quester-qwen3-4B

QueStER Qwen3 Keyword Model

LoRA adapter for Qwen3-4B that generates keyword-based query expansions for retrieval.

Paper

Accepted to EACL 2026

This model is associated with the paper QueStER: Query Specification for Generative Keyword-Based Retrieval. Training code github

Install

pip install -U transformers  accelerate torch safetensors

Load and run

import torch
from transformers import AutoModelForCausalLM
from transformers import AutoTokenizer

model_id = "Arthur-75/quester-qwen3-4B"

system_prompt = "Generate relevant single-word keywords to improve retrieval performance. Only output unique keywords, separated by commas."

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    trust_remote_code=True,
)
model.eval()


def clean_output(text: str) -> str:
    text = text.strip()
    if "</think>" in text:
        text = text.split("</think>", 1)[-1].strip()
    if "<think>" in text:
        text = text.split("<think>", 1)[-1].strip()
    if text.lower().startswith("keywords:"):
        text = text[len("keywords:"):].strip()
    if text.lower().startswith("keyword:"):
        text = text[len("keyword:"):].strip()
    if "\nassistant" in text:               
        text = text.split("\nassistant", 1)[-1].strip()
    if "assistant\n" in text:              
        text = text.split("assistant\n", 1)[-1].strip()
    return text


def generate_keywords(query: str) -> str:
    messages =  [
            {"role": "system", "content": system_prompt},
            {"role": "user", "content":"[QUERY]: "+ query+" [KEYWORDS]:"},
        ]

    prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
        enable_thinking=False,
    )

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    with torch.inference_mode():
        output = model.generate(
            **inputs,
            max_new_tokens=64,
            do_sample=False,
        )

    new_tokens = output[:, inputs["input_ids"].shape[1]:]
    text = tokenizer.decode(new_tokens[0], skip_special_tokens=True)
    return clean_output(text)


query = "What are the symptoms of vitamin D deficiency?"
print(generate_keywords(query))

Citation

@inproceedings{satouf-etal-2026-quester,
    title = "{Q}ue{S}t{ER}: Query Specification for Generative Keyword-Based Retrieval",
    author = "Satouf, Arthur  and
      Zong, Yuxuan  and
      Boubacar, Habiboulaye Amadou  and
      Piantanida, Pablo  and
      Piwowarski, Benjamin",
    editor = "Demberg, Vera  and
      Inui, Kentaro  and
      Marquez, Llu{\'i}s",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EACL} 2026",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-eacl.312/",
    doi = "10.18653/v1/2026.findings-eacl.312",
    pages = "5957--5968",
    ISBN = "979-8-89176-386-9",
    abstract = "Generative retrieval (GR) differs from the traditional index{--}then{--}retrieve pipeline by storing relevance in model parameters and generating retrieval cues directly from the query, but it can be brittle out of domain and expensive to scale. We introduce QueStER (QUEry SpecificaTion for gEnerative Keyword-Based Retrieval), which bridges GR and query reformulation by learning to generate explicit keyword-based search specifications. Given a user query, a lightweight LLM produces a keyword query that is executed by a standard retriever (BM25), combining the generalization benefits of generative query rewriting with the efficiency and scalability of lexical indexing. We train the rewriting policy with reinforcement learning techniques. Across in- and out-of-domain evaluations, QueStER consistently improves over BM25 and is competitive with neural IR baselines, while maintaining strong efficiency."
}