搜索技能.../

机器学习 | Skills Pool

分类机器学习

机器学习

搜索查看全部分类

AI 推荐

机器学习

分类

机器学习调试销售与营销大语言模型与 AI 系统管理效率与集成代码质量 Git 工作流架构模式文档项目管理测试金融与投资安全知识库技术文档后端前端教育 CI/CD 自动化工具内容创作数据工程游戏开发学术脚本 CMS 与平台移动端云计算 IDE 插件打包与分发数据分析框架内部媒体设计全栈房产与法律生物信息学命令行工具 SQL 数据库计算化学容器实验室工具数据库工具健康与健身智能合约监控域名与 DNS 工具 Web3 工具健康养生支付文学与写作电商开发哲学与伦理天文与物理电子商务占卜与神秘学手工艺术科学计算商业应用 DeFi NoSQL 数据库烹饪艺术 $Undefined

查看全部分类

机器学习调试销售与营销大语言模型与 AI 系统管理效率与集成代码质量 Git 工作流架构模式文档项目管理测试金融与投资安全知识库技术文档后端前端教育 CI/CD 自动化工具内容创作数据工程游戏开发学术脚本 CMS 与平台移动端云计算 IDE 插件打包与分发数据分析框架内部媒体设计全栈房产与法律生物信息学命令行工具 SQL 数据库计算化学容器实验室工具数据库工具健康与健身智能合约监控域名与 DNS 工具 Web3 工具健康养生支付文学与写作电商开发哲学与伦理天文与物理电子商务占卜与神秘学手工艺术科学计算商业应用 DeFi NoSQL 数据库烹饪艺术 $Undefined

机器学习

Regression Result Analyzer

Imported from modelsim/VIP/16_REGRESSION_RESULT_ANALYZER_SKILL.md (Regression Result Analyzer Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Previous 1…507 508 509…519 Next

Understand V4

Inner Polya cycle for context engineering. Iteratively transforms a WorldModel through classify/enrich/branch/refine/validate/compact strategies until the model is sufficient or budget is exhausted.

Move Tools

Define move-tool behaviors for object manipulation and viewport navigation with predictable selection and resize ergonomics.

Consensus Pattern

Implementation pattern for D-041 multi-model consensus via LiteLLM + Instructor. Use when implementing any LLM inference call that makes content decisions (genre, author, science scope, etc.). Never use single-LLM calls for content classification.

Adversarial Self Refine

Iteratively improve any solution through blind assertive critique until the model defends or converges. Uses isolated sub-agents (CRITIC + AUTHOR) so critique is never biased by authoring context. Falls back to single-thread when sub-agents unavailable (DEGRADED). Use when refining drafts, code, prompts, designs, plans, or any artifact that benefits from iterative quality improvement. Applies a recursive functor pattern with emergent termination detection.

Adaptive Reasoning v1.0

Single-entry classifier and inline reasoning executor. Classifies a task across four observable dimensions and routes to one of three inline reasoning modes: direct-exec, adversarial-review (two-pass defect discovery), or bounded-synthesis (A/B/AB comparison for pre-implementation refinement). All reasoning executes INLINE — no delegation to sub-agents, no separate skills. This skill does NOT decide code acceptance (deterministic validators do) and does NOT own persistence.

Coverage Plan Writer

Imported from modelsim/VIP/09_COVERAGE_PLAN_WRITER_SKILL.md (Coverage Plan Writer Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Rtl Analysis

Imported from modelsim/VIP/01_RTL_ANALYSIS_SKILL.md (RTL Static Analysis Skill: Connection Check, Signal Integrity & CDC Verification). Use when this verification workflow is needed.

qingyin-alice-zhong0

Assertion Coverage Reviewer

Imported from modelsim/VIP/14_ASSERTION_COVERAGE_REVIEWER_SKILL.md (Assertion Coverage Reviewer Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Coverage Mapper

Imported from modelsim/VIP/11_COVERAGE_MAPPER_SKILL.md (Code ↔ Functional Coverage Mapper Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Jira Bug Triage

Imported from modelsim/VIP/20_JIRA_BUG_TRIAGE_SKILL.md (Jira Bug Triage & Filing Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Sequence Scenario Generator

Imported from modelsim/VIP/07_SEQUENCE_SCENARIO_GENERATOR_SKILL.md (Sequence and Scenario Generator Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Testplan Traceability

Imported from modelsim/VIP/19_TESTPLAN_TRACEABILITY_SKILL.md (Testplan ↔ Coverage Traceability Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Uvm Component Builder

Imported from modelsim/VIP/06_UVM_COMPONENT_BUILDER_SKILL.md (UVM Component Builder Skill). Use when this verification workflow is needed.

qingyin-alice-zhong0

Dpo Training

DPO (Direct Preference Optimization) full fine-tuning pipeline for safety-aligned language models. Uses TRL DPOTrainer with pairwise preference data (prompt/chosen/rejected). No LoRA — full parameter update for maximum alignment signal. Supports any HuggingFace causal LM (Qwen2, LLaMA-3, LLaMA-2, etc.).

C²-Eval Universal Pipeline

Universal C²-Eval (Course-Correction Evaluation) pipeline for any HuggingFace model. Measures Corr@k — the probability that a model self-corrects after being forced to start generating harmful content. Each model gets its own isolated run directory. Use when evaluating a new model with C²-Eval, interpreting Corr@k results, or setting up the evaluation pipeline from scratch. No model registration required — pass model path directly.

Blackwell Vlm Deployer

Deployment, optimization, and serving of Vision-Language Models (VLMs) and Large Language Models (LLMs) on NVIDIA Blackwell GB200/B200 hardware. Use when Gemini CLI needs to assess hardware capacity, estimate VRAM needs, configure multi-GPU serving (vLLM/TP), or setup interactive web interfaces for large models like Qwen3-VL.

Fine Tuning

Fine-tune pretrained models with LoRA, QLoRA, or full fine-tuning using HF Trainer and TRL for instruction, chat, and preference alignment.

Gpu Provisioner

Provision and manage cost-efficient GPU instances for from-scratch LLM training with strict spend confirmation gates.

Inference Deploy

Deploy trained models for inference with vLLM, TGI, Ollama, or llama.cpp including quantization, API setup, and latency benchmarking.

Model Publisher

Package trained checkpoints, generate model cards, and publish releases to Hugging Face Hub with optional GGUF export.

Tokenizer Trainer

Train and evaluate BPE tokenizers for from-scratch LLM pretraining with HF-compatible outputs.

Freezed Patterns

Freezed ile immutable model ve union type pattern'leri. Model/entity olusturma ve sealed class icin kullan.

Vol3 Compute Engine

Use when scaffolding, repairing, or extending a Vol III compute engine, test oracle, executable witness, or manuscript-backed numerical claim. Trigger on compute engine, scaffold tests, hardcoded value, oracle, executable witness, or add compute support.

Logic Cleaner

Language-agnostic refactoring engine. Reduces cyclomatic complexity, flattens control flow, and enforces strict naming and readability standards.

Deepdrivewe Example

Use this skill when building a new example or custom workflow with DeepDriveWE (weighted ensemble MD on the Academy multi-agent framework). Covers project layout, agent subclassing, Pydantic config models, binner/recycler/resampler selection, Parsl compute configs, and verification steps.

Nanobanana 2 Image Generation

Use this skill when users ask to generate, edit, or compose images with Nano Banana (Gemini image models), including text-to-image, image editing, multi-image composition, grounding, and output sizing/saving controls.

Code Thinker

嚴格的複雜程式任務思考與執行技能，用於複雜邏輯修改、核心除錯、跨檔重構、架構敏感變更與任何容易讓模型衝動寫碼的任務。此技能要求先做靜默內部審查，再執行最小且可驗證的修改；嚴禁對使用者外顯 thinking tags、chain-of-thought 或逐條內部推理。

Understand V2

Inner Polya cycle for context engineering. Iteratively transforms a WorldModel through classify/enrich/branch/refine/validate/compact strategies until the model is sufficient or budget is exhausted.

Understand V3

Inner Polya cycle for context engineering. Iteratively transforms a WorldModel through classify/enrich/branch/refine/validate/compact strategies until the model is sufficient or budget is exhausted.