semantic_search.ipynb

Kernel: Unknown Kernel

This notebook regroups the code sample of the video below, which is a part of the Hugging Face course.

#@title
from IPython.display import HTML

HTML('<iframe width="560" height="315" src="https://www.youtube.com/embed/OATCgQtNX2o?rel=0&amp;controls=0&amp;showinfo=0" frameborder="0" allowfullscreen></iframe>')

Install the Transformers and Datasets libraries to run this notebook.

! pip install datasets transformers[sentencepiece]

import torch
from transformers import AutoTokenizer, AutoModel

sentences = [
    "I took my dog for a walk",
    "Today is going to rain",
    "I took my cat for a walk",
]

model_ckpt = "sentence-transformers/all-MiniLM-L6-v2"
tokenizer = AutoTokenizer.from_pretrained(model_ckpt)
model = AutoModel.from_pretrained(model_ckpt)

encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors="pt")

with torch.no_grad():
    model_output = model(**encoded_input)
    
    
token_embeddings = model_output.last_hidden_state
print(f"Token embeddings shape: {token_embeddings.size()}")

import torch.nn.functional as F


def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output.last_hidden_state
    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
    )


sentence_embeddings = mean_pooling(model_output, encoded_input["attention_mask"])
# Normalize the embeddings
sentence_embeddings = F.normalize(sentence_embeddings, p=2, dim=1)
print(f"Sentence embeddings shape: {sentence_embeddings.size()}")

import numpy as np
from sklearn.metrics.pairwise import cosine_similarity

sentence_embeddings = sentence_embeddings.detach().numpy()

scores = np.zeros((sentence_embeddings.shape[0], sentence_embeddings.shape[0]))

for idx in range(sentence_embeddings.shape[0]):
    scores[idx, :] = cosine_similarity([sentence_embeddings[idx]], sentence_embeddings)[0]

from datasets import load_dataset

squad = load_dataset("squad", split="validation").shuffle(seed=42).select(range(100))


def get_embeddings(text_list):
    encoded_input = tokenizer(
        text_list, padding=True, truncation=True, return_tensors="pt"
    )
    encoded_input = {k: v for k, v in encoded_input.items()}
    with torch.no_grad():
        model_output = model(**encoded_input)
    return mean_pooling(model_output, encoded_input["attention_mask"])


squad_with_embeddings = squad.map(
    lambda x: {"embeddings": get_embeddings(x["context"]).cpu().numpy()[0]}
)

squad_with_embeddings.add_faiss_index(column="embeddings")

question = "Who headlined the halftime show for Super Bowl 50?"
question_embedding = get_embeddings([question]).cpu().detach().numpy()

scores, samples = squad_with_embeddings.get_nearest_examples(
    "embeddings", question_embedding, k=3
)