This content is for v0.11.0. Switch to the latest version for up-to-date documentation.
tigris inspect
Read what an ONNX model declares, or what a compiled .tgrs plan records,
without compiling or executing anything.
tigris inspect MODEL [OPTIONS]The format is detected from the file contents, so the extension does not matter.
Options
Section titled “Options”| Flag | Type | Required | Description |
|---|---|---|---|
MODEL |
path | yes | ONNX model or .tgrs plan |
-v, --verbose |
flag | no | Add operators, tensors and execution-plan details |
--json |
flag | no | Print complete, versioned metadata as JSON |
ONNX models
Section titled “ONNX models”For an ONNX file, inspect reports the declared graph: IR version, opsets,
inputs and outputs with their declared shapes, the operator mix and the
initializers.
tigris inspect model.onnx╭───────────────────────────── TiGrIS Inspect - g ─────────────────────────────╮│ Format ONNX graph ││ File size 991 B ││ IR version 13 ││ Opsets ai.onnx: 13 ││ Input x float32 [1, 3, 32, 32] ││ Output y float32 [1, 8, 32, 32] ││ Operators 1: Conv ││ Initializers 1 dense, 0 sparse, 0 external dense │╰──────────────────────────────────────────────────────────────────────────────╯Declared graph metadata. Shapes are not inferred; external tensor data is notloaded.Shapes are shown as the model declares them. inspect does not infer shapes and
does not load external tensor data. To check whether a model fits a memory
budget, use tigris analyze.
Compiled plans
Section titled “Compiled plans”For a .tgrs plan, inspect reports the plan schema, the interface in stored
axis order and declared dtype, the operators, the schedule and the memory the
compiler recorded. This is the interface tigris run
expects its inputs in.
tigris inspect downloaded-model/model.tgrs╭─────────────────────────── TiGrIS Inspect - model ───────────────────────────╮│ Format TiGrIS execution plan ││ File size 16.19 KiB ││ Schema 7 ││ Input input float32 (stored: float32 [1, 168], model order) ││ Output output float32 (stored: float32 [1, 24], model order) ││ Operators 1: Gemm ││ Schedule 1 stage, 0 tiled, 0 chains │╰──────────────────────────────────────────────────────────────────────────────╯╭─────────────────────────────────── Memory ───────────────────────────────────╮│ Fast budget 1.00 KiB ││ Recorded graph peak 768 B ││ Weights (including bias) 15.84 KiB ││ Weight storage Read in place (XIP), uncompressed │╰──────────────────────────────────────────────────────────────────────────────╯Memory values are compiler records, not measured runtime peak or total RAM.The memory panel shows what the compiler recorded in the plan. It is not a measured runtime peak and not the total RAM of an application. A plan’s schema number says which plan format it uses; it does not by itself name a runtime version.
JSON output
Section titled “JSON output”--json prints everything inspect reads, with exact byte counts and without
weight values. The top-level inspection_version field versions the document,
so scripts can check it before reading further.
tigris inspect model.tgrs --jsonWhat inspect does not do
Section titled “What inspect does not do”- It does not compile, execute or validate kernel numerics.
- It does not infer shapes or load external ONNX tensor data.
- It needs no native library, so it works on every platform
tigris-mlinstalls on. See Installation.