diff --git a/.github/actions/compile-models/requirements.in b/.github/actions/compile-models/requirements.in
index 21e87b47df..4c60c14515 100644
--- a/.github/actions/compile-models/requirements.in
+++ b/.github/actions/compile-models/requirements.in
@@ -10,13 +10,13 @@ onnxruntime==1.23.2
 onnxruntime-tools>=1.7.0
 
 # Hugging Face Hub
-huggingface_hub==0.36.0
-hf-xet==1.2.0
+huggingface_hub==0.36.2
+hf-xet==1.4.2
 
 # Utils
 pillow>=12.1.1
 requests>=2.32.5
 tqdm>=4.67.3
-accelerate>=1.12.0
+accelerate>=1.13.0
 urllib3>=2.6.3
-filelock>=3.21.2
+filelock>=3.25.2
diff --git a/.github/actions/compile-models/requirements.txt b/.github/actions/compile-models/requirements.txt
index e1f3888fc4..e48aa58cef 100644
--- a/.github/actions/compile-models/requirements.txt
+++ b/.github/actions/compile-models/requirements.txt
@@ -2,13 +2,13 @@
 # This file is autogenerated by pip-compile with Python 3.13
 # by the following command:
 #
-#    pip-compile --allow-unsafe --generate-hashes --pip-args='--extra-index-url https://download.pytorch.org/whl/cpu' .github/actions/compile-models/requirements.in
+#    pip-compile --allow-unsafe --generate-hashes --pip-args='--extra-index-url https://download.pytorch.org/whl/cpu' '.github\actions\compile-models\requirements.in'
 #
 --extra-index-url https://download.pytorch.org/whl/cpu
 
-accelerate==1.12.0 \
-    --hash=sha256:3e2091cd341423207e2f084a6654b1efcd250dc326f2a37d6dde446e07cabb11 \
-    --hash=sha256:70988c352feb481887077d2ab845125024b2a137a5090d6d7a32b57d03a45df6
+accelerate==1.13.0 \
+    --hash=sha256:cf1a3efb96c18f7b152eb0fa7490f3710b19c3f395699358f08decca2b8b62e0 \
+    --hash=sha256:d631b4e0f5b3de4aff2d7e9e6857d164810dfc3237d54d017f075122d057b236
     # via -r .github/actions/compile-models/requirements.in
 certifi==2025.4.26 \
     --hash=sha256:0a816057ea3cdefcef70270d2c515e4506bbc954f417fa5ade2021213bb8f0c6 \
@@ -118,9 +118,9 @@ coloredlogs==15.0.1 \
     # via
     #   onnxruntime
     #   onnxruntime-tools
-filelock==3.21.2 \
-    --hash=sha256:cfd218cfccf8b947fce7837da312ec3359d10ef2a47c8602edd59e0bacffb708 \
-    --hash=sha256:d6cd4dbef3e1bb63bc16500fc5aa100f16e405bbff3fb4231711851be50c1560
+filelock==3.25.2 \
+    --hash=sha256:b64ece2b38f4ca29dd3e810287aa8c48182bbecd1ae6e9ae126c9b35f1382694 \
+    --hash=sha256:ca8afb0da15f229774c9ad1b455ed96e85a81373065fb10446672f64444ddf70
     # via
     #   -r .github/actions/compile-models/requirements.in
     #   huggingface-hub
@@ -136,33 +136,36 @@ fsspec==2025.3.2 \
     # via
     #   huggingface-hub
     #   torch
-hf-xet==1.2.0 \
-    --hash=sha256:10bfab528b968c70e062607f663e21e34e2bba349e8038db546646875495179e \
-    --hash=sha256:210d577732b519ac6ede149d2f2f34049d44e8622bf14eb3d63bbcd2d4b332dc \
-    --hash=sha256:27df617a076420d8845bea087f59303da8be17ed7ec0cd7ee3b9b9f579dff0e4 \
-    --hash=sha256:293a7a3787e5c95d7be1857358a9130694a9c6021de3f27fa233f37267174382 \
-    --hash=sha256:29c8fc913a529ec0a91867ce3d119ac1aac966e098cf49501800c870328cc090 \
-    --hash=sha256:2a212e842647b02eb6a911187dc878e79c4aa0aa397e88dd3b26761676e8c1f8 \
-    --hash=sha256:30e06daccb3a7d4c065f34fc26c14c74f4653069bb2b194e7f18f17cbe9939c0 \
-    --hash=sha256:3651fd5bfe0281951b988c0facbe726aa5e347b103a675f49a3fa8144c7968fd \
-    --hash=sha256:46740d4ac024a7ca9b22bebf77460ff43332868b661186a8e46c227fdae01848 \
-    --hash=sha256:4c1428c9ae73ec0939410ec73023c4f842927f39db09b063b9482dac5a3bb737 \
-    --hash=sha256:66e159cbfcfbb29f920db2c09ed8b660eb894640d284f102ada929b6e3dc410a \
-    --hash=sha256:6de1fc44f58f6dd937956c8d304d8c2dea264c80680bcfa61ca4a15e7b76780f \
-    --hash=sha256:7d40b18769bb9a8bc82a9ede575ce1a44c75eb80e7375a01d76259089529b5dc \
-    --hash=sha256:9c91d5ae931510107f148874e9e2de8a16052b6f1b3ca3c1b12f15ccb491390f \
-    --hash=sha256:a55558084c16b09b5ed32ab9ed38421e2d87cf3f1f89815764d1177081b99865 \
-    --hash=sha256:a8c27070ca547293b6890c4bf389f713f80e8c478631432962bb7f4bc0bd7d7f \
-    --hash=sha256:b70218dd548e9840224df5638fdc94bd033552963cfa97f9170829381179c813 \
-    --hash=sha256:cd3a6027d59cfb60177c12d6424e31f4b5ff13d8e3a1247b3a584bf8977e6df5 \
-    --hash=sha256:ceeefcd1b7aed4956ae8499e2199607765fbd1c60510752003b6cc0b8413b649 \
-    --hash=sha256:d06fa97c8562fb3ee7a378dd9b51e343bc5bc8190254202c9771029152f5e08c \
-    --hash=sha256:e6584a52253f72c9f52f9e549d5895ca7a471608495c4ecaa6cc73dba2b24d69 \
-    --hash=sha256:f182f264ed2acd566c514e45da9f2119110e48a87a327ca271027904c70c5832
+hf-xet==1.4.2 \
+    --hash=sha256:09b138422ecbe50fd0c84d4da5ff537d27d487d3607183cd10e3e53f05188e82 \
+    --hash=sha256:163aab46854ccae0ab6a786f8edecbbfbaa38fcaa0184db6feceebf7000c93c0 \
+    --hash=sha256:2f45c712c2fa1215713db10df6ac84b49d0e1c393465440e9cb1de73ecf7bbf6 \
+    --hash=sha256:32c012286b581f783653e718c1862aea5b9eb140631685bb0c5e7012c8719a87 \
+    --hash=sha256:41659966020d59eb9559c57de2cde8128b706a26a64c60f0531fa2318f409418 \
+    --hash=sha256:4642a6cf249c09da8c1f87fe50b24b2a3450b235bf8adb55700b52f0ea6e2eb6 \
+    --hash=sha256:5b2e0132c56d7ee1bf55bdb638c4b62e7106f6ac74f0b786fed499d5548c5570 \
+    --hash=sha256:5c588e21d80010119458dd5d02a69093f0d115d84e3467efe71ffb2c67c19146 \
+    --hash=sha256:6b0932eb8b10317ea78b7da6bab172b17be03bbcd7809383d8d5abd6a2233e04 \
+    --hash=sha256:6d2f8ee39fa9fba9af929f8c0d0482f8ee6e209179ad14a909b6ad78ffcb7c81 \
+    --hash=sha256:6d53df40616f7168abfccff100d232e9d460583b9d86fa4912c24845f192f2b8 \
+    --hash=sha256:71f02d6e4cdd07f344f6844845d78518cc7186bd2bc52d37c3b73dc26a3b0bc5 \
+    --hash=sha256:769431385e746c92dc05492dde6f687d304584b89c33d79def8367ace06cb555 \
+    --hash=sha256:77e8c180b7ef12d8a96739a4e1e558847002afe9ea63b6f6358b2271a8bdda1c \
+    --hash=sha256:949dcf88b484bb9d9276ca83f6599e4aa03d493c08fc168c124ad10b2e6f75d7 \
+    --hash=sha256:9d9a634cc929cfbaf2e1a50c0e532ae8c78fa98618426769480c58501e8c8ac2 \
+    --hash=sha256:a296744d771a8621ad1d50c098d7ab975d599800dae6d48528ba3944e5001ba0 \
+    --hash=sha256:ac8202ae1e664b2c15cdfc7298cbb25e80301ae596d602ef7870099a126fcad4 \
+    --hash=sha256:ad185719fb2e8ac26f88c8100562dbf9dbdcc3d9d2add00faa94b5f106aea53f \
+    --hash=sha256:b7457b6b482d9e0743bd116363239b1fa904a5e65deede350fbc0c4ea67c71ea \
+    --hash=sha256:c3b3c6a882016b94b6c210957502ff7877802d0dbda8ad142c8595db8b944271 \
+    --hash=sha256:c9dd1c1bc4cc56168f81939b0e05b4c36dd2d28c13dc1364b17af89aa0082496 \
+    --hash=sha256:e9b38d876e94d4bdcf650778d6ebbaa791dd28de08db9736c43faff06ede1b5a \
+    --hash=sha256:f563f7efe49588b7d0629d18d36f46d1658fe7e08dce3fa3d6526e1c98315e2d \
+    --hash=sha256:fca58a2ae4e6f6755cc971ac6fcdf777ea9284d7e540e350bb000813b9a3008d
     # via -r .github/actions/compile-models/requirements.in
-huggingface-hub==0.36.0 \
-    --hash=sha256:47b3f0e2539c39bf5cde015d63b72ec49baff67b6931c3d97f3f84532e2b8d25 \
-    --hash=sha256:7bcc9ad17d5b3f07b57c78e79d527102d08313caa278a641993acddcb894548d
+huggingface-hub==0.36.2 \
+    --hash=sha256:1934304d2fb224f8afa3b87007d58501acfda9215b334eed53072dd5e815ff7a \
+    --hash=sha256:48f0c8eac16145dfce371e9d2d7772854a4f591bcb56c9cf548accf531d54270
     # via
     #   -r .github/actions/compile-models/requirements.in
     #   accelerate
diff --git a/.github/actions/download-models/action.yml b/.github/actions/download-models/action.yml
index ff884b66d5..9f84b56939 100644
--- a/.github/actions/download-models/action.yml
+++ b/.github/actions/download-models/action.yml
@@ -21,7 +21,7 @@ runs:
     - name: Setup python version
       uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
       with:
-        python-version: '3.13'
+        python-version: '3.14'
 
     - name: Install dependencies
       shell: bash
diff --git a/.github/actions/download-models/requirements.in b/.github/actions/download-models/requirements.in
index aba5bee02d..8613b83471 100644
--- a/.github/actions/download-models/requirements.in
+++ b/.github/actions/download-models/requirements.in
@@ -1,8 +1,8 @@
 # Hugging Face Hub
-huggingface_hub>=1.4.1
-hf-xet>=1.2.0
+huggingface_hub==1.7.2
+hf-xet==1.4.2
 
 # Utils
 requests>=2.32.5
 urllib3>=2.6.3
-filelock>=3.21.2
+filelock>=3.25.2
diff --git a/.github/actions/download-models/requirements.txt b/.github/actions/download-models/requirements.txt
index b6c1b86cd6..b442df0e38 100644
--- a/.github/actions/download-models/requirements.txt
+++ b/.github/actions/download-models/requirements.txt
@@ -4,6 +4,10 @@
 #
 #    pip-compile --generate-hashes .github/actions/download-models/requirements.in
 #
+annotated-doc==0.0.4 \
+    --hash=sha256:571ac1dc6991c450b25a9c2d84a3705e2ae7a53467b5d111c24fa8baabbed320 \
+    --hash=sha256:fbcda96e87e9c92ad167c2e53839e57503ecfda18804ea28102353485033faa4
+    # via typer
 anyio==4.11.0 \
     --hash=sha256:0287e96f4d26d4149305414d4e3bc32f0dcd0862365a4bddea19d7a1ec38c4fc \
     --hash=sha256:82a8d0b81e318cc5ce71a5f1f8b5c4e63619620b63141ef8c995fa0db95a57c4
@@ -112,16 +116,16 @@ charset-normalizer==3.4.1 \
 click==8.3.0 \
     --hash=sha256:9b9f285302c6e3064f4330c05f05b81945b2a39544279343e6e7c5f27a9baddc \
     --hash=sha256:e7b8232224eba16f4ebe410c25ced9f7875cb5f3263ffc93cc3e8da705e229c4
-    # via typer-slim
+    # via typer
 colorama==0.4.6 \
     --hash=sha256:08695f5cb7ed6e0531a20572697297273c47b8cae5a63ffc6d6ed5c201be6e44 \
     --hash=sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6
     # via
     #   click
     #   tqdm
-filelock==3.21.2 \
-    --hash=sha256:cfd218cfccf8b947fce7837da312ec3359d10ef2a47c8602edd59e0bacffb708 \
-    --hash=sha256:d6cd4dbef3e1bb63bc16500fc5aa100f16e405bbff3fb4231711851be50c1560
+filelock==3.25.2 \
+    --hash=sha256:b64ece2b38f4ca29dd3e810287aa8c48182bbecd1ae6e9ae126c9b35f1382694 \
+    --hash=sha256:ca8afb0da15f229774c9ad1b455ed96e85a81373065fb10446672f64444ddf70
     # via
     #   -r .github/actions/download-models/requirements.in
     #   huggingface-hub
@@ -133,29 +137,32 @@ h11==0.16.0 \
     --hash=sha256:4e35b956cf45792e4caa5885e69fba00bdbc6ffafbfa020300e549b208ee5ff1 \
     --hash=sha256:63cf8bbe7522de3bf65932fda1d9c2772064ffb3dae62d55932da54b31cb6c86
     # via httpcore
-hf-xet==1.2.0 \
-    --hash=sha256:10bfab528b968c70e062607f663e21e34e2bba349e8038db546646875495179e \
-    --hash=sha256:210d577732b519ac6ede149d2f2f34049d44e8622bf14eb3d63bbcd2d4b332dc \
-    --hash=sha256:27df617a076420d8845bea087f59303da8be17ed7ec0cd7ee3b9b9f579dff0e4 \
-    --hash=sha256:293a7a3787e5c95d7be1857358a9130694a9c6021de3f27fa233f37267174382 \
-    --hash=sha256:29c8fc913a529ec0a91867ce3d119ac1aac966e098cf49501800c870328cc090 \
-    --hash=sha256:2a212e842647b02eb6a911187dc878e79c4aa0aa397e88dd3b26761676e8c1f8 \
-    --hash=sha256:30e06daccb3a7d4c065f34fc26c14c74f4653069bb2b194e7f18f17cbe9939c0 \
-    --hash=sha256:3651fd5bfe0281951b988c0facbe726aa5e347b103a675f49a3fa8144c7968fd \
-    --hash=sha256:46740d4ac024a7ca9b22bebf77460ff43332868b661186a8e46c227fdae01848 \
-    --hash=sha256:4c1428c9ae73ec0939410ec73023c4f842927f39db09b063b9482dac5a3bb737 \
-    --hash=sha256:66e159cbfcfbb29f920db2c09ed8b660eb894640d284f102ada929b6e3dc410a \
-    --hash=sha256:6de1fc44f58f6dd937956c8d304d8c2dea264c80680bcfa61ca4a15e7b76780f \
-    --hash=sha256:7d40b18769bb9a8bc82a9ede575ce1a44c75eb80e7375a01d76259089529b5dc \
-    --hash=sha256:9c91d5ae931510107f148874e9e2de8a16052b6f1b3ca3c1b12f15ccb491390f \
-    --hash=sha256:a55558084c16b09b5ed32ab9ed38421e2d87cf3f1f89815764d1177081b99865 \
-    --hash=sha256:a8c27070ca547293b6890c4bf389f713f80e8c478631432962bb7f4bc0bd7d7f \
-    --hash=sha256:b70218dd548e9840224df5638fdc94bd033552963cfa97f9170829381179c813 \
-    --hash=sha256:cd3a6027d59cfb60177c12d6424e31f4b5ff13d8e3a1247b3a584bf8977e6df5 \
-    --hash=sha256:ceeefcd1b7aed4956ae8499e2199607765fbd1c60510752003b6cc0b8413b649 \
-    --hash=sha256:d06fa97c8562fb3ee7a378dd9b51e343bc5bc8190254202c9771029152f5e08c \
-    --hash=sha256:e6584a52253f72c9f52f9e549d5895ca7a471608495c4ecaa6cc73dba2b24d69 \
-    --hash=sha256:f182f264ed2acd566c514e45da9f2119110e48a87a327ca271027904c70c5832
+hf-xet==1.4.2 \
+    --hash=sha256:09b138422ecbe50fd0c84d4da5ff537d27d487d3607183cd10e3e53f05188e82 \
+    --hash=sha256:163aab46854ccae0ab6a786f8edecbbfbaa38fcaa0184db6feceebf7000c93c0 \
+    --hash=sha256:2f45c712c2fa1215713db10df6ac84b49d0e1c393465440e9cb1de73ecf7bbf6 \
+    --hash=sha256:32c012286b581f783653e718c1862aea5b9eb140631685bb0c5e7012c8719a87 \
+    --hash=sha256:41659966020d59eb9559c57de2cde8128b706a26a64c60f0531fa2318f409418 \
+    --hash=sha256:4642a6cf249c09da8c1f87fe50b24b2a3450b235bf8adb55700b52f0ea6e2eb6 \
+    --hash=sha256:5b2e0132c56d7ee1bf55bdb638c4b62e7106f6ac74f0b786fed499d5548c5570 \
+    --hash=sha256:5c588e21d80010119458dd5d02a69093f0d115d84e3467efe71ffb2c67c19146 \
+    --hash=sha256:6b0932eb8b10317ea78b7da6bab172b17be03bbcd7809383d8d5abd6a2233e04 \
+    --hash=sha256:6d2f8ee39fa9fba9af929f8c0d0482f8ee6e209179ad14a909b6ad78ffcb7c81 \
+    --hash=sha256:6d53df40616f7168abfccff100d232e9d460583b9d86fa4912c24845f192f2b8 \
+    --hash=sha256:71f02d6e4cdd07f344f6844845d78518cc7186bd2bc52d37c3b73dc26a3b0bc5 \
+    --hash=sha256:769431385e746c92dc05492dde6f687d304584b89c33d79def8367ace06cb555 \
+    --hash=sha256:77e8c180b7ef12d8a96739a4e1e558847002afe9ea63b6f6358b2271a8bdda1c \
+    --hash=sha256:949dcf88b484bb9d9276ca83f6599e4aa03d493c08fc168c124ad10b2e6f75d7 \
+    --hash=sha256:9d9a634cc929cfbaf2e1a50c0e532ae8c78fa98618426769480c58501e8c8ac2 \
+    --hash=sha256:a296744d771a8621ad1d50c098d7ab975d599800dae6d48528ba3944e5001ba0 \
+    --hash=sha256:ac8202ae1e664b2c15cdfc7298cbb25e80301ae596d602ef7870099a126fcad4 \
+    --hash=sha256:ad185719fb2e8ac26f88c8100562dbf9dbdcc3d9d2add00faa94b5f106aea53f \
+    --hash=sha256:b7457b6b482d9e0743bd116363239b1fa904a5e65deede350fbc0c4ea67c71ea \
+    --hash=sha256:c3b3c6a882016b94b6c210957502ff7877802d0dbda8ad142c8595db8b944271 \
+    --hash=sha256:c9dd1c1bc4cc56168f81939b0e05b4c36dd2d28c13dc1364b17af89aa0082496 \
+    --hash=sha256:e9b38d876e94d4bdcf650778d6ebbaa791dd28de08db9736c43faff06ede1b5a \
+    --hash=sha256:f563f7efe49588b7d0629d18d36f46d1658fe7e08dce3fa3d6526e1c98315e2d \
+    --hash=sha256:fca58a2ae4e6f6755cc971ac6fcdf777ea9284d7e540e350bb000813b9a3008d
     # via
     #   -r .github/actions/download-models/requirements.in
     #   huggingface-hub
@@ -167,9 +174,9 @@ httpx==0.28.1 \
     --hash=sha256:75e98c5f16b0f35b567856f597f06ff2270a374470a5c2392242528e3e3e42fc \
     --hash=sha256:d909fcccc110f8c7faf814ca82a9a4d816bc5a6dbfea25d6591d6985b8ba59ad
     # via huggingface-hub
-huggingface-hub==1.4.1 \
-    --hash=sha256:9931d075fb7a79af5abc487106414ec5fba2c0ae86104c0c62fd6cae38873d18 \
-    --hash=sha256:b41131ec35e631e7383ab26d6146b8d8972abc8b6309b963b306fbcca87f5ed5
+huggingface-hub==1.7.2 \
+    --hash=sha256:288f33a0a17b2a73a1359e2a5fd28d1becb2c121748c6173ab8643fb342c850e \
+    --hash=sha256:7f7e294e9bbb822e025bdb2ada025fa4344d978175a7f78e824d86e35f7ab43b
     # via -r .github/actions/download-models/requirements.in
 idna==3.10 \
     --hash=sha256:12f65c9b470abda6dc35cf8e63cc574b1c52b11df2c86030af0ac09b01b13ea9 \
@@ -178,10 +185,22 @@ idna==3.10 \
     #   anyio
     #   httpx
     #   requests
+markdown-it-py==4.0.0 \
+    --hash=sha256:87327c59b172c5011896038353a81343b6754500a08cd7a4973bb48c6d578147 \
+    --hash=sha256:cb0a2b4aa34f932c007117b194e945bd74e0ec24133ceb5bac59009cda1cb9f3
+    # via rich
+mdurl==0.1.2 \
+    --hash=sha256:84008a41e51615a49fc9966191ff91509e3c40b939176e643fd50a5c2196b8f8 \
+    --hash=sha256:bb413d29f5eea38f31dd4754dd7377d4465116fb207585f97bf925588687c1ba
+    # via markdown-it-py
 packaging==24.2 \
     --hash=sha256:09abb1bccd265c01f4a3aa3f7a7db064b36514d2cba19a2f694fe6150451a759 \
     --hash=sha256:c228a6dc5e932d346bc5739379109d49e8853dd8223571c7c5b55260edc0b97f
     # via huggingface-hub
+pygments==2.19.2 \
+    --hash=sha256:636cb2477cec7f8952536970bc533bc43743542f70392ae026374600add5b887 \
+    --hash=sha256:86540386c03d588bb81d44bc3928634ff26449851e99741617ecb9037ee5ec0b
+    # via rich
 pyyaml==6.0.2 \
     --hash=sha256:01179a4a8559ab5de078078f37e5c1a30d76bb88519906844fd7bdea1b7729ff \
     --hash=sha256:0833f8694549e586547b576dcfaba4a6b55b9e96098b36cdc7ebefe667dfed48 \
@@ -241,10 +260,14 @@ requests==2.32.5 \
     --hash=sha256:2462f94637a34fd532264295e186976db0f5d453d1cdd31473c85a6a161affb6 \
     --hash=sha256:dbba0bac56e100853db0ea71b82b4dfd5fe2bf6d3754a8893c3af500cec7d7cf
     # via -r .github/actions/download-models/requirements.in
+rich==14.3.3 \
+    --hash=sha256:793431c1f8619afa7d3b52b2cdec859562b950ea0d4b6b505397612db8d5362d \
+    --hash=sha256:b8daa0b9e4eef54dd8cf7c86c03713f53241884e814f4e2f5fb342fe520f639b
+    # via typer
 shellingham==1.5.4 \
     --hash=sha256:7ecfff8f2fd72616f7481040475a65b2bf8af90a56c89140852d1120324e8686 \
     --hash=sha256:8dbca0739d487e5bd35ab3ca4b36e11c4078f3a234bfce294b0a0291363404de
-    # via huggingface-hub
+    # via typer
 sniffio==1.3.1 \
     --hash=sha256:2f6da418d1f1e0fddd844478f41680e794e6051915791a034ff65e5f100525a2 \
     --hash=sha256:f4324edc670a0f49750a81b895f35c3adb843cca46f0530f79fc1babb23789dc
@@ -253,16 +276,14 @@ tqdm==4.67.1 \
     --hash=sha256:26445eca388f82e72884e0d580d5464cd801a3ea01e63e5601bdff9ba6a48de2 \
     --hash=sha256:f8aef9c52c08c13a65f30ea34f4e5aac3fd1a34959879d7e59e63027286627f2
     # via huggingface-hub
-typer-slim==0.20.0 \
-    --hash=sha256:9fc6607b3c6c20f5c33ea9590cbeb17848667c51feee27d9e314a579ab07d1a3 \
-    --hash=sha256:f42a9b7571a12b97dddf364745d29f12221865acef7a2680065f9bb29c7dc89d
+typer==0.24.1 \
+    --hash=sha256:112c1f0ce578bfb4cab9ffdabc68f031416ebcc216536611ba21f04e9aa84c9e \
+    --hash=sha256:e39b4732d65fbdcde189ae76cf7cd48aeae72919dea1fdfc16593be016256b45
     # via huggingface-hub
 typing-extensions==4.13.2 \
     --hash=sha256:a439e7c04b49fec3e5d3e2beaa21755cadbbdc391694e28ccdd36ca4a1408f8c \
     --hash=sha256:e6c81219bd689f51865d9e372991c540bda33a0379d5573cddb9a3a23f7caaef
-    # via
-    #   huggingface-hub
-    #   typer-slim
+    # via huggingface-hub
 urllib3==2.6.3 \
     --hash=sha256:1b62b6884944a57dbe321509ab94fd4d3b307075e0c2eae991ac71ee15ad38ed \
     --hash=sha256:bf272323e553dfb2e87d9bfd225ca7b0f467b919d7bbd355436d3fd37cb0acd4
diff --git a/.github/requirements-dev.in b/.github/requirements-dev.in
index acfc646ac6..24e54a2e1d 100644
--- a/.github/requirements-dev.in
+++ b/.github/requirements-dev.in
@@ -1,4 +1,4 @@
-setuptools==80.9.0
+setuptools<80.10
 wheel>=0.46.3
-build==1.4.0
+build==1.4.1
 patchelf>=0.16.1.0,<=0.17.2.4; sys_platform == 'linux' and platform_machine == 'x86_64'
diff --git a/.github/requirements-dev.txt b/.github/requirements-dev.txt
index e8cb251b4a..e47ee123b6 100644
--- a/.github/requirements-dev.txt
+++ b/.github/requirements-dev.txt
@@ -4,9 +4,9 @@
 #
 #    pip-compile --allow-unsafe --generate-hashes .github/requirements-dev.in
 #
-build==1.4.0 \
-    --hash=sha256:6a07c1b8eb6f2b311b96fcbdbce5dab5fe637ffda0fd83c9cac622e927501596 \
-    --hash=sha256:f1b91b925aa322be454f8330c6fb48b465da993d1e7e7e6fa35027ec49f3c936
+build==1.4.1 \
+    --hash=sha256:21c81f7a0fa423f0da229335c5c2a605967fbfc9af3c4b6ecd368265ed59c6bc \
+    --hash=sha256:30adeb28821e573a49b556030d8c84186d112f6a38b12fa5476092c4544ae55a
     # via -r .github/requirements-dev.in
 colorama==0.4.6 \
     --hash=sha256:08695f5cb7ed6e0531a20572697297273c47b8cae5a63ffc6d6ed5c201be6e44 \
diff --git a/.github/workflows/clang-format.yml b/.github/workflows/clang-format.yml
index d920a03cef..4cb49ceddc 100644
--- a/.github/workflows/clang-format.yml
+++ b/.github/workflows/clang-format.yml
@@ -22,15 +22,15 @@ jobs:
     runs-on: ubuntu-latest-8-cores
     steps:
       - name: Checkout NPU Compiler
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
           fetch-depth: 0
           lfs: true
 
       - name: Setup python version
-        uses: actions/setup-python@83679a892e2d95755f2dac6acb0bfd1e9ac5d548 # v6.1.0
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
         with:
-            python-version: '3.13'
+            python-version: '3.14'
 
       - name: Run clang-format via pre-commit (pull_request)
         if: github.event_name == 'pull_request'
diff --git a/.github/workflows/codeql.yml b/.github/workflows/codeql.yml
index a2c1722529..f3a15057d3 100644
--- a/.github/workflows/codeql.yml
+++ b/.github/workflows/codeql.yml
@@ -31,16 +31,16 @@ jobs:
           build-mode: none
     steps:
     - name: Checkout repository
-      uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+      uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
 
     - name: Initialize CodeQL
-      uses: github/codeql-action/init@5d4e8d1aca955e8d8589aabd499c5cae939e33c7 # v4.31.9
+      uses: github/codeql-action/init@38697555549f1db7851b81482ff19f1fa5c4fedc # v4.34.1
       with:
         languages: ${{ matrix.language }}
         build-mode: ${{ matrix.build-mode }}
 
     - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@5d4e8d1aca955e8d8589aabd499c5cae939e33c7 # v4.31.9
+      uses: github/codeql-action/analyze@38697555549f1db7851b81482ff19f1fa5c4fedc # v4.34.1
       with:
         category: "/language:${{matrix.language}}"
 
diff --git a/.github/workflows/job_build_cid.yml b/.github/workflows/job_build_cid.yml
index 9f5553057f..0a1a225b95 100644
--- a/.github/workflows/job_build_cid.yml
+++ b/.github/workflows/job_build_cid.yml
@@ -118,7 +118,7 @@ jobs:
           cmake -E make_directory "${CID_PACKAGE_ARTIFACTS_DIR}"
 
       - name: Checkout actions
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
           path: ${{ env.NPU_ACTIONS_DIR }}
           sparse-checkout: |
@@ -177,14 +177,14 @@ jobs:
       - name: Restore CiD artifacts from cache
         if: ${{ inputs.build-cache }}
         id: cache-restore
-        uses: actions/cache/restore@9255dc7a253b0ccc959486e2bca901246202afeb # v5.0.1
+        uses: actions/cache/restore@668228422ae6a00e4ad889ee87cd7109ec5666a7 # v5.0.4
         with:
           key: ${{ steps.cache-key.outputs.cache-key }}
           path: ${{ env.CID_PACKAGE_ARTIFACTS_DIR }}
 
       - name: Clone OpenVINO
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         timeout-minutes: 15
         with:
           repository: ${{ steps.versions.outputs.openvino-repository }}
@@ -194,7 +194,7 @@ jobs:
 
       - name: Checkout NPU Compiler
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
           path: ${{ env.NPU_COMPILER_REPO }}
           submodules: true
@@ -202,13 +202,13 @@ jobs:
 
       - name: System info
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: openvinotoolkit/openvino/.github/actions/system_info@82bbf0292c5bd490bc25a383e63d6461d3ee05c5 # 2025.4.1
+        uses: openvinotoolkit/openvino/.github/actions/system_info@9185925763612da451e6c5dd2819670bec249ea5 # 2026.0.1
 
       - name: Setup python version
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/setup-python@83679a892e2d95755f2dac6acb0bfd1e9ac5d548 # v6.1.0
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
         with:
-          python-version: '3.13'
+          python-version: '3.14'
 
       - name: Install python deps
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
@@ -351,7 +351,7 @@ jobs:
 
       - name: Cache CiD artifacts
         if: ${{ inputs.build-cache && !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/cache/save@9255dc7a253b0ccc959486e2bca901246202afeb # v5.0.1
+        uses: actions/cache/save@668228422ae6a00e4ad889ee87cd7109ec5666a7 # v5.0.4
         with:
           path: ${{ env.CID_PACKAGE_ARTIFACTS_DIR }}
           key: ${{ steps.cache-key.outputs.cache-key }}
diff --git a/.github/workflows/job_build_drv_linux.yml b/.github/workflows/job_build_drv_linux.yml
index 4ebf43cd38..e736d39967 100644
--- a/.github/workflows/job_build_drv_linux.yml
+++ b/.github/workflows/job_build_drv_linux.yml
@@ -64,14 +64,14 @@ jobs:
       - name: Restore Driver package from cache
         if: ${{ inputs.build-cache }}
         id: cache-restore
-        uses: actions/cache/restore@9255dc7a253b0ccc959486e2bca901246202afeb # v5.0.1
+        uses: actions/cache/restore@668228422ae6a00e4ad889ee87cd7109ec5666a7 # v5.0.4
         with:
           key: ${{ steps.cache-key.outputs.cache-key }}
           path: ${{ env.NPU_DRIVER_INSTALL_DIR }}
 
       - name: Checkout versions action
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
           path: ${{ env.NPU_ACTIONS_DIR }}
           sparse-checkout: |
@@ -85,7 +85,7 @@ jobs:
 
       - name: Clone Linux NPU Driver
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         timeout-minutes: 15
         with:
           repository: ${{ steps.versions.outputs.npu-linux-driver-repository }}
@@ -95,7 +95,7 @@ jobs:
 
       - name: System info
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: openvinotoolkit/openvino/.github/actions/system_info@82bbf0292c5bd490bc25a383e63d6461d3ee05c5 # 2025.4.1
+        uses: openvinotoolkit/openvino/.github/actions/system_info@9185925763612da451e6c5dd2819670bec249ea5 # 2026.0.1
 
       - name: Install system packages
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
@@ -145,7 +145,7 @@ jobs:
 
       - name: Cache artifacts
         if: ${{ inputs.build-cache && !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/cache/save@9255dc7a253b0ccc959486e2bca901246202afeb # v5.0.1
+        uses: actions/cache/save@668228422ae6a00e4ad889ee87cd7109ec5666a7 # v5.0.4
         with:
           path: ${{ env.NPU_DRIVER_INSTALL_DIR }}
           key: ${{ steps.cache-key.outputs.cache-key }}
diff --git a/.github/workflows/job_build_plugin_linux.yml b/.github/workflows/job_build_plugin_linux.yml
index 392f3675b1..d397bbd391 100644
--- a/.github/workflows/job_build_plugin_linux.yml
+++ b/.github/workflows/job_build_plugin_linux.yml
@@ -88,14 +88,14 @@ jobs:
       - name: Restore OpenVINO/NPU package from cache
         if: ${{ inputs.build-cache }}
         id: cache-restore
-        uses: actions/cache/restore@9255dc7a253b0ccc959486e2bca901246202afeb # v5.0.1
+        uses: actions/cache/restore@668228422ae6a00e4ad889ee87cd7109ec5666a7 # v5.0.4
         with:
           key: ${{ steps.cache-key.outputs.cache-key }}
           path: ${{ env.OPENVINO_INSTALL_DIR }}
 
       - name: Checkout versions action
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
           path: ${{ env.NPU_ACTIONS_DIR }}
           sparse-checkout: |
@@ -109,7 +109,7 @@ jobs:
 
       - name: Clone OpenVINO
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         timeout-minutes: 15
         with:
           repository: ${{ steps.versions.outputs.openvino-repository }}
@@ -119,7 +119,7 @@ jobs:
 
       - name: Clone NPU Compiler
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         timeout-minutes: 15
         with:
           path: ${{ env.NPU_COMPILER_REPO }}
@@ -128,13 +128,13 @@ jobs:
 
       - name: System info
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: openvinotoolkit/openvino/.github/actions/system_info@82bbf0292c5bd490bc25a383e63d6461d3ee05c5 # 2025.4.1
+        uses: openvinotoolkit/openvino/.github/actions/system_info@9185925763612da451e6c5dd2819670bec249ea5 # 2026.0.1
 
       - name: Setup python version
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/setup-python@83679a892e2d95755f2dac6acb0bfd1e9ac5d548 # v6.1.0
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6.2.0
         with:
-          python-version: '3.13'
+          python-version: '3.14'
 
       - name: Install python dependencies
         if: ${{ !steps.cache-restore.outputs.cache-hit }}
@@ -152,7 +152,7 @@ jobs:
 
       - name: Initialize CodeQL
         if: ${{ !steps.cache-restore.outputs.cache-hit && inputs.with-codeql }}
-        uses: github/codeql-action/init@5d4e8d1aca955e8d8589aabd499c5cae939e33c7 # v4.31.9
+        uses: github/codeql-action/init@38697555549f1db7851b81482ff19f1fa5c4fedc # v4.34.1
         with:
           languages: c-cpp
           build-mode: manual
@@ -239,7 +239,7 @@ jobs:
       - name: Perform CodeQL Analysis
         id: codeql-analyze
         if: ${{ !steps.cache-restore.outputs.cache-hit && inputs.with-codeql }}
-        uses: github/codeql-action/analyze@5d4e8d1aca955e8d8589aabd499c5cae939e33c7 # v4.31.9
+        uses: github/codeql-action/analyze@38697555549f1db7851b81482ff19f1fa5c4fedc # v4.34.1
         with:
           category: "/language:c-cpp"
           output: ${{ env.CODEQL_OUTPUTS }}
@@ -262,7 +262,7 @@ jobs:
 
       - name: Upload CodeQL SARIF
         if: ${{ !steps.cache-restore.outputs.cache-hit && inputs.with-codeql }}
-        uses: github/codeql-action/upload-sarif@5d4e8d1aca955e8d8589aabd499c5cae939e33c7 # v4.31.9
+        uses: github/codeql-action/upload-sarif@38697555549f1db7851b81482ff19f1fa5c4fedc # v4.34.1
         with:
           sarif_file: ${{ env.CODEQL_OUTPUTS }}/filtered-results.sarif
           checkout_path: ${{ env.NPU_COMPILER_REPO }}
@@ -277,7 +277,7 @@ jobs:
 
       - name: Cache artifacts
         if: ${{ inputs.build-cache && !steps.cache-restore.outputs.cache-hit }}
-        uses: actions/cache/save@9255dc7a253b0ccc959486e2bca901246202afeb # v5.0.1
+        uses: actions/cache/save@668228422ae6a00e4ad889ee87cd7109ec5666a7 # v5.0.4
         with:
           path: ${{ env.OPENVINO_INSTALL_DIR }}
           key: ${{ steps.cache-key.outputs.cache-key }}
diff --git a/.github/workflows/job_tests_compilation_linux.yml b/.github/workflows/job_tests_compilation_linux.yml
index 5bc9f887b2..82aa212597 100644
--- a/.github/workflows/job_tests_compilation_linux.yml
+++ b/.github/workflows/job_tests_compilation_linux.yml
@@ -53,6 +53,7 @@ jobs:
           - NPU.3720
           - NPU.4000
           - NPU.5010
+          - NPU.5020
         compiler-mode: 
           - ${{ inputs.compiler-type }}
         include:
@@ -62,20 +63,22 @@ jobs:
             models-config: ./validation/public_models_4000_${{ inputs.compiler-type }}.json
           - npu-platform: NPU.5010
             models-config: ./validation/public_models_5010_${{ inputs.compiler-type }}.json
+          - npu-platform: NPU.5020
+            models-config: ./validation/public_models_5020_${{ inputs.compiler-type }}.json
     steps:
       - name: Checkout
-        if: ${{ !(matrix.npu-platform == 'NPU.5010' && inputs.compiler-type == 'DRIVER') }}
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        if: ${{ !(matrix.npu-platform == 'NPU.5020' && inputs.compiler-type == 'DRIVER') }}
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
 
       - name: Download artifacts
-        if: ${{ !(matrix.npu-platform == 'NPU.5010' && inputs.compiler-type == 'DRIVER') }}
+        if: ${{ !(matrix.npu-platform == 'NPU.5020' && inputs.compiler-type == 'DRIVER') }}
         uses: actions/download-artifact@018cc2cf5baa6db3ef3c5f8a56943fffe632ef53  # v6.0.0
         with:
           name: ${{ inputs.install-package }}
           path: ${{ env.INSTALL_PACKAGE_DIR }}
 
       - name: Unpack Driver artifacts
-        if: ${{ inputs.compiler-type == 'DRIVER' && !(matrix.npu-platform == 'NPU.5010') }}
+        if: ${{ inputs.compiler-type == 'DRIVER' && !(matrix.npu-platform == 'NPU.5020') }}
         run: |
           cd ${INSTALL_PACKAGE_DIR}
           dpkg --extract external/intel-driver-compiler-npu_*.deb .
@@ -85,13 +88,13 @@ jobs:
           find . -type f
 
       - name: Install system packages
-        if: ${{ inputs.compiler-type == 'DRIVER' && !(matrix.npu-platform == 'NPU.5010') }}
+        if: ${{ inputs.compiler-type == 'DRIVER' && !(matrix.npu-platform == 'NPU.5020') }}
         run: |
           sudo apt update
           sudo apt install -y libtbb12
 
       - name: Download models for ${{ matrix.npu-platform }}
-        if: ${{ !(matrix.npu-platform == 'NPU.5010' && inputs.compiler-type == 'DRIVER') }}
+        if: ${{ !(matrix.npu-platform == 'NPU.5020' && inputs.compiler-type == 'DRIVER') }}
         uses: ./.github/actions/download-models
         with:
           compiler-type: ${{ inputs.compiler-type }}
@@ -99,7 +102,7 @@ jobs:
           models-dir: ${{ env.MODELS_DIR }}
 
       - name: Compile models for ${{ matrix.npu-platform }}
-        if: ${{ !(matrix.npu-platform == 'NPU.5010' && inputs.compiler-type == 'DRIVER') }}
+        if: ${{ !(matrix.npu-platform == 'NPU.5020' && inputs.compiler-type == 'DRIVER') }}
         uses: ./.github/actions/compile-models
         with:
           compiler-type: ${{ inputs.compiler-type }}
diff --git a/.github/workflows/scorecard.yml b/.github/workflows/scorecard.yml
index 330b2ad4a8..9a098c26da 100644
--- a/.github/workflows/scorecard.yml
+++ b/.github/workflows/scorecard.yml
@@ -23,7 +23,7 @@ jobs:
 
     steps:
       - name: "Checkout code"
-        uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8 # v6.0.1
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
         with:
           persist-credentials: false
 
@@ -42,6 +42,6 @@ jobs:
           retention-days: 5
 
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@5d4e8d1aca955e8d8589aabd499c5cae939e33c7 # v4.31.9
+        uses: github/codeql-action/upload-sarif@38697555549f1db7851b81482ff19f1fa5c4fedc # v4.34.1
         with:
           sarif_file: results.sarif
diff --git a/.github/workflows/ubuntu_24.yml b/.github/workflows/ubuntu_24.yml
index 1cfb838ceb..6ac77a65bd 100644
--- a/.github/workflows/ubuntu_24.yml
+++ b/.github/workflows/ubuntu_24.yml
@@ -32,7 +32,7 @@ jobs:
       build-cache: true
       build-cache-key-suffix: ${{ github.sha }}
       build-plugin: true
-      build-driver: false
+      build-driver: true
       build-cid: true
       run-unit-tests: true
       run-lit-tests: true
diff --git a/validation/public_models_5020_PLUGIN.json b/validation/public_models_5020_PLUGIN.json
new file mode 100644
index 0000000000..2e21783aac
--- /dev/null
+++ b/validation/public_models_5020_PLUGIN.json
@@ -0,0 +1,600 @@
+{
+  "models": [
+    {
+      "name": "resnet50-onnx_FP32_LATENCY_PLUGIN",
+      "repository": "Qdrant/resnet50-onnx",
+      "repository_type": "huggingface",
+      "repository_path": "model.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC",
+        "shape": "input[1,3,224,224]"
+      }
+    },
+    {
+      "name": "resnet50-onnx_FP32_THROUGHPUT_PLUGIN",
+      "repository": "Qdrant/resnet50-onnx",
+      "repository_type": "huggingface",
+      "repository_path": "model.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "THROUGHPUT"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC",
+        "shape": "input[1,3,224,224]"
+      }
+    },
+    {
+      "name": "resnet50-v1-12_FP32_LATENCY_PLUGIN",
+      "repository": "onnx/models",
+      "repository_type": "github",
+      "repository_branch": "main",
+      "repository_path": "validated/vision/classification/resnet/model/resnet50-v1-12.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC"
+      }
+    },
+    {
+      "name": "resnet50-v1-12-qdq_FP32-INT8_LATENCY_PLUGIN",
+      "repository": "onnx/models",
+      "repository_type": "github",
+      "repository_branch": "main",
+      "repository_path": "validated/vision/classification/resnet/model/resnet50-v1-12-qdq.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC"
+      }
+    },
+    {
+      "name": "resnet-50-pytorch_FP32-INT8_LATENCY_PLUGIN",
+      "repository": "microsoft/resnet-50",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "resnet",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "image_url": "http://images.cocodataset.org/val2017/000000039769.jpg",
+          "input_names": ["pixel_values"],
+          "output_names": ["logits"],
+          "dynamic_axes": {
+              "pixel_values": {"0": "batch_size"},
+              "logits": {"0": "batch_size"}
+          }
+      },
+      "Quantize": {
+          "quant_type": "QInt8"
+      },
+      "Compile": {
+          "input_precision": "U8",
+          "output_precision": "FP32",
+          "shape": "pixel_values[1,3,224,224]"
+      }
+    },
+    {
+      "name": "mobilenet_v2_1.0_224_FP32_LATENCY_PLUGIN",
+      "repository": "onnx-community/mobilenet_v2_1.0_224",
+      "repository_type": "huggingface",
+      "repository_path": "onnx/model.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC"
+      }
+    },
+    {
+      "name": "mobilenet_v2_1.0_224_FP16_LATENCY_PLUGIN",
+      "repository": "onnx-community/mobilenet_v2_1.0_224",
+      "repository_type": "huggingface",
+      "repository_path": "onnx/model_fp16.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC"
+      }
+    },
+    {
+      "name": "mobilenet_v2_1.0_224_FP32-INT8_LATENCY_PLUGIN",
+      "repository": "onnx-community/mobilenet_v2_1.0_224",
+      "repository_type": "huggingface",
+      "repository_path": "onnx/model_quantized.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC"
+      }
+    },
+    {
+      "name": "mobilenetv2_140_Opset18_timm_FP32_LATENCY_PLUGIN",
+      "repository": "onnx/models",
+      "repository_type": "github",
+      "repository_branch": "main",
+      "repository_path": "Computer_Vision/mobilenetv2_140_Opset18_timm/mobilenetv2_140_Opset18.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC"
+      }
+    },
+    {
+      "name": "mobilenetv2-12-qdq_FP32-INT8_LATENCY_PLUGIN",
+      "repository": "onnx/models",
+      "repository_type": "github",
+      "repository_branch": "main",
+      "repository_path": "validated/vision/classification/mobilenet/model/mobilenetv2-12-qdq.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NCHW",
+        "output_layout": "NC",
+        "model_input_layout": "NCHW",
+        "model_output_layout": "NC"
+      }
+    },
+    {
+      "name": "yolov4_FP32_LATENCY_PLUGIN",
+      "repository": "onnx/models",
+      "repository_type": "github",
+      "repository_branch": "main",
+      "repository_path": "validated/vision/object_detection_segmentation/yolov4/model/yolov4.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NHWC",
+        "model_input_layout": "NHWC"
+      }
+    },
+    {
+      "name": "yolov4_FP32-INT8_LATENCY_PLUGIN",
+      "repository": "onnx/models",
+      "repository_type": "github",
+      "repository_branch": "main",
+      "repository_path": "validated/vision/object_detection_segmentation/yolov4/model/yolov4.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Quantize": {
+          "quant_type": "QInt8"
+      },
+      "Compile": {
+        "input_precision": "FP16",
+        "output_precision": "FP16",
+        "input_layout": "NHWC",
+        "model_input_layout": "NHWC"
+      }
+    },
+    {
+      "name": "bert-base-uncased_FP32_LATENCY_PLUGIN",
+      "repository": "google-bert/bert-base-uncased",
+      "repository_type": "huggingface",
+      "repository_path": "model.onnx",
+      "framework": "onnx",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Compile": {
+        "input_precision": "I32",
+        "output_precision": "FP32",
+        "shape": "input_ids[1,128],attention_mask[1,128],token_type_ids[1,128]"
+      }
+    },
+    {
+      "name": "bert-base-multilingual-cased_FP32_LATENCY_PLUGIN",
+      "repository": "google-bert/bert-base-multilingual-cased",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "transformer",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+        "NPU_COMPILER_TYPE": "PLUGIN",
+        "NPU_PLATFORM": "VPU5020",
+        "DEVICE_ID": "5020",
+        "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+        "text1": "What is the meaning of your existence?",
+        "input_names": ["input_ids", "attention_mask"],
+        "output_names": ["last_hidden_state", "pooler_output"],
+        "dynamic_axes": {
+            "input_ids": {"0": "batch_size", "1": "sequence_length"},
+            "attention_mask": {"0": "batch_size", "1": "sequence_length"},
+            "last_hidden_state": {"0": "batch_size", "1": "sequence_length"},
+            "pooler_output": {"0": "batch_size"}
+        }
+      },
+      "Compile": {
+        "input_precision": "I32",
+        "output_precision": "FP32",
+        "shape": "input_ids[1,128],attention_mask[1,128]"
+      }
+    },
+    {
+      "name": "bert-base-uncased-mrpc_FP32_LATENCY_PLUGIN",
+      "repository": "Intel/bert-base-uncased-mrpc",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "transformer",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "text1": "What is the meaning of your existence?",
+          "text2": "Hello world!",
+          "input_names": ["input_ids", "attention_mask","token_type_ids"],
+          "output_names": ["logits"],
+          "dynamic_axes": {
+              "input_ids": {"0": "batch_size", "1": "sequence_length"},
+              "attention_mask": {"0": "batch_size", "1": "sequence_length"},
+              "token_type_ids": {"0": "batch_size", "1": "sequence_length"},
+              "logits": {"0": "batch_size"}
+          }
+      },
+      "Compile": {
+          "input_precision": "I32",
+          "output_precision": "FP32",
+          "shape": "input_ids[1,128],attention_mask[1,128],token_type_ids[1,128]"
+      }
+    },
+    {
+      "name": "clip-vit-base-patch16_FP32_LATENCY_PLUGIN",
+      "repository": "openai/clip-vit-base-patch16",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "clip",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "text1": "a photo of a cat",
+          "image_url": "http://images.cocodataset.org/val2017/000000039769.jpg",
+          "input_names": ["input_ids", "attention_mask", "pixel_values"],
+          "output_names": ["logits_per_image", "logits_per_text", "text_embeds", "image_embeds"],
+          "dynamic_axes": {
+              "input_ids": {"0": "batch_size", "1": "seq_len"},
+              "attention_mask": {"0": "batch_size", "1": "seq_len"},
+              "pixel_values": {"0": "batch_size"},
+              "logits_per_image": {"0": "batch_size", "1": "batch_size"},
+              "logits_per_text": {"0": "batch_size", "1": "batch_size"},
+              "text_embeds": {"0": "batch_size"},
+              "image_embeds": {"0": "batch_size"}
+          }
+      },
+      "Compile": {
+          "input_precision": "I32",
+          "output_precision": "FP32",
+          "shape": "input_ids[1,77],attention_mask[1,77],pixel_values[1,3,224,224]"
+      }
+    },
+    {
+      "name": "clip-vit-base-patch32_FP32_LATENCY_PLUGIN",
+      "repository": "openai/clip-vit-base-patch32",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "clip",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "text1": "a photo of a cat",
+          "image_url": "http://images.cocodataset.org/val2017/000000039769.jpg",
+          "input_names": ["input_ids", "attention_mask", "pixel_values"],
+          "output_names": ["logits_per_image", "logits_per_text", "text_embeds", "image_embeds"],
+          "dynamic_axes": {
+              "input_ids": {"0": "batch_size", "1": "seq_len"},
+              "attention_mask": {"0": "batch_size", "1": "seq_len"},
+              "pixel_values": {"0": "batch_size"},
+              "logits_per_image": {"0": "batch_size", "1": "batch_size"},
+              "logits_per_text": {"0": "batch_size", "1": "batch_size"},
+              "text_embeds": {"0": "batch_size"},
+              "image_embeds": {"0": "batch_size"}
+          }
+      },
+      "Compile": {
+          "input_precision": "I32",
+          "output_precision": "FP32",
+          "shape": "input_ids[1,77],attention_mask[1,77],pixel_values[1,3,224,224]"
+      }
+    },
+    {
+      "name": "CLIP-ViT-B-32-laion2B-s34B-b79K_FP32_LATENCY_PLUGIN",
+      "repository": "laion/CLIP-ViT-B-32-laion2B-s34B-b79K",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "clip",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "text1": "a photo of a cat",
+          "image_url": "http://images.cocodataset.org/val2017/000000039769.jpg",
+          "input_names": ["input_ids", "attention_mask", "pixel_values"],
+          "output_names": ["logits_per_image", "logits_per_text", "text_embeds", "image_embeds"],
+          "dynamic_axes": {
+              "input_ids": {"0": "batch_size", "1": "seq_len"},
+              "attention_mask": {"0": "batch_size", "1": "seq_len"},
+              "pixel_values": {"0": "batch_size"},
+              "logits_per_image": {"0": "batch_size", "1": "batch_size"},
+              "logits_per_text": {"0": "batch_size", "1": "batch_size"},
+              "text_embeds": {"0": "batch_size"},
+              "image_embeds": {"0": "batch_size"}
+          }
+      },
+      "Compile": {
+          "input_precision": "I32",
+          "output_precision": "FP32",
+          "shape": "input_ids[1,77],attention_mask[1,77],pixel_values[1,3,224,224]"
+      }
+    },
+    {
+      "name": "sam-vit-base_image_encoder_FP32_LATENCY_PLUGIN",
+      "disabled": true,
+      "repository": "facebook/sam-vit-base",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "sam_image_encoder",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "image_url": "https://huggingface.co/ybelkada/segment-anything/resolve/main/assets/car.png",
+          "input_names": ["pixel_values"],
+          "output_names": ["image_embeddings"],
+          "dynamic_axes": {
+              "pixel_values": {"0": "batch"},
+              "image_embeddings": {"0": "batch"}
+          }
+      },
+      "Compile": {
+          "input_precision": "U8",
+          "output_precision": "FP32",
+          "shape": "pixel_values[1,3,1024,1024]"
+      }
+    },
+    {
+      "name": "sam-vit-base_prompt_encoder_FP32_LATENCY_PLUGIN",
+      "repository": "facebook/sam-vit-base",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "sam_prompt_encoder",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "point_coords": [512.0,512.0],
+          "point_labels": [1],
+          "mask_input": [1,1,256,256],
+          "input_names": ["point_coords", "point_labels", "mask_input"],
+          "output_names": ["sparse_embeddings", "dense_embeddings"],
+          "dynamic_axes": {
+              "point_coords": {"0": "batch", "1": "num_points"},
+              "point_labels": {"0": "batch", "1": "num_points"},
+              "mask_input": {"0": "batch"},
+              "sparse_embeddings": {"0": "batch", "1": "num_points"},
+              "dense_embeddings": {"0": "batch"}
+          }
+      },
+      "Compile": {
+          "input_precision": "FP32",
+          "output_precision": "FP32",
+          "shape": "point_coords[1,1,1,2],point_labels[1,1,1],mask_input[1,1,256,256]"
+      }
+    },
+    {
+      "name": "sam-vit-base_mask_decoder_FP32_LATENCY_PLUGIN",
+      "repository": "facebook/sam-vit-base",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "sam_mask_decoder",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "image_embeddings_shape": [1,256,64,64],
+          "sparse_embeddings_shape": [1,1,2,256],
+          "dense_embeddings_shape": [1,256,64,64],
+          "input_names": ["image_embeddings", "sparse_embeddings", "dense_embeddings", "multimask_output"],
+          "output_names": ["low_res_masks", "iou_predictions"],
+          "dynamic_axes": {
+              "image_embeddings": {"0": "batch"},
+              "sparse_embeddings": {"0": "batch", "1": "num_points"},
+              "dense_embeddings": {"0": "batch"},
+              "multimask_output": {"0": "batch"},
+              "low_res_masks": {"0": "batch"},
+              "iou_predictions": {"0": "batch"}
+          }
+      },
+      "Compile": {
+          "input_precision": "FP32",
+          "output_precision": "FP32",
+          "shape": "image_embeddings[1,256,64,64],sparse_embeddings[1,1,2,256],dense_embeddings[1,256,64,64]"
+      }
+    },
+    {
+      "name": "vit-base-patch16-224_FP32_LATENCY_PLUGIN",
+      "repository": "google/vit-base-patch16-224",
+      "repository_type": "huggingface",
+      "repository_path": "pytorch_model.bin",
+      "framework": "pytorch",
+      "model_type": "vision_transformer",
+      "category": "PLUGIN/VPU5020/SILICON",
+      "extra_config": {
+          "NPU_COMPILER_TYPE": "PLUGIN",
+          "NPU_PLATFORM": "VPU5020",
+          "DEVICE_ID": "5020",
+          "PERFORMANCE_HINT": "LATENCY"
+      },
+      "Convert": {
+          "image_url": "http://images.cocodataset.org/val2017/000000039769.jpg",
+          "input_names": ["pixel_values"],
+          "output_names": ["logits"],
+          "dynamic_axes": {
+              "pixel_values": {"0": "batch_size"},
+              "logits": {"0": "batch_size"}
+          }
+      },
+      "Compile": {
+          "input_precision": "U8",
+          "output_precision": "FP32",
+          "shape": "pixel_values[1,3,224,224]"
+      }
+    }
+  ]
+}