Sync Upstream
nyaruka/phonenumbers
Sync this Go port with a new upstream google/libphonenumber release — regenerate the embedded metadata and reconcile the ported Java logic.
A skill your agent uses when validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues, memory/concurrency mismatches, per-op…
Install Aic Silicon Align
$ npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install ai-dynamo/aiconfigurator aic-silicon-align --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/ai-dynamo/aiconfigurator.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/aic-silicon-align .claude/skills/aic-silicon-align && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "aic-silicon-align" agent skill from https://github.com/ai-dynamo/aiconfigurator/tree/main/.claude/skills/aic-silicon-align into .claude/skills/aic-silicon-align/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "aic-silicon-align", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/ai-dynamo/aiconfigurator/tree/main/.claude/skills/aic-silicon-alignType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install ai-dynamo/aiconfigurator aic-silicon-align --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/ai-dynamo/aiconfigurator.git skills-src && mkdir -p .agents/skills && cp -r skills-src/.claude/skills/aic-silicon-align .agents/skills/aic-silicon-align && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "aic-silicon-align" agent skill from https://github.com/ai-dynamo/aiconfigurator/tree/main/.claude/skills/aic-silicon-align into .agents/skills/aic-silicon-align/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "aic-silicon-align", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install ai-dynamo/aiconfigurator aic-silicon-align --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/ai-dynamo/aiconfigurator.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/.claude/skills/aic-silicon-align .cursor/skills/aic-silicon-align && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "aic-silicon-align" agent skill from https://github.com/ai-dynamo/aiconfigurator/tree/main/.claude/skills/aic-silicon-align into .cursor/skills/aic-silicon-align/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "aic-silicon-align", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/ai-dynamo/aiconfigurator.git --path .claude/skills/aic-silicon-align--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install ai-dynamo/aiconfigurator aic-silicon-align --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/ai-dynamo/aiconfigurator.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/.claude/skills/aic-silicon-align .gemini/skills/aic-silicon-align && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "aic-silicon-align" agent skill from https://github.com/ai-dynamo/aiconfigurator/tree/main/.claude/skills/aic-silicon-align into .gemini/skills/aic-silicon-align/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "aic-silicon-align", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install ai-dynamo/aiconfigurator aic-silicon-alignInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/ai-dynamo/aiconfigurator.git skills-src && mkdir -p .github/skills && cp -r skills-src/.claude/skills/aic-silicon-align .github/skills/aic-silicon-align && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "aic-silicon-align" agent skill from https://github.com/ai-dynamo/aiconfigurator/tree/main/.claude/skills/aic-silicon-align into .github/skills/aic-silicon-align/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "aic-silicon-align", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install ai-dynamo/aiconfigurator aic-silicon-align --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/ai-dynamo/aiconfigurator.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/.claude/skills/aic-silicon-align .opencode/skills/aic-silicon-align && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "aic-silicon-align" agent skill from https://github.com/ai-dynamo/aiconfigurator/tree/main/.claude/skills/aic-silicon-align into .opencode/skills/aic-silicon-align/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "aic-silicon-align", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
Facts
- Skill name
aic-silicon-align- GitHub stars
- 457
- Token cost
- ~1.2k tokens
- SKILL.md length
- 620 words
- Files
- 1
- Skills in repo
- 5
- Repo updated
- First seen
- Licence
- Apache-2.0
At a glance
A skill your agent uses when validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues, memory/concurrency mismatches, per-op…
- Works in 5 steps: Reproduce the prediction (CPU-only):… → Consistency arithmetic before GPUs:… → Engine memory ledger (dummy weights… → …
- Validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues
- SKILL.md covers The ladder, What dummy weights do and…, Measurement parity checklist… and Perf-DB cross-checks, plus 2 more sections
- Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md
What it does
Aic Silicon Align is an agent skill from ai-dynamo/aiconfigurator. Use when validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues, memory/concurrency mismatches, per-op latency divergence). Covers dummy-weight methodology, measurement parity traps, component-isolation ladders, and ledger reconciliation.
Its SKILL.md is about 1.2k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.
It sits in Business, Finance & HR, covering Accounting and bookkeeping. The repository describes itself as: Offline optimization of your disaggregated Dynamo graph. The licence is Apache-2.0.
When your agent uses it
- Validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues
- Memory/concurrency mismatches
- Per-op latency divergence)
Example prompts
- “/aic-silicon-align”
Workflow steps
5 steps, taken from the first numbered list in SKILL.md.
- Reproduce the prediction (CPU-only): same yaml/version/flags. If the
- Consistency arithmetic before GPUs: `throughput = concurrency x
- Engine memory ledger (dummy weights suffice): serve with
- Standalone per-role benchmarks (prefill worker, decode worker) at
- Deployment-faithful e2e: exact images, generated scripts, patches,
What it can do on your machine
Read from SKILL.md and the folder at commit f254959. 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
No scripts in the folder and no shell commands in SKILL.md.
From the folder's file list and the shell code blocks in SKILL.md.
Network
No URLs in SKILL.md.
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
Aic Silicon Align loads about 1.2k tokens when it runs. Until then it costs about 82 tokens; SKILL.md has 620 words of instructions outside code blocks.
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
Auto-check passed
The automated check found no risky patterns in SKILL.md.
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 ai-dynamo/aiconfigurator at commit f254959, republished under its Apache-2.0 licence (© ai-dynamo). 620 words, ~1,249 tokens.
Download SKILL.mdSave it as
.claude/skills/aic-silicon-align/SKILL.md (or your agent's skills folder).- name
- aic-silicon-align
- description
- Use when validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues, memory/concurrency mismatches, per-op latency divergence). Covers dummy-weight methodology, measurement parity traps, component-isolation ladders, and ledger reconciliation.
AIC Silicon Alignment
Root-cause a prediction-vs-measurement gap by isolating layers and comparing adjacent ones only. Never attribute a residual across more than one rung.
The ladder
- Reproduce the prediction (CPU-only): same yaml/version/flags. If the reported numbers don't reproduce, stop — it's a config diff, not fidelity.
- Consistency arithmetic before GPUs:
throughput = concurrency x per-user speed. Derive the effective concurrency from the report; a gap vs configured concurrency localizes the problem to admission/memory, not kernel speed. - Engine memory ledger (dummy weights suffice): serve with
--load-format dummyand read the engine's own lines — weights GiB, KV tokens, max-concurrency. Compare against AIC's per-component memory dict, component by component, never totals only (errors cancel: one model showed -8/+2.3/-1.4 GiB netting to a small total delta). - Standalone per-role benchmarks (prefill worker, decode worker) at KV-feasible batch sizes; compare against AIC's per-role rows.
- Deployment-faithful e2e: exact images, generated scripts, patches, and flags the reporter used. Only this rung may be compared to the reported end-to-end numbers.
What dummy weights do and don't preserve
- Usually preserve: memory footprint and data-independent kernel timing — but this is loader-specific, NOT a guarantee. Before relying on it, verify against the real checkpoint: parameter dtypes/shapes/layouts and quant scales after loading, post-load weight transforms, and that the framework dispatches the same kernels (some dummy loaders allocate a different dtype than the checkpoint ships, or skip scale tensors that change dispatch). If any of these differ, validate the affected measurement with real weights before using it.
- Do NOT preserve: anything routed by data — e.g. MoE expert distributions collapse (near-identical hidden states -> few unique experts -> weight reads shrink several-fold). Mark such measurements as biased and state the direction. To still use them: sweep the hidden axis synthetically in the collector (controlled router logits, controlled unique-expert count) to build a calibration curve, then invert the engine measurement onto it.
Show full SKILL.md (313 more words)Show less
Measurement parity checklist (each one has flipped a conclusion)
- CUDA-graph capture sizes must match the deployment — coverage swung a small-batch decode step 2.4x in one case. AIC-generated deploys pass an explicit list; your benchmark must too.
- Standalone decode ITL is polluted by chunked-prefill mixing when
max-num-batched-tokensis small; a disagg decode worker never prefills. Prefer deployment-faithful setups or report medians with the caveat. - torch-profiler distorts wall time and large-kernel durations (a single stream showing >100% busy is the tell). Anchor on unprofiled ITL; trust only small-kernel durations from traces.
- EP/TP ranks run in lockstep: stragglers' wait is absorbed into other ranks' NCCL kernel durations. Account per-step across all ranks, never one rank's kernels in isolation.
- Discard first-run numbers (JIT/autotune warmup skews percentiles).
- Ops that model compute+comm jointly (overlap ops) must be compared against the same joint quantity on the engine side.
Perf-DB cross-checks
- Re-collect the suspect datapoint on the same silicon: separates methodology error from machine/version skew.
- Kernel-diff the collector's timed region against an engine trace: same kernel families and algos? Extra eager glue ops are a harness bug (see the fused-ops fix); missing ops are an accounting gap.
- Collector data is only as true as its op dispatch: verify against
framework source (file:line@version), per
.claude/rules/collector/.
Model/checkpoint facts are per-checkpoint, not per-family
Read quantization ignore/exclude_modules from the checkpoint (e.g. NVFP4
releases keep attention BF16; native FP8 ones quantize it). Framework
defaults (memory fractions, capture sizes) come from framework source with
citations, never from memory.
Discipline
- Keep a running ledger (prediction | measured | delta | attribution) and update it every experiment; record retracted claims with the reason.
- Fix coupled accounting items together, or verify that a partial fix does not regress feasibility (a correct-but-lone weight increase can push a model into "infeasible" because other components over-count).
- A fidelity fix lands with: the measured ledger it was validated against, framework citations, and the residuals it deliberately leaves.
© ai-dynamo, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Files
Just SKILL.md in .claude/skills/aic-silicon-align of ai-dynamo/aiconfigurator.
Open the folder on GitHubat commit f254959
Compare with similar skills
Aic Silicon Align next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Aic Silicon Align this skillai-dynamo/aiconfigurator | 457 | — | ~1.2k | Automated safety check: Pass | Apache-2.0 | |
| Sync Upstreamnyaruka/phonenumbers | 1.6k | — | ~2.8k | Automated safety check: Pass | MIT | |
| Radiology Tablehuang-sir1/radiology-skills | 1.9k | — | ~1.3k | Automated safety check: Pass | Custom licence | |
| ERPClaw ERP Controlleravansaber/erpclaw | 114 | — | ~18k | Automated safety check: Pass | GPL-3.0 | |
| Odoo Agency Fleet Reviewerpipe-org/mcp-odoo | 421 | — | ~699 | Automated safety check: Pass | MIT | |
| Beancount Closebex-co/beancount-io | 296 | — | ~1.4k | Automated safety check: Pass | MIT |
Similar skills
Radiology Table
huang-sir1/radiology-skills
Create/audit editable publication tables with source reconciliation; not figures or statistical inference.
ERPClaw ERP Controller
avansaber/erpclaw
Operates the ERPClaw self-hosted ERP in plain language: accounting, invoicing, inventory, purchasing, tax, HR, payroll and reports, treating the ERP as the single source of truth.
Odoo Agency Fleet Review
erpipe-org/mcp-odoo
Review many client Odoo databases at once through odoo-mcp's cross-instance tools — fleet-wide accounting health, per-client aging, partial-failure triage — for agencies and partners managing 5–50…
Beancount Close
bex-co/beancount-io
Close an accounting period in a Beancount ledger by reconciling each active account through beancount-reconcile, checking assertions and recurring gaps, reviewing flags, then proposing a commit with…
Forward Implementation First
Vuk97/forward-implementation-first
Keeps an agent building and validating real output instead of servicing its own bookkeeping.
More from ai-dynamo/aiconfigurator
Adapt Server Config
ai-dynamo/aiconfigurator
Convert InferenceX DB config/benchmark pairs, Dynamo recipe YAML, or confirmed custom serving configs into validated aiconfigurator estimate requests.
Aic Collector Op Development
ai-dynamo/aiconfigurator
Design, add, review, or modify AIC Collector operations and their case population.
Aic Auto Collect
ai-dynamo/aiconfigurator
A skill your agent uses when running long AIC/aiconfigurator GPU perf auto-collection for a specific GPU/framework/frameworkversion, including draft-PR checkpoints, resumable collectxx.py runs…
Aic Version Slots
ai-dynamo/aiconfigurator
A skill your agent uses when adding perf data for a new framework version, deciding which backendversion to query or pin in tests, bumping the maintained (current/previous) versions, retiring old…
Related
Categories
Questions about Aic Silicon Align
What does Aic Silicon Align do?
A skill your agent uses when validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues, memory/concurrency mismatches, per-op…. Aic Silicon Align is an agent skill from ai-dynamo/aiconfigurator. Use when validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues, memory/concurrency mismatches, per-op latency divergence).
When should I use Aic Silicon Align?
Aic Silicon Align fits situations like: validating AIC predictions against real GPU serving measurements and root-causing fidelity gaps (prediction-vs-measured issues; memory/concurrency mismatches; per-op latency divergence).
How do I install Aic Silicon Align in Claude Code?
Run `npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a claude-code`. Or copy the skill folder (.claude/skills/aic-silicon-align in ai-dynamo/aiconfigurator) into .claude/skills/aic-silicon-align in your project. Claude Code loads it when a task matches its description.
How do I install Aic Silicon Align in Codex?
Run `npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a codex`. Or copy the skill folder (.claude/skills/aic-silicon-align in ai-dynamo/aiconfigurator) into .agents/skills/aic-silicon-align in your project. Codex loads it when a task matches its description.
Can I use Aic Silicon Align in Cursor, Gemini CLI or GitHub Copilot?
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add ai-dynamo/aiconfigurator --skill aic-silicon-align -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/aic-silicon-align, .gemini/skills/aic-silicon-align, .github/skills/aic-silicon-align and .opencode/skills/aic-silicon-align in your project.
What does Aic Silicon Align need to run?
SKILL.md names no scripts, command-line tools or credentials: Aic Silicon Align is instructions for the agent only.
Does Aic Silicon Align access the network?
SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.
Is Aic Silicon Align safe to install?
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
What licence does Aic Silicon Align use?
Aic Silicon Align is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
How many tokens does Aic Silicon Align use?
About 1.2k tokens (SKILL.md is roughly 5k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.
What are the alternatives to Aic Silicon Align?
Skills that share tags, products or a category with Aic Silicon Align: Sync Upstream (nyaruka/phonenumbers, 1.6k stars), Radiology Table (huang-sir1/radiology-skills, 1.9k stars), ERPClaw ERP Controller (avansaber/erpclaw, 114 stars) and Odoo Agency Fleet Review (erpipe-org/mcp-odoo, 421 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
Who maintains Aic Silicon Align?
ai-dynamo (a GitHub organization) maintains it in ai-dynamo/aiconfigurator, which has 457 GitHub stars. The repository holds 5 skills in this directory. The repository was last updated on September 18, 2026.
Source: ai-dynamo/aiconfigurator on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.