Quark Onnx Subgraph Partitioner

What it does

When your agent uses it

• The user wants to understand a models high-level structure
• Document its architectural blocks
• Prepare a partition for downstream workflows (mixed-precision quantization
• Layer-wise profiling

Example prompts

• “partition my ONNX model”
• “generate a subgraph JSON”
• “which nodes belong to the attention blocks”
• “/quark-onnx-subgraph-partitioner”

Requirements

• Python 3

Workflow steps

7 steps, taken from the step headings in SKILL.md.

1. 1Load and inspect the model
2. 2Identify structural landmarks
3. 3Assign blocks
4. 4Choose start and end nodes
5. 5Validate the partition
6. 6Emit the JSON file
7. 7Summary table

What it can do on your machine

Read from SKILL.md and the folder at commit 313cb0b. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Ships script files (Python), which the agent can run.

  Shell commands in SKILL.md call:

  • python3
  • pip

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  No URLs in SKILL.md. Its commands use pip, which can reach the network depending on how they are called.

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names no API keys, tokens, secrets or passwords.

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Quark Onnx Subgraph Partitioner loads about 2.7k tokens when it runs. Until then it costs about 244 tokens; SKILL.md has 1,075 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~244

When it runs · the whole SKILL.md, loaded when a task matches

~2.7k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from amd/Quark at commit 313cb0b, republished under its MIT licence (© amd). 1,075 words, ~2,672 tokens.

name

quark-onnx-subgraph-partitioner

description

Partition an ONNX model graph into named functional subgraphs and emit a subgraph_partition.json file. Use when the user wants to understand a model's high-level structure, document its architectural blocks, or prepare a partition for downstream workflows (mixed-precision quantization, layer-wise profiling, partial deployment, graph visualization). Triggers on "partition my ONNX model", "generate a subgraph JSON", "which nodes belong to the attention blocks", "show me the model architecture", "split the model into blocks", "group layers for AMP", or any request that needs a named block decomposition of an ONNX graph. Handles any ONNX architecture: CNNs (ResNet, EfficientNet, YOLOv5/v8, DenseNet, MobileNet), Transformers (BERT, ViT, Swin, LLMs), multi-modal (CLIP, BLIP), detection models with FPN/PAN necks, BEV perception models (BEVFormer), and custom or unfamiliar architectures identified from node-name prefixes and op sequences.

layer

l1-atomic

primary_artifact

subgraph_partition.json

source_knowledge

quark/onnx/algorithm/mprecision/subgraph_parser.py, quark/onnx/algorithm/mprecision/mprecision_config.py…

SKILL_DIR=skills/_legacy_impl/l1-atomic/onnx/quark-onnx-subgraph-partitioner

{
  "quantized":     false,
  "num_subgraphs": 4,
  "subgraphs": [
    {
      "name": "backbone_stem",
      "description": "Initial stride-2 Conv → SiLU activation, produces P1 features",
      "start_nodes": ["/model.0/conv/Conv"],
      "end_nodes":   ["/model.0/act/Mul"]
    },
    {
      "name": "attention_layer_0",
      "description": "Full encoder layer 0: self-attention + FFN + LayerNorm + residuals",
      "start_nodes": ["/encoder/layer.0/attention/self/query/MatMul"],
      "end_nodes":   ["/encoder/layer.0/output/LayerNorm"]
    }
  ]
}

python3 "$SKILL_DIR/inspect_model.py" <MODEL_PATH>

grep -n 'op_type:\|  name:\|input:\|output:' <MODEL_PATH> | head -400

python3 "$SKILL_DIR/validate_partition.py" <MODEL_PATH> <DRAFT_JSON>

© amd, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 2 other files in skills/_legacy_impl/l1-atomic/onnx/quark-onnx-subgraph-partitioner of amd/Quark.

• SKILL.md
• inspect_model.py
• validate_partition.py

Open the folder on GitHubat commit 313cb0b