tinymaix-integration
/tinymaix-integration
Use when integrating, porting, configuring, or debugging TinyMaix MCU inference, model loading, tensor shapes, memory, quantization, or TinyML runtime issues
- Author
- easyzoom ↗
- Repository
- aix-skills ↗
- Licence
- MIT
- Last repository change
- 17.07.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 17.07.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Use when integrating, porting, configuring, or debugging TinyMaix MCU inference, model loading, tensor shapes, memory, quantization, or TinyML runtime issues
Original author description
Use when integrating, porting, configuring, or debugging TinyMaix MCU inference, model loading, tensor shapes, memory, quantization, or TinyML runtime issues
How it works
- Overview
- When To Use
- First Questions
- Integration Checklist
- Common Failures
- Verification
- Example
SKILL.md preview
--- name: tinymaix-integration description: Use when integrating, porting, configuring, or debugging TinyMaix MCU inference, model loading, tensor shapes, memory, quantization, or TinyML runtime issues --- # TinyMaix Integration ## Overview Use this skill to integrate TinyMaix by proving the model, tensor shapes, preprocessing, memory buffers, and target backend before optimizing inference speed. TinyML failures usually come from mismatched input format, quantization, or insufficient memory. ## When To Use Use this skill when: - The user wants to run TinyMaix on an MCU. - The issue involves model conversion, `tm_load`, `tm_preprocess`, `tm_run`, tensor dimensions, quantized data, RAM/flash limits, or wrong inference results. - The target has strict RAM, flash, CPU, or accelerator constraints. Do not use this skill for full ML framework training or model design beyond embedded deployment checks. ## First Questions Ask for: - Target MCU/core, RAM/flash, compiler, and whether SIMD/FPU/DSP extensions exist. - TinyMaix version/source and model format. - Input shape, data type, quantization, preprocessing, and expected output. - Memory allocation strategy and inference buffer sizes. - Current symptom: compile error, load error, run error, wrong output, or too slow. ## Integration Checklist 1. Confirm model compatibility. Verify operator set, quantization type, input/output shapes, and converted model files. 1. Prove preprocessing. Normalize, resize, color order, layout, and quantization must match training/export assumptions. 1. Budget memory. Account for model, activations, input/output tensors, stack, and any temporary buffers. 1. Bring up one known sample. Run a golden input with expected output before using live sensor data. 1. Optimize only after correctness. CPU extensions, fixed-point paths, and memory placement come after correct inference. ## Common Failures - Wrong NHWC/NCHW layout or color order. - Input values are float-scaled but model expects int8/uint8 quantized data. - Activation buffers overflow RAM. - Model uses unsupported operations. - Output interpretation ignores quantization scale/zero point. - Benchmark uses live noisy data before golden-vector validation. ## Verification Before claiming TinyMaix works: - State model, input/output shapes, data type, quantization, and memory budget. - Confirm a golden input produces expected class/value within tolerance. - Confirm stack/heap/activation usage fits the target. - Report inference time if performance is part of the task. ## Example User: ```text TinyMaix 跑起来了但识别结果全错。 ``` Agent: 1. Asks for model format, input shape, preprocessing, quantization, and golden sample. 1. Checks layout, scale/zero-point, and output decoding. 1. Verifies with a known input before optimizing speed.
Versions
The database has no other skills named tinymaix-integration.
Database snapshot: 13.11.2026 · package version 678f25b2ac24 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100flashdb-integration
Use when integrating, porting, configuring, or debugging FlashDB KVDB or TSDB on MCU flash, filesystems, RTOS, or bare-metal projects - 100tjpgd-integration
Use when integrating, porting, or debugging TJpgDec, Tiny JPEG Decompressor, embedded JPEG decoding, MCU image assets, display callbacks, or memory-constrained rendering - 90edge-inference
Edge / on-device AI inference architecture: running models on MCUs, NPUs, mobile, and mini-PCs; quantization for edge, TOPS/memory/energy budgets, TinyML, and the latency case for on-device vs cloud. Architect-level. USE WHEN: designing on-device/edge AI, "edge inference",… - 100edge-ai
Use this skill when the user says 'edge AI', 'on-device inference', 'TensorRT', 'ONNX Runtime', 'TFLite', 'LiteRT', 'TinyML', 'quantization int8', 'Jetson', 'Coral', 'NPU', 'ExecuTorch', 'model on microcontroller'. Deploys trained models to edge hardware: export path (PyTorch/TF… - 96cuda-streams
Manage tensor lifetimes across CUDA streams and decide whether record_stream is avoidable. Use when using a tensor on a stream other than the one it was allocated on, adding or reviewing Tensor.record_stream / CUDACachingAllocator::recordStream, side streams, wait_stream, or…