القائمة
TTC router skill for inference scaling decisions. Use when training, evaluating, or benchmarking the router that decides between low/medium/high/superthink compute modes.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Transform HuggingFace/arXiv research papers into trainable engineering features, skills, scripts and configs for Qwen 3.5 0.8B. Use when user provides a paper URL or wants to convert research into implementation. Creates complete skill ecosystems with ASCII diagrams.
TTC distillation skill for knowledge transfer from teacher to student models. Use when distilling large models to smaller ones with TTC awareness. Supports medium and high profiles.
Hyperparameter tuning skill for TTC training using Optuna. Use when optimizing learning rate, batch size, LoRA rank, and other training parameters.
Judge model training skill for TTC evaluation. Use when training models to evaluate and score outputs from other models. Supports reward model training and judge datasets.
TTC-aware training skill for Qwen models. Use when training models with Test-Time Compute awareness, including CPT → SFT → RLVR → TTC pipeline stages. Supports Kaggle and HF Jobs backends.
| name | ttc-router |
| description | TTC router skill for inference scaling decisions. Use when training, evaluating, or benchmarking the router that decides between low/medium/high/superthink compute modes. |
| compatibility | Requires router model and benchmark datasets |
The TTC router decides which compute level to use for each query:
┌─────────────┐
│ Router │
│ Classifier │
└──────┬──────┘
│
┌────────────────┼────────────────┐
▼ ▼ ▼
┌──────────┐ ┌──────────┐ ┌──────────┐
│ LOW │ │ MEDIUM │ │ HIGH │
│ 1 sample │ │ 3 samples│ │ 10 samples│
└──────────┘ └──────────┘ └──────────┘
python scripts/train_router.py \
--base-model Qwen/Qwen2.5-0.5B \
--trace-data path/to/traces.jsonl \
--output models/ttc-router \
--backend kaggle
{"query": "What is 2+2?", "complexity": "low", "optimal_samples": 1}
{"query": "Solve this math problem...", "complexity": "high", "optimal_samples": 10}
python scripts/eval_router.py \
--router models/ttc-router \
--benchmark mmlu \
--modes low,medium,high
| Metric | Description |
|---|---|
| Accuracy | Correct routing decisions |
| Efficiency | Average compute per query |
| Coverage | Distribution across modes |
| Benchmark | Low | Medium | High | Router Choice |
|---|---|---|---|---|
| MMLU | 45% | 58% | 62% | Mixed |
| GSM8K | 30% | 55% | 70% | Pred. High |
| Simple QA | 85% | 86% | 87% | Pred. Low |
For maximum reasoning:
python scripts/superthink.py \
--model Qwen/Qwen2.5-7B-Instruct \
--query "Complex problem..." \
--samples 50 \
--aggregation best_of_n
User: Train a router that knows when to use more thinking
Agent will:
1. Collect inference traces with varying compute
2. Train classifier on complexity labels
3. Evaluate on held-out benchmark
4. Return router model