Quick Links
bart-large-aeslc-rouge-3-loss-differentiable-10-cnt-supervised
This model is a fine-tuned version of facebook/bart-large on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 5.1776
- Rouge1: 0.2418
- Rouge2: 0.1239
- Rougel: 0.2314
- Rougelsum: 0.2311
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: 1e-05
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- 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
- lr_scheduler_warmup_ratio: 0.06
- num_epochs: 20
Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum |
|---|---|---|---|---|---|---|---|
| 3.8234 | 3.4 | 10 | 5.0940 | 0.2049 | 0.1039 | 0.1952 | 0.1954 |
| 1.7867 | 6.8 | 20 | 5.1776 | 0.2418 | 0.1239 | 0.2314 | 0.2311 |
| 1.3308 | 10.0 | 30 | 5.4160 | 0.2345 | 0.1139 | 0.2216 | 0.2215 |
| 0.7055 | 13.4 | 40 | 5.7111 | 0.236 | 0.1141 | 0.2235 | 0.2235 |
| 0.5036 | 16.8 | 50 | 5.9660 | 0.2347 | 0.1118 | 0.2221 | 0.2222 |
Framework versions
- Transformers 4.57.3
- Pytorch 2.9.1+cu128
- Datasets 3.6.0
- Tokenizers 0.22.1
- Downloads last month
- 3
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for taiypeo/bart-large-aeslc-rouge-3-loss-differentiable-10-cnt-supervised
Base model
facebook/bart-large
# Load model directly from transformers import AutoTokenizer, AutoModelForSeq2SeqLM tokenizer = AutoTokenizer.from_pretrained("taiypeo/bart-large-aeslc-rouge-3-loss-differentiable-10-cnt-supervised") model = AutoModelForSeq2SeqLM.from_pretrained("taiypeo/bart-large-aeslc-rouge-3-loss-differentiable-10-cnt-supervised")