Borrowing it
Nothing to install: this file belongs to Tele-AI/TeleFuser. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/Tele-AI/TeleFuser/main/.claude/skills/profile-pipeline/SKILL.mdgit clone --depth 1 https://github.com/Tele-AI/TeleFuserWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/tele-ai/telefuser/profile-pipeline)<a href="https://agentmods.dev/skills/tele-ai/telefuser/profile-pipeline"><img src="https://agentmods.dev/badge/skills/tele-ai/telefuser/profile-pipeline/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/tele-ai/telefuser/profile-pipeline"><img src="https://agentmods.dev/badge/skills/tele-ai/telefuser/profile-pipeline.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00049 | $0.00537 |
| Opus 5 | $0.00024 | $0.00269 |
| Sonnet 5 | $0.00010 | $0.00107 |
| Haiku 4.5 | $0.00005 | $0.00054 |
Grade A, and why
profile-pipeline scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 37 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Profile a Pipeline
Use the current profiler implementation and docs/en/profiler.md as the API and command source of truth. Do not reproduce profiler helpers or create new profiling environment variables.
Capture a reproducible workload
Record the commit, model/checkpoint, task, input shape, frame count, diffusion steps, dtype, attention backend, devices, warmup, synchronization points, and whether compilation or caches are warm. Define the metric that matters: latency distribution, throughput, output cadence, or peak memory.
Check GPU availability and memory before profiling. Avoid concurrent workloads that invalidate the trace.
Progress from broad to narrow
- Stage timing: use the existing profiler flags and pipeline metrics documented in
docs/en/profiler.md. Identify the stage dominating the target metric. - Isolated stage analysis: use the repository's
StageBenchHarnessand captured I/O signature when the stage can be reproduced faithfully. Verify that the harness input shapes and runtime state match the full pipeline. - Kernel analysis: inspect the PyTorch/Chrome trace and kernel breakdown for the isolated bottleneck. Account for synchronization, communication, memory copies, and launch overhead rather than ranking kernels by duration alone.
- NCU deep dive: use Nsight Compute only when a specific reproducible kernel question remains. Keep launch counts small and document the selected kernel and metric set.
Stop when the evidence is sufficient to select or reject an optimization. Do not require a user checkpoint between layers unless the user requested stage-by-stage confirmation or the next layer is materially expensive.
Interpret cautiously
- Compare warmed runs using identical inputs and configuration.
- Separate CPU launch overhead, GPU execution, communication, transfers, and synchronization.
- Do not infer resource groups from device placement when stages can overlap.
- For streaming, compare p95 chunk latency with the media duration represented by a chunk and retain transport/encoding margin.
- Do not assign a memory-bound or compute-bound diagnosis from one utilization percentage alone; use the relevant throughput, occupancy, stalls, and launch context.
- Do not claim a speedup until the proposed change is benchmarked on the same workload.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 37 lines · 49 tokens per session scan A b0dc76bf602b
profile-pipeline is a skill published in the GitHub repository Tele-AI/TeleFuser (26 stars, last pushed yesterday), licensed Apache-2.0. It adds 49 tokens to every session and 537 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
ml-training-recipes
Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning…
ml-training-recipes
Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning…
ml-training-recipes
Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning…
tensorboard
Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's ML visualization toolkit.
ray-train
Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning with Ray Tune, fault tolerance, elastic scaling. Use when training massive models across multiple machines or running distributed hyperparameter sweeps.
torchforge-rl-training
Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions, easy algorithm experimentation, or scalable training with Monarch and TorchTitan.