par microsoft
Open source · 41k downloads · 139 likes
Le modèle *harrier oss v1 270m* est un modèle d'embeddings multilingues développé par Microsoft, conçu pour générer des représentations vectorielles denses de textes. Il excelle dans des tâches variées comme la recherche d'informations, le clustering, l'évaluation de similarité sémantique, la classification ou encore le réordonnancement de résultats. Grâce à une architecture optimisée et un entraînement contrastif sur des jeux de données multilingues, il atteint des performances de pointe sur des benchmarks comme le *Multilingual MTEB v2*. Ce qui le distingue particulièrement, c'est sa capacité à s'adapter à différents scénarios via des instructions en langage naturel, permettant une personnalisation fine des embeddings sans nécessiter de réentraînement. Il supporte un large éventail de langues, couvrant les principales langues européennes, asiatiques et moyen-orientales, ce qui en fait un outil polyvalent pour des applications globales. Son approche par *last-token pooling* et normalisation L2 garantit des embeddings robustes et comparables, tout en restant accessible via des bibliothèques comme Sentence Transformers.
harrier-oss-v1 is a family of multilingual text embedding models developed by Microsoft. The models use decoder-only architectures with last-token pooling and L2 normalization to produce dense text embeddings. They can be applied to a wide range of tasks, including but not limited to retrieval, clustering, semantic similarity, classification, bitext mining, and reranking. The models achieve state-of-the-art results on the Multilingual MTEB v2 benchmark as of the release date.
| Model | Parameters | Embedding Dimension | Max Tokens | MTEB v2 Score |
|---|---|---|---|---|
| harrier-oss-v1-270m | 270M | 640 | 32,768 | 66.5 |
| harrier-oss-v1-0.6b | 0.6B | 1,024 | 32,768 | 69.0 |
| harrier-oss-v1-27b | 27B | 5,376 | 32,768 | 74.3 |
All models are trained with contrastive learning objectives on a large-scale mixture of multilingual datasets covering diverse tasks. The 270m and 0.6b variants are additionally trained with knowledge distillation from larger embedding models.
Below is an example to encode queries and passages from the MS-MARCO passage ranking dataset.
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("microsoft/harrier-oss-v1-270m", model_kwargs={"dtype": "auto"})
queries = [
"how much protein should a female eat",
"summit define",
]
documents = [
"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."
]
query_embeddings = model.encode(queries, prompt_name="web_search_query")
document_embeddings = model.encode(documents)
scores = (query_embeddings @ document_embeddings.T) * 100
print(scores.tolist())
Have a look at config_sentence_transformers.json for the prompts that are pre-configured, such as web_search_query, sts_query, and bitext_query. You can also use a custom instruction directly via e.g. model.encode(queries, prompt="Instruct: Retrieve semantically similar text\nQuery: ").
import torch
import torch.nn.functional as F
from torch import Tensor
from transformers import AutoTokenizer, AutoModel
def last_token_pool(last_hidden_states: Tensor, attention_mask: Tensor) -> Tensor:
left_padding = (attention_mask[:, -1].sum() == attention_mask.shape[0])
if left_padding:
return last_hidden_states[:, -1]
else:
sequence_lengths = attention_mask.sum(dim=1) - 1
batch_size = last_hidden_states.shape[0]
return last_hidden_states[torch.arange(batch_size, device=last_hidden_states.device), sequence_lengths]
def get_detailed_instruct(task_description: str, query: str) -> str:
return f'Instruct: {task_description}\nQuery: {query}'
# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'
queries = [
get_detailed_instruct(task, 'how much protein should a female eat'),
get_detailed_instruct(task, 'summit define')
]
# No need to add instruction for retrieval documents
documents = [
"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."
]
input_texts = queries + documents
tokenizer = AutoTokenizer.from_pretrained('microsoft/harrier-oss-v1-270m')
model = AutoModel.from_pretrained('microsoft/harrier-oss-v1-270m', dtype='auto')
model.eval()
model.cuda()
max_length = 32768
# Tokenize the input texts
batch_dict = tokenizer(input_texts, max_length=max_length, padding=True, truncation=True, return_tensors='pt')
batch_dict = {k: v.cuda() for k, v in batch_dict.items()}
outputs = model(**batch_dict)
embeddings = last_token_pool(outputs.last_hidden_state, batch_dict['attention_mask'])
# normalize embeddings
embeddings = F.normalize(embeddings, p=2, dim=1)
scores = (embeddings[:2] @ embeddings[2:].T) * 100
print(scores.tolist())
The models are trained on multilingual data and support a wide range of languages, including but not limited to: Arabic, Bulgarian, Catalan, Czech, Danish, German, Greek, English, Spanish, Estonian, Persian, Finnish, French, Hebrew, Hindi, Croatian, Hungarian, Indonesian, Italian, Japanese, Korean, Lithuanian, Latvian, Macedonian, Malay, Dutch, Norwegian, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Albanian, Serbian, Swedish, Thai, Turkish, Ukrainian, Urdu, Vietnamese, and Chinese.
Please follow the mteb repository on how to reproduce our scores. The evaluation prompts used for each task are also available at mteb_v2_eval_prompts.json.
1. Do I need to add instructions to the query?
Yes, this is how the model is trained, otherwise you will see a performance degradation. The task definition should be a one-sentence instruction that describes the task. This is a way to customize text embeddings for different scenarios through natural language instructions.
On the other hand, there is no need to add instructions to the document side.
2. Why are my reproduced results slightly different from reported in the model card?
Different versions of transformers and pytorch could cause negligible but non-zero performance differences.
3. What pooling strategy does this model use?
The model uses last-token pooling — the embedding of the last non-padding token is used as the sentence representation. The embedding is then L2-normalized. This is handled automatically when using Sentence Transformers.