Prompt Cards

LoRA/QLoRA Strategy
Recommend when to use LoRA/QLoRA vs full finetune. Define rank search, target layers, and quantization-aware adapters. Include memory/perf tables per GPU class.
Tags: LLM, LoRA, QLoRA, finetuning, adapters, GPU
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Safety Red Team & Taxonomy
Create a safety taxonomy (harm classes) and a multilingual red-team plan with auto-generation of adversarial prompts. Provide block/transform policies and human review paths.
Tags: LLM, safety, red-team, taxonomy, policy, multilingual
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Privacy: DP-SGD & Redaction
Outline a privacy strategy: DP-SGD variants for SFT, selective redaction layers, privacy evals (membership inference), and logging minimization.
Tags: LLM, privacy, DP-SGD, redaction, membership-inference, logging
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Multilingual Alignment @ Scale
Design a multilingual alignment plan (zh/ja/hi/id/pt/en): shared subword policy, cross-lingual instructions, and locale-specific refusal tuning. Provide leakage checks.
Tags: LLM, multilingual, alignment, tokenization, refusal-tuning
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Multi-Task Multi-Domain Evals
Create a senior-grade eval battery: reasoning (math/code), instruction-following, safety, multilingual QA, and tool-use. Include uncertainty intervals and power analysis for A/Bs.
Tags: LLM, evaluation, multidomain, statistics, AB-testing
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Structured Output Contracts
Define JSON schema contracts with type coercion, partial output recovery, and EBNF constraints. Provide test-time correction and repair strategies.
Tags: LLM, structured-output, JSON, EBNF, validation, repair
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Agents: Planner–Executor–Critic
Specify a lightweight agent loop with decomposition, execution, and critique. Provide termination conditions, trace logging, and loop unroll limits.
Tags: LLM, agents, planning, critique, traces, governance
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Toolformer-Style Tool Use
Design a tool-use curriculum: function signatures, schema discovery, tool reliability scoring, and retry/backoff policy. Include sandboxing and cost guards.
Tags: LLM, tools, function-calling, reliability, sandbox, cost
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Hallucination Detection & Abstain
Create a hallucination detector using entailment+attribution signals. Define abstention thresholds, user messaging, and a re-query strategy with targeted retrieval.
Tags: LLM, hallucination, entailment, abstention, UX, grounding
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Retrieval Eval Harness
Build an eval harness: recall@k, calibrated precision, answer faithfulness, and human-time-to-verify. Include topic-aware test buckets and data drift alarms.
Tags: LLM, retrieval, eval, faithfulness, drift, metrics
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
RAG 2.0: Freshness & Faithfulness
Architect a retrieval stack with hybrid search, temporal decay, dedup, and passage-level citation anchors. Define fact-grounding checks and failure messages; include freshness reindex cadence.
Tags: LLM, RAG, hybrid, temporal-decay, citations, freshness
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:
Synthetic Data: Self-Play & Critique
Propose a self-play generation strategy where a teacher model drafts, a critic model scores, and a curator enforces diversity/novelty. Provide leakage and drift monitors.
Tags: LLM, synthetic-data, self-play, critic, curation, drift
Author: Assistant
Created at: 2025-12-18 00:00:00
Average Rating:
Total Ratings:

Curio AI Brain

Available in Chrome Web Store!