<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kronis-etal-2026-wmt26">
    <titleInfo>
        <title>WMT26 Model Compression Task: Quantising and Distilling TildeOpen</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Martins</namePart>
        <namePart type="family">Kronis</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Toms</namePart>
        <namePart type="family">Bergmanis</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ingus</namePart>
        <namePart type="family">Pretkalniņš</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Marcis</namePart>
        <namePart type="family">Pinnis</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the Eleventh Conference on Machine Translation</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Barry</namePart>
            <namePart type="family">Haddow</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Tom</namePart>
            <namePart type="family">Kocmi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Philipp</namePart>
            <namePart type="family">Koehn</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Christof</namePart>
            <namePart type="family">Monz</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-474-3</identifier>
    </relatedItem>
    <abstract>This paper describes Tilde’s submission to the unconstrained track of the WMT~2026 Model Compression task in the Czech-German direction. Our baseline is TildeOpen-15B-64k, a model distilled from the TildeOpen-30B-64k European foundation model. We adapt it for CS-DE translation, using supervised fine-tuning followed by GRPO-based reinforcement learning. From this baseline, we explore two axes of compression: post-training quantisation and knowledge distillation. For quantisation, we use LLM Compressor to produce a GPTQ-calibrated model with 4-bit NVFP4 weights (10.1\textbackslash,GB, $\textbackslashtimes$3 smaller than the 30.3\textbackslash,GB footprint of the 15B baseline) and a data-free RTN model with FP8 weights and activations (16.3\textbackslash,GB). In parallel, we prune and distil the base 15B model into an 8B student and fine-tune it with the same supervised fine-tuning recipe. Applying the two quantisation schemes to the 8B model yields our smallest models at 8.9\textbackslash,GB and 5.7\textbackslash,GB. The latter is over $\textbackslashtimes$5 more compact and $\textbackslashtimes$1.8 faster than the 15B baseline. Evaluation on four test sets across a variety of metrics shows that quantisation and distillation have minimal impact on translation quality. The NVFP4-quantised 15B model offers the best quality/size/speed trade-off and is our primary submission for this competition. All six models are publicly released.</abstract>
    <identifier type="citekey">kronis-etal-2026-wmt26</identifier>
    <location>
        <url>https://aclanthology.org/2026.wmt-main.137/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>1999</start>
            <end>2008</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T WMT26 Model Compression Task: Quantising and Distilling TildeOpen
%A Kronis, Martins
%A Bergmanis, Toms
%A Pretkalniņš, Ingus
%A Pinnis, Marcis
%Y Haddow, Barry
%Y Kocmi, Tom
%Y Koehn, Philipp
%Y Monz, Christof
%S Proceedings of the Eleventh Conference on Machine Translation
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-474-3
%F kronis-etal-2026-wmt26
%X This paper describes Tilde’s submission to the unconstrained track of the WMT~2026 Model Compression task in the Czech-German direction. Our baseline is TildeOpen-15B-64k, a model distilled from the TildeOpen-30B-64k European foundation model. We adapt it for CS-DE translation, using supervised fine-tuning followed by GRPO-based reinforcement learning. From this baseline, we explore two axes of compression: post-training quantisation and knowledge distillation. For quantisation, we use LLM Compressor to produce a GPTQ-calibrated model with 4-bit NVFP4 weights (10.1\textbackslash,GB, $\textbackslashtimes$3 smaller than the 30.3\textbackslash,GB footprint of the 15B baseline) and a data-free RTN model with FP8 weights and activations (16.3\textbackslash,GB). In parallel, we prune and distil the base 15B model into an 8B student and fine-tune it with the same supervised fine-tuning recipe. Applying the two quantisation schemes to the 8B model yields our smallest models at 8.9\textbackslash,GB and 5.7\textbackslash,GB. The latter is over $\textbackslashtimes$5 more compact and $\textbackslashtimes$1.8 faster than the 15B baseline. Evaluation on four test sets across a variety of metrics shows that quantisation and distillation have minimal impact on translation quality. The NVFP4-quantised 15B model offers the best quality/size/speed trade-off and is our primary submission for this competition. All six models are publicly released.
%U https://aclanthology.org/2026.wmt-main.137/
%P 1999-2008

Markdown (Informal)

[WMT26 Model Compression Task: Quantising and Distilling TildeOpen](https://aclanthology.org/2026.wmt-main.137/) (Kronis et al., WMT 2026)

• WMT26 Model Compression Task: Quantising and Distilling TildeOpen (Kronis et al., WMT 2026)

ACL

• Martins Kronis, Toms Bergmanis, Ingus Pretkalniņš, and Marcis Pinnis. 2026. WMT26 Model Compression Task: Quantising and Distilling TildeOpen. In Proceedings of the Eleventh Conference on Machine Translation, pages 1999–2008, Budapest, Hungary. Association for Computational Linguistics.