MiniCPM-V Demo — iOS, Android & HarmonyOS

English | 中文

This demo runs the MiniCPM-V family of multimodal models fully on-device on iOS, Android, and HarmonyOS NEXT. Currently supported:

• MiniCPM-V 2.6
• MiniCPM-V 4.0
• MiniCPM-V 4.6
• MiniCPM-V 4.7 35B-A3B (MoE, streams experts from SSD)
• MiniCPM-V 4.7 1B (dense)
• MiniCPM5-1B (text-only)
• MiniCPM5-2B (text-only)
• VoxCPM2 (text-to-speech, TTS)

• MiniCPM-V-demo/ — iOS demo (Xcode project)
• MiniCPM-V-demo-Android/ — Android demo (Gradle / Kotlin)
• MiniCPM-V-demo-HarmonyOS/ — HarmonyOS NEXT demo (DevEco Studio / ArkTS)

NOTE: This project bundles llama.cpp as a git submodule. The upstream fork tc-mb/llama.cpp-omni carries a dozen unrelated branches and a full clone weighs ~350 MB, so shallow = true is set in .gitmodules by default. The recommended shallow + single-branch clone is:

>

> # one-shot (parent repo + submodules, all shallow)

> Or, if you've already cloned the parent repo and want to init the submodule afterwards:

>

> git submodule update --init --recursive --depth 1 --single-branch

> This only pulls a single commit of the ios-omni branch (~tens of MB) instead of the full llama.cpp-omni fork history. Developers who need to push to tc-mb/llama.cpp-omni:ios-omni can run git fetch --unshallow inside the submodule to lift the shallow restriction.

The README is organised in two parts:

• Part 1 — Platform setup: how to build and run the demo on iOS, Android and HarmonyOS.
• Part 2 — GGUF model files: where to get the model weights for each MiniCPM-V version, and the minimum on-device hardware needed to run them.

Just want to install the app? Pre-built packages are below; step-by-step instructions (requirements, signing notes, MD5) are in DOWNLOAD.md. The rest of this README is only needed if you want to build from source.

| Platform | Install | | --- | --- | | 🍎 iOS / iPadOS | TestFlight — public link | | 🤖 Android | APK — latest release | | 📱 HarmonyOS | HAP — latest release |

Part 1. Platform Setup

1.1 iOS Demo

NOTE: To deploy and test the app on an iOS device, you may need an Apple Developer account.

1.1.1 Install Xcode & command-line tools

• Download Xcode from the App Store (verified on Xcode 26.1; project deployment target = iOS 16.4)
• Install the Command Line Tools:

xcode-select --install

• Agree to the software license agreement:

sudo xcodebuild -license

• CMake ≥ 3.28 (needed by the xcframework build in the next step):

brew install cmake

1.1.2 Build llama.xcframework (required on first build)

This repo does not track any compiled artefacts, so the prebuilt llama.xcframework (~189 MB) needs to be produced locally from the llama.cpp-omni submodule and dropped into MiniCPM-V-demo/thirdparty/ for Xcode to link against. A one-shot script is provided — by default it only builds the two slices the demo actually links (real device + simulator), which takes ~2-3 min on a modern M-series Mac:

./scripts/build_xcframework.sh

The output is installed at MiniCPM-V-demo/thirdparty/llama.xcframework/.

If you need a different build scope (simulator-only, full multi-platform, …), the script forwards MINIMAL_MODE:

MINIMAL_MODE=ios-sim    ./scripts/build_xcframework.sh   # simulator only        (~3 min)
MINIMAL_MODE=ios-device ./scripts/build_xcframework.sh   # device only           (~3 min)
MINIMAL_MODE=ios        ./scripts/build_xcframework.sh   # device + simulator (default, ~3 min)
MINIMAL_MODE=all        ./scripts/build_xcframework.sh   # iOS + macOS + tvOS + xrOS (~25 min)

The equivalent manual commands, if you'd rather not use the script:

cd llama.cpp-omni
MINIMAL_MODE=ios ./build-xcframework.sh
cp -r ./build-apple/llama.xcframework ../MiniCPM-V-demo/thirdparty/

During the build you will see warnings like ignoring duplicate libraries and skipping debug map object with duplicate name and timestamp — these come from llama.cpp's mtmd module having identically-named .o files across different model architectures. They are harmless and the resulting framework works correctly.

When do I need to rebuild?

- The parent repo bumped the llama.cpp-omni submodule pointer (git submodule status shows a commit different from your last local build).

- You edited any source under llama.cpp-omni/ that affects the framework.

1.1.3 Open in Xcode and run

Open MiniCPM-V-demo/MiniCPM-V-demo.xcodeproj with Xcode. It may take a moment for Xcode to automatically download the required dependencies. Select the target device at the top, then click the "Run" (triangle) button to launch the demo.

If Xcode fails with There is no XCFramework found at '.../llama.xcframework', you skipped §1.1.2 — go back and run the build.

1.2 Android Demo

Requirements:

• Android Studio (Giraffe or newer)
• Android SDK + NDK (the project pins NDK 27.0.12077973 and CMake 3.22.1)
• A physical device with a 64-bit ARM SoC (arm64-v8a)
• Device RAM: see the per-model requirements in Part 2

cd MiniCPM-V-demo-Android
./gradlew assembleDebug

Or open MiniCPM-V-demo-Android/ directly in Android Studio and click Run.

The first launch will download the GGUF model files into the app's external storage. You can also sideload model files manually via adb push — see in-app Model Manager for the expected directory layout.

1.3 HarmonyOS Demo

Requirements:

• DevEco Studio 5.0 or newer (with the HarmonyOS Native SDK / NDK)
• A real device or emulator running HarmonyOS API 12+ (e.g. nova 14 vitality / Mate 60 / Pura 70)
• 64-bit ARM architecture (arm64-v8a)
• Device RAM: see the per-model requirements in Part 2

1. Open MiniCPM-V-demo-HarmonyOS/ in DevEco Studio.
2. File → Project Structure → Signing Configs and tick Automatically generate signature (requires a Huawei developer account; this only needs to be done once).
3. Connect the device with USB debugging enabled, then click Run (the green triangle).

The HarmonyOS port shares the exact same llama.cpp-omni submodule, model catalogue, OBS direct-link URLs and MD5 hashes with the iOS / Android demos.

Part 2. GGUF Model Files

Hardware requirements

The on-device memory needed to run a model is roughly (model file size) + KV cache + a few hundred MB of working memory for the vision encoder and llama.cpp internals. The recommended values below leave enough headroom for the OS and the demo app itself.

| Model | LLM params | Recommended quant | LLM file | mmproj (f16) | Total download | Recommended device RAM | | --- | --- | --- | --- | --- | --- | --- | | MiniCPM-V 2.6 | 8B | Q4_K_M | ~4.4 GB | ~1.0 GB | ~5.4 GB | ≥ 8 GB | | MiniCPM-V 4.0 | 4.1B | Q4_K_M | ~2.0 GB | ~0.9 GB | ~2.9 GB | ≥ 6 GB | | MiniCPM-V 4.6 | 1.3B | Q4_K_M | ~0.5 GB | ~1.1 GB | ~1.6 GB | ≥ 6 GB | | MiniCPM-V 4.7 35B-A3B | 35B (3B active) | q3gud-im (Q3_K experts) | ~1.7 GB | ~1.1 GB | ~16.6 GB | ≥ 12 GB | | MiniCPM-V 4.7 1B | 1B | Q4_K_M | ~0.7 GB | ~1.1 GB | ~1.8 GB | ≥ 4 GB | | MiniCPM5-1B | 1B | Q4_K_M | ~0.5 GB | — | ~0.5 GB | ≥ 4 GB | | MiniCPM5-2B | 2B | Q4_K_M | ~1.5 GB | — | ~1.5 GB | ≥ 6 GB | | VoxCPM2 (TTS) | ~2B | Q4_K_M + F16 | ~1.0 GB | — | ~2.8 GB | ≥ 6 GB |

Notes:

• mmproj is the vision projector + ViT weights; it is shipped in f16 because quantising the visual tower hurts perception quality noticeably more than quantising the LLM.
• MiniCPM5-1B and MiniCPM5-2B are text-only models and do not need an mmproj file.
• VoxCPM2 requires two GGUF files (BaseLM + Acoustic) and does not use an mmproj either — it is a pure TTS model.
• MiniCPM-V 4.7 35B-A3B uses a Q3_K expert quant (mix_q3gud_im) rather than Q4, and its routed experts are split out of the main .gguf into a separate experts/ directory. The ~16.6 GB above is the main file + all 40 expert shards + mmproj; plan for ~17 GB of free storage. Because the experts are streamed from SSD on demand rather than held in RAM, this variant also needs fast internal storage — see §2.4.
• All three demos default to a context window of 4K tokens. Larger contexts will increase the KV-cache footprint roughly linearly, so on a borderline device you may need to lower it.
• On Android / HarmonyOS, devices with 8 GB+ RAM are strongly recommended for V 2.6. On iOS, V 2.6 has been validated on iPhone 15 Pro / 16 series and recent iPads with M-series chips; older 6 GB devices may swap heavily.
• MiniCPM-V 4.7 35B-A3B has been validated on a 12 GB iPhone 17 Pro Max. On devices with less RAM the app automatically falls back to a smaller expert cache, which works but is considerably slower.

2.1 MiniCPM-V 2.6 GGUF Files

Download Official GGUF Files

• HuggingFace: https://huggingface.co/openbmb/MiniCPM-V-2_6-gguf
• ModelScope: https://modelscope.cn/models/OpenBMB/MiniCPM-V-2_6-gguf

2.2 MiniCPM-V 4.0 GGUF Files

Download Official GGUF Files

• HuggingFace: https://huggingface.co/openbmb/MiniCPM-V-4-gguf
• ModelScope: https://modelscope.cn/models/OpenBMB/MiniCPM-V-4-gguf

2.3 MiniCPM-V 4.6 GGUF Files

Download Official GGUF Files

• HuggingFace: https://huggingface.co/openbmb/MiniCPM-V-4.6-gguf
• ModelScope: https://modelscope.cn/models/OpenBMB/MiniCPM-V-4.6-gguf

2.4 MiniCPM-V 4.7 GGUF Files

MiniCPM-V 4.7 ships in two sizes:

• MiniCPM-V 4.7 35B-A3B — MoE (35B total / 3B active). Uses MoE-SSD: the routed experts are

• MiniCPM-V 4.7 1B — dense, small enough to keep everything resident.

Where to get the GGUF files

There are two sets, for two different purposes:

• Standard quants — desktop / server. The official repos carry the usual whole-model quant range

* HuggingFace: openbmb/MiniCPM-V-4.7-35B-A3B-gguf · openbmb/MiniCPM-V-4.7-1B-gguf * ModelScope: openbmb/MiniCPM-V-4.7-35B-A3B-gguf · openbmb/MiniCPM-V-4.7-1B-gguf

• On-device split — for this demo, on a phone. The -les main file plus its experts/ directory

* HuggingFace: tc-mb/MiniCPM-V-Apps-gguf * ModelScope: zerg13/MiniCPM-V-Apps-gguf

Why a separate split exists. The official repos do not carry a -les variant, and this demo

cannot load their 35B mix_q3gud_im.gguf: that file keeps the routed experts inline, so a phone would

need the whole ~15.5 GB as a single Metal buffer (mmap) or as resident memory (no-mmap) — either way

the OS kills it. Stripping the experts into experts/ lets them be paged in from SSD on demand, which

is what makes 35B-class inference possible on a phone at all. The split is therefore an artifact made

for this demo, not a general-purpose release. 1B needs no split; it is hosted above only because

that is the revision this demo was validated against.

Files are grouped into per-model sub-directories:

| Variant | Files | On-device location | | --- | --- | --- | | 35B-A3B | MiniCPM-V-4.7-35B-A3B/MiniCPM-V-4_7-35B-A3B-mix_q3gud_im-les.gguf
MiniCPM-V-4.7-35B-A3B/mmproj-model-f16.gguf
MiniCPM-V-4.7-35B-A3B/experts/ (40 shards + manifest.json) | Documents/MiniCPMV47/ | | 1B | MiniCPM-V-4.7-1B/MiniCPM-V-4_7-1B-Q4_K_M.gguf
MiniCPM-V-4.7-1B/mmproj-model-f16.gguf | Documents/MiniCPMV47_1B/ |

The experts/ directory must stay together with its main file. The -les variant carries no

routed-expert weights of its own, so a mismatched pair fails the on-disk consistency check and the

model refuses to load. The reliable way to get this right is to download inside the app

(Settings → tap the model → Download), which fetches every file and verifies the shard sizes

for you.

2.5 MiniCPM5-1B GGUF Files (text-only)

Download Official GGUF Files

• HuggingFace: https://huggingface.co/openbmb/MiniCPM5-1B-GGUF
• ModelScope: https://modelscope.cn/models/OpenBMB/MiniCPM5-1B-GGUF

2.6 MiniCPM5-2B GGUF Files (text-only)

Download Official GGUF Files

• HuggingFace: https://huggingface.co/openbmb/MiniCPM5-2B-GGUF
• ModelScope: https://modelscope.cn/models/OpenBMB/MiniCPM5-2B-GGUF

2.7 VoxCPM2 GGUF Files (text-to-speech)

VoxCPM2 is a ~2B-parameter text-to-speech model. It consists of two GGUF files:

• BaseLM (VoxCPM2-BaseLM-Q4_K_M.gguf, ~956 MB) — the MiniCPM-4 based language model
• Acoustic (VoxCPM2-Acoustic-F16.gguf, ~1.74 GB) — the acoustic decoder (FSQ + LocEnc + LocDiT + AudioVAE)

Download Official GGUF Files

• HuggingFace: https://huggingface.co/tc-mb/MiniCPM-V-Apps-gguf
• ModelScope: https://www.modelscope.cn/models/zerg13/MiniCPM-V-Apps-gguf