Agora board builds (Model Farm, verified on real boards)

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+targets/QCS6490/W8A16/decoder.ctx.bin.aidem filter=lfs diff=lfs merge=lfs -text
+targets/QCS6490/W8A16/visual.ctx.bin.aidem filter=lfs diff=lfs merge=lfs -text
+targets/QCS8550/W8A16/decoder.ctx.bin.aidem filter=lfs diff=lfs merge=lfs -text
+targets/QCS8550/W8A16/visual.ctx.bin.aidem filter=lfs diff=lfs merge=lfs -text

NOTICE.md

@@ -0,0 +1,5 @@
+# Provenance
+
+Resources are copied without binary changes from the APLUX Model Farm build `CLIPSeg` that Agora verified on real boards (decoded sample check, then measured), pinned by sha256 in SHA256SUMS.
+
+Agora arranges files under targets/<SoC>/<precision>/ and adds model.json. AidLite, AidGen, AidVoice and the Qualcomm QNN runtime are separate dependencies with their own terms.

README.md

@@ -0,0 +1,38 @@
+---
+license: other
+license_name: aplux-model-farm
+license_link: https://huggingface.co/APLUX-Agora/CLIPSeg/blob/main/LICENSE_MODEL_FARM.pdf
+pipeline_tag: image-segmentation
+tags: [Agora, QNN, AIoT, Qualcomm]
+---
+
+# CLIPSeg model resources
+
+按英文目标描述分割图片，返回原图大小的掩码和完整特征。
+
+Board builds that [Agora](https://mega.aidlux.com) runs and measures on real Qualcomm boards. Try it online, without a board: https://mega.aidlux.com/models/clipseg
+
+| SoC · board | Precisions | Measured on Agora (default precision) |
+| --- | --- | --- |
+| QCS8550 · 犀牛派 X1 (Rhino Pi X1) | W8A16 | 61.0 ms |
+| QCS6490 · 犀牛派 A1 (Rhino Pi A1) | W8A16 | 211.0 ms |
+
+Task: 分割. Inputs: image (image), text (text). Outputs: masks (masks), out (json).
+
+## Download
+
+```bash
+export AGORA_WORKSPACE="$HOME/my-clipseg"
+export AGORA_SOC='QCS8550'
+hf download APLUX-Agora/CLIPSeg \
+  --include "targets/$AGORA_SOC/*" --local-dir "$AGORA_WORKSPACE/resources"
+```
+
+Each `targets/<SoC>/` holds `model.json` (default precision and variant paths) and one directory per precision. Context binaries are named `model.bin`; multi-file packages are kept as published by Model Farm.
+
+Pre- and post-processing, the runner and an exported project: open the model on Agora and choose “部署指南”, or export any workflow that uses it.
+
+## Licence and provenance
+
+Source model licence: see the original model's terms. 
+Converted, deployable resources: APLUX Model Farm License (LICENSE_MODEL_FARM.pdf). See NOTICE.md and SHA256SUMS.

SHA256SUMS

@@ -0,0 +1,14 @@
+144a143f84b117dc0b7b37b75edc4057256b22ca61195bdb6266cf701cb4c8ad  targets/QCS6490/W8A16/composite.json
+de9df12b2b3fc61f1d5050c8fe1900ca9148deed7842133d68397671e22454be  targets/QCS6490/W8A16/decoder.ctx.bin.aidem
+9fd691f7c8039210e0fced15865466c65820d09b63988b0174bfe25de299051a  targets/QCS6490/W8A16/merges.txt
+4fb09ebcfd7651205ca8299b993c30088e5535ef350a72d46c5c4580eeac0440  targets/QCS6490/W8A16/preprocessor.json
+2286c1fe27d35cf56a0d6c1820c68c07d7a3377baf3d3853f2e8241a084ed429  targets/QCS6490/W8A16/text.onnx
+6ffd501cd89b0d8f84c8fed1b790bab13ae0487f0f470fc73c3a872253750888  targets/QCS6490/W8A16/visual.ctx.bin.aidem
+e089ad92ba36837a0d31433e555c8f45fe601ab5c221d4f607ded32d9f7a4349  targets/QCS6490/W8A16/vocabulary.json
+2e95a15c995b3c12a342a050c93866115acf8c20c6b3bf862cbd4dccf288b08d  targets/QCS8550/W8A16/composite.json
+1b9d62fa9c48f5d5eaddddbe8f7cc2b4d43ee2e4816ea51ec62a8942ef8176cf  targets/QCS8550/W8A16/decoder.ctx.bin.aidem
+9fd691f7c8039210e0fced15865466c65820d09b63988b0174bfe25de299051a  targets/QCS8550/W8A16/merges.txt
+4fb09ebcfd7651205ca8299b993c30088e5535ef350a72d46c5c4580eeac0440  targets/QCS8550/W8A16/preprocessor.json
+2286c1fe27d35cf56a0d6c1820c68c07d7a3377baf3d3853f2e8241a084ed429  targets/QCS8550/W8A16/text.onnx
+da832a2020e8abb8c476856c26e8c037547e4000c914d8ad21d16dfd8d3b3a4a  targets/QCS8550/W8A16/visual.ctx.bin.aidem
+e089ad92ba36837a0d31433e555c8f45fe601ab5c221d4f607ded32d9f7a4349  targets/QCS8550/W8A16/vocabulary.json

targets/QCS6490/W8A16/composite.json

@@ -0,0 +1,50 @@
+{
+  "assets": {
+    "merges": "merges.txt",
+    "preprocessor": "preprocessor.json",
+    "vocabulary": "vocabulary.json"
+  },
+  "components": {
+    "decoder": "decoder.ctx.bin.aidem",
+    "text": "text.onnx",
+    "visual": "visual.ctx.bin.aidem"
+  },
+  "kind": "clipseg",
+  "source": {
+    "files": [
+      {
+        "path": "models/QCS6490/W8A16/clipseg_352_image_qcs6490_w8a16.qnn231.ctx.bin.aidem",
+        "sha256": "6ffd501cd89b0d8f84c8fed1b790bab13ae0487f0f470fc73c3a872253750888",
+        "size": 80640536
+      },
+      {
+        "path": "models/QCS6490/W8A16/clipseg_352_decoder_qcs6490_w8a16.qnn231.ctx.bin.aidem",
+        "sha256": "de9df12b2b3fc61f1d5050c8fe1900ca9148deed7842133d68397671e22454be",
+        "size": 1571954
+      },
+      {
+        "path": "models/QCS6490/W8A16/clipseg_352_text_int8.onnx",
+        "sha256": "2286c1fe27d35cf56a0d6c1820c68c07d7a3377baf3d3853f2e8241a084ed429",
+        "size": 67807017
+      },
+      {
+        "path": "code/python/clipseg_rd64_refined/merges.txt",
+        "sha256": "9fd691f7c8039210e0fced15865466c65820d09b63988b0174bfe25de299051a",
+        "size": 524619
+      },
+      {
+        "path": "code/python/clipseg_rd64_refined/vocab.json",
+        "sha256": "e089ad92ba36837a0d31433e555c8f45fe601ab5c221d4f607ded32d9f7a4349",
+        "size": 1059962
+      },
+      {
+        "path": "code/python/clipseg_rd64_refined/preprocessor_config.json",
+        "sha256": "4fb09ebcfd7651205ca8299b993c30088e5535ef350a72d46c5c4580eeac0440",
+        "size": 380
+      }
+    ],
+    "projectId": 1254,
+    "ref": "d59adbba14ec959ea61503d74a018952f505169f"
+  },
+  "version": 1
+}

targets/QCS6490/W8A16/decoder.ctx.bin.aidem

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de9df12b2b3fc61f1d5050c8fe1900ca9148deed7842133d68397671e22454be
+size 1571954

targets/QCS6490/W8A16/preprocessor.json

@@ -0,0 +1,23 @@
+{
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "feature_extractor_type": "ViTImageProcessor",
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "processor_class": "CLIPSegProcessor",
+  "resample": 2,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "height": 352,
+    "width": 352
+  }
+}

targets/QCS6490/W8A16/text.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2286c1fe27d35cf56a0d6c1820c68c07d7a3377baf3d3853f2e8241a084ed429
+size 67807017

targets/QCS6490/W8A16/visual.ctx.bin.aidem

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ffd501cd89b0d8f84c8fed1b790bab13ae0487f0f470fc73c3a872253750888
+size 80640536

targets/QCS6490/model.json

@@ -0,0 +1,7 @@
+{
+  "soc": "QCS6490",
+  "default_precision": "W8A16",
+  "variants": {
+    "W8A16": "W8A16/"
+  }
+}

targets/QCS8550/W8A16/composite.json

@@ -0,0 +1,50 @@
+{
+  "assets": {
+    "merges": "merges.txt",
+    "preprocessor": "preprocessor.json",
+    "vocabulary": "vocabulary.json"
+  },
+  "components": {
+    "decoder": "decoder.ctx.bin.aidem",
+    "text": "text.onnx",
+    "visual": "visual.ctx.bin.aidem"
+  },
+  "kind": "clipseg",
+  "source": {
+    "files": [
+      {
+        "path": "models/QCS8550/W8A16/clipseg_352_image_qcs8550_w8a16.qnn231.ctx.bin.aidem",
+        "sha256": "da832a2020e8abb8c476856c26e8c037547e4000c914d8ad21d16dfd8d3b3a4a",
+        "size": 78477849
+      },
+      {
+        "path": "models/QCS8550/W8A16/clipseg_352_decoder_qcs8550_w8a16.qnn231.ctx.bin.aidem",
+        "sha256": "1b9d62fa9c48f5d5eaddddbe8f7cc2b4d43ee2e4816ea51ec62a8942ef8176cf",
+        "size": 1604659
+      },
+      {
+        "path": "models/QCS8550/W8A16/clipseg_352_text_int8.onnx",
+        "sha256": "2286c1fe27d35cf56a0d6c1820c68c07d7a3377baf3d3853f2e8241a084ed429",
+        "size": 67807017
+      },
+      {
+        "path": "code/python/clipseg_rd64_refined/merges.txt",
+        "sha256": "9fd691f7c8039210e0fced15865466c65820d09b63988b0174bfe25de299051a",
+        "size": 524619
+      },
+      {
+        "path": "code/python/clipseg_rd64_refined/vocab.json",
+        "sha256": "e089ad92ba36837a0d31433e555c8f45fe601ab5c221d4f607ded32d9f7a4349",
+        "size": 1059962
+      },
+      {
+        "path": "code/python/clipseg_rd64_refined/preprocessor_config.json",
+        "sha256": "4fb09ebcfd7651205ca8299b993c30088e5535ef350a72d46c5c4580eeac0440",
+        "size": 380
+      }
+    ],
+    "projectId": 1254,
+    "ref": "d59adbba14ec959ea61503d74a018952f505169f"
+  },
+  "version": 1
+}

targets/QCS8550/W8A16/decoder.ctx.bin.aidem

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b9d62fa9c48f5d5eaddddbe8f7cc2b4d43ee2e4816ea51ec62a8942ef8176cf
+size 1604659

targets/QCS8550/W8A16/preprocessor.json

@@ -0,0 +1,23 @@
+{
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "feature_extractor_type": "ViTImageProcessor",
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "processor_class": "CLIPSegProcessor",
+  "resample": 2,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "height": 352,
+    "width": 352
+  }
+}

targets/QCS8550/W8A16/text.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2286c1fe27d35cf56a0d6c1820c68c07d7a3377baf3d3853f2e8241a084ed429
+size 67807017

targets/QCS8550/W8A16/visual.ctx.bin.aidem

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da832a2020e8abb8c476856c26e8c037547e4000c914d8ad21d16dfd8d3b3a4a
+size 78477849

targets/QCS8550/model.json

@@ -0,0 +1,7 @@
+{
+  "soc": "QCS8550",
+  "default_precision": "W8A16",
+  "variants": {
+    "W8A16": "W8A16/"
+  }
+}