{"id":1295,"date":"2023-12-23T14:52:51","date_gmt":"2023-12-23T21:52:51","guid":{"rendered":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/"},"modified":"2023-12-23T14:52:51","modified_gmt":"2023-12-23T21:52:51","slug":"ferret-an-terminate-to-terminate-mllm-by-apple","status":"publish","type":"post","link":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/","title":{"rendered":"Ferret: An Terminate-to-Terminate MLLM by Apple"},"content":{"rendered":"<div data-target=\"readme-toc.content\">\n<article itemprop=\"text\">\n<h2 tabindex=\"-1\" dir=\"auto\"><a target=\"_blank\" rel=\"noopener noreferrer\" href=\"http:\/\/github.com\/apple\/ml-ferret\/blob\/main\/figs\/ferret_icon.png\"><img loading=\"lazy\" decoding=\"async\" src=\"http:\/\/github.com\/apple\/ml-ferret\/raw\/main\/figs\/ferret_icon.png\" alt=\"Alt text for the image\" width=\"40\" height=\"45\"><\/a>    Ferret: Refer and Floor The rest Wherever at Any Granularity<\/h2>\n<p dir=\"auto\"><em>An Terminate-to-Terminate MLLM that Gain Any-Fabricate Referring and Floor The rest in Response.<\/em> [[[[<a href=\"https:\/\/arxiv.org\/abs\/2310.07704\" rel=\"nofollow\">Paper<\/a>]<\/p>\n<p dir=\"auto\"><a href=\"https:\/\/hxyou.github.io\/\" rel=\"nofollow\">Haoxuan You*<\/a>, <a href=\"https:\/\/haotian-zhang.github.io\/\" rel=\"nofollow\">Haotian Zhang*<\/a>, <a href=\"https:\/\/zhegan27.github.io\/\" rel=\"nofollow\">Zhe Gan<\/a>, <a href=\"https:\/\/scholar.google.com\/citations?user=l1hP40AAAAAJ&amp;hl=en\" rel=\"nofollow\">Xianzhi Du<\/a>, <a href=\"https:\/\/zbwglory.github.io\/\" rel=\"nofollow\">Bowen Zhang<\/a>, <a href=\"https:\/\/www.cs.cmu.edu\/~ziruiw\/\" rel=\"nofollow\">Zirui Wang<\/a>, <a href=\"http:\/\/llcao.net\/\" rel=\"nofollow\">Liangliang Cao<\/a>, <a href=\"https:\/\/www.ee.columbia.edu\/~sfchang\/\" rel=\"nofollow\">Shih-Fu Chang<\/a>, <a href=\"https:\/\/sites.google.com\/site\/yinfeiyang\/\" rel=\"nofollow\">Yinfei Yang<\/a>[*: equal contribution]<\/p>\n<h2 tabindex=\"-1\" dir=\"auto\">Overview<\/h2>\n<p dir=\"auto\">    <a target=\"_blank\" rel=\"noopener noreferrer\" href=\"http:\/\/github.com\/apple\/ml-ferret\/blob\/main\/figs\/ferret_fig_diagram_v2.png\"><img decoding=\"async\" src=\"http:\/\/github.com\/apple\/ml-ferret\/raw\/main\/figs\/ferret_fig_diagram_v2.png\" width=\"100%\"><\/a> <br \/>                Plan of Ferret Mannequin.<\/p>\n<p dir=\"auto\">Key Contributions:<\/p>\n<ul dir=\"auto\">\n<li>Ferret Mannequin &#8211; <strong>Hybrid Set aside Representation + Spatial-mindful Visible Sampler<\/strong> allow gorgeous-grained and delivery-vocabulary referring and grounding in MLLM.<\/li>\n<li>GRIT Dataset (~1.1M) &#8211; A <strong>Neat-scale, Hierarchical, Sturdy<\/strong> ground-and-refer instruction tuning dataset.<\/li>\n<li>Ferret-Bench &#8211; A multimodal evaluate benchmark that collectively requires <strong>Referring\/Grounding, Semantics, Knowledge, and Reasoning<\/strong>.<\/li>\n<\/ul>\n<h2 tabindex=\"-1\" dir=\"auto\">Liberate<\/h2>\n<ul dir=\"auto\">\n<li>[12\/14]    &#128293; We released the <a href=\"http:\/\/github.com\/#checkpoints\">checkpoints(7B, 13B)<\/a>.<\/li>\n<li>[10\/30]    &#128293; We released the code of <strong>FERRET<\/strong> mannequin and <a href=\"http:\/\/github.com\/apple\/ml-ferret\/blob\/main\/ferret\/eval\/ferret_gpt4_data\">Ferret-Bench<\/a>.<\/li>\n<\/ul>\n<p dir=\"auto\"><strong>Utilization and License Notices<\/strong>: The records, and code is supposed and licensed for learn utilize handiest. They are furthermore restricted to uses that follow the license settlement of LLaMA, Vicuna and GPT-4. The dataset is CC BY NC 4.0 (permitting handiest non-commercial utilize) and items trained utilizing the dataset must composed now not be ancient delivery air of learn functions.<\/p>\n<h2 tabindex=\"-1\" dir=\"auto\">Contents<\/h2>\n<ul dir=\"auto\">\n<li><a href=\"http:\/\/github.com\/#install\">Set up<\/a><\/li>\n<li><a href=\"http:\/\/github.com\/#train\">Declare<\/a><\/li>\n<li><a href=\"http:\/\/github.com\/#evaluation\">Review<\/a><\/li>\n<li><a href=\"http:\/\/github.com\/#demo\">Demo<\/a><\/li>\n<\/ul>\n<h2 tabindex=\"-1\" dir=\"auto\">Set up<\/h2>\n<ol dir=\"auto\">\n<li>Clone this repository and navigate to FERRET folder<\/li>\n<\/ol>\n<div dir=\"auto\" data-snippet-clipboard-copy-content=\"git clone https:\/\/github.com\/apple\/ml-ferretcd ml-ferret\">\n<pre>git clone https:\/\/github.com\/apple\/ml-ferret<span>cd<\/span> ml-ferret<\/pre>\n<\/div>\n<ol start=\"2\" dir=\"auto\">\n<li>Set up Equipment<\/li>\n<\/ol>\n<div dir=\"auto\" data-snippet-clipboard-copy-content=\"conda create -n ferret python=3.10 -yconda activate ferretpip install --upgrade pip  # enable PEP 660 supportpip install -e .pip install pycocotoolspip install protobuf==3.20.0\">\n<pre>conda create -n ferret python=3.10 -yconda activate ferretpip install --upgrade pip  <span><span>#<\/span> enable PEP 660 support<\/span>pip install -e <span>.<\/span>pip install pycocotoolspip install protobuf==3.20.0<\/pre>\n<\/div>\n<ol start=\"3\" dir=\"auto\">\n<li>Set up extra packages for coaching instances<\/li>\n<\/ol>\n<div data-snippet-clipboard-copy-content=\"pip install ninjapip install flash-attn --no-build-isolation\">\n<pre><code>pip install ninjapip install flash-attn --no-build-isolation<\/code><\/pre>\n<\/div>\n<h2 tabindex=\"-1\" dir=\"auto\">Declare<\/h2>\n<p dir=\"auto\">FERRET is trained on 8 A100 GPUs with 80GB memory. To coach on fewer GPUs, you would possibly scale back the <code>per_device_train_batch_size<\/code> and originate greater the <code>gradient_accumulation_steps<\/code> accordingly. Continually place the realm batch dimension the same: <code>per_device_train_batch_size<\/code> x <code>gradient_accumulation_steps<\/code> x <code>num_gpus<\/code>.<\/p>\n<h3 tabindex=\"-1\" dir=\"auto\">Hyperparameters<\/h3>\n<p dir=\"auto\">We utilize a identical location of hyperparameters as LLaVA(Vicuna) in finetuning.<\/p>\n<table readabilitydatatable=\"1\">\n<thead>\n<tr>\n<th>Hyperparameter<\/th>\n<th>Global Batch Dimension<\/th>\n<th>Studying price<\/th>\n<th>Epochs<\/th>\n<th>Max length<\/th>\n<th>Weight decay<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>FERRET-7B<\/td>\n<td>128<\/td>\n<td>2e-5<\/td>\n<td>3<\/td>\n<td>2048<\/td>\n<td>0<\/td>\n<\/tr>\n<tr>\n<td>FERRET-13B<\/td>\n<td>128<\/td>\n<td>2e-5<\/td>\n<td>3<\/td>\n<td>2048<\/td>\n<td>0<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h3 tabindex=\"-1\" dir=\"auto\">Prepare Vicuna checkpoint and LLaVA&#8217;s projector<\/h3>\n<p dir=\"auto\">Sooner than you delivery, put collectively our unpleasant mannequin Vicuna, which is an instruction-tuned chatbot. Please obtain its weights following the directions <a href=\"https:\/\/github.com\/lm-sys\/FastChat#model-weights\">here<\/a>.  Vicuna v1.3 is ancient in FERRET.<\/p>\n<p dir=\"auto\">Then obtain LLaVA&#8217;s first-stage pre-trained projector weight (<a href=\"https:\/\/huggingface.co\/liuhaotian\/llava-336px-pretrain-vicuna-7b-v1.3\" rel=\"nofollow\">7B<\/a>, <a href=\"https:\/\/huggingface.co\/liuhaotian\/llava-336px-pretrain-vicuna-13b-v1.3\" rel=\"nofollow\">13B<\/a>).<\/p>\n<h3 tabindex=\"-1\" dir=\"auto\">FERRET Training<\/h3>\n<p dir=\"auto\">The scripts are equipped (<a href=\"http:\/\/github.com\/apple\/ml-ferret\/blob\/main\/experiments\/ferret_7b_train.sh\">7B<\/a>, <a href=\"http:\/\/github.com\/apple\/ml-ferret\/blob\/main\/experiments\/ferret_13b_train.sh\">13B<\/a>).<\/p>\n<h2 tabindex=\"-1\" dir=\"auto\">Review<\/h2>\n<p dir=\"auto\">Please explore this <a href=\"http:\/\/github.com\/apple\/ml-ferret\/blob\/main\/EVAL.md\">doc<\/a> for the necessary points.<\/p>\n<h2 tabindex=\"-1\" dir=\"auto\">Checkpoints<\/h2>\n<p dir=\"auto\">We extracted the <code>delta<\/code> between our pre-trained mannequin and Vicuna. Please first obtain weights of Vicuna following the <a href=\"http:\/\/github.com\/#prepare-vicuna-checkpoint-and-llavas-projector\">outdated instruction<\/a>. Then obtain our ready offsets of weights: <a href=\"https:\/\/docs-assets.developer.apple.com\/ml-research\/models\/ferret\/ferret-7b\/ferret-7b-delta.zip\" rel=\"nofollow\">7B<\/a>, <a href=\"https:\/\/docs-assets.developer.apple.com\/ml-research\/models\/ferret\/ferret-13b\/ferret-13b-delta.zip\" rel=\"nofollow\">13B<\/a> utilizing <code>wget<\/code> or <code>curl<\/code>and unzip the downloaded offsets. Lastly, apply the offset to the Vicuna&#8217;s weight by working the next script:<\/p>\n<div dir=\"auto\" data-snippet-clipboard-copy-content=\"# 7Bpython3 -m ferret.model.apply_delta     --base .\/model\/vicuna-7b-v1-3     --target .\/model\/ferret-7b-v1-3     --delta path\/to\/ferret-7b-delta# 13Bpython3 -m ferret.model.apply_delta     --base .\/model\/vicuna-13b-v1-3     --target .\/model\/ferret-13b-v1-3     --delta path\/to\/ferret-13b-delta\">\n<pre><span><span>#<\/span> 7B<\/span>python3 -m ferret.model.apply_delta     --base .\/model\/vicuna-7b-v1-3     --target .\/model\/ferret-7b-v1-3     --delta path\/to\/ferret-7b-delta<span><span>#<\/span> 13B<\/span>python3 -m ferret.model.apply_delta     --base .\/model\/vicuna-13b-v1-3     --target .\/model\/ferret-13b-v1-3     --delta path\/to\/ferret-13b-delta<\/pre>\n<\/div>\n<p dir=\"auto\"><strong>Notices<\/strong>: Apple&#8217;s rights in the connected weight differentials are hereby licensed under the CC-BY-NC license. Apple makes no representations as regards to LLaMa or every other third event tool, that are discipline to their very fetch terms.<\/p>\n<p dir=\"auto\">Please consult with the next half about the suitable approach to location up a neighborhood demo with pre-trained weight.<\/p>\n<h2 tabindex=\"-1\" dir=\"auto\">Demo<\/h2>\n<p dir=\"auto\">To dart our demo, it be vital to mutter FERRET and utilize the checkpoints in the community. Gradio net UI is ancient. Please dart the next commands one after the opposite.<\/p>\n<h4 tabindex=\"-1\" dir=\"auto\">Launch a controller<\/h4>\n<div dir=\"auto\" data-snippet-clipboard-copy-content=\"python -m ferret.serve.controller --host 0.0.0.0 --port 10000\">\n<pre>python -m ferret.serve.controller --host 0.0.0.0 --port 10000<\/pre>\n<\/div>\n<h4 tabindex=\"-1\" dir=\"auto\">Launch a gradio net server.<\/h4>\n<div dir=\"auto\" data-snippet-clipboard-copy-content=\"python -m ferret.serve.gradio_web_server --controller http:\/\/localhost:10000 --model-list-mode reload --add_region_feature\">\n<pre>python -m ferret.serve.gradio_web_server --controller http:\/\/localhost:10000 --model-list-mode reload --add_region_feature<\/pre>\n<\/div>\n<h4 tabindex=\"-1\" dir=\"auto\">Launch a mannequin worker<\/h4>\n<p dir=\"auto\">Right here&#8217;s the worker that load the ckpt and produce the inference on the GPU.  Each and every worker is accountable for a single mannequin specified in <code>--model-path<\/code>.<\/p>\n<div dir=\"auto\" data-snippet-clipboard-copy-content=\"CUDA_VISIBLE_DEVICES=0 python -m ferret.serve.model_worker --host 0.0.0.0 --controller http:\/\/localhost:10000 --port 40000 --worker http:\/\/localhost:40000 --model-path .\/checkpoints\/FERRET-13B-v0 --add_region_feature\">\n<pre>CUDA_VISIBLE_DEVICES=0 python -m ferret.serve.model_worker --host 0.0.0.0 --controller http:\/\/localhost:10000 --port 40000 --worker http:\/\/localhost:40000 --model-path .\/checkpoints\/FERRET-13B-v0 --add_region_feature<\/pre>\n<\/div>\n<p dir=\"auto\">Wait till the route of finishes loading the mannequin and also you explore &#8220;Uvicorn working on &#8230;&#8221;.  Now, refresh your Gradio net UI, and also it&#8217;s likely you&#8217;ll per chance explore the mannequin you correct launched in the mannequin list.<\/p>\n<p dir=\"auto\">    <a target=\"_blank\" rel=\"noopener noreferrer\" href=\"http:\/\/github.com\/apple\/ml-ferret\/blob\/main\/figs\/ferret_demo.png\"><img decoding=\"async\" src=\"http:\/\/github.com\/apple\/ml-ferret\/raw\/main\/figs\/ferret_demo.png\" width=\"105%\"><\/a> <br \/>                Example of Ferret Interactive Demo.<\/p>\n<h2 tabindex=\"-1\" dir=\"auto\">Citation<\/h2>\n<p dir=\"auto\">Ought to you scrutinize Ferret precious, please cite utilizing this BibTeX:<\/p>\n<div dir=\"auto\" data-snippet-clipboard-copy-content=\"@article{you2023ferret,  title={Ferret: Refer and Ground Anything Anywhere at Any Granularity},  author={You, Haoxuan and Zhang, Haotian and Gan, Zhe and Du, Xianzhi and Zhang, Bowen and Wang, Zirui and Cao, Liangliang and Chang, Shih-Fu and Yang, Yinfei},  journal={arXiv preprint arXiv:2310.07704},  year={2023}}\">\n<pre><span>@article<\/span>{<span>you2023ferret<\/span>,  <span>title<\/span>=<span><span>{<\/span>Ferret: Refer and Ground Anything Anywhere at Any Granularity<span>}<\/span><\/span>,  <span>author<\/span>=<span><span>{<\/span>You, Haoxuan and Zhang, Haotian and Gan, Zhe and Du, Xianzhi and Zhang, Bowen and Wang, Zirui and Cao, Liangliang and Chang, Shih-Fu and Yang, Yinfei<span>}<\/span><\/span>,  <span>journal<\/span>=<span><span>{<\/span>arXiv preprint arXiv:2310.07704<span>}<\/span><\/span>,  <span>year<\/span>=<span><span>{<\/span>2023<span>}<\/span><\/span>}<\/pre>\n<\/div>\n<h2 tabindex=\"-1\" dir=\"auto\">Acknowledgement<\/h2>\n<ul dir=\"auto\">\n<li><a href=\"https:\/\/github.com\/haotian-liu\/LLaVA\">The lava<\/a>: the codebase we built upon.<\/li>\n<li><a href=\"https:\/\/github.com\/lm-sys\/FastChat\">Vicuna<\/a>: the LLM codebase.<\/li>\n<\/ul>\n<\/article>\n<\/div>\n<p><a href=\"https:\/\/github.com\/apple\/ml-ferret\" class=\"button purchase\" rel=\"nofollow noopener\" target=\"_blank\">Read More<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Ferret: Refer and Floor The rest Wherever at Any Granularity An Terminate-to-Terminate MLLM that Gain&#8230;<\/p>\n","protected":false},"author":1,"featured_media":1296,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[485,486],"tags":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v23.5 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Ferret: An Terminate-to-Terminate MLLM by Apple - CBS26 Arizona<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/\" \/>\n<meta property=\"og:locale\" content=\"en_GB\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Ferret: An Terminate-to-Terminate MLLM by Apple - CBS26 Arizona\" \/>\n<meta property=\"og:description\" content=\"Ferret: Refer and Floor The rest Wherever at Any Granularity An Terminate-to-Terminate MLLM that Gain...\" \/>\n<meta property=\"og:url\" content=\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/\" \/>\n<meta property=\"og:site_name\" content=\"CBS26 Arizona\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/CBSNews\/\" \/>\n<meta property=\"article:published_time\" content=\"2023-12-23T21:52:51+00:00\" \/>\n<meta property=\"og:image\" content=\"http:\/\/github.com\/apple\/ml-ferret\/raw\/main\/figs\/ferret_icon.png\" \/>\n<meta name=\"author\" content=\"cbs26.com\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@CBSNews\" \/>\n<meta name=\"twitter:site\" content=\"@CBSNews\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"cbs26.com\" \/>\n\t<meta name=\"twitter:label2\" content=\"Estimated reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"4 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/\"},\"author\":{\"name\":\"cbs26.com\",\"@id\":\"https:\/\/cbs26.com\/#\/schema\/person\/6f199387dd167b788a14f8b99f8ff352\"},\"headline\":\"Ferret: An Terminate-to-Terminate MLLM by Apple\",\"datePublished\":\"2023-12-23T21:52:51+00:00\",\"dateModified\":\"2023-12-23T21:52:51+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/\"},\"wordCount\":584,\"commentCount\":4,\"publisher\":{\"@id\":\"https:\/\/cbs26.com\/#organization\"},\"image\":{\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg\",\"articleSection\":[\"Ferret\",\"Terminate-to-Terminate\"],\"inLanguage\":\"en-GB\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/\",\"url\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/\",\"name\":\"Ferret: An Terminate-to-Terminate MLLM by Apple - CBS26 Arizona\",\"isPartOf\":{\"@id\":\"https:\/\/cbs26.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg\",\"datePublished\":\"2023-12-23T21:52:51+00:00\",\"dateModified\":\"2023-12-23T21:52:51+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#breadcrumb\"},\"inLanguage\":\"en-GB\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-GB\",\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage\",\"url\":\"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg\",\"contentUrl\":\"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg\",\"width\":1177,\"height\":720,\"caption\":\"Ferret: An Terminate-to-Terminate MLLM by Apple\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/cbs26.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Ferret: An Terminate-to-Terminate MLLM by Apple\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/cbs26.com\/#website\",\"url\":\"https:\/\/cbs26.com\/\",\"name\":\"CBS26 Arizona\",\"description\":\"The Spirit of Arizona\",\"publisher\":{\"@id\":\"https:\/\/cbs26.com\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/cbs26.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-GB\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/cbs26.com\/#organization\",\"name\":\"CBS26 Arizona\",\"url\":\"https:\/\/cbs26.com\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-GB\",\"@id\":\"https:\/\/cbs26.com\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/cbs26.fw_-1-e1703104627272.png\",\"contentUrl\":\"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/cbs26.fw_-1-e1703104627272.png\",\"width\":296,\"height\":96,\"caption\":\"CBS26 Arizona\"},\"image\":{\"@id\":\"https:\/\/cbs26.com\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/CBSNews\/\",\"https:\/\/x.com\/CBSNews\"]},{\"@type\":\"Person\",\"@id\":\"https:\/\/cbs26.com\/#\/schema\/person\/6f199387dd167b788a14f8b99f8ff352\",\"name\":\"cbs26.com\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-GB\",\"@id\":\"https:\/\/cbs26.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/e5c693e6a74bdbc1461a75db724fe465?s=96&d=mm&r=r\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/e5c693e6a74bdbc1461a75db724fe465?s=96&d=mm&r=r\",\"caption\":\"cbs26.com\"},\"sameAs\":[\"http:\/\/cbs26.com\"],\"url\":\"https:\/\/cbs26.com\/index.php\/author\/cbs26-com\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Ferret: An Terminate-to-Terminate MLLM by Apple - CBS26 Arizona","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/","og_locale":"en_GB","og_type":"article","og_title":"Ferret: An Terminate-to-Terminate MLLM by Apple - CBS26 Arizona","og_description":"Ferret: Refer and Floor The rest Wherever at Any Granularity An Terminate-to-Terminate MLLM that Gain...","og_url":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/","og_site_name":"CBS26 Arizona","article_publisher":"https:\/\/www.facebook.com\/CBSNews\/","article_published_time":"2023-12-23T21:52:51+00:00","og_image":[{"url":"http:\/\/github.com\/apple\/ml-ferret\/raw\/main\/figs\/ferret_icon.png"}],"author":"cbs26.com","twitter_card":"summary_large_image","twitter_creator":"@CBSNews","twitter_site":"@CBSNews","twitter_misc":{"Written by":"cbs26.com","Estimated reading time":"4 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#article","isPartOf":{"@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/"},"author":{"name":"cbs26.com","@id":"https:\/\/cbs26.com\/#\/schema\/person\/6f199387dd167b788a14f8b99f8ff352"},"headline":"Ferret: An Terminate-to-Terminate MLLM by Apple","datePublished":"2023-12-23T21:52:51+00:00","dateModified":"2023-12-23T21:52:51+00:00","mainEntityOfPage":{"@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/"},"wordCount":584,"commentCount":4,"publisher":{"@id":"https:\/\/cbs26.com\/#organization"},"image":{"@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage"},"thumbnailUrl":"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg","articleSection":["Ferret","Terminate-to-Terminate"],"inLanguage":"en-GB","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/","url":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/","name":"Ferret: An Terminate-to-Terminate MLLM by Apple - CBS26 Arizona","isPartOf":{"@id":"https:\/\/cbs26.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage"},"image":{"@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage"},"thumbnailUrl":"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg","datePublished":"2023-12-23T21:52:51+00:00","dateModified":"2023-12-23T21:52:51+00:00","breadcrumb":{"@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#breadcrumb"},"inLanguage":"en-GB","potentialAction":[{"@type":"ReadAction","target":["https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/"]}]},{"@type":"ImageObject","inLanguage":"en-GB","@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#primaryimage","url":"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg","contentUrl":"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/1295-ferret-an-terminate-to-terminate-mllm-by-apple.jpg","width":1177,"height":720,"caption":"Ferret: An Terminate-to-Terminate MLLM by Apple"},{"@type":"BreadcrumbList","@id":"https:\/\/cbs26.com\/index.php\/2023\/12\/23\/ferret-an-terminate-to-terminate-mllm-by-apple\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/cbs26.com\/"},{"@type":"ListItem","position":2,"name":"Ferret: An Terminate-to-Terminate MLLM by Apple"}]},{"@type":"WebSite","@id":"https:\/\/cbs26.com\/#website","url":"https:\/\/cbs26.com\/","name":"CBS26 Arizona","description":"The Spirit of Arizona","publisher":{"@id":"https:\/\/cbs26.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/cbs26.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-GB"},{"@type":"Organization","@id":"https:\/\/cbs26.com\/#organization","name":"CBS26 Arizona","url":"https:\/\/cbs26.com\/","logo":{"@type":"ImageObject","inLanguage":"en-GB","@id":"https:\/\/cbs26.com\/#\/schema\/logo\/image\/","url":"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/cbs26.fw_-1-e1703104627272.png","contentUrl":"https:\/\/cbs26.com\/wp-content\/uploads\/2023\/12\/cbs26.fw_-1-e1703104627272.png","width":296,"height":96,"caption":"CBS26 Arizona"},"image":{"@id":"https:\/\/cbs26.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/CBSNews\/","https:\/\/x.com\/CBSNews"]},{"@type":"Person","@id":"https:\/\/cbs26.com\/#\/schema\/person\/6f199387dd167b788a14f8b99f8ff352","name":"cbs26.com","image":{"@type":"ImageObject","inLanguage":"en-GB","@id":"https:\/\/cbs26.com\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/e5c693e6a74bdbc1461a75db724fe465?s=96&d=mm&r=r","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/e5c693e6a74bdbc1461a75db724fe465?s=96&d=mm&r=r","caption":"cbs26.com"},"sameAs":["http:\/\/cbs26.com"],"url":"https:\/\/cbs26.com\/index.php\/author\/cbs26-com\/"}]}},"_links":{"self":[{"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/posts\/1295"}],"collection":[{"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/comments?post=1295"}],"version-history":[{"count":0,"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/posts\/1295\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/media\/1296"}],"wp:attachment":[{"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/media?parent=1295"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/categories?post=1295"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/cbs26.com\/index.php\/wp-json\/wp\/v2\/tags?post=1295"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}