pythia-160m-hq-emails-v4
This model is a fine-tuned version of EleutherAI/pythia-160m-deduped on the postbot/multi-emails-hq dataset. It achieves the following results on the evaluation set:
- Loss: 2.2856
- Accuracy: 0.6113
- perplexity: 9.8313
Model description
this is v4
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: 0.0006
- train_batch_size: 4
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 32
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.05
- num_epochs: 4.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| 2.412 | 0.99 | 76 | 2.5027 | 0.5458 |
| 1.9702 | 1.99 | 152 | 2.2757 | 0.5850 |
| 1.4628 | 2.99 | 228 | 2.2162 | 0.6082 |
| 1.1662 | 3.99 | 304 | 2.2856 | 0.6113 |
Framework versions
- Transformers 4.27.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.8.0
- Tokenizers 0.13.1
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 25.12 |
| ARC (25-shot) | 23.12 |
| HellaSwag (10-shot) | 30.05 |
| MMLU (5-shot) | 26.58 |
| TruthfulQA (0-shot) | 45.51 |
| Winogrande (5-shot) | 50.28 |
| GSM8K (5-shot) | 0.0 |
| DROP (3-shot) | 0.31 |