@media all and (-ms-high-contrast: none), (-ms-high-contrast: active) {
.typography:not(##):not(##):not(##) .markup table.image-wrapper img,
.typography.editor:not(##):not(##):not(##) .markup table.image-wrapper img,
.typography:not(##):not(##):not(##) .markup table.kindle-wrapper img,
.typography.editor:not(##):not(##):not(##) .markup table.kindle-wrapper img {
max-width: 550px;
}
}
@media screen and (max-width: 650px) {
.typography .markup div.youtube-overlay,
.typography.editor .markup div.youtube-overlay,
.typography .markup div.vimeo-overlay,
.typography.editor .markup div.vimeo-overlay {
display: none !important;
}
}
@media screen and (max-width: 370px) {
.typography:not(##):not(##):not(##) .markup div.tiktok-wrap,
.typography.editor:not(##):not(##):not(##) .markup div.tiktok-wrap {
width: calc(95vw – 32px);
height: calc((95vw – 32px – 2px) / 0.485714);
}
}
@media screen and (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup div.embedded-publication-wrap .embedded-publication.show-subscribe,
.typography.editor:not(##):not(##):not(##) .markup div.embedded-publication-wrap .embedded-publication.show-subscribe {
padding: 24px;
}
}
@media screen and (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup div.subscription-widget-wrap .subscription-widget.show-subscribe,
.typography.editor:not(##):not(##):not(##) .markup div.subscription-widget-wrap .subscription-widget.show-subscribe,
.typography:not(##):not(##):not(##) .markup div.subscription-widget-wrap-editor .subscription-widget.show-subscribe,
.typography.editor:not(##):not(##):not(##) .markup div.subscription-widget-wrap-editor .subscription-widget.show-subscribe,
.typography:not(##):not(##):not(##) .markup div.captioned-button-wrap .subscription-widget.show-subscribe,
.typography.editor:not(##):not(##):not(##) .markup div.captioned-button-wrap .subscription-widget.show-subscribe {
padding: 0px 24px;
}
}
@media screen and (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup div.subscription-widget-wrap .subscription-widget.show-subscribe .subscription-widget-subscribe .button,
.typography.editor:not(##):not(##):not(##) .markup div.subscription-widget-wrap .subscription-widget.show-subscribe .subscription-widget-subscribe .button,
.typography:not(##):not(##):not(##) .markup div.subscription-widget-wrap-editor .subscription-widget.show-subscribe .subscription-widget-subscribe .button,
.typography.editor:not(##):not(##):not(##) .markup div.subscription-widget-wrap-editor .subscription-widget.show-subscribe .subscription-widget-subscribe .button,
.typography:not(##):not(##):not(##) .markup div.captioned-button-wrap .subscription-widget.show-subscribe .subscription-widget-subscribe .button,
.typography.editor:not(##):not(##):not(##) .markup div.captioned-button-wrap .subscription-widget.show-subscribe .subscription-widget-subscribe .button {
padding: 10px 12px;
min-width: 110px;
}
}
@media (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup .tweet,
.typography.editor:not(##):not(##):not(##) .markup .tweet {
padding: 12px;
}
}
@media (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup .tweet .tweet-text,
.typography.editor:not(##):not(##):not(##) .markup .tweet .tweet-text {
font-size: 14px;
line-height: 20px;
}
}
@media (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.two,
.typography.editor:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.two,
.typography:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.three,
.typography.editor:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.three,
.typography:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.four,
.typography.editor:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.four {
height: 200px;
}
}
@media (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup .tweet a.expanded-link .expanded-link-img,
.typography.editor:not(##):not(##):not(##) .markup .tweet a.expanded-link .expanded-link-img {
max-height: 180px;
}
}
@media (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup .tweet a.expanded-link .expanded-link-description,
.typography.editor:not(##):not(##):not(##) .markup .tweet a.expanded-link .expanded-link-description {
display: none;
}
}
@media screen and (max-width: 650px) {
.typography:not(##):not(##):not(##) .markup .apple-podcast-container,
.typography.editor:not(##):not(##):not(##) .markup .apple-podcast-container {
width: unset;
}
}
@media (max-width: 420px) {
.typography:not(##):not(##):not(##) .markup .install-substack-app-embed img.install-substack-app-embed-img,
.typography.editor:not(##):not(##):not(##) .markup .install-substack-app-embed img.install-substack-app-embed-img {
margin: 0 auto 16px auto;
}
}
@media (max-width: 420px) {
.typography:not(##):not(##):not(##) .markup .install-substack-app-embed .install-substack-app-embed-text,
.typography.editor:not(##):not(##):not(##) .markup .install-substack-app-embed .install-substack-app-embed-text {
margin: 0 0 12px 0;
max-width: 100%;
width: auto;
text-align: center;
}
}
@media (max-width: 420px) {
.typography:not(##):not(##):not(##) .markup .install-substack-app-embed .install-substack-app-embed-link,
.typography.editor:not(##):not(##):not(##) .markup .install-substack-app-embed .install-substack-app-embed-link {
display: flex;
justify-content: center;
}
}
@media screen and (min-width: 481px) {
.share-button-container:not(##):not(##):not(##) {
height: 38px;
}
}
@media screen and (min-width: 481px) {
.share-button-container:not(##):not(##):not(##) a.comment {
height: 38px;
line-height: 38px;
padding-right: 10px;
}
}
@media screen and (max-width: 480px) {
.share-button-container:not(##):not(##):not(##) .separator {
display: block;
margin: 0;
height: 8px;
border-left: none;
}
}
@media screen and (max-width: 480px) {
.share-button-container:not(##):not(##):not(##) a.share.first img {
padding-left: 0;
}
}
@media screen and (min-width: 481px) {
.share-button-container a.mobile {
display: none !important;
}
}
@media screen and (min-width: 541px) {
.settings-add-pub-modal-wrapper:not(##):not(##):not(##) .container .add-recommending-pub-modal-container {
padding: 36px;
height: 680px;
}
}
@media screen and (min-width: 541px) {
.settings-add-pub-modal-wrapper:not(##):not(##):not(##) .container .add-recommending-pub-modal-container .footer {
position: absolute;
bottom: 36px;
margin: 0px;
}
}
@media screen and (max-width: 650px) {
.header-anchor-parent:not(##):not(##):not(##) {
display: none;
}
}
@media screen and (max-width: 768px) {
.post:not(##):not(##):not(##) {
padding: 16px 0 0 0;
}
}
@media screen and (max-width: 650px) {
.post:not(##):not(##):not(##) .post-header .post-label {
margin-top: 8px;
}
}
@media screen and (max-width: 650px) {
.post:not(##):not(##):not(##) .post-header .meta-author-wrap.alternative-meta .meta-right-column .post-meta {
margin-top: 6px;
}
}
@media screen and (max-width: 650px) {
.post:not(##):not(##):not(##) .footer-facepile-container {
height: 64px;
padding: 0 16px;
display: flex;
align-items: center;
justify-content: flex-start;
width: 100%;
}
}
@media screen and (max-width: 650px) {
.post:not(##):not(##):not(##) .post-footer.use-separators {
justify-content: center;
}
}
@media screen and (max-width: 650px) {
.post:not(##):not(##):not(##) .post-footer.next-prev {
height: 64px;
justify-content: space-between;
box-sizing: border-box;
}
}
@media screen and (max-width: 650px) {
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-table {
display: block;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-table-row {
display: flex;
flex-direction: row;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-userhead-cell,
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-body-cell {
display: block;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-body-cell {
flex-grow: 1;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-body-table {
display: block;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-body-table-row {
display: block;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-copy-cell,
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-controls-cell {
display: block;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-copy-cell {
margin: 0 0 16px 0;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-controls-cell {
width: auto;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-controls {
margin: auto;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-controls .button.primary {
width: 100%;
}
.post-contributor-footer:not(##):not(##):not(##) .post-contributor-bio-text {
font-size: 14px;
}
}
@media screen and (min-width: 768px) {
.post-silhouette:not(##):not(##):not(##) {
padding: 32px 0;
}
}
@media screen and (max-width: 650px) {
.post-silhouette:not(##):not(##):not(##) .post-silhouette-title {
margin-top: 10.44225025px;
height: 120px;
}
}
@media screen and (max-width: 650px) {
.post-silhouette:not(##):not(##):not(##) .post-silhouette-meta {
width: 75%;
}
}
@media screen and (max-width: 650px) {
.post-silhouette:not(##):not(##):not(##) .post-silhouette-meta.with-byline-image {
margin: 20px 0;
}
}
@media screen and (max-width: 650px) {
.use-theme-bg:not(##):not(##):not(##) .post-meta.alternative-meta .post-meta-item,
.post-meta.alternative-meta:not(##):not(##):not(##) .post-meta-item {
padding-right: 16px;
}
}
@media screen and (max-width: 370px) {
.use-theme-bg:not(##):not(##):not(##) .post-meta.alternative-meta .post-meta-item,
.post-meta.alternative-meta:not(##):not(##):not(##) .post-meta-item {
font-size: 14px;
}
}
@media screen and (max-width: 650px) {
.use-theme-bg:not(##):not(##):not(##) .post-meta.alternative-meta .post-meta-item.guest-author-publication,
.post-meta.alternative-meta:not(##):not(##):not(##) .post-meta-item.guest-author-publication {
display: none;
}
}
@media screen and (max-width: 370px) {
.post-meta .post-meta-item .post-meta-button {
height: 36px !important;
}
.post-meta:not(##):not(##):not(##) .post-meta-item .post-meta-button {
/* important to override in-line height style on emails */
}
.post-meta:not(##):not(##):not(##) .post-meta-item .post-meta-button .meta-button-label {
display: none;
}
.post-meta:not(##):not(##):not(##) .post-meta-item .post-meta-button > svg {
margin-right: 0;
}
}
@media screen and (max-width: 370px) {
.post-meta:not(##):not(##):not(##) .post-meta-item {
font-size: 12px;
}
}
@media screen and (max-width: 650px) {
.post:not(##):not(##):not(##) .floating-subscribe-button {
bottom: 20px;
right: 20px;
}
}
@media all and (-ms-high-contrast: none), (-ms-high-contrast: active) {
body:not(##):not(##):not(##) .markup table.image-wrapper img,
body:not(##):not(##):not(##) .markup table.kindle-wrapper img {
max-width: 550px;
}
}
@media screen and (max-width: 650px) {
body .markup div.youtube-overlay,
body .markup div.vimeo-overlay {
display: none !important;
}
}
@media screen and (max-width: 370px) {
body:not(##):not(##):not(##) .markup div.tiktok-wrap {
width: calc(95vw – 32px);
height: calc((95vw – 32px – 2px) / 0.485714);
}
}
@media screen and (max-width: 650px) {
body:not(##):not(##):not(##) .markup div.embedded-publication-wrap .embedded-publication.show-subscribe {
padding: 24px;
}
}
@media screen and (max-width: 650px) {
body:not(##):not(##):not(##) .markup div.subscription-widget-wrap .subscription-widget.show-subscribe,
body:not(##):not(##):not(##) .markup div.subscription-widget-wrap-editor .subscription-widget.show-subscribe,
body:not(##):not(##):not(##) .markup div.captioned-button-wrap .subscription-widget.show-subscribe {
padding: 0px 24px;
}
}
@media screen and (max-width: 650px) {
body:not(##):not(##):not(##) .markup div.subscription-widget-wrap .subscription-widget.show-subscribe .subscription-widget-subscribe .button,
body:not(##):not(##):not(##) .markup div.subscription-widget-wrap-editor .subscription-widget.show-subscribe .subscription-widget-subscribe .button,
body:not(##):not(##):not(##) .markup div.captioned-button-wrap .subscription-widget.show-subscribe .subscription-widget-subscribe .button {
padding: 10px 12px;
min-width: 110px;
}
}
@media (max-width: 650px) {
body:not(##):not(##):not(##) .markup .tweet {
padding: 12px;
}
}
@media (max-width: 650px) {
body:not(##):not(##):not(##) .markup .tweet .tweet-text {
font-size: 14px;
line-height: 20px;
}
}
@media (max-width: 650px) {
body:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.two,
body:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.three,
body:not(##):not(##):not(##) .markup .tweet .tweet-photos-container.four {
height: 200px;
}
}
@media (max-width: 650px) {
body:not(##):not(##):not(##) .markup .tweet a.expanded-link .expanded-link-img {
max-height: 180px;
}
}
@media (max-width: 650px) {
body:not(##):not(##):not(##) .markup .tweet a.expanded-link .expanded-link-description {
display: none;
}
}
@media screen and (max-width: 650px) {
body:not(##):not(##):not(##) .markup .apple-podcast-container {
width: unset;
}
}
@media (max-width: 420px) {
body:not(##):not(##):not(##) .markup .install-substack-app-embed img.install-substack-app-embed-img {
margin: 0 auto 16px auto;
}
}
@media (max-width: 420px) {
body:not(##):not(##):not(##) .markup .install-substack-app-embed .install-substack-app-embed-text {
margin: 0 0 12px 0;
max-width: 100%;
width: auto;
text-align: center;
}
}
@media (max-width: 420px) {
body:not(##):not(##):not(##) .markup .install-substack-app-embed .install-substack-app-embed-link {
display: flex;
justify-content: center;
}
}
@media screen and (min-width: 500px) {
body:not(##):not(##):not(##) .header a.logo {
width: 42px;
height: 42px;
border-radius: 12px;
}
}
@media screen and (max-width: 420px) {
body .subscription-receipt table:first-of-type .subscription-amount .subscription-discount {
width: 72px !important;
}
}
@media screen and (min-width: 481px) {
body:not(##):not(##):not(##) .share-button-container {
height: auto;
}
}
@media screen and (max-width: 480px) {
body .share-button-container .separator {
display: block !important;
margin: 0 !important;
height: 8px !important;
border-left: none !important;
}
}
@media screen and (max-width: 650px) {
.digest:not(##):not(##):not(##) .item .post-meta-item.audience {
display: none;
}
}
@media screen and (min-width: 500px) {
.digest-publication:not(##):not(##):not(##) .logo img {
width: 42px;
height: 42px;
border-radius: 8px;
}
}
@media screen and (max-width: 650px) {
.comments-page:not(##):not(##):not(##) .container .comment-list .collapsed-reply {
margin-left: calc(10 + 32px – 24px);
}
}
@media screen and (max-width: 650px) {
.comment:not(##):not(##):not(##) > .comment-list {
padding-left: 24px;
}
}
@media screen and (max-width: 650px) {
.finish-magic-login-modal:not(##):not(##):not(##) .modal-content .container {
padding: 24px 0;
}
}
@media (max-width: 650px) {
.reader2-text-b3:not(##):not(##):not(##) {
line-height: 24px;
}
}
@media screen and (max-width: 650px) {
.reader2-text-h4:not(##):not(##):not(##) {
line-height: 24px;
}
}
@media screen and (min-width: 541px) {
.user-profile-modal:not(##):not(##):not(##) {
padding-left: 12px;
padding-right: 12px;
}
}
@media screen and (max-width: 650px) {
.subscribe-widget:not(##):not(##):not(##) form.form .sideBySideWrap button.rightButton {
padding: 10px 12px;
}
}
@media screen and (min-width: 541px) {
.pub-icon:hover:not(##):not(##):not(##) .logo-hover,
.feed-item-icon:hover:not(##):not(##):not(##) .logo-hover {
display: block;
}
}
@media screen and (max-width: 650px) {
.post-ufi.single-full-width-button:not(##):not(##):not(##) .post-ufi-button-wrapper {
width: 100%;
padding: 16px;
}
.post-ufi.single-full-width-button:not(##):not(##):not(##) .post-ufi-button-wrapper:empty {
display: none;
}
.post-ufi.single-full-width-button:not(##):not(##):not(##) .post-ufi-button {
width: 100%;
justify-content: center;
}
}
@media screen and (max-width: 768px) {
.file-embed-wrapper:not(##):not(##):not(##) {
padding: 0;
}
}
@media screen and (max-width: 768px) {
.file-embed-wrapper-editor:not(##):not(##):not(##) {
padding: 0;
}
}
@media screen and (max-width: 768px) {
.file-embed-wrapper-editor:active:not(##):not(##):not(##) {
padding: 0;
}
}
@media only screen and (max-width: 650px) {
.file-embed-button.wide:not(##):not(##):not(##),
.file-embed-error-button.wide:not(##):not(##):not(##) {
display: none;
}
}
@media only screen and (min-width: 630px) {
.file-embed-button.narrow:not(##):not(##):not(##),
.file-embed-error-button.narrow:not(##):not(##):not(##) {
display: none;
}
}
@media screen and (min-width: 541px) {
.audio-player-wrapper:not(##):not(##):not(##) .audio-player {
min-width: 500px;
}
}
@media screen and (max-width: 650px) {
.audio-player-wrapper:not(##):not(##):not(##) .audio-player .audio-player-progress {
border-left-width: 16px;
border-right-width: 16px;
}
}
@media screen and (max-width: 650px) {
.audio-player-wrapper:not(##):not(##):not(##) .audio-player .audio-player-progress .audio-player-progress-bar .audio-player-progress-bar-popup {
top: -54px;
}
}
@media screen and (max-width: 650px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-progress {
border-left-width: 16px;
border-right-width: 16px;
}
}
@media screen and (max-width: 650px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-progress .audio-player-progress-bar .audio-player-progress-bar-popup {
top: -54px;
}
}
@media (min-width: 250px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player {
padding: 32px;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group {
display: flex;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group .button:last-of-type {
display: block;
}
}
@media (min-width: 300px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group {
display: block;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group .button:first-of-type {
display: block;
}
}
@media (min-width: 350px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-substack-logo {
display: block;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-title {
margin-top: 16px;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-hero-image-container {
padding-top: 15%;
width: 15%;
display: block;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group .button:first-of-type {
display: block;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-substack-logo {
display: block;
}
}
@media (min-width: 350px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-hero-image-container {
padding-top: 25%;
width: 25%;
display: block;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group {
display: flex;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group .button:first-of-type {
display: block;
}
}
@media (min-width: 400px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-hero-image-container {
padding-top: 40%;
width: 40%;
}
}
@media (max-width: 400px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group {
margin-top: 12px;
}
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .btn-group .button {
font-size: 13px;
padding: 6px 12px;
height: auto;
margin-top: 10px;
}
}
@media (min-width: 600px) {
.audio-player-wrapper-fancy:not(##):not(##):not(##) .audio-player .audio-player-hero-image-container {
padding-top: 55%;
width: 55%;
}
}
@media (max-width: 650px) {
.poll-editor-modal:not(##):not(##):not(##) {
min-width: calc(100% – 20px);
}
}
@media (max-width: 750px) {
.poll-embed:not(##):not(##):not(##) .poll-anchor-target .poll-anchor-copy-button {
left: 8px;
top: 45px;
}
}
@media screen and (min-width: 541px) {
.poll-embed:not(##):not(##):not(##) .poll-wrapper.poll-web .poll-dialog .modal-table .modal-row .modal-content > .container {
width: 552px;
padding: 26px 24px;
}
}
@media screen and (max-width: 650px) {
.poll-embed:not(##):not(##):not(##) .poll-wrapper.poll-web .poll-dialog .modal-table .modal-row .modal-content > .container {
padding: 40px 0;
}
}
@media screen and (max-width: 650px) {
.poll-embed:not(##):not(##):not(##) .poll-wrapper.poll-web .poll-dialog .modal-row .modal-cell .modal-exit-btn {
margin-right: -20px;
}
}
Deep Learning Weekly: Issue 384Introducing Gemini 2.0, MiniVLA: A Better VLA with a Smaller Footprint, a paper on Large Concept Models: Language Modeling in a Sentence Representation Space, and many more!
This week in deep learning, we bring you Introducing Gemini 2.0: our new AI model for the agentic era, MiniVLA: A Better VLA with a Smaller Footprint, and a paper on Large Concept Models: Language Modeling in a Sentence Representation Space. You may also enjoy A fully open-source model for predicting biomolecular structures, FACTS Grounding: A new benchmark for evaluating the factuality of large language models, a paper on An Evolved Universal Transformer Memory, and more! As always, happy reading and hacking. If you have something you think should be in next week’s issue, find us on Twitter: @dl_weekly. Until next week! IndustryIntroducing Gemini 2.0: our new AI model for the agentic era Google released the first model in the Gemini 2.0 family of models: an experimental version of Gemini 2.0 Flash with native multimodal capabilities. MIT researchers introduce Boltz-1, a fully open-source model for predicting biomolecular structures MIT scientists have released a powerful, open-source AI model, called Boltz-1, that could significantly accelerate biomedical research and drug development. Bringing AI Agents to Enterprises with Google Agentspace Google introduced Google Agentspace, which unlocks enterprise expertise with agents that bring together Gemini’s advanced reasoning, Google-quality search, and enterprise data. Introducing Phi-4: Microsoft’s Newest Small Language Model Specializing in Complex Reasoning Microsoft introduced Phi-4, a state-of-the-art small language model (SLM) that excels at complex reasoning, in addition to conventional language processing. Sharing new research, models, and datasets from Meta FAIR Meta FAIR released several new research artifacts that highlight recent innovations in developing agents, robustness and safety, and architectures that facilitate machine learning. NVIDIA made microservices for multilingual generative AI, such as NVIDIA NeMo Retriever and NVIDIA NIM reranking, available on their API catalog. Liquid AI raises $250M led by AMD to build a new type of generative AI model Liquid AI, an MIT spinoff developing generative AI models based on a fundamentally different architecture than transformers, announced that it has raised $250 million in a funding round led by AMD. MLOps & LLMOpsA tutorial that shows you how to build an agentic workflow that can extract key patient information, check clinical guidelines, and produce a human-readable case summary. A post about the release of Weaviate 1.28, with features like role-based access control, enhanced async indexing, and native Japanese language support. Making it easier to build human-in-the-loop agents with interrupt A blog post exploring the “interrupt” feature in LangGraph, designed to make it easier to build human-in-the-loop agents. LearningFACTS Grounding: A new benchmark for evaluating the factuality of large language models A Google DeepMind article introducing FACTS Grounding, a benchmark for evaluating the factuality of LLMs. Fine-Tuning Small Language Models to Optimize Code Review Accuracy A post that introduces an automated fine-tuning approach that addresses cost, performance, and privacy challenges for generative AI by incorporating curriculum learning. MiniVLA: A Better VLA with a Smaller Footprint A technical blog post introducing MiniVLA, a smaller and more efficient version of OpenVLA for robotics. Satellite powered estimation of global solar potential An informative blog post discussing the expansion of Google Maps Platform Solar API, now utilizing satellite imagery for estimating global solar potential. Libraries & CodeA Python library for decoding videos into PyTorch tensors, on CPU and CUDA GPU. cshaitao/awesome-llms-as-judges The official repo for the paper, LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods. Papers & PublicationsLarge Concept Models: Language Modeling in a Sentence Representation Space Abstract: LLMs have revolutionized the field of artificial intelligence and have emerged as the de-facto tool for many tasks. The current established technology of LLMs is to process input and generate output at the token level. This is in sharp contrast to humans who operate at multiple levels of abstraction well beyond single words, to analyze information and to generate creative content. In this paper, we present an attempt at an architecture which operates on an explicit higher-level semantic representation, which we name a concept. Concepts are language- and modality-agnostic and represent a higher level idea or action in a flow. Hence, we build a “Large Concept Model”. In this study, as proof of feasibility, we assume that a concept corresponds to a sentence, and use an existing sentence embedding space, SONAR, which supports up to 200 languages in both text and speech modalities. The Large Concept Model is trained to perform autoregressive sentence prediction in an embedding space. We explore multiple approaches, namely MSE regression, variants of diffusion-based generation, and models operating in a quantized SONAR space. These explorations are performed using 1.6B parameter models and training data in the order of 1.3T tokens. We then scale one architecture to a model size of 7B parameters and training data of about 2.7T tokens. We perform an experimental evaluation on several generative tasks, namely summarization and a new task of summary expansion. Finally, we show that our model exhibits impressive zero-shot generalization performance to many languages, outperforming existing LLMs of the same size. The training code of our models is freely available. An Evolved Universal Transformer Memory Abstract: Prior methods propose to offset the escalating costs of modern foundation models by dropping specific parts of their contexts with hand-designed rules, while attempting to preserve their original performance. We overcome this trade-off with Neural Attention Memory Models (NAMMs), introducing a learned network for memory management that improves both the performance and efficiency of transformers. We evolve NAMMs atop pre-trained transformers to provide different latent contexts focusing on the most relevant information for individual layers and attention heads. NAMMs are universally applicable to any model using self-attention as they condition exclusively on the values in the produced attention matrices. Learning NAMMs on a small set of problems, we achieve substantial performance improvements across multiple long-context benchmarks while cutting the model’s input contexts up to a fraction of the original sizes. We show the generality of our conditioning enables zero-shot transfer of NAMMs trained only on language to entirely new transformer architectures even across input modalities, with their benefits carrying over to vision and reinforcement learning. Abstract: Transformer based models have provided significant performance improvements in monaural speech separation. However, there is still a performance gap compared to a recent proposed upper bound. The major limitation of the current dual-path Transformer models is the inefficient modeling of long-range elemental interactions and local feature patterns. In this work, we achieve the upper bound by proposing a gated single-head transformer architecture with convolution-augmented joint self-attentions, named textit{MossFormer} (textit{Mo}naural textit{s}peech textit{s}eparation Transtextit{Former}). To effectively solve the indirect elemental interactions across chunks in the dual-path architecture, MossFormer employs a joint local and global self-attention architecture that simultaneously performs a full-computation self-attention on local chunks and a linearised low-cost self-attention over the full sequence. The joint attention enables MossFormer model full-sequence elemental interaction directly. In addition, we employ a powerful attentive gating mechanism with simplified single-head self-attentions. Besides the attentive long-range modeling, we also augment MossFormer with convolutions for the position-wise local pattern modeling. As a consequence, MossFormer significantly outperforms the previous models and achieves the state-of-the-art results on WSJ0-2/3mix and WHAM!/WHAMR! benchmarks. Our model achieves the SI-SDRi upper bound of 21.2 dB on WSJ0-3mix and only 0.3 dB below the upper bound of 23.1 dB on WSJ0-2mix. Thanks for reading Deep Learning Weekly! Subscribe for free to receive new posts and support my work.
Did someone share this email? Get your own free subscription at Deep Learning Weekly. Deep Learning Weekly is sponsored by Comet ML.
|
