E5-mistral-7b-instruct

Improving Text Embeddings with Large Language Models. Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, Furu Wei, arXiv 2024

This model has 32 layers and the embedding size is 4096.

Check MTEB leaderboard.

Model description

Please, refer to the original model card for more details on E5-mistral-7b-instruct.

Use with mlx-llm

Download weights from files section and install mlx-llm from GitHub.

git clone https://github.com/riccardomusmeci/mlx-llm
cd mlx-llm
pip install .

Run

import mlx.core as mx
import numpy as np
from mlx_llm.model import create_model
from transformers import AutoTokenizer

model = create_model(
  "e5-mistral-7b-instruct",
  weights_path="path/to/weights.npz",
  strict=False
)

def get_detailed_instruct(task_description: str, query: str) -> str:
    return f'Instruct: {task_description}\nQuery: {query}'

def last_token_pool(embeds: mx.array, attn_mask: mx.array) -> mx.array:
    left_padding = (attn_mask[:, -1].sum() == attn_mask.shape[0])
    if left_padding:
        return embeds[:, -1]
    else:
        sequence_lengths = attn_mask.sum(axis=1) - 1
        batch_size = embeds.shape[0]
        return embeds[mx.arange(batch_size), sequence_lengths]

task = 'Given a web search query, retrieve relevant passages that answer the query'

input_texts = [
    get_detailed_instruct(task, 'how much protein should a female eat'),
    # get_detailed_instruct(task, 'summit define'),
    "As a general guideline, the CDC's average requirement of protein for women ages 19 to 70 is 46 grams per day. But, as you can see from this chart, you'll need to increase that if you're expecting or training for a marathon. Check out the chart below to see how much protein you should be eating each day.",
    "Definition of summit for English Language Learners. : 1  the highest point of a mountain : the top of a mountain. : 2  the highest level. : 3  a meeting or series of meetings between the leaders of two or more governments."
]

tokenizer = AutoTokenizer.from_pretrained('intfloat/e5-mistral-7b-instruct')

# prepare input and attn_mask
max_length = 4096
batch_dict = tokenizer(
    input_texts, 
    max_length=max_length - 1, 
    return_attention_mask=False, 
    padding=False, 
    truncation=True
)

batch_dict['input_ids'] = [
    input_ids + [tokenizer.eos_token_id] for input_ids in batch_dict['input_ids']
]

batch_dict = tokenizer.pad(
    batch_dict, 
    padding=True, 
    return_attention_mask=True, 
    return_tensors='np'
)

x = mx.array(batch_dict["input_ids"].tolist())
attn_mask = mx.array(batch_dict["attention_mask"].tolist())

# compute embed
embeds = model.embed(x)
mx.eval(embeds)

embeds = np.array(last_token_pool(embeds, attn_mask))

# Normalize embeds
norm_den = np.linalg.norm(embeds, axis=-1)
norm_embeds = embeds / norm_den[:, None]

scores = (norm_embeds @ norm_embeds.T) * 100
print(scores)