Qwen2-7B-Instruct-embed-base

Model Details

Qwen2 is a language model series including decoder language models of different model sizes. For each size, we release the base language model and the aligned chat model. It is based on the Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, etc. Additionally, we have an improved tokenizer adaptive to multiple natural languages and codes.

Requirements

The code of Qwen2 has been in the latest Hugging face transformers and we advise you to install transformers>=4.37.0, or you might encounter the following error:

KeyError: 'qwen2'

Usage

The 'lm_head' layer of this model has been removed, which means it can be used for embeddings. It will not perform greatly, as it needs to be further fine-tuned, as shown by intfloat/e5-mistral-7b-instruct.

Inference

from sentence_transformers import SentenceTransformer
import torch

# 1. Load a pretrained Sentence Transformer model
model = SentenceTransformer("ssmits/Qwen2-7B-embed-base") # device = "cpu" when <= 24 GB VRAM

# The sentences to encode
sentences = [
    "The weather is lovely today.",
    "It's so sunny outside!",
    "He drove to the stadium.",
]

# 2. Calculate embeddings by calling model.encode()
embeddings = model.encode(sentences)
print(embeddings.shape)
# (3, 3584)

# 3. Calculate the embedding similarities
# Assuming embeddings is a numpy array, convert it to a torch tensor
embeddings_tensor = torch.tensor(embeddings)

# Using torch to compute cosine similarity matrix
similarities = torch.nn.functional.cosine_similarity(embeddings_tensor.unsqueeze(0), embeddings_tensor.unsqueeze(1), dim=2)

print(similarities)
# tensor([[1.0000, 0.8608, 0.6609],
#         [0.8608, 1.0000, 0.7046],
#         [0.6609, 0.7046, 1.0000]])

Note: In my tests it utilizes more than 24GB (RTX 4090), so an A100 or A6000 would be required for inference.

Inference (HuggingFace Transformers)

Without sentence-transformers, you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.

from transformers import AutoTokenizer, AutoModel
import torch

#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)

# Sentences we want sentence embeddings for
sentences = ['This is an example sentence', 'Each sentence is converted']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('ssmits/Qwen2-7B-Instruct-embed-base')
model = AutoModel.from_pretrained('ssmits/Qwen2-7B-Instruct-embed-base') # device = "cpu" when <= 24 GB VRAM

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

How to enable Multi-GPU

from transformers import AutoModel
from torch.nn import DataParallel

model = AutoModel.from_pretrained("ssmits/Qwen2-7B-Instruct-embed-base")
for module_key, module in model._modules.items():
    model._modules[module_key] = DataParallel(module)