<?xml version="1.0"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="fr">
	<id>https://datafranca.org/wiki/index.php?action=history&amp;feed=atom&amp;title=Multi-Token_Projection</id>
	<title>Multi-Token Projection - Historique des versions</title>
	<link rel="self" type="application/atom+xml" href="https://datafranca.org/wiki/index.php?action=history&amp;feed=atom&amp;title=Multi-Token_Projection"/>
	<link rel="alternate" type="text/html" href="https://datafranca.org/wiki/index.php?title=Multi-Token_Projection&amp;action=history"/>
	<updated>2026-05-06T01:35:19Z</updated>
	<subtitle>Historique des versions pour cette page sur le wiki</subtitle>
	<generator>MediaWiki 1.39.5</generator>
	<entry>
		<id>https://datafranca.org/wiki/index.php?title=Multi-Token_Projection&amp;diff=113181&amp;oldid=prev</id>
		<title>Arianne le 26 juin 2025 à 16:27</title>
		<link rel="alternate" type="text/html" href="https://datafranca.org/wiki/index.php?title=Multi-Token_Projection&amp;diff=113181&amp;oldid=prev"/>
		<updated>2025-06-26T16:27:24Z</updated>

		<summary type="html">&lt;p&gt;&lt;/p&gt;
&lt;table style=&quot;background-color: #fff; color: #202122;&quot; data-mw=&quot;interface&quot;&gt;
				&lt;col class=&quot;diff-marker&quot; /&gt;
				&lt;col class=&quot;diff-content&quot; /&gt;
				&lt;col class=&quot;diff-marker&quot; /&gt;
				&lt;col class=&quot;diff-content&quot; /&gt;
				&lt;tr class=&quot;diff-title&quot; lang=&quot;fr&quot;&gt;
				&lt;td colspan=&quot;2&quot; style=&quot;background-color: #fff; color: #202122; text-align: center;&quot;&gt;← Version précédente&lt;/td&gt;
				&lt;td colspan=&quot;2&quot; style=&quot;background-color: #fff; color: #202122; text-align: center;&quot;&gt;Version du 26 juin 2025 à 12:27&lt;/td&gt;
				&lt;/tr&gt;&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-lineno&quot; id=&quot;mw-diff-left-l2&quot;&gt;Ligne 2 :&lt;/td&gt;
&lt;td colspan=&quot;2&quot; class=&quot;diff-lineno&quot;&gt;Ligne 2 :&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Définition ==&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Définition ==&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;−&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;del style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;xxxxxxx&lt;/del&gt;&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;Technique qui permet au modèle de prédire plusieurs &lt;/ins&gt;&amp;#039;&amp;#039;&amp;#039;[[&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;Segment textuel|segments textuels&lt;/ins&gt;]]&amp;#039;&amp;#039;&amp;#039; &lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;en un seul passage. Elle permet aussi de planifier &lt;/ins&gt;et &lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;de générer stratégiquement des représentations qui facilitent une prédiction plus précise et potentiellement plus rapide des futurs segments textuels. Elle est utilisée dans les modèles &lt;/ins&gt;&amp;#039;&amp;#039;&amp;#039;[[&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;DeepSeek&lt;/ins&gt;]]&amp;#039;&amp;#039;&amp;#039; &lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;et fonctionne en ajoutant des modules spécialisés qui prédisent non seulement le segment textuel de base, mais aussi plusieurs segments textuels à venir dans la séquence.&lt;/ins&gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;−&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt; &lt;/div&gt;&lt;/td&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-side-added&quot;&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;−&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;del style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;Voir aussi &lt;/del&gt;&amp;#039;&amp;#039;&amp;#039;[[&lt;del style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;apprentissage par renforcement avec rétroaction humaine&lt;/del&gt;]]&amp;#039;&amp;#039;&amp;#039; et &amp;#039;&amp;#039;&amp;#039;[[&lt;del style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;réseau autoattentif&lt;/del&gt;]]&amp;#039;&amp;#039;&amp;#039;&lt;/div&gt;&lt;/td&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-side-added&quot;&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Compléments ==&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Compléments ==&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;−&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&amp;#039;&amp;#039;&lt;del style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;à faire&lt;/del&gt;&amp;#039;&amp;#039;&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;Cette technique est utilisée entre autres par &lt;/ins&gt;&amp;#039;&amp;#039;&amp;#039;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;[[ChatGPT]]&lt;/ins&gt;&amp;#039;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;&amp;#039;&amp;#039;-4, &amp;#039;&amp;#039;&amp;#039;[[Claude]]&amp;#039;&amp;#039;&amp;#039; 3, &amp;#039;&amp;#039;&amp;#039;[[LLaMA]]&amp;#039;&amp;#039;&amp;#039; 3, &amp;#039;&amp;#039;&amp;#039;[[BERT]]&amp;#039;&amp;#039;&amp;#039;.&lt;/ins&gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-side-deleted&quot;&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;&amp;lt;hr/&amp;gt;&lt;/ins&gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-side-deleted&quot;&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;Les applications de la projection de segments textuels multiple sont nombreuses. On la retrouve dans les &amp;#039;&amp;#039;&amp;#039;[[Robot conversationnel|robots conversationnels]]&amp;#039;&amp;#039;&amp;#039;, les &amp;#039;&amp;#039;&amp;#039;[[Assistant virtuel|assistants virtuels]]&amp;#039;&amp;#039;&amp;#039; ainsi qu&amp;#039;en &amp;#039;&amp;#039;&amp;#039;[[traduction automatique neuronale]]&amp;#039;&amp;#039;&amp;#039;.&lt;/ins&gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-side-deleted&quot;&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;&amp;lt;hr/&amp;gt;&lt;/ins&gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-side-deleted&quot;&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;Cette technique se repose aussi sur de nombreuses avancées clés, telles que les &amp;#039;&amp;#039;&amp;#039;[[Réseau autoattentif|réseaux autoattentifs]]&amp;#039;&amp;#039;&amp;#039;, les &amp;#039;&amp;#039;&amp;#039;[[Modèle autorégressif|modèles autorégressifs]]&amp;#039;&amp;#039;&amp;#039;, les &amp;#039;&amp;#039;&amp;#039;modèles bidirectionnels&amp;#039;&amp;#039;&amp;#039;, le &amp;#039;&amp;#039;&amp;#039;[[peaufinage]]&amp;#039;&amp;#039;&amp;#039; et l&amp;#039;&amp;#039;&amp;#039;&amp;#039;[[apprentissage par renforcement avec rétroaction humaine]]&amp;#039;&amp;#039;&amp;#039;.&lt;/ins&gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Français ==&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Français ==&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-lineno&quot; id=&quot;mw-diff-left-l18&quot;&gt;Ligne 18 :&lt;/td&gt;
&lt;td colspan=&quot;2&quot; class=&quot;diff-lineno&quot;&gt;Ligne 20 :&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&amp;#039;&amp;#039;&amp;#039; MTP&amp;#039;&amp;#039;&amp;#039;&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&amp;#039;&amp;#039;&amp;#039; MTP&amp;#039;&amp;#039;&amp;#039;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;−&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #ffe49c; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&amp;lt;!--Technique that enables the model to predict multiple token un a single forward pass and to strategically pre-plan and generate representations that facilitate more accurate and potentially faster prediction of future tokens. It is used in DeepSeek models and it works by adding specialized &lt;del style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;moules &lt;/del&gt;that predict not only the nest token but also several tokens ahead in the sequence.--&amp;gt;&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&amp;lt;!--Technique that enables the model to predict multiple token un a single forward pass and to strategically pre-plan and generate representations that facilitate more accurate and potentially faster prediction of future tokens. It is used in DeepSeek models and it works by adding specialized &lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;modules &lt;/ins&gt;that predict not only the nest token but also several tokens ahead in the sequence.--&amp;gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Sources ==&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Sources ==&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;/table&gt;</summary>
		<author><name>Arianne</name></author>
	</entry>
	<entry>
		<id>https://datafranca.org/wiki/index.php?title=Multi-Token_Projection&amp;diff=113167&amp;oldid=prev</id>
		<title>Arianne le 25 juin 2025 à 19:00</title>
		<link rel="alternate" type="text/html" href="https://datafranca.org/wiki/index.php?title=Multi-Token_Projection&amp;diff=113167&amp;oldid=prev"/>
		<updated>2025-06-25T19:00:19Z</updated>

		<summary type="html">&lt;p&gt;&lt;/p&gt;
&lt;table style=&quot;background-color: #fff; color: #202122;&quot; data-mw=&quot;interface&quot;&gt;
				&lt;col class=&quot;diff-marker&quot; /&gt;
				&lt;col class=&quot;diff-content&quot; /&gt;
				&lt;col class=&quot;diff-marker&quot; /&gt;
				&lt;col class=&quot;diff-content&quot; /&gt;
				&lt;tr class=&quot;diff-title&quot; lang=&quot;fr&quot;&gt;
				&lt;td colspan=&quot;2&quot; style=&quot;background-color: #fff; color: #202122; text-align: center;&quot;&gt;← Version précédente&lt;/td&gt;
				&lt;td colspan=&quot;2&quot; style=&quot;background-color: #fff; color: #202122; text-align: center;&quot;&gt;Version du 25 juin 2025 à 15:00&lt;/td&gt;
				&lt;/tr&gt;&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-lineno&quot; id=&quot;mw-diff-left-l18&quot;&gt;Ligne 18 :&lt;/td&gt;
&lt;td colspan=&quot;2&quot; class=&quot;diff-lineno&quot;&gt;Ligne 18 :&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&amp;#039;&amp;#039;&amp;#039; MTP&amp;#039;&amp;#039;&amp;#039;&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&amp;#039;&amp;#039;&amp;#039; MTP&amp;#039;&amp;#039;&amp;#039;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td colspan=&quot;2&quot; class=&quot;diff-side-deleted&quot;&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot; data-marker=&quot;+&quot;&gt;&lt;/td&gt;&lt;td style=&quot;color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #a3d3ff; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;&lt;ins style=&quot;font-weight: bold; text-decoration: none;&quot;&gt;&amp;lt;!--Technique that enables the model to predict multiple token un a single forward pass and to strategically pre-plan and generate representations that facilitate more accurate and potentially faster prediction of future tokens. It is used in DeepSeek models and it works by adding specialized moules that predict not only the nest token but also several tokens ahead in the sequence.--&gt;&lt;/ins&gt;&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;br/&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Sources ==&lt;/div&gt;&lt;/td&gt;&lt;td class=&quot;diff-marker&quot;&gt;&lt;/td&gt;&lt;td style=&quot;background-color: #f8f9fa; color: #202122; font-size: 88%; border-style: solid; border-width: 1px 1px 1px 4px; border-radius: 0.33em; border-color: #eaecf0; vertical-align: top; white-space: pre-wrap;&quot;&gt;&lt;div&gt;== Sources ==&lt;/div&gt;&lt;/td&gt;&lt;/tr&gt;
&lt;/table&gt;</summary>
		<author><name>Arianne</name></author>
	</entry>
	<entry>
		<id>https://datafranca.org/wiki/index.php?title=Multi-Token_Projection&amp;diff=113141&amp;oldid=prev</id>
		<title>Arianne : Page créée avec « == en construction ==  == Définition == xxxxxxx  Voir aussi &#039;&#039;&#039;apprentissage par renforcement avec rétroaction humaine&#039;&#039;&#039; et &#039;&#039;&#039;réseau autoattentif&#039;&#039;&#039;  == Compléments == &#039;&#039;à faire&#039;&#039;  == Français == &#039;&#039;&#039; XXXXXX&#039;&#039;&#039;  == Anglais == &#039;&#039;&#039; Multi-Token Projection&#039;&#039;&#039;  &#039;&#039;&#039; Multiple-Token Projection&#039;&#039;&#039;  &#039;&#039;&#039; MTP&#039;&#039;&#039;  == Sources == [https://arxiv.org/html/2505.07608v1   Source : arxiv]  [https://datascientest.com/multi-token-prediction-tout-savoir   Source : DataSc... »</title>
		<link rel="alternate" type="text/html" href="https://datafranca.org/wiki/index.php?title=Multi-Token_Projection&amp;diff=113141&amp;oldid=prev"/>
		<updated>2025-06-20T20:08:15Z</updated>

		<summary type="html">&lt;p&gt;Page créée avec « == en construction ==  == Définition == xxxxxxx  Voir aussi &amp;#039;&amp;#039;&amp;#039;&lt;a href=&quot;/wiki/Apprentissage_par_renforcement_avec_r%C3%A9troaction_humaine&quot; title=&quot;Apprentissage par renforcement avec rétroaction humaine&quot;&gt;apprentissage par renforcement avec rétroaction humaine&lt;/a&gt;&amp;#039;&amp;#039;&amp;#039; et &amp;#039;&amp;#039;&amp;#039;&lt;a href=&quot;/wiki/R%C3%A9seau_autoattentif&quot; title=&quot;Réseau autoattentif&quot;&gt;réseau autoattentif&lt;/a&gt;&amp;#039;&amp;#039;&amp;#039;  == Compléments == &amp;#039;&amp;#039;à faire&amp;#039;&amp;#039;  == Français == &amp;#039;&amp;#039;&amp;#039; XXXXXX&amp;#039;&amp;#039;&amp;#039;  == Anglais == &amp;#039;&amp;#039;&amp;#039; Multi-Token Projection&amp;#039;&amp;#039;&amp;#039;  &amp;#039;&amp;#039;&amp;#039; Multiple-Token Projection&amp;#039;&amp;#039;&amp;#039;  &amp;#039;&amp;#039;&amp;#039; MTP&amp;#039;&amp;#039;&amp;#039;  == Sources == [https://arxiv.org/html/2505.07608v1   Source : arxiv]  [https://datascientest.com/multi-token-prediction-tout-savoir   Source : DataSc... »&lt;/p&gt;
&lt;p&gt;&lt;b&gt;Nouvelle page&lt;/b&gt;&lt;/p&gt;&lt;div&gt;== en construction ==&lt;br /&gt;
&lt;br /&gt;
== Définition ==&lt;br /&gt;
xxxxxxx&lt;br /&gt;
&lt;br /&gt;
Voir aussi &amp;#039;&amp;#039;&amp;#039;[[apprentissage par renforcement avec rétroaction humaine]]&amp;#039;&amp;#039;&amp;#039; et &amp;#039;&amp;#039;&amp;#039;[[réseau autoattentif]]&amp;#039;&amp;#039;&amp;#039;&lt;br /&gt;
&lt;br /&gt;
== Compléments ==&lt;br /&gt;
&amp;#039;&amp;#039;à faire&amp;#039;&amp;#039;&lt;br /&gt;
&lt;br /&gt;
== Français ==&lt;br /&gt;
&amp;#039;&amp;#039;&amp;#039; XXXXXX&amp;#039;&amp;#039;&amp;#039;&lt;br /&gt;
&lt;br /&gt;
== Anglais ==&lt;br /&gt;
&amp;#039;&amp;#039;&amp;#039; Multi-Token Projection&amp;#039;&amp;#039;&amp;#039;&lt;br /&gt;
&lt;br /&gt;
&amp;#039;&amp;#039;&amp;#039; Multiple-Token Projection&amp;#039;&amp;#039;&amp;#039;&lt;br /&gt;
&lt;br /&gt;
&amp;#039;&amp;#039;&amp;#039; MTP&amp;#039;&amp;#039;&amp;#039;&lt;br /&gt;
&lt;br /&gt;
== Sources ==&lt;br /&gt;
[https://arxiv.org/html/2505.07608v1   Source : arxiv]&lt;br /&gt;
&lt;br /&gt;
[https://datascientest.com/multi-token-prediction-tout-savoir   Source : DataScientest]&lt;br /&gt;
&lt;br /&gt;
[https://deepwiki.com/deepseek-ai/DeepSeek-V3/3.4-multi-token-prediction-(mtp)   Source : DeepWiki]&lt;br /&gt;
&lt;br /&gt;
[https://medium.com/data-science-collective/deepseek-explained-4-multi-token-prediction-33f11fe2b868   Source : Medium]&lt;br /&gt;
&lt;br /&gt;
[[Catégorie:vocabulary]]&lt;/div&gt;</summary>
		<author><name>Arianne</name></author>
	</entry>
</feed>