Skip to content

Commit

Permalink
Updated readme
Browse files Browse the repository at this point in the history
  • Loading branch information
ParishadBehnam committed Apr 4, 2024
1 parent fd91850 commit 1407f91
Showing 1 changed file with 51 additions and 0 deletions.
51 changes: 51 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,3 +6,54 @@
<img src="https://github.com/McGill-NLP/llm2vec/assets/12207571/48efd48a-431b-4625-8e0f-248a442e3839?raw=true" width="75%" alt="LLM2Vec_figure1"/>
</p>

## Instrallation
To use LLM2Vec, first install the llm2vec package from PyPI.

```bash
pip install llm2vec
```
You can also directly install it from our code by cloning the repository and:

```bash
pip install -e .
```

## Getting Started
LLM2Vec is a generic model, which takes a `tokenizer` and a `model`. First, we define the model and tokenizer using `transformers` library:

```python
import torch
from transformers import AutoTokenizer, AutoModel, AutoConfig
config = AutoConfig.from_pretrained("McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp", trust_remote_code=True)
model = AutoModel.from_pretrained("McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp", trust_remote_code=True, config=config, torch_dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained("McGill-NLP/LLM2Vec-Sheared-LLaMA-mntp")
```

Then, we define our llm2vec model as follows:

```python
from llm2vec import LLM2Vec

l2v = LLM2Vec(model, tokenizer)
```

This model now returns the text embedding for any input in the form of `[[instruction, text]]`.

```python
inputs = [
['Retrieve duplicate questions from StackOverflow forum', 'Python (Numpy) array sorting'],
['', 'Sort a list in python'],
['', 'Sort an array in Java'],
]
repr = l2v.encode(inputs, convert_to_tensor=True)
sim_pos = torch.nn.functional.cosine_similarity(repr[0].unsqueeze(0), repr[1].unsqueeze(0)) # tensor([0.5987])
sim_neg = torch.nn.functional.cosine_similarity(repr[0].unsqueeze(0), repr[2].unsqueeze(0)) # tensor([0.5585])
```

# Model List

# Training
Training code will be available soon.

# Bugs or questions?
If you have any question about the code, feel free to email Parishad (`[email protected]`) and Vaibhav (`[email protected]`).

0 comments on commit 1407f91

Please sign in to comment.