.5-turbo-1106` summaries spanning multiple domains, including "random" long-context examples from redpajama, the pile, etc.
|
| 22 |
+
|
| 23 |
+
- Note: this model has **not** been fine-tuned on any other summarization datasets, just the `synthsumm` data
|
| 24 |
+
|
| 25 |
+
Try it out in the [gradio demo](https://huggingface.co/spaces/pszemraj/document-summarization)
|
| 26 |
+
|
| 27 |
+
## Model description
|
| 28 |
+
|
| 29 |
+
This model is a fine-tuned version of [google/long-t5-tglobal-base](https://huggingface.co/google/long-t5-tglobal-base) on the None dataset.
|
| 30 |
+
It achieves the following results on the evaluation set:
|
| 31 |
+
- Loss: 1.4378
|
| 32 |
+
- Rouge1: 48.0918
|
| 33 |
+
- Rouge2: 21.2531
|
| 34 |
+
- Rougel: 34.4307
|
| 35 |
+
- Rougelsum: 43.0271
|
| 36 |
+
- Gen Len: 84.5231
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
## Training procedure
|
| 40 |
+
|
| 41 |
+
### Training hyperparameters
|
| 42 |
+
|
| 43 |
+
The following hyperparameters were used during training:
|
| 44 |
+
- learning_rate: 0.0003
|
| 45 |
+
- train_batch_size: 1
|
| 46 |
+
- eval_batch_size: 1
|
| 47 |
+
- seed: 26605
|
| 48 |
+
- gradient_accumulation_steps: 8
|
| 49 |
+
- total_train_batch_size: 8
|
| 50 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 51 |
+
- lr_scheduler_type: inverse_sqrt
|
| 52 |
+
- lr_scheduler_warmup_ratio: 0.03
|
| 53 |
+
- num_epochs: 2.0
|
| 54 |
+
|
| 55 |
+
### Training results
|
| 56 |
+
|
| 57 |
+
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|
| 58 |
+
|:-------------:|:-----:|:----:|:---------------:|:-------:|:-------:|:-------:|:---------:|:--------:|
|
| 59 |
+
| 1.9183 | 0.38 | 125 | 1.5762 | 38.7221 | 15.0873 | 28.3123 | 34.9655 | 129.2154 |
|
| 60 |
+
| 1.8815 | 0.77 | 250 | 1.5230 | 44.3531 | 17.9384 | 31.7417 | 39.5563 | 87.3538 |
|
| 61 |
+
| 1.7264 | 1.15 | 375 | 1.4735 | 45.7781 | 20.102 | 33.329 | 41.4737 | 101.9231 |
|
| 62 |
+
| 1.8545 | 1.54 | 500 | 1.4505 | 47.0134 | 20.6159 | 33.6118 | 41.6579 | 88.2308 |
|
| 63 |
+
| 1.7444 | 1.92 | 625 | 1.4378 | 48.0918 | 21.2531 | 34.4307 | 43.0271 | 84.5231 |
|
| 64 |
+
|
| 65 |
+
|
| 66 |
+
### Framework versions
|
| 67 |
+
|
| 68 |
+
- Transformers 4.36.0.dev0
|
| 69 |
+
- Pytorch 2.1.0
|
| 70 |
+
- Datasets 2.15.0
|
| 71 |
+
- Tokenizers 0.15.0
|