Menu
Hyperparameter tuning skill for TTC training using Optuna. Use when optimizing learning rate, batch size, LoRA rank, and other training parameters.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
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.
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.
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-hparam |
| description | Hyperparameter tuning skill for TTC training using Optuna. Use when optimizing learning rate, batch size, LoRA rank, and other training parameters. |
| compatibility | Requires Optuna installed and compute backend |
Automated hyperparameter search for TTC training pipelines using Optuna.
pip install optuna optuna-integration
| Parameter | Range | Default |
|---|---|---|
learning_rate | 1e-6 to 1e-3 | 2e-5 |
batch_size | 1, 2, 4, 8, 16 | 4 |
lora_r | 4, 8, 16, 32, 64 | 16 |
lora_alpha | 8, 16, 32 | 32 |
warmup_ratio | 0.0 to 0.2 | 0.1 |
weight_decay | 0.0 to 0.1 | 0.01 |
python scripts/optuna_tune.py \
--model Qwen/Qwen3.5-0.8B \
--dataset path/to/data \
--n-trials 20 \
--backend kaggle \
--metric eval_loss
python scripts/optuna_tune.py \
--model Qwen/Qwen3.5-0.8B \
--dataset path/to/data \
--n-trials 100 \
--backend hf-jobs \
--metric eval_loss \
--pruner median \
--sampler tpe
import optuna
def objective(trial):
lr = trial.suggest_float("learning_rate", 1e-6, 1e-3, log=True)
batch_size = trial.suggest_categorical("batch_size", [1, 2, 4, 8])
lora_r = trial.suggest_categorical("lora_r", [4, 8, 16, 32])
# Run training with these params
loss = train_and_evaluate(lr, batch_size, lora_r)
return loss
After tuning:
optuna_study.db - SQLite database with all trialsbest_params.json - Optimal hyperparametersoptuna_dashboard.html - Visualization reportUser: Tune hyperparameters for my SFT training on Qwen 0.8B
Agent will:
1. Define search space
2. Launch Optuna study
3. Run trials on backend
4. Report best parameters
5. Save study results