F2LLM-v2-4B

F2LLM-v2-4B

F2LLM-v2 is a family of general-purpose, multilingual embedding models in 8 distinct sizes ranging from 80M to 14B. Trained on a curated composite of 60 million publicly available high-quality data, F2LLM-v2 supports more than 200 languages, with a particular emphasis on previously underserved mid- and low-resource languages.

F2LLM-v2 is fully open. We release base models in 5 sizes, instruct models in 8 sizes, the training data, the training code, and intermediate checkpoints. The three smallest instruct models are pruned and trained from the 0.6B base model.

Performance

The F2LLM-v2 family set a new state-of-the-art on a wide range of MTEB benchmarks, including Code, European, Scandinavian, German, French, Spanish, Polish, Dutch, Japanese, Vietnamese, Thai, Indic, Persian, among others.

<img src="img/performance.png" width="100%" alt="Performance">

For details, refer to the MTEB leaderboard.

Usage

With Sentence Transformers

To encode text with the Sentence Transformers library:

from sentence_transformers import SentenceTransformer
model = SentenceTransformer("codefuse-ai/F2LLM-v2-4B", device="cuda:0", model_kwargs={"torch_dtype": "bfloat16"})
# Some sample query and documents
query = "What is F2LLM used for?"
documents = [
    'We present F2LLM, a family of fully open embedding LLMs that achieve a strong balance between model size, training data, and embedding performance.',
    'F2LLM is a model for computing text embeddings that can be used for various NLP tasks such as information retrieval, semantic search, and text classification.',
    'F2LLM 是 CodeFuse 开源的系列嵌入模型。',
    'F2LLM — это модель вычисления встраивания текста, которую можно использовать для различных задач НЛП, таких как поиск информации, семантический поиск и классификация текста.'
]
# Encode the query and documents separately. The encode_query method uses the query prompt
query_embedding = model.encode_query(query)
document_embeddings = model.encode_document(documents)
print(query_embedding.shape, document_embeddings.shape)
# (2560,) (4, 2560)
# Compute cosine similarity between the query and documents
similarity = model.similarity(query_embedding, document_embeddings)
print(similarity)
# tensor([[0.6348, 0.8547, 0.7168, 0.8356]])

With Transformers

Or directly with the Transformers library:

from transformers import AutoModel, AutoTokenizer
import torch
import torch.nn.functional as F
model_path = "codefuse-ai/F2LLM-v2-4B"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModel.from_pretrained(model_path, torch_dtype=torch.bfloat16, device_map={'': 0})
query = "What is F2LLM used for?"
query_prompt = "Instruct: Given a question, retrieve passages that can help answer the question.\nQuery: "
documents = [
    'We present F2LLM, a family of fully open embedding LLMs that achieve a strong balance between model size, training data, and embedding performance.',
    'F2LLM is a model for computing text embeddings that can be used for various NLP tasks such as information retrieval, semantic search, and text classification.',
    'F2LLM 是 CodeFuse 开源的系列嵌入模型。',
    'F2LLM — это модель вычисления встраивания текста, которую можно использовать для различных задач НЛП, таких как поиск информации, семантический поиск и классификация текста.'
]
def encode(sentences):
    batch_size = len(sentences)
    # the tokenizer will automatically add eos token
    tokenized_inputs = tokenizer(sentences, padding=True, return_tensors='pt').to(model.device)
    last_hidden_state = model(**tokenized_inputs).last_hidden_state
    eos_positions = tokenized_inputs.attention_mask.sum(dim=1) - 1
    embeddings = last_hidden_state[torch.arange(batch_size, device=model.device), eos_positions]
    embeddings = F.normalize(embeddings, p=2, dim=1)
    return embeddings
# Encode the query and documents
query_embedding = encode([query_prompt + query])
document_embeddings = encode(documents)
print(query_embedding.shape, document_embeddings.shape)
# torch.Size([1, 2560]) torch.Size([4, 2560])
# Compute cosine similarity between the query and documents
similarity = query_embedding @ document_embeddings.T
print(similarity)
# tensor([[0.6328, 0.8555, 0.7148, 0.8398]], device='cuda:0',
#        dtype=torch.bfloat16, grad_fn=<MmBackward0>)

Prompts

The model supports custom instructions in the following format:

Instruct: your_instruction
Query:

In general, for retrieval and reranking tasks:

• use the prompt for queries
• do not prepend the prompt to documents/passages

For symmetric tasks such as STS, clustering, and bitext mining, you can encode the documents either with or without prompts. The model is trained to support both scenarios.

MRL Support

This model is trained with Matryoshka Representation Learning (MRL), allowing for a superior tradeoff between performance and embedding size. You can truncate the embeddings to keep only the first d dimensions to reduce storage and speed up vector search in downstream systems. The model is trained with a smallest Matryoshka dimension of 8.

<img src="img/mrl.png" width="80%" alt="MRL results">

Example:

embedding = embedding[..., :128]
embedding = torch.nn.functional.normalize(embedding, p=2, dim=-1)

Note: you need to apply normalization after trucation, not the other way around.

If you are using sentence transformer, you can also simply pass truncate_dim=128 to the encode interface.

Intermediate Checkpoints

To facilitate future research, we release intermediate checkpoints in the intermediate_checkpoints branch.

Citation

@misc{f2llm-v2,
      title={F2LLM-v2: Inclusive, Performant, and Efficient Embeddings for a Multilingual World}, 
      author={Ziyin Zhang and Zihan Liao and Hang Yu and Peng Di and Rui Wang},
      year={2026},
      eprint={2603.19223},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2603.19223}, 
}