Results for “reward-modeling”

Train or fine-tune language and vision models using TRL or Unsloth on Hugging Face Jobs cloud infrastructure, with support for SFT, DPO, GRPO, and reward modeling, plus GGUF conversion for local deployment.

TRAK: Attributing Model Behavior at Scale

Use to design support incentives that improve behaviour without destroying the metric — pairing pay with guardrails, naming gaming modes, and choosing measures that survive Goodhart pressure. Trigger for "incentive plan", "agent bonus scheme", "SPIFF design", "pay for QA score", "what metric should we bonus", CSAT incentives, or reviewing whether a comp change is driving gaming.

Add account state (P&L, win rate, drawdown) to RL observations + drawdown penalty in rewards. Trigger when: (1) model needs account awareness, (2) training should penalize drawdowns, (3) upgrading obs_dim 5300→5600.

Fix HOLD bias in RL reward function. Trigger when: (1) model learns to always HOLD, (2) trade rate is too low (<10%), (3) slippage penalty exceeds typical price moves.

Guides users through participating in the SOMA decentralized training network, covering data submission, model training, reward claiming, and strategic optimization.

Recommend the right AI model for a task by scoring candidates across six dimensions (Reasoning, Engineering, Speed, Breadth, Reliability, Governance) and displaying a spider-chart profile.

Every metric encodes an opinion about which mistake hurts.

Use when the team needs pricing, packaging, free-tier, trial, usage metering, or business-model guidance for a software product. Trigger on requests to design how the product should make money or control value access.

Design habit-forming product loops using the Hook Model (Trigger, Action, Variable Reward, Investment) to analyze and improve user engagement, retention, and re-engagement strategies.

Cross-model benchmark for gstack skills. (gstack)

Entrena modelos de machine learning con Scikit-learn, LightGBM y XGBoost, desde un baseline hasta un modelo productivo con validación robusta y explicabilidad.

Turn model work into a production ML system with data contracts, reproducible training, quality gates, deployable artifacts, and monitoring.

Plan which model tier handles which work BEFORE execution begins — a high-cognition model deeply understands the problem, lays the foundations, then emits a modular plan assigning each module the cheapest tier that can safely execute it, with escalation tripwires and one-way-door protection. Advisory only: it announces "next module → tier X / model Y" at each boundary and the HUMAN switches models — harnesses like Cursor cannot switch mid-run. Load when the user asks which model to use, wants a model plan, model tiers, model-tier routing, assign models to tasks or modules, says "cheap model got stuck", "which model for this task", "cost-efficient model choice", or when implementation-plan / problem-to-plan need a model: tier column. NOT dynamic-routing (plan-path selection after failure) — this skill assigns cognition tiers to work.

Benchmarks language model agents on the C2C multi-agent negotiation task, reporting win rate across starting positions.

Recommend model families and validation strategy based on data, constraints, and objective. Use when: (1) choosing algorithms, (2) balancing bias/variance, (3) planning benchmark baselines. NOT for: final legal/compliance sign-off.

Design habit-forming product loops using the Hook Model (Trigger, Action, Variable Reward, Investment). Use when the user mentions "users arent coming back", "engagement loops", "habit formation", "push notifications", "variable rewards", "daily active users", "habit zone", or "user retention loops". Also trigger when designing notification strategies, building streaks or progress systems, or analyzing why users stop using a product after initial signup. Covers ethics evaluation and onboarding for habits. For friction reduction and B=MAP, see improve-retention. For viral sharing, see contagious.

Surface beliefs, assumptions, stories, and values shaping a system. Use when deeper mental models need examining with care and evidence.

Improves AI skills for activation, clarity, and cross-model reliability through benchmarking, salience tuning, and regression triage.

Agent validation overhaul: reward weight overrides, fitness decline gate, pinned data, staged experiments

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.

Provides production-grade techniques for evaluating LLM outputs using LLMs as judges, covering direct scoring, pairwise comparison, bias mitigation, rubric generation, and confidence calibration.

Trains language models with Direct Preference Optimization using preference pairs, covering DPOTrainer setup, dataset preparation, and beta tuning for stable preference learning without explicit reward models.

Train AI models to be harmless through self-critique and AI feedback using a set of constitutional principles, without requiring human labels for harmful outputs.