القائمة
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.
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 router skill for inference scaling decisions. Use when training, evaluating, or benchmarking the router that decides between low/medium/high/superthink compute modes.
| name | ttc-training |
| description | 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. |
| compatibility | Requires kaggle-cli or huggingface-cli configured |
Training pipeline for Qwen models with TTC (Test-Time Compute) awareness.
# Ensure kaggle CLI is configured
kaggle competitions list
# Or for HF Jobs
huggingface-cli login
# From ttc-inference-scaling-stack repo
python scripts/cpt_train.py \
--model Qwen/Qwen2.5-0.5B \
--dataset path/to/corpus \
--backend kaggle \
--output-dir models/cpt-qwen
python scripts/sft_train.py \
--base-model models/cpt-qwen \
--dataset path/to/sft_data \
--lora-r 16 \
--backend kaggle
python scripts/rlvr_train.py \
--sft-model models/sft-qwen \
--verifier qwen-verifier \
--backend hf-jobs
python scripts/ttc_train.py \
--rlvr-model models/rlvr-qwen \
--router-profile balanced \
--backend hf-jobs
| Profile | Compute | Use Case |
|---|---|---|
low | ~1 GPU-hour | Quick experiments, smoke tests |
medium | ~5 GPU-hours | Standard training, Kaggle |
high | ~20+ GPU-hours | Production runs, HF Jobs |
After training completion:
models/<stage>-qwen/final_model/ - Model weightsmodels/<stage>-qwen/train_summary.json - Training metricsmodels/<stage>-qwen/config.json - Training configurationUser: Train Qwen 3.5 0.8B on my instruction dataset with SFT, use Kaggle backend
Agent will:
1. Load skill instructions (this file)
2. Prepare Kaggle notebook configuration
3. Launch training job
4. Monitor and report results
5. Save model to specified location