Upload models for the Android app

README.md

@@ -0,0 +1,20 @@
+---
+license: apache-2.0
+tags:
+- onnx
+- inpainting
+- android
+- qnn
+---
+
+# Moebius inpainting — ONNX models prepared for Android
+
+Derived from [simonw/Moebius-ONNX](https://huggingface.co/simonw/Moebius-ONNX) (ONNX export of [hustvl/Moebius](https://github.com/hustvl/Moebius), Apache-2.0)
+for the native Android app in the `onnx-android` project. Files are downloaded by the app and checked against sha256 values compiled into the APK.
+
+| Set | Files | For |
+|---|---|---|
+| `htp` | `hyb.json` + `hyb.s*.cpu.onnx` (CPU segments) + `hyb.s*.htp_ctx.onnx` / `_ctx_qnn.bin` (QNN HTP context binaries), `vae_*.static_ctx.onnx` / `_ctx_qnn.bin` | Snapdragon 8 Elite (SM8750, HTP v79) only. UNet split into 13 segments: 64×64 attention blocks and the timestep embedding on CPU, the rest on the NPU. Context binaries compiled offline with onnxruntime-qnn 2.7.0 (QNN 2.51); the device needs qnn-runtime ≥ 2.51 |
+| `cpu` | `unet.einsum.onnx`, `vae_encoder.static.onnx`, `vae_decoder.static.onnx` | Any device, CPU EP, fp32. Static shapes (UNet batch 2), Einsum rewritten to MatMul, condition embedding replaced by a `context` input (2×10×3072, shipped in the APK) |
+
+Changes from the original graphs: static shapes, `input_ids` → `context` input, Einsum → Transpose/Reshape/MatMul, normalization pre-scaling (HTP fp16 range), graph splitting. Weights are unchanged.

hyb.json

@@ -0,0 +1,226 @@
+{
+ "segments": [
+  {
+   "file": "hyb.s0.htp.onnx",
+   "device": "htp",
+   "nodes": [
+    0,
+    81
+   ],
+   "inputs": [
+    "context",
+    "latent"
+   ],
+   "outputs": [
+    "/m/diff_model/encoder_hid_proj/Add_output_0",
+    "/m/diff_model/up_blocks.2/attentions.2/transformer_blocks.0/attn2/Reshape_4_output_0",
+    "/m/diff_model/up_blocks.2/attentions.2/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/up_blocks.2/attentions.1/transformer_blocks.0/attn2/Reshape_4_output_0",
+    "/m/diff_model/up_blocks.2/attentions.1/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/up_blocks.2/attentions.0/transformer_blocks.0/attn2/Reshape_4_output_0",
+    "/m/diff_model/up_blocks.2/attentions.0/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/down_blocks.0/attentions.1/transformer_blocks.0/attn2/Reshape_4_output_0",
+    "/m/diff_model/down_blocks.0/attentions.1/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/down_blocks.0/attentions.0/transformer_blocks.0/attn2/Reshape_4_output_0",
+    "/m/diff_model/down_blocks.0/attentions.0/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/conv_in/conv_pw/Conv_output_0",
+    "/m/diff_model/down_blocks.0/resnets.0/conv1/Add_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s1.cpu.onnx",
+   "device": "cpu",
+   "nodes": [
+    82,
+    95
+   ],
+   "inputs": [
+    "timesteps"
+   ],
+   "outputs": [
+    "/m/diff_model/time_embedding/linear_2/Gemm_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s2.htp.onnx",
+   "device": "htp",
+   "nodes": [
+    96,
+    114
+   ],
+   "inputs": [
+    "/m/diff_model/time_embedding/linear_2/Gemm_output_0",
+    "/m/diff_model/down_blocks.0/resnets.0/conv1/Add_output_0",
+    "/m/diff_model/conv_in/conv_pw/Conv_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/down_blocks.0/resnets.0/nonlinearity_1/Mul_output_0",
+    "/m/diff_model/down_blocks.0/resnets.0/Add_1_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s3.cpu.onnx",
+   "device": "cpu",
+   "nodes": [
+    115,
+    198
+   ],
+   "inputs": [
+    "/m/diff_model/down_blocks.0/resnets.0/Add_1_output_0",
+    "/m/diff_model/down_blocks.0/attentions.0/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/down_blocks.0/attentions.0/transformer_blocks.0/attn2/Reshape_4_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/down_blocks.0/attentions.0/Add_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s4.htp.onnx",
+   "device": "htp",
+   "nodes": [
+    199,
+    227
+   ],
+   "inputs": [
+    "/m/diff_model/down_blocks.0/attentions.0/Add_output_0",
+    "/m/diff_model/down_blocks.0/resnets.0/nonlinearity_1/Mul_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/down_blocks.0/resnets.1/Add_1_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s5.cpu.onnx",
+   "device": "cpu",
+   "nodes": [
+    228,
+    310
+   ],
+   "inputs": [
+    "/m/diff_model/down_blocks.0/resnets.1/Add_1_output_0",
+    "/m/diff_model/down_blocks.0/attentions.1/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/down_blocks.0/attentions.1/transformer_blocks.0/attn2/Reshape_4_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/down_blocks.0/attentions.1/Add_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s6.htp.onnx",
+   "device": "htp",
+   "nodes": [
+    311,
+    1631
+   ],
+   "inputs": [
+    "/m/diff_model/down_blocks.0/attentions.1/Add_output_0",
+    "/m/diff_model/down_blocks.0/resnets.0/nonlinearity_1/Mul_output_0",
+    "/m/diff_model/encoder_hid_proj/Add_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/up_blocks.2/resnets.0/Add_1_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s7.cpu.onnx",
+   "device": "cpu",
+   "nodes": [
+    1632,
+    1714
+   ],
+   "inputs": [
+    "/m/diff_model/up_blocks.2/resnets.0/Add_1_output_0",
+    "/m/diff_model/up_blocks.2/attentions.0/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/up_blocks.2/attentions.0/transformer_blocks.0/attn2/Reshape_4_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/up_blocks.2/attentions.0/Add_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s8.htp.onnx",
+   "device": "htp",
+   "nodes": [
+    1715,
+    1744
+   ],
+   "inputs": [
+    "/m/diff_model/up_blocks.2/attentions.0/Add_output_0",
+    "/m/diff_model/down_blocks.0/attentions.0/Add_output_0",
+    "/m/diff_model/down_blocks.0/resnets.0/nonlinearity_1/Mul_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/up_blocks.2/resnets.1/Add_1_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s9.cpu.onnx",
+   "device": "cpu",
+   "nodes": [
+    1745,
+    1827
+   ],
+   "inputs": [
+    "/m/diff_model/up_blocks.2/resnets.1/Add_1_output_0",
+    "/m/diff_model/up_blocks.2/attentions.1/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/up_blocks.2/attentions.1/transformer_blocks.0/attn2/Reshape_4_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/up_blocks.2/attentions.1/Add_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s10.htp.onnx",
+   "device": "htp",
+   "nodes": [
+    1828,
+    1857
+   ],
+   "inputs": [
+    "/m/diff_model/up_blocks.2/attentions.1/Add_output_0",
+    "/m/diff_model/conv_in/conv_pw/Conv_output_0",
+    "/m/diff_model/down_blocks.0/resnets.0/nonlinearity_1/Mul_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/up_blocks.2/resnets.2/Add_1_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s11.cpu.onnx",
+   "device": "cpu",
+   "nodes": [
+    1858,
+    1942
+   ],
+   "inputs": [
+    "/m/diff_model/up_blocks.2/resnets.2/Add_1_output_0",
+    "/m/diff_model/up_blocks.2/attentions.2/transformer_blocks.0/attn2/Einsum_1/in2_R",
+    "/m/diff_model/up_blocks.2/attentions.2/transformer_blocks.0/attn2/Reshape_4_output_0"
+   ],
+   "outputs": [
+    "/m/diff_model/up_blocks.2/attentions.2/Add_output_0"
+   ]
+  },
+  {
+   "file": "hyb.s12.htp.onnx",
+   "device": "htp",
+   "nodes": [
+    1943,
+    1953
+   ],
+   "inputs": [
+    "/m/diff_model/up_blocks.2/attentions.2/Add_output_0"
+   ],
+   "outputs": [
+    "noise"
+   ]
+  }
+ ],
+ "final": "noise",
+ "graph_inputs": [
+  "latent",
+  "timesteps",
+  "context"
+ ],
+ "source": "unet.ns.onnx"
+}

hyb.s0.htp_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85a38fd492685bc234f39c63271e0aab6988574f885b1a46e0eb32773376cfff
+size 3048

hyb.s0.htp_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14d594fc7eb8f1962c721ac9a7c08102ae76399112458e23c8f4740c4c5a3505
+size 6656000

hyb.s1.cpu.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52ca5f2d075e78bc817abc3b1a529a678cb8ba31761c096a9b11f68e9ceb8cdd
+size 8207345

hyb.s10.htp_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71b793201d91088165e5c78125798846b1bc8a5cfc50c470044e4e8015304596
+size 1171

hyb.s10.htp_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3191c169c31045a0fcae347a54a5d8cb5914abee0dac005f79d79fe7ef08133e
+size 3497984

hyb.s11.cpu.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:161a6fba36c0dd3501d90bae60dd5fe29d7ca56294a4f1c2889825fb6752eae1
+size 11532733

hyb.s12.htp_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3c27c58e9451041ad12367ed3c8dc130b5c478190220bf0a9a17a55e30b6952
+size 808

hyb.s12.htp_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72b2f694a27711da6e4dad140a234d1106801e0f5db0fb0cc76cc0b3eab28a3a
+size 540672

hyb.s2.htp_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2c445d0c4a107dc3c77c6922525bb131d315cfc6acc579d4d7b559e609836e9
+size 1307

hyb.s2.htp_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9969dfaad79c0da3108bdb9acd0931a99a861d91f7b4a62178093234c6f918fe
+size 1716224

hyb.s3.cpu.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c44e3a8d70a000193710f852eb64cbf3bc3c39c43b186d61558ba5e3b6d0c9bd
+size 11533031

hyb.s4.htp_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b53c90a674718939cb5526fa80d393085856ab807bf0182d74e8fe04093cf13c
+size 1059

hyb.s4.htp_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2c1f0521e39506f17f215801b802f6c398fd6c6d3351353c31152718345bba3
+size 2347008

hyb.s5.cpu.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ccf1ab479d093f8fa7200ce48cf4a1a18383e3e078827dc7f53b3b018ea4253
+size 11532277

hyb.s6.htp_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:125e0ead3c788560982ef4af6473297adf91d751504fef1a3c8e2f3ba8d48491
+size 1168

hyb.s6.htp_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d25a2b3958ff43467559f09d1bc7ec34ad63b93b006f135e83b66190eb138557
+size 492933120

hyb.s7.cpu.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37e835ac7a7ae833b2919b80e7abd92e234d9f04f15e9731061025d28afc661a
+size 11531477

hyb.s8.htp_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4b0e53b99ca29db870b8a912bb0ffc3db5e19e1185028a6648d2d5e499b7136
+size 1188

hyb.s8.htp_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f680e82df385db1f4e248043a0afcf1c9bfe7185759494d04e8a670f9e81cec
+size 3502080

hyb.s9.cpu.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b98768bda6f9d8cecd8b9dee343e62f5e406278f213d4788c23c412653bb8482
+size 11531477

unet.einsum.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b1d9c08d9b38a97171dc72d68c01b8caee5c6fa5ddc0ef42d5b3b7914820b6a
+size 904522395

vae_decoder.static.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20e26ae7333e4374e685932afa4986441b3c14712e256fc74e9555fee5957e2f
+size 198090837

vae_decoder.static_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:290b0db465b37598ed12cb14eaa944b07f241018bb548d8085784186b047562d
+size 726

vae_decoder.static_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:701d86e0529dc8cb004256cf0890a5ed71f1d7e090550081a76f5fa8215662e6
+size 111640576

vae_encoder.static.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:390dfee84e3b3361b4eb290a829f42984f1e129c583c08abc917accb7be0f5b6
+size 136758234

vae_encoder.static_ctx.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5946bfbd701c8aa3878475c89a9e739e4750bbb84587e57931d7204e285db4a
+size 730

vae_encoder.static_ctx_qnn.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b233315cc79790ef62854b2fbe0323b39797829f20b2af81ec86d13ac5becd2e
+size 76587008