README.md

---
license: apache-2.0
language:
- en
pipeline_tag: text-generation
tags:
- text-generation
- openlm
- silo
---

# Silo Language Models: Isolating Legal Risk in a Datastore

This is Silo-PD, first introduced in [Silo Language Models]() by researchers at University of Washington, UC Berkeley, and the Allen Institute for AI.

### NOTE: Dependencies

To use the model, you need to install a specific transformers fork:

```
pip install git+https://github.com/kernelmachine/transformers@openlm#egg=transformers
```

The model also depends on `xformers`, install via 

```
pip install xformers
```

### Model Description


Silo-PD is a 1.3B parameter, decoder-only language model trained on data in the public domain from [the Open License Corpus (OLC)](https://huggingface.co/datasets/kernelmachine/open-license-corpus). 

The model is based on the LLaMA architecture as implemented in (OpenLM)[].

The model is trained with 128 A100 GPUs across 16 nodes.


### Model and Training Hyperparameters

We follow the model architecture of LLaMa, and we use the GPT-NeoX-20B tokenizer, with 50432 BPE types. 

During training, we use 2,048 token sequences that are packed across document boundaries, and we pre-pend a beginning-of-text token to every document. 

We use weight decay of 0.1, the Adam optimizer with beta_2 of 0.95, 2,000 steps of warmup, with a cosine learning rate scheduler. 


| Model  | #L  | #H  | d_model | LR     | Batch  |
|--------|-----|-----|-------------|--------|--------|
| 1.3B   | 24  | 16  | 2048        | 1e-3   | 2.6M   |



### Training data
Specifically, it was trained on the following domain proportions (please see the OLC repository for more details on the data sources for each domain):


| Domain          | Tokens (B) | %     |
|-----------------|------------|-------|
| Legal           | 27.1       | 86.2  | 
| Books           | 2.9        | 9.3   |
| Science         | 1.2        | 3.8   |
| News            | 0.2        | 0.7   |
| Total           | 31.4       | 100.0 | 

We train with early stopping for 60B tokens in total, for a total of 2 epochs of training over this subset 

Since the distribution of OLC is highly skewed, we perform a simple upweighting scheme where we upsample all data that accounts for less than 5% of the corpus by a factor of 3x, which we found to work well after a sweep of different settings.

### Intended Uses and Limitations

This model can be used for prompting for evaluation of downstream tasks as well as text generation. 

### How to use


You can use this model directly with a pipeline for text generation.


```python
from transformers import pipeline
generator = pipeline('text-generation', model="kernelmachine/silo-pd-1.3b", device='cuda')
generator("Hello")
[{'generated_text': 'Hello, my dear," said the old man, "I have been waiting for you\na long'}]
```

By default, generation is deterministic. In order to use the top-k sampling, please set do_sample to True.


```python
from transformers import pipeline, set_seed
set_seed(42)
generator = pipeline('text-generation', model="kernelmachine/silo-pd-1.3b", device='cuda', do_sample=True)
generator("Hello")
[{'generated_text': 'Hello, Mother," he called.\n\n"Hello, Son.  Have you got a car'}]
```

### Limitations and Bias

Silo-PD inherits the biases and limitations of public domain data, which carry risks of toxic or otherwise unfair output, due to the prevalence of older copyright-expired text.

Silo-PD may also output personally identifiable information, because we did not filter that out of training data.