Known platform issues

Limits that depend on the device, OS version or driver rather than on the feature. For which runtime supports which feature, see the support matrix.

Apple platforms#

• Native LiteRT-LM iOS artifacts omit the Gemma FST constraint provider. Gemma 3/4 and FunctionGemma conversations must disable constrained decoding; Dart already does so. Ordinary generation, thinking and best-effort tool formatting remain available. Automatic tool loops and strict structured output remain unsupported.
• Native LiteRT-LM on iOS: iOS 16.4 is the declared deployment floor, not a tested one. The companion Swift package requires an iOS 16.4 app target, but nothing has been run on iOS 16.4. On a device, model load and generation have run only on an iPhone 16 Pro with iOS 18.3.2 (CPU and GPU); simulator evidence is iOS 26.4, CPU only. No other iOS version is device-verified (#514).
• Flutter macOS LiteRT-LM needs llamadart_litert_lm_flutter to put the runtime in the app bundle. Without it, an unsandboxed app run on the build machine from the project loads the runtime from the hook cache, and a sandboxed (the flutter create default), copied or distributed app throws Failed to load dynamic library 'libCLiteRTLM_mac.dylib' on the first CPU or GPU generation. Add the companion, or the LiteRT-LM build phase.
• macOS x86_64: no LiteRT-LM runtime is published (litert-lm-native#72). Universal apps launch on an Intel Mac and under Rosetta, and LiteRT-LM throws LlamaUnsupportedException there; llama.cpp still runs. See Intel Macs.
• iOS x86_64 simulator: no LiteRT-LM runtime is published. Apps that include LiteRT-LM must exclude the x86_64 simulator architecture; llama.cpp still builds for it. LiteRT-LM throws LlamaUnsupportedException there.
• iOS simulator: llama.cpp on Metal returns incorrect output. On an arm64 simulator (iPhone 17 Pro, iOS 26.5, Apple silicon host) Qwen3.5 0.8B (Q4_K_M) generates garbage text on Metal and answers correctly on the CPU, and Qwen2.5 0.5B fails on Metal the same way (#950); the cause is not known. On an iOS simulator auto therefore runs llama.cpp on the CPU with 0 GPU layers. GpuBackend.metal and ComputeDevice.gpu still load on Metal there, and the load logs a warning: use them only to check a model on the simulator's Metal device. Under ComputeDevice.auto, another explicit GPU backend, such as GpuBackend.vulkan, also loads its GPU layers on Metal and warns, since the Apple runtime has no other GPU backend. The x86_64 (Rosetta) simulator behaves the same. iOS devices and Apple silicon macOS are unchanged: auto uses Metal, and on an iPhone 16 Pro its output matches the CPU's. Intel macOS differs: see the next entry.
• macOS x86_64: upstream llama.cpp Metal returns garbled output on Intel Macs with AMD GPUs (llama.cpp#30702), and upstream's own macOS x64 builds turn Metal off. In an x86_64 process, auto therefore runs llama.cpp on the CPU with 0 GPU layers (#1597). An x86_64 process cannot tell an Intel Mac from Rosetta on Apple silicon, so a Rosetta process also uses the CPU. GpuBackend.metal and ComputeDevice.gpu still load on Metal, and the load logs a warning. Under ComputeDevice.auto, another explicit GPU backend, such as GpuBackend.vulkan, also loads its GPU layers on Metal and warns, since the Apple runtime has no other GPU backend. gpuDevices and tensorSplit need one of them there, since auto would load on the CPU. arm64 macOS is unchanged.

Linux and Windows#

• Linux x64 LiteRT-LM GPU needs a hardware Vulkan driver. On Mesa llvmpipe alone it answers a few prompts, then crashes the process; use cpu on such hosts (#572).
• Windows x64 LiteRT-LM GPU needs dxil.dll and dxcompiler.dll, which the pinned runtime bundles. Its GPU accelerator and sampler load webgpu_dawn.dll, which the hook bundles. Qwen3 0.6B ran on an NVIDIA L4 (Direct3D 12) through dart run and a dart build cli bundle; a Flutter Windows app has not been run with it.

Android#

• Android LiteRT-LM on Adreno 750 (Galaxy S24): ComputeDevice.auto, the default, runs LiteRT-LM on the GPU, which cannot run Qwen3 0.6B. The adapter's 128 MiB storage-buffer binding limit is smaller than one of the model's weight buffers, so expect the same from any model and adapter with that mismatch. GPU engine creation fails: on a Galaxy S24 the first use throws LlamaUnsupportedException when ComputeDevice.gpu was requested; auto retries on the CPU with a warning (#1034). Gemma 4 E2B runs on that GPU. ModelParams(device: ComputeDevice.cpu) skips the GPU attempt and answers correctly on the same device (upstream LiteRT-LM#3866).

• Android NPU depends on the device SoC, the .litertlm bundle and the LiteRT dispatch libraries (LiteRtLmModelParams.dispatchLibDir). If native LiteRT-LM cannot create an NPU engine, use cpu or gpu.

• Android llama.cpp Vulkan is experimental and device-dependent. auto runs llama.cpp on the CPU there; Vulkan runs only when the app asks for ComputeDevice.gpu or GpuBackend.vulkan. Previously observed runtime failures: on a Pixel 9 Pro (Mali-G715), a prompt longer than about 32 tokens evaluated in one batch returns wrong text, and a request with a grammar or tool call can then abort the process, confirmed so far with one model (Qwen3.5-0.8B Q4_0) on that one device (#948); a Galaxy A35 (Mali-G68) is reported to time out in llama_decode (#1092); a Galaxy S24 (Adreno 750) crashes on the first generation (llamadart-native#79). The first of these is a llama.cpp defect at a Vulkan subgroup size of 16 (llama.cpp#28637), last observed on llama.cpp v0.6.0 and not re-checked on v0.7.0. The affected code also mis-tiles subgroup sizes below 8, which no device run has confirmed. On Android, llamadart reads the subgroup size of the Vulkan GPUs when it creates a context: on such a GPU, and on any GPU whose subgroup size it cannot read, the context decodes a text prompt in micro-batches of at most 8 tokens by default (n_ubatch itself is unchanged); on other GPUs it decodes the prompt in full micro-batches. On a Pixel 9 Pro, prompt evaluation with 8-token micro-batches ran at about the speed uncapped decoding did (about 34 tokens/s against about 30 tokens/s for a 59-token prompt); other GPUs are not measured. An app that has validated Vulkan on its target devices can set ModelParams.microBatchSize to choose the micro-batch size itself: an explicit value is used as given, and one above 32 brings the wrong text back on affected GPUs. The cap applies to text-prompt decoding only: it does not cover prompts with image or audio input, embeddings, decision models, text-to-speech, or speculative-decoding verification batches during generation. Those batches can exceed 32 tokens, so speculative decoding can still produce wrong output on an affected GPU. llama.cpp's Vulkan backend also needs Vulkan 1.2 from the Vulkan loader and the GPU driver. When the Vulkan GPU a load would run on reports an older version, ComputeDevice.gpu throws LlamaUnsupportedException naming the GPU and both versions, and ComputeDevice.auto loads the model on the CPU and logs a warning. A device with a Vulkan 1.1 driver takes that path. When other GPUs of the load are usable, such as a discrete GPU beside an old one, the load leaves the old one out and logs a warning; a GPU llama.cpp would not select, like an integrated GPU beside a discrete one, changes nothing. On Android, the exact Qualcomm vendor 0x5143, Adreno (TM) 750 driver 2150604839 has shader-compiler crashes and incorrect quantized results, including common Q4_K_M models. When this device is registered, GPU loads are refused even if another GPU was selected: projector and decision-head GPU selection can independently choose the affected device. ComputeDevice.auto loads on CPU with a warning, including when preferredBackend requests Vulkan; ComputeDevice.gpu throws LlamaUnsupportedException. CPU loads do not probe Vulkan. Other driver versions are unchanged. A shader crash workaround alone does not qualify this driver; re-enabling a model requires qualification of the final native artifact on that exact driver. See native issue #79.

  Keep auto or cpu on Android unless the app has validated Vulkan on its target devices.

Vulkan and the web#

• Some Vulkan drivers crash in the cooperative-matrix path; see GPU crash or device loss.
• WebGPU readiness depends on the browser, device, bridge assets and model size; see WebGPU bridge.