Claude Skills by thedixitjain

Use when testing the golden_jupyter golden build

'Configure with gradient clipping helper operations. Auto-activating skill for ML Training. Triggers on: gradient clipping helper, gradient clipping helper Part of the ML Training skill category. Use when working with gradient clipping helper functionality. Trigger with phrases like \"gradient clipping helper\", \"gradient helper\", \"gradient\". '

Train reasoning and verifiable-task behavior with GRPO and reinforcement learning from verifiable rewards (RLVR). Use when task success is algorithmically checkable (math, code, tool calls, structured output), when designing GRPO reward functions, or when a GRPO run diverges or reward-hacks.

Display project statistics — phases, plans, requirements, git metrics, and timeline

Lightweight WSI tile extraction and preprocessing. Use for basic slide processing, tissue detection, tile extraction, and stain normalization for H&E images. Best for simple pipelines, dataset preparation, and quick tile-based analysis. For advanced spatial proteomics, multiplexed imaging, or deep learning pipelines use pathml.

Use when packaging an HPCA artifact for the voluntary post-acceptance evaluation on the separate AE HotCRP: scoping reproducibility tiers, preparing simulator- and silicon-heavy workflows for cold-start evaluators, budgeting their wall-clock time, and earning IEEE reproducibility badges.

Use when auditing an HPCA evaluation: declaring the fidelity contract (simulator, configuration, workloads, sampled regions, validation), matching each claim to its instrument, capturing real-silicon machine state, tuning baselines honestly, and reporting per-workload distributions rather than a single mean.

Use when planning an HPCA project timeline from venue fit through the two-step July registration and paper gate, the autumn review wait, the single rebuttal/revision window, notification, IEEE camera-ready, artifact evaluation, and the February/March talk, with backward-planning offsets for a hardware-evaluated paper.

Use when packaging the artifacts of an accepted ACM/IEEE HRI paper — code, robot behaviors, de-identified data, and study materials — for openness and any ACM badges the edition offers, while being honest about the limits of reproducing an embodied human-subjects study.

Use when writing the single ACM/IEEE HRI full-paper rebuttal that responds to three external reviewers and two area chairs — triaging study-design and statistics critiques, correcting misreadings, promising only what a camera-ready can deliver, staying anonymous and within the length limit, and giving the online reviewer discussion reasons to raise scores.

Use when designing or auditing the human-subjects study at the heart of an ACM/IEEE HRI full paper — choosing between/within-subjects designs, running Wizard-of-Oz honestly, powering the sample, reporting statistics with effect sizes and qualitative rigor, selecting validated scales, adding manipulation checks, pre-registering, and meeting HRI's human-participants ethics obligations.

Hugging Face Dataset Viewer

Add and manage evaluation results in Hugging Face model cards. Supports extracting eval tables from README content, importing scores from Artificial Analysis API, and running custom model evaluations with vLLM/lighteval. Works with the model-index metadata format.

Add and manage evaluation results in Hugging Face model cards. Supports extracting eval tables from README content, importing scores from Artificial Analysis API, and running custom model evaluations with vLLM/lighteval. Works with the model-index metadata format.

This skill should be used when users want to train or fine-tune language models using TRL (Transformer Reinforcement Learning) on Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment. Includes guidance on the TRL Jobs package, UV scripts with PEP 723 format, dataset preparation and validation, hardware selection, cost estimation, Trackio monitoring, Hub authentication, and model persistence. Should be invoked f...

Train or fine-tune language and vision models using TRL (Transformer Reinforcement Learning) or Unsloth with Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment. Includes guidance on the TRL Jobs...

Trains and fine-tunes vision models for object detection (D-FINE, RT-DETR v2, DETR, YOLOS), image classification (timm models — MobileNetV3, MobileViT, ResNet, ViT/DINOv3 — plus any Transformers classifier), and SAM/SAM2 segmentation using Hugging Face Transformers on Hugging Face Jobs...

Run PyTorch training across GPUs with minimal changes.

Use when the user asks about finding the best, top, or recommended model for a task, wants to know what AI model to use, or wants to compare models by benchmark scores. Triggers on: \"best model for X\", \"what model should I use for\", \"top models for [task]\", \"which model runs on my...

Build and publish a Gradio demo on Hugging Face Spaces for a user-provided LoRA. Use when someone asks to create, generate, ship, or publish a Space, demo, Gradio app, or playground for a LoRA — including LoRAs for Qwen-Image, Qwen-Image-Edit, LTX-Video, Wan, FLUX, SDXL, or other...

Fast BPE/WordPiece tokenization and custom vocab training.

Identify command-and-control beaconing patterns in network traffic by applying statistical frequency analysis, jitter calculation, and coefficient of variation scoring to detect periodic callbacks from compromised endpoints.

'Manage hyperparameter tuner operations. Auto-activating skill for ML Training. Triggers on: hyperparameter tuner, hyperparameter tuner Part of the ML Training skill category. Use when working with hyperparameter tuner functionality. Trigger with phrases like \"hyperparameter tuner\", \"hyperparameter tuner\", \"hyperparameter\". '

Structured hypothesis formulation, experiment design, and results interpretation for Product Managers. Use when the user needs to validate an assumption, design an A/B test, evaluate experiment results, or decide whether to ship based on data. Triggers include \"hypothesis\", \"A/B test\", \"experiment\", \"validate assumption\", \"test this\", \"should we ship\", or when making a decision that should be data-informed.

Use to understand why ICALP (EATCS) has no artifact-evaluation track or badge scheme, and what plays the equivalent role for a pure-theory paper — the full version with complete proofs, reproducible computational certificates, and optional machine formalization — so authors coming from a systems/ML venue do not waste effort building an artifact ICALP does not evaluate.

Use when packaging ICASSP code, data, audio or image samples, model checkpoints, scoring scripts, seeds, and logs, even though ICASSP has no formal artifact badge. Covers what signal-processing reviewers actually open, how single-blind review lets artifacts be public from the start, and how to make a task's measurement reproducible turnkey.

Use when packaging code, models, or data for an ICCV paper at review or release time, covering anonymous in-archive code under the do-not-cite-your-repo rule, the June-to-October release runway after acceptance, weight and license decisions in a field whose landmark releases became infrastructure, and reuse-grade packaging.

Use when designing the experimental program for an ICCV paper against the early-March deadline, covering benchmark-drift audits across the two-year gap, baseline fairness in the foundation-model era, ablations that isolate mechanisms, qualitative and failure evidence, and sequencing runs so the decisive result lands before the deadline, not after.

Use when packaging IEEE ICDE code, data, workload generators, and logs as supplemental material whose availability reviewers score, and for any post-acceptance reproducibility or badge process the edition runs. Covers what a builder-heavy committee inspects first, making a data-systems benchmark turnkey, and single-blind packaging.

Use when packaging code, data, and logs for an ICDM (IEEE International Conference on Data Mining) paper - building the anonymized, history-scrubbed repository that the PDF must cite for a triple-blind Research Track submission, how the single-blind Applied Track changes what may be revealed, and the smoke checks that make an ICDM artifact reviewer-usable.

Use when preparing the \"artifact\" of an ICDT (International Conference on Database Theory) paper, which at a pure-theory venue is the complete-proofs full version rather than a code package — covering why ICDT has no ACM-style artifact-badging or code-artifact track, what the marked appendix and the archived arXiv full version must contain, and (only for the rare algorithmic paper) how any code should be handled.

Use when packaging ICLR code, data, checkpoints, demos, logs, and reproduction instructions for reviewers or post-acceptance release, including anonymized links and private OpenReview discussion-period sharing. Use when a reviewer asks for a missing repro path, when a public comment questions whether claims can be verified, or when converting an anonymous supplement into a durable post-acceptance release for the permanent ICLR record.

Use when designing or auditing ICLR experiments, including baselines, ablations, scaling laws, robustness, statistics, benchmarks, human evaluation, and compute reporting. Use when a reviewer questions whether a representation-learning or model gain is real, when you must isolate one mechanism with an ablation, or when preparing a small compute-matched control that can be posted inline during the public discussion period.

Use when packaging ICML artifacts - code, data, model weights, simulators, benchmarks, proof scripts, notebooks, anonymous repositories, and supplementary code/data ZIPs - for both the double-blind review package and the public release that accompanies accepted PMLR papers. Use when checking anonymity, decision relevance, licensing, and the OpenReview code URL field under current ICML rules.

Use when packaging the artifacts behind an ICRA paper — ROS packages, controllers, simulation environments, trained policies, CAD and PCB files, datasets, and trial logs — into something a robotics reviewer or reader can actually run or audit, given that ICRA has no formal artifact-badging track to certify it for you.

Use when designing or auditing the experimental section of an ICRA paper — real-robot versus simulation-only evidence, trial counts and success-rate reporting, task distributions and resets, baseline fairness on shared hardware, sim-to-real transfer claims, failure-mode analysis, and the statistics robotics reviewers actually expect.

Use when preparing an artifact for the ICSE Artifact Evaluation track after paper acceptance, covering the ACM badge system (Available, Reusable, Results Reproduced/Replicated), evaluator-proof packaging, archival requirements, open licensing, and the January-window timeline anchored to recent cycles.

Use when designing or auditing the evaluation of an ICSE research-track paper, covering subject and benchmark selection, baseline fairness, statistical tests and effect sizes as SE reviewers expect them, qualitative-methods rigor, ablations for AI-based techniques, and threats-driven study design.

Use when reasoning about how an ICSE research-track submission is evaluated, covering double-anonymous PC review, the four posted criteria, the Accept / Major Revision / Reject decision model, the revision-and-final-decision timeline, and what recent ICSE acceptance statistics imply for strategy.

Use when packaging an IEEE ICSME artifact for the Joint Artifact Evaluation Track and ROSE Festival, covering the IEEE \"Open Research Object\" and \"Research Object Reviewed\" badges, what evaluators check first, DOI-issuing archives, evaluator-proof documentation, the shared track with SCAM and VISSOFT, and the separate post-acceptance deadline.

Use when designing or auditing IEEE ICSME empirical evaluations, covering real evolving subject systems, mining-software-repositories provenance, fair baselines, SE-standard statistics and effect sizes, change-history and survivorship confounds, qualitative rigor, contamination-aware LLM ablations, and matching evidence to the shape of each maintenance/evolution claim.

Use when planning an IEEE ICSME project timeline from venue and track choice through abstract registration, paper submission, the early-decision cut and author-response rebuttal, notification, IEEE Xplore camera-ready, the ROSE-Festival artifact evaluation, and presentation, with backward-planning offsets for a maintenance/evolution paper and honest handling of the single-annual-round calendar.

Use when targeting IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) or deciding whether a computer vision or machine learning manuscript fits this archival IEEE venue. Encodes the journal's fit, framing, method-and-evidence bar, house style, official-submission re-check, and desk-reject heuristics.

Use when preparing an IEEE S&P (Oakland) artifact-evaluation submission after acceptance, including choosing among the Available, Functional, and Results Reproduced badges, DOI-backed deposits, packaging exploits and malware-adjacent code responsibly, and scoping what evaluators can actually re-run.

Use when designing or auditing the evaluation of an IEEE S&P (Oakland) paper, including end-to-end attack demonstration, adaptive-adversary evaluation of defenses, measurement sampling and validity, baselines and ablations, statistical reporting of attack success, and the ethics constraints that shape what experiments are permissible.

Use when packaging IJCAI or IJCAI-ECAI code, data, proofs, models, and appendices as reproducibility evidence or supplementary material, especially when there is no separate artifact-evaluation badge but reviewers need convincing, anonymous, deadline-safe evidence.

Use when designing or auditing IJCAI or IJCAI-ECAI experiments, baselines, ablations, statistical evidence, hyperparameter reporting, compute descriptions, dataset handling, ethics risks, and reproducibility evidence for AI papers.

Use when the method choice, baselines, and computational-experiment design need alignment for an INFORMS Journal on Computing (IJOC) manuscript — before the experiments are run at scale. Designs a fair, reproducible experimental protocol; it does not write the algorithm proofs (see ijoc-theory-development).

Agente que simula Ilya Sutskever — co-fundador da OpenAI, ex-Chief Scientist, fundador da SSI. Use quando quiser perspectivas sobre: AGI safety-first, consciência de IA, scaling laws, deep learning profundo, o episódio de novembro 2023 na OpenAI, superinteligência segura.

Use when preparing an ACM IMC artifact for release, covering the artifact-availability declaration, post-acceptance availability shepherding, DOI-issuing dataset archives, documenting measurement provenance and vantage points, Community Contribution Award eligibility, and the difference between the anonymized review artifact and the public release.