Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,42 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
library_name: peft
|
| 4 |
+
tags:
|
| 5 |
+
- llama-factory
|
| 6 |
+
- lora
|
| 7 |
+
base_model: lmsys/vicuna-13b-v1.5-16k
|
| 8 |
+
model-index:
|
| 9 |
+
- name: CHAST
|
| 10 |
+
results: []
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
# CHAST (Finetuned Vicuna13b_16K_yaml_lora32)
|
| 15 |
+
|
| 16 |
+
This model is a fine-tuned version of [lmsys/vicuna-13b-v1.5-16k](https://huggingface.co/lmsys/vicuna-13b-v1.5-16k).
|
| 17 |
+
|
| 18 |
+
## Model description
|
| 19 |
+
|
| 20 |
+
Computes Covert Harms and Social Threats (CHAST) metrics for conversational data.
|
| 21 |
+
|
| 22 |
+
### Training hyperparameters
|
| 23 |
+
|
| 24 |
+
The following hyperparameters were used during training:
|
| 25 |
+
- learning_rate: 5e-05
|
| 26 |
+
- train_batch_size: 8
|
| 27 |
+
- eval_batch_size: 64
|
| 28 |
+
- seed: 42
|
| 29 |
+
- gradient_accumulation_steps: 4
|
| 30 |
+
- total_train_batch_size: 32
|
| 31 |
+
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 32 |
+
- lr_scheduler_type: cosine
|
| 33 |
+
- num_epochs: 2.0
|
| 34 |
+
- mixed_precision_training: Native AMP
|
| 35 |
+
|
| 36 |
+
### Framework versions
|
| 37 |
+
|
| 38 |
+
- PEFT 0.9.0
|
| 39 |
+
- Transformers 4.37.2
|
| 40 |
+
- Pytorch 2.2.1+cu121
|
| 41 |
+
- Datasets 2.18.0
|
| 42 |
+
- Tokenizers 0.15.2
|