Skip to main content

​
Wizard Commands Reference

The AITraining wizard supports various commands to help you navigate, search, and configure your training job. These commands work at any prompt:
CommandShortcutDescription
:backGo back to the previous step
:help?, :hShow detailed help for the current prompt
:exit:quitCancel the wizard and exit

​
Using :back

You can go back at any point to change previous answers: 
Model (number, HF ID, or command): :back
↩️ Going back to dataset selection...

Dataset (number, HF ID, or command):

​
Using :help

Every prompt has contextual help: 
Training split name [train]: :help

ℹ️  Help
  Dataset splits are named subsets of your data.

  Common split names:
    • 'train' - Training data (most common)
    • 'test' - Test/evaluation data
    • 'validation' or 'valid' - Validation data

  Note: This is NOT asking for a percentage split (like 80/20).
  It's asking for the exact name of the split in your dataset.

Training split name [train]:

​
Catalog Commands

These commands work when browsing models or datasets:
CommandDescription
/search <query>Search for models/datasets by name
/sortChange sorting (trending, downloads, likes, recent)
/filterFilter models by size (models only)
/refreshClear cache and reload the list

​
/search

Find specific models or datasets: 
Model (number, HF ID, or command): /search llama

Popular models (trending):
  1. meta-llama/Llama-3.2-1B (1B)
  2. meta-llama/Llama-3.2-3B (3B)
  3. meta-llama/Llama-3.1-8B (8B)
  4. meta-llama/Llama-3.1-70B (70B)
  ...
Search examples:
  • /search gemma - Find Gemma models
  • /search code - Find code-focused models
  • /search alpaca - Find Alpaca-style datasets
  • /search conversation - Find conversation datasets

​
/sort

Change how results are ordered: 
Model (number, HF ID, or command): /sort
Sort options: [T]rending [D]ownloads [L]ikes [R]ecent
Sort by [T]: D
Sort OptionKeyDescription
TrendingTWhat’s popular right now
DownloadsDMost downloaded all-time
LikesLMost liked by the community
RecentRNewest additions

​
/filter

Filter models by parameter count (only works for models, not datasets): 
Model (number, HF ID, or command): /filter
Filter size: [A]ll [S]mall(<3B) [M]edium(3-10B) [L]arge(>10B)
Filter size [A]: S
FilterKeySize RangeTypical Hardware
AllANo filterAny
SmallS< 3B parametersMacBook, consumer GPU
MediumM3B - 10B parametersGaming GPU, workstation
LargeL> 10B parametersCloud GPU, multi-GPU

​
/refresh

Clear the cache and fetch fresh data: 
Model (number, HF ID, or command): /refresh
Cache cleared.

Popular models (trending):
  ...

​
Selection Methods

When choosing a model or dataset, you have several options:

​
By Number

Select from the displayed list: 
Popular models (trending):
  1. google/gemma-3-270m (270M)
  2. google/gemma-2-2b (2B)
  3. meta-llama/Llama-3.2-1B (1B)

Model (number, HF ID, or command): 2
✓ Model: google/gemma-2-2b

​
By HuggingFace ID

Type the full model/dataset ID: 
Model (number, HF ID, or command): mistralai/Mistral-7B-v0.3
✓ Model: mistralai/Mistral-7B-v0.3

​
By Local Path

Point to a local directory: 
Dataset (number, HF ID, or command): ./my_training_data
✓ Dataset: ./my_training_data

​
Input Conventions

​
Defaults

Values in [brackets] are defaults. Press Enter to accept: 
Project name [my-llm-project]: ↵
✓ Project: my-llm-project

​
Required Fields

Fields marked [REQUIRED] must be filled: 
Prompt column name [REQUIRED] [prompt]: ↵
❌ This field is required for DPO/ORPO training.
Prompt column name [REQUIRED] [prompt]: instruction
✓

​
Yes/No Questions

Answer with y/yes or n/no: 
Configure advanced parameters? [y/N]: y

Enable LoRA? [Y/n]: ↵
✓ LoRA enabled (default)
Capitalized letter indicates the default:
  • [Y/n] - Default is Yes
  • [y/N] - Default is No

​
Keyboard Shortcuts

KeyAction
EnterAccept default or confirm input
Ctrl+CCancel wizard (same as :exit)
Arrow Up/DownScroll through numbered options (if supported)

​
Advanced Parameters

When configuring advanced parameters, the wizard groups them: 
⚙️  Training Hyperparameters

Configure Training Hyperparameters parameters? [y/N]: y

epochs [1]:
batch_size [2]:
lr [3e-5]:
Each group can be configured independently:
GroupContains
Training Hyperparametersepochs, batch_size, lr, warmup_ratio
PEFT/LoRApeft, lora_r, lora_alpha, quantization
DPO/ORPOdpo_beta, max_prompt_length
Hub Integrationpush_to_hub, username, token
Knowledge Distillationteacher_model, distill_temperature
Hyperparameter Sweepuse_sweep, sweep_n_trials
Enhanced Evaluationuse_enhanced_eval, eval_metrics
Reinforcement Learningrl_reward_model_path (PPO only)

​
Tips

Every single prompt has detailed help. If you’re unsure what something means, type :help.
Made a wrong choice? Use :back to return to previous steps. Your other answers are preserved.
Instead of scrolling through hundreds of models, use /search llama or /search 7b to narrow down.
Not sure which models will work? Use /filter → S (small) to see only models that fit consumer hardware.
On your first training, accept most defaults. Get something working, then customize.

​
Command Quick Reference

# Navigation
:back          Go to previous step
:help          Show help for current prompt
:exit          Cancel and exit

# Catalog (models/datasets)
/search query  Search by name
/sort          Change sort order
/filter        Filter by size (models only)
/refresh       Reload list

# Selection
1, 2, 3...     Select by number
google/gemma   Type HuggingFace ID
./my_data      Type local path

# Input
Enter          Accept default
y/n            Yes/No answers
Ctrl+C         Cancel