Model Card for Mimir-7B-Instruct

How to Get Started with the Model

Below you can find an example of model usage. To facilitate its usage, we recommend to follow these steps:

huggingface-cli download mimir-lcm/Mimir-7B-Instruct --local-dir mimir-lcm/Mimir-7B-Instruct 
git clone https://github.com/facebookresearch/large_concept_model.git
mv large_concept_model/lcm .
pip install torch==2.5.1 --extra-index-url https://download.pytorch.org/whl/cu121 --upgrade
pip install fairseq2==v0.3.0rc1 --pre --extra-index-url  https://fair.pkg.atmeta.com/fairseq2/whl/rc/pt2.5.1/cu121 --upgrade
pip install omegaconf==2.3.0
pip install sonar-space==0.3.2
pip install wtpsplit==2.1.2

Now you should be able to run the following:

import lcm
import torch
from pathlib import Path

from lcm.models.two_tower_diffusion_lcm.builder import (
    create_two_tower_diffusion_lcm_model,
)
from lcm.models.two_tower_diffusion_lcm.archs import two_tower_diffusion_lcm_7B
from lcm.inference.two_tower_diffusion_lcm.generator import (
    TwoTowerDiffusionLCMGenerator,
    DiffusionLCMGeneratorOptions,
)
from lcm.datasets.batch import EmbeddingsBatch
from sonar.inference_pipelines.text import TextToEmbeddingModelPipeline, EmbeddingToTextModelPipeline

from wtpsplit import SaT

lcm.setup_fairseq2()

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

from lcm.models.two_tower_diffusion_lcm.builder import TwoTowerDiffusionLCModel

_original_sample_fn = TwoTowerDiffusionLCModel.sample_initial_noise_vectors

def _patched_sample_fn(self, batch_size: int):
    latents = _original_sample_fn(self, batch_size)
    return latents.to(dtype=self.dtype)

TwoTowerDiffusionLCModel.sample_initial_noise_vectors = _patched_sample_fn

CHECKPOINT_PATH = "mimir-lcm/Mimir-7B-Instruct/model.pt"
INFERENCE_DTYPE = torch.float16

TEXT_DECODER = EmbeddingToTextModelPipeline(decoder="text_sonar_basic_decoder", tokenizer="text_sonar_basic_decoder", device=torch.device(DEVICE))
TEXT_EMBEDDER = TextToEmbeddingModelPipeline(encoder="text_sonar_basic_encoder", tokenizer="text_sonar_basic_encoder", device=torch.device(DEVICE))

def decode_embeddings(embeddings):
    
    embeddings = embeddings.to(device=DEVICE, dtype=torch.float32)
    
    print("Decoding...")
    results = TEXT_DECODER.predict(
        embeddings, 
        target_lang="eng_Latn"
    )
    
    return results

def get_eos_vector():
    return TEXT_EMBEDDER.predict(["End of text."], source_lang="eng_Latn").squeeze().to(device=DEVICE, dtype=INFERENCE_DTYPE)

def load_two_tower_model(checkpoint_path, device="cuda"):
    
    config = two_tower_diffusion_lcm_7B()

    print("Building model structure...")
    model = create_two_tower_diffusion_lcm_model(
        config, 
        device=torch.device(device),
        dtype=INFERENCE_DTYPE
    )

    print(f"Loading weights from {checkpoint_path}...")
    state_dict = torch.load(checkpoint_path, map_location=device)
    
    if "model" in state_dict:
        state_dict = state_dict["model"]
        
    model.load_state_dict(state_dict, strict=True)
    
    model.eval()
    model.to(device=DEVICE, dtype=INFERENCE_DTYPE)
    print("Model loaded successfully.")
    return model

def run_inference(model, prompt_embeddings, device="cuda"):

    options = DiffusionLCMGeneratorOptions(
        eos_threshold=0.9,
        inference_timesteps=40,
        initial_noise_scale=0.6,
        guidance_scale=1.5,
        guidance_rescale=0.7,
        epsilon_scaling=1.00045,
        stop_on_repetition_cosine_threshold=0.9,
        seed=42,
    )
    
    generator = TwoTowerDiffusionLCMGenerator(model, options, eos_vec=get_eos_vector())

    seqs = prompt_embeddings.to(device)
    batch_input = EmbeddingsBatch(seqs=seqs, padding_mask=None)

    print("Running generation...")
    output = generator(batch_input)
    
    return output

if __name__ == "__main__":

    raw_prompt_text = "User turn.\n\nJohn lives in his house and loves to play soccer.\n\nGive a brief definition of the word \"house\" in the sentence given as input. Generate only the definition.\n\nAssistant turn."

    model = load_two_tower_model(CHECKPOINT_PATH, DEVICE)
    
    with torch.no_grad():

        sat_model = SaT("segment-any-text/sat-3l")
        if torch.cuda.is_available():
            sat_model.half().to(DEVICE)
            
        split_outputs = list(sat_model.split([raw_prompt_text], threshold=0.02))
        sentences = [s.strip() for s in split_outputs[0] if s.strip()]

        print(sentences)

        prompt = TEXT_EMBEDDER.predict(sentences, source_lang="eng_Latn", batch_size=1024)
        prompt = prompt.to(device=DEVICE, dtype=INFERENCE_DTYPE)
        prompt = prompt.unsqueeze(0)
    
    results = run_inference(model, prompt, DEVICE)
    
    for j, hyp in enumerate(results.hypotheses[0]):
        print(decode_embeddings(hyp.seq)[prompt.shape[1]:])

Make sure to change the source language from "eng_Latn" to the one you want to perform inference with.

Furthermore, for the instruct model, we used the following mappings for the "User turn." and "Assistant turn." strings.

USER_TRANSLATION = {
    "arb_Arab": "دور المستخدم.", 
    "bel_Cyrl": "Ход карыстальніка.", 
    "ben_Beng": "ব্যবহারকারীর পালা।",
    "bos_Latn": "Red korisnika.", 
    "bul_Cyrl": "Ред е на потребителя.", 
    "cat_Latn": "Torn de l'usuari.",
    "ces_Latn": "Je řada na uživateli.", 
    "cym_Latn": "Tro'r defnyddiwr.", 
    "dan_Latn": "Brugerens tur.",
    "deu_Latn": "Der Benutzer ist am Zug.", 
    "eng_Latn": "User turn.", 
    "fra_Latn": "C'est au tour de l'utilisateur.",
    "heb_Hebr": "תור המשתמש.", 
    "hin_Deva": "उपयोगकर्ता की बारी।", 
    "hrv_Latn": "Poteg korisnika.",
    "ind_Latn": "Giliran pengguna.", 
    "jpn_Jpan": "ユーザーの番です。", 
    "ita_Latn": "È il turno dell'utente.",
    "kan_Knda": "ಬಳಕೆದಾರರ ತಿರುವು.", 
    "kor_Hang": "사용자 차례입니다.", 
    "lvs_Latn": "Lietotāja kārta.",
    "mal_Mlym": "ഉപയോക്താവിന്റെ ഊഴം",
    "mar_Deva": "वापरकर्त्याची पाळी.",
    "mkd_Cyrl": "Потребен е корисник.",
    "npi_Deva": "प्रयोगकर्ताको पालो।",
    "nld_Latn": "De gebruiker is aan de beurt.",
    "ory_Orya": "ୟୁଜର୍ ଟର୍ନ୍ |",
    "pol_Latn": "Ruch użytkownika.",
    "por_Latn": "É a vez do utilizador.",
    "ron_Latn": "E rândul utilizatorului.",
    "rus_Cyrl": "Ход пользователя.",
    "slk_Latn": "Je na ťa.",
    "slv_Latn": "Na vrsti je uporabnik.",
    "srp_Cyrl": "Кориснички потез.",
    "spa_Latn": "Turno del usuario.",
    "swe_Latn": "Användarens tur.",
    "swh_Latn": "Mzunguko wa mtumiaji.",
    "tam_Taml": "பயனர் முறை.",
    "tel_Telu": "వినియోగదారుని వంతు",
    "tha_Thai": "ผู้ใช้เทิร์น",
    "tur_Latn": "Kullanıcı sırası.",
    "ukr_Cyrl": "Хід користувача.",
    "urd_Arab": "صارف کی باری۔",
    "vie_Latn": "Đến lượt người dùng.",
    "zho_Hans": "轮到用户了。"
}

ASSISTANT_TRANSLATION = {
    "arb_Arab": "دور المساعد.", 
    "bel_Cyrl": "Памочнік, свой ход.", 
    "ben_Beng": "সহকারী পালা।",
    "bos_Latn": "Pomoćni potez.", 
    "bul_Cyrl": "Ред е на помощника.", 
    "cat_Latn": "Torn de l'ajudant.",
    "ces_Latn": "Na řadě je asistent.", 
    "cym_Latn": "Tro'r cynorthwyydd.", 
    "dan_Latn": "Assistenten er på tur.",
    "deu_Latn": "Der Assistent ist an der Reihe.", 
    "eng_Latn": "Assistant turn.", 
    "fra_Latn": "C'est au tour de l'assistant.",
    "heb_Hebr": "תורו של העוזר.", 
    "hin_Deva": "सहायक की बारी।", 
    "hrv_Latn": "Pomoćni potez.",
    "ind_Latn": "Giliran asisten.", 
    "jpn_Jpan": "アシスタントの番。", 
    "ita_Latn": "È il turno dell'assistente.",
    "kan_Knda": "ಸಹಾಯಕ ತಿರುವು.", 
    "kor_Hang": "조수 차례.", 
    "lvs_Latn": "Palīga kārta.",
    "mal_Mlym": "സഹായിയുടെ ഊഴം",
    "mar_Deva": "सहाय्यक पालवी.",
    "mkd_Cyrl": "Помошник-ред.",
    "npi_Deva": "सहायक पालो।",
    "nld_Latn": "De assistent is aan de beurt.",
    "ory_Orya": "ଆସିଷ୍ଟାଣ୍ଟ ଟର୍ନ୍ |",
    "pol_Latn": "Rzuty pomocników.",
    "por_Latn": "É a vez do assistente.",
    "ron_Latn": "E rândul asistentului.",
    "rus_Cyrl": "Очередь помощника.",
    "slk_Latn": "Na rade je asistent.",
    "slv_Latn": "Na vrsti je pomočnik.",
    "srp_Cyrl": "Помоћни круг.",
    "spa_Latn": "Turno del asistente.",
    "swe_Latn": "Assistenten är på tur.",
    "swh_Latn": "Mzunguko wa msaidizi.",
    "tam_Taml": "துணைவரின் முறை.",
    "tel_Telu": "సహాయకుడి వంతు",
    "tha_Thai": "ผู้ช่วยเลี้ยว",
    "tur_Latn": "Sıra asistanında.",
    "ukr_Cyrl": "Черга помічника.",
    "urd_Arab": "معاون کی باری۔",
    "vie_Latn": "Đến lượt trợ lý.",
    "zho_Hans": "助手的回合。"
}

Evaluation Results

Below you can find plots comparing model performance between the 1.6B version of Mimir and the 7B version on MLQA and XL-Sum using the ROUGE-L metric.

MLQA

XL-Sum

Citation

If you use this model in your research, please cite the following:

@inproceedings{musacchio-etal-2026-mimir,
    title = "Mimir: Large-Scale Multilingual Concept Modeling",
    author = "Musacchio, Elio  and
      Siciliani, Lucia  and
      Basile, Pierpaolo",
    editor = "Lee, Hung-Yi  and
      Dev, Sunipa  and
      Li, Jessy  and
      Elliott, Desmond",
    booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
    month = oct,
    year = "2026",
    address = "Budapest, Hungary",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.emnlp-main.569/",
    pages = "13051--13073",
    ISBN = "979-8-89176-459-0",
    abstract = "Current language modeling approaches are built around tokens. Text corpora are split into tokens, and models are trained by performing computations on these tokens, such as predicting the next token given the preceding ones as context. This paradigm has become the standard in modern language modeling, especially given the outstanding performance obtained by token-based architectures. However, recent works have not only begun to question how language models process and understand meaning from tokens, but also to question whether using higher levels of granularity could advance the research field. This led to the idea of Concept Modeling, that is, to directly train models for next-concept prediction rather than next-token prediction. In this work, we introduce Mimir, a 1.6B Large Concept Model trained for multilingual concept understanding and generation. We leverage a large-scale multilingual pre-training corpus (38,883,987,240 sentences) spanning 46 languages and a large-scale multi-turn and multilingual instruction-tuning dataset (66,816,428 sentences) covering a total of 35 languages. We extensively evaluate model performance against a language model with a comparable number of parameters."
}