Update README.md
Browse files
README.md
CHANGED
|
@@ -3,3 +3,34 @@ license: other
|
|
| 3 |
license_name: sggs
|
| 4 |
license_link: LICENSE
|
| 5 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
license_name: sggs
|
| 4 |
license_link: LICENSE
|
| 5 |
---
|
| 6 |
+
pythia-160m-hq-emails-v4
|
| 7 |
+
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:
|
| 8 |
+
|
| 9 |
+
Loss: 2.2856
|
| 10 |
+
Accuracy: 0.6113
|
| 11 |
+
perplexity: 9.8313
|
| 12 |
+
Model description
|
| 13 |
+
this is v4
|
| 14 |
+
|
| 15 |
+
Intended uses & limitations
|
| 16 |
+
More information needed
|
| 17 |
+
|
| 18 |
+
Training and evaluation data
|
| 19 |
+
More information needed
|
| 20 |
+
|
| 21 |
+
Training procedure
|
| 22 |
+
Training hyperparameters
|
| 23 |
+
The following hyperparameters were used during training:
|
| 24 |
+
|
| 25 |
+
learning_rate: 0.0006
|
| 26 |
+
train_batch_size: 4
|
| 27 |
+
eval_batch_size: 1
|
| 28 |
+
seed: 42
|
| 29 |
+
distributed_type: multi-GPU
|
| 30 |
+
gradient_accumulation_steps: 32
|
| 31 |
+
total_train_batch_size: 128
|
| 32 |
+
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 33 |
+
lr_scheduler_type: cosine
|
| 34 |
+
lr_scheduler_warmup_ratio: 0.05
|
| 35 |
+
num_epochs: 4.0
|
| 36 |
+
mixed_precision_training: Native AMP
|