Mirai Labs
Official@trymirai · United States of America
On-device AI infrastructure. Inference, models, optimizations.
Agent Skills by Mirai Labs
Showing 8 vetted skills indexed across 1 GitHub repositories.
trymiraivalidate-inference
Validates inference outputs by comparing them with reference results via Rust cargo test harness.
trymirailist-downloaded-models
List downloaded models by inspecting the ./models/{ENGINE_VERSION} directory structure.
trymiraiarchitecture
Document the core architecture and component interactions of the uzu inference engine.
trymiraidownload-model
Download test models by REPO_ID into ./models/{ENGINE_VERSION}/{MODEL_NAME}.
trymiraitrace-model
Validate per-layer AI model inference outputs against traces.safetensors.
trymiraimetal-gpu-debug
Automate Metal GPU trace collection, parsing, and analysis on macOS.
trymirailist-models
List models supported by the inference engine via the Python helper.
trymiraibenchmark-model
Benchmark AI model inference latency, throughput, and resource usage via CLI.
Frequently Asked Questions About Mirai Labs
FAQPage SchemaWhat specific tasks does Mirai Labs enable for model engineers?▼
Engineers can validate inference outputs against reference results, benchmark latency and throughput, and perform per-layer trace analysis. The infrastructure supports managing local model directories and debugging GPU-specific execution paths on macOS hardware.
Which hardware and software environments are required for these operations?▼
These operations require macOS environments capable of running Metal GPU workloads. The infrastructure utilizes a Rust-based test harness for validation and relies on specific directory structures for model storage and versioning.
How are model inference outputs verified for accuracy?▼
Accuracy is verified by comparing inference outputs against reference results using a Rust-based test harness. Additionally, per-layer outputs are validated against traces.safetensors files to ensure numerical parity during the execution process.