File size: 3,541 Bytes
03cbb29 81ca873 |
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 |
---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: flant5
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. -->
# flant5
This model is a fine-tuned version of [google/flan-t5-base](https://huggingface.co/google/flan-t5-base) on an unknown dataset.
## 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: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2
### Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2
## Model Recycling
[Evaluation on 36 datasets](https://ibm.github.io/model-recycling/model_gain_chart?avg=9.03&mnli_lp=nan&20_newsgroup=4.19&ag_news=1.36&amazon_reviews_multi=0.23&anli=14.13&boolq=17.27&cb=23.12&cola=9.97&copa=29.50&dbpedia=6.50&esnli=5.11&financial_phrasebank=18.16&imdb=0.52&isear=1.43&mnli=11.97&mrpc=13.44&multirc=5.70&poem_sentiment=19.42&qnli=3.74&qqp=7.12&rotten_tomatoes=3.64&rte=25.34&sst2=0.09&sst_5bins=4.72&stsb=20.65&trec_coarse=4.15&trec_fine=9.53&tweet_ev_emoji=13.59&tweet_ev_emotion=4.90&tweet_ev_hate=1.07&tweet_ev_irony=7.25&tweet_ev_offensive=2.16&tweet_ev_sentiment=1.88&wic=12.97&wnli=9.44&wsc=7.45&yahoo_answers=3.38&model_name=talhaa%2Fflant5&base_name=google%2Ft5-v1_1-base) using talhaa/flant5 as a base model yields average score of 77.86 in comparison to 68.82 by google/t5-v1_1-base.
The model is ranked 1st among all tested models for the google/t5-v1_1-base architecture as of 10/01/2023
Results:
| 20_newsgroup | ag_news | amazon_reviews_multi | anli | boolq | cb | cola | copa | dbpedia | esnli | financial_phrasebank | imdb | isear | mnli | mrpc | multirc | poem_sentiment | qnli | qqp | rotten_tomatoes | rte | sst2 | sst_5bins | stsb | trec_coarse | trec_fine | tweet_ev_emoji | tweet_ev_emotion | tweet_ev_hate | tweet_ev_irony | tweet_ev_offensive | tweet_ev_sentiment | wic | wnli | wsc | yahoo_answers |
|---------------:|----------:|-----------------------:|--------:|--------:|--------:|--------:|-------:|----------:|--------:|-----------------------:|-------:|--------:|--------:|--------:|----------:|-----------------:|--------:|--------:|------------------:|--------:|--------:|------------:|--------:|--------------:|------------:|-----------------:|-------------------:|----------------:|-----------------:|---------------------:|---------------------:|--------:|-------:|--------:|----------------:|
| 87.0685 | 89.5333 | 67.14 | 52.1875 | 82.844 | 78.5714 | 80.1534 | 70 | 77.2667 | 90.6963 | 84.9 | 93.512 | 72.4902 | 87.4797 | 86.2745 | 61.8399 | 87.5 | 93.1173 | 90.7173 | 89.6811 | 85.9206 | 93.8073 | 56.5611 | 89.4438 | 97.4 | 91.6 | 47.054 | 80.5067 | 52.5926 | 74.8724 | 84.7674 | 71.76 | 68.8088 | 56.338 | 55.7692 | 72.6333 |
For more information, see: [Model Recycling](https://ibm.github.io/model-recycling/)
|