Debug Model
Debug a MAX model that loads, compiles, serves, and generates tokens but whose output disagrees with a reference implementation. Use whenever parity debugging stalls on scalar taps, the model returns gibberish or wrong greedy tokens, logit cosine is high but argmax differs, or generation is coherent then diverges. Applies during an architecture port, a quantization bring-up, a multi-GPU conversion, or after a MAX upgrade. Triggers on "parity failure", "silent corruption", "logits match but tokens diverge", "top-1 mismatch", "greedy divergence", and "model serves but generates garbage". Not for
Install
/plugin marketplace add modular/skillsThe repo is a Claude Code plugin marketplace. Run /plugin afterwards to pick the plugin that includes this skill.
npx skills add modular/skills --skill debug-modelgit clone --depth 1 https://github.com/modular/skills.git /tmp/skills
mkdir -p ~/.claude/skills
cp -r /tmp/skills/debug-model ~/.claude/skills/debug-modelPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Parity/coherence failure protocol The model runs without errors but output is wrong. Scalar `F.print` taps hide directional bugs, and each new tap costs a full recompile. Build a per-layer tensor-dump comparator first. Every later check becomes a numpy read from disk. **Use this skill when** MAX output disagrees with a PyTorch reference you can run and hook. The primary case is a custom-architecture port that serves but fails parity or coherence checks. The same protocol covers a quantized variant of a working port, a multi-GPU conversion of a working single-GPU port, and a regression after a MAX upgrade. It applies anywhere a trusted reference exists. **Don't use this skill when:** - The server crashes on load → fix config, weights, module (`import-model`) - You haven't finished implementing the module → `import-model` Phase 2 - An already-verified model needs logit-comparison tolerances tuned → that's threshold calibration, which this skill doesn't cover ## References | File | Read when | |---------------------------------------------------------------------------|-----------------------------------------------------------------| | [comparator-build.md](references/comparator-build.md) | Building HF/MAX dumpers and the comparator |
Details
- Skill name
debug-model- Repository
- modular/skills
- Path
debug-model/SKILL.md- Repo updated
- 18 November 2026
FAQ
What does the Debug Model skill do?
Debug a MAX model that loads, compiles, serves, and generates tokens but whose output disagrees with a reference implementation. Use whenever parity debugging stalls on scalar taps, the model returns gibberish or wrong greedy tokens, logit cosine is high but argmax differs, or generation is coherent then diverges. Applies during an architecture port, a quantization bring-up, a multi-GPU conversion, or after a MAX upgrade. Triggers on "parity failure", "silent corruption", "logits match but tokens diverge", "top-1 mismatch", "greedy divergence", and "model serves but generates garbage". Not for
How do I install the Debug Model skill?
Claude Code plugin: /plugin marketplace add modular/skills. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Debug Model popular?
It lives in modular/skills, which has 213 GitHub stars and 22 forks.
Related skills
Benchmark Model
213modular/skills
Benchmark a model served on MAX with the `max benchmark` command: measure throughput (tokens/sec), latency (TTFT, TPOT, inter-token latency), and GPU utilization by driving load against a running `max serve` endpoint. Use this whenever the user wants to benchmark, load-test, or measure the performance of a MAX model, get tokens-per-second / TTFT / TPOT numbers, run a concurrency or request-rate sweep, compare latency vs throughput, size a deployment, or produce benchmark JSON, even if they don't say "benchmark" by name. Also use when a `max benchmark` run fails to connect or reports zero/garba
Eval Model
213modular/skills
Measures the task accuracy of text models served by MAX using standard benchmarks such as GSM8K, MMLU, HellaSwag, ARC, AIME, GPQA, TruthfulQA, WinoGrande, and BABILong. Use when benchmarking a served model, comparing it with model-card or reference scores, verifying that a new MAX model produces correct answers, or running repeatable dataset evaluations against a MAX OpenAI-compatible endpoint.
Import Model
213modular/skills
Use when importing a new model architecture into MAX from a Hugging Face model ID. Triggers on: "import a model into MAX", "add model to MAX", "bring up <HF model> in MAX". Workflow: inspect the Hugging Face config and modeling code, then scaffold from a similar ModuleV3 architecture. Implement each layer as a max.nn Module that matches HF, serve, then verify against the Hugging Face reference. When the server runs but output is wrong (gibberish, greedy mismatch, coherent-then-diverges), load debug-model for the divergence hunt.
Drawio MCP Diagramming
206thomast1906/github-copilot-agent-skills
Create and edit diagrams using the Draw.io MCP server — any shape, any vendor. USE FOR: draw me a diagram, create an architecture diagram, add Azure/AWS/GCP/Cisco/Kubernetes icons to a diagram, convert Mermaid to draw.io, fix overlapping arrows, edit a .drawio file, network topology diagrams, CI/CD pipeline diagrams, auth flow diagrams. Supports XML, Mermaid, and CSV. Uses drawio/search_shapes to find any of 10,000+ shapes across all vendor and icon libraries. DO NOT USE FOR: Excalidraw output (use excalidraw-mcp-diagramming skill).
Waxa Eval
423mizchi/skills
Use when iterating on a skill's prompt with the waxa CLI (https://github.com/mizchi/skills/tree/main/tools/waxa) — authoring scenarios, choosing graders, interpreting unclear-points, advancing a ledger, and judging convergence. Encodes the four-stage iteration pattern observed in real iter loops (structural fix → grader breadth → surface-form coverage → residual unclear) and the scenario-design pitfalls (blank-slate executor's network limit, prompt expectation explicitness, regex coverage of Japanese/English surface forms). Meta-skill: do NOT auto-invoke for routine skill edits; fires only whe
Excalidraw MCP Diagramming
206thomast1906/github-copilot-agent-skills
Create and edit diagrams on a live Excalidraw canvas using the Excalidraw MCP server. Use when asked to draw, diagram, sketch, or visualise architectures, workflows, data flows, system designs, flowcharts, mind maps, or sequence diagrams. Trigger phrases include "create an excalidraw", "draw me a diagram", "make a flowchart", "visualise the system", "diagram this architecture", "export to PNG/SVG". Can export to PNG, SVG, .excalidraw file, or a shareable URL. Do NOT use for Draw.io or diagrams.net output (use drawio-mcp-diagramming instead).
Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.