README.md

---
license: cc-by-nc-4.0
language:
- ace
- acm
- acq
- aeb
- af
- ajp
- ak
- am
- apc
- ar
- ars
- ary
- arz
- as
- ast
- awa
- ay
- azb
- azj
- ba
- bm
- ban
- be
- bem
- bn
- bho
- bjn
- bo
- bs
- bug
- bg
- ca
- ceb
- cs
- cjk
- ckb
- crh
- cy
- da
- de
- dik
- dyu
- dz
- el
- en
- eo
- et
- eu
- ee
- fo
- fa
- fj
- fi
- fon
- fr
- fur
- ff
- gd
- ga
- gl
- gn
- gu
- ht
- ha
- he
- hi
- hne
- hr
- hu
- hy
- ig
- ilo
- id
- is
- it
- jv
- ja
- kab
- kac
- kam
- kn
- ks
- ka
- kr
- kk
- kbp
- kea
- km
- ki
- rw
- ky
- kmb
- kg
- ko
- kmr
- lo
- lv
- lij
- li
- ln
- lt
- lmo
- ltg
- lb
- lua
- lg
- luo
- lus
- mag
- mai
- ml
- mr
- min
- mk
- plt
- mt
- mni
- mn
- mos
- mi
- ms
- my
- nl
- nn
- nb
- ne
- nso
- nus
- ny
- oc
- gaz
- ory
- pag
- pa
- pap
- pl
- pt
- prs
- pbt
- qu
- ro
- rn
- ru
- sg
- sa
- sat
- scn
- shn
- si
- sk
- sl
- sm
- sn
- sd
- so
- st
- es
- als
- sc
- sr
- ss
- su
- sv
- sw
- szl
- ta
- tt
- te
- tg
- tl
- th
- ti
- taq
- tpi
- tn
- ts
- tk
- tum
- tr
- tw
- tzm
- ug
- uk
- umb
- ur
- uz
- vec
- vi
- war
- wo
- xh
- yi
- yo
- yue
- zh
- zu
language_details: >-
  ace_Arab, ace_Latn, acm_Arab, acq_Arab, aeb_Arab, afr_Latn, ajp_Arab,
  aka_Latn, amh_Ethi, apc_Arab, arb_Arab, ars_Arab, ary_Arab, arz_Arab,
  asm_Beng, ast_Latn, awa_Deva, ayr_Latn, azb_Arab, azj_Latn, bak_Cyrl,
  bam_Latn, ban_Latn, bel_Cyrl, bem_Latn, ben_Beng, bho_Deva, bjn_Arab,
  bod_Tibt, bos_Latn, bug_Latn, bul_Cyrl, cat_Latn, ceb_Latn, ces_Latn,
  cjk_Latn, ckb_Arab, crh_Latn, cym_Latn, dan_Latn, deu_Latn, dik_Latn,
  dyu_Latn, dzo_Tibt, ell_Grek, eng_Latn, epo_Latn, est_Latn, eus_Latn,
  ewe_Latn, fao_Latn, pes_Arab, fij_Latn, fin_Latn, fon_Latn, fra_Latn,
  fur_Latn, fuv_Latn, gla_Latn, gle_Latn, glg_Latn, grn_Latn, guj_Gujr,
  hat_Latn, hau_Latn, heb_Hebr, hin_Deva, hne_Deva, hrv_Latn, hun_Latn,
  hye_Armn, ibo_Latn, ilo_Latn, ind_Latn, isl_Latn, ita_Latn, jav_Latn,
  jpn_Jpan, kab_Latn, kac_Latn, kam_Latn, kan_Knda, kas_Arab, kas_Deva,
  kat_Geor, knc_Arab, knc_Latn, kaz_Cyrl, kbp_Latn, kea_Latn, khm_Khmr,
  kik_Latn, kin_Latn, kir_Cyrl, kmb_Latn, kon_Latn, kor_Hang, kmr_Latn,
  lao_Laoo, lvs_Latn, lij_Latn, lim_Latn, lin_Latn, lit_Latn, lmo_Latn,
  ltg_Latn, ltz_Latn, lua_Latn, lug_Latn, luo_Latn, lus_Latn, mag_Deva,
  mai_Deva, mal_Mlym, mar_Deva, min_Latn, mkd_Cyrl, plt_Latn, mlt_Latn,
  mni_Beng, khk_Cyrl, mos_Latn, mri_Latn, zsm_Latn, mya_Mymr, nld_Latn,
  nno_Latn, nob_Latn, npi_Deva, nso_Latn, nus_Latn, nya_Latn, oci_Latn,
  gaz_Latn, ory_Orya, pag_Latn, pan_Guru, pap_Latn, pol_Latn, por_Latn,
  prs_Arab, pbt_Arab, quy_Latn, ron_Latn, run_Latn, rus_Cyrl, sag_Latn,
  san_Deva, sat_Beng, scn_Latn, shn_Mymr, sin_Sinh, slk_Latn, slv_Latn,
  smo_Latn, sna_Latn, snd_Arab, som_Latn, sot_Latn, spa_Latn, als_Latn,
  srd_Latn, srp_Cyrl, ssw_Latn, sun_Latn, swe_Latn, swh_Latn, szl_Latn,
  tam_Taml, tat_Cyrl, tel_Telu, tgk_Cyrl, tgl_Latn, tha_Thai, tir_Ethi,
  taq_Latn, taq_Tfng, tpi_Latn, tsn_Latn, tso_Latn, tuk_Latn, tum_Latn,
  tur_Latn, twi_Latn, tzm_Tfng, uig_Arab, ukr_Cyrl, umb_Latn, urd_Arab,
  uzn_Latn, vec_Latn, vie_Latn, war_Latn, wol_Latn, xho_Latn, ydd_Hebr,
  yor_Latn, yue_Hant, zho_Hans, zho_Hant, zul_Latn
pipeline_tag: sentence-similarity
---

# BLASER Ref (Ported)  

This is a **ported version of the BLASER quality estimation (REF) model** originally developed in [BLASER: Bilingual Language-Agnostic Sentence Representations](https://huggingface.co/facebook/blaser-2.0-ref).  

- **Ported to Hugging Face Transformers**: no dependency on Fairseq.  
- **Uses embeddings from the ported SONAR 200 multilingual text encoder** ([cointegrated/SONAR_200_text_encoder](https://huggingface.co/cointegrated/SONAR_200_text_encoder)).  
- **Supports the same 202 languages** as SONAR / NLLB-200.  
- **Outputs BLASER scores on a 1–5 scale** for a source–MT–REF triplet.

> ⚠️ This is **not the original implementation**. Attribution goes to the original BLASER authors.

---

## How to compute Ref scores

```python
# !pip install transformers sentencepiece  torch -q
import torch
from transformers import AutoTokenizer, AutoModel
from transformers.models.m2m_100.modeling_m2m_100 import M2M100Encoder

# 1. Load SONAR encoder
sonar_model_name = "cointegrated/SONAR_200_text_encoder"
encoder = M2M100Encoder.from_pretrained(sonar_model_name)
tokenizer = AutoTokenizer.from_pretrained(sonar_model_name)

def encode_mean_pool(texts, tokenizer, encoder, lang='eng_Latn', norm=False):
    tokenizer.src_lang = lang
    with torch.inference_mode():
        batch = tokenizer(texts, return_tensors='pt', padding=True)
        seq_embs = encoder(**batch).last_hidden_state
        mask = batch.attention_mask
        mean_emb = (seq_embs * mask.unsqueeze(-1)).sum(1) / mask.unsqueeze(-1).sum(1)
        if norm:
            mean_emb = torch.nn.functional.normalize(mean_emb)
    return mean_emb

# Example sentences
src_sentences = ["Le chat s'assit sur le tapis."]
mt_sentences = ["The cat sat down on the carpet."]  # Example MT output
ref_sentences = ["The cat sat on the mat."]  # Example reference translation

# Encode source and MT sentences
src_embs = encode_mean_pool(src_sentences, tokenizer, encoder, lang="fra_Latn")
mt_embs = encode_mean_pool(mt_sentences, tokenizer, encoder, lang="eng_Latn")
ref_embs = encode_mean_pool(ref_sentences, tokenizer, encoder, lang="eng_Latn")

# 2. Load BLASER Ref model (ported)
ref_model_name = "oist/blaser-2.0-ref-ported"
ref_model = AutoModel.from_pretrained(qe_model_name, trust_remote_code=True)
ref_model.eval()  # set to evaluation mode

# 3. Compute Ref scores
with torch.inference_mode():
    ref_scores = ref_model(src_embs, mt_embs, ref_embs)  # expects source and MT embeddings, and ref embeddings
    print("Blaser score shape:", ref_scores.shape)
    print("Blaser scores:", ref_scores[0])