Loading...
Complete reference for all Forge CLI commands.
Launch the interactive optimization wizard.
When run without arguments, Forge launches an interactive wizard that guides you through selecting input type, choosing target GPU, configuring settings, and running the optimization.
| Option | Description |
|---|---|
| -v, --version | Display version number |
| --no-color | Disable colored output |
| --json | Output results as JSON |
| --verbose | Enable verbose logging |
Authenticate with your RightNow account. Opens your browser for secure authentication.
Sign out of your account and remove local authentication tokens.
View your credit balance and usage.
| forge credits claim | Claim monthly bonus credits (Pro/Mega only) |
Start a CUDA kernel optimization session.
| Option | Description |
|---|---|
| -t, --task <id> | KernelBench task ID (e.g., 25 or level2/50) |
| --huggingface <model> | HuggingFace model ID (e.g., meta-llama/Llama-3-8B) |
| --layers <types> | HuggingFace layers to optimize (attention,mlp) |
| --file <path> | PyTorch .py file path |
| Option | Description | Cost |
|---|---|---|
| --turbo | Fast mode (16 pop, 10 gen, early stop) | 1 credit |
| (default) | Balanced mode (32 pop, 20 gen, early stop) | 1 credit |
| --quality | Deep mode (64 pop, 50 gen, no early stop) | 2 credits |
| --gpu <type> | Target GPU: H100, B200, A100_80GB, A100_40GB, L40S, A10, L4, T4 |
| --population <n> | Number of candidates per generation (default: 32) |
| --generations <n> | Maximum generations to run (default: 20) |
| --target <speedup> | Target speedup multiplier (e.g., 5.0 for 5x) |
| --no-early-stop | Disable early stopping when target is reached |
| --format <type> | Output format: triton (default) or cuda |
| --output <path> | Save optimized kernel to file |
| --demo | Run in demo mode (no GPU required) |
Browse and search KernelBench tasks.
| -l, --level <number> | Filter by level (1-4) |
| -s, --search <query> | Search tasks by name or description |
| -t, --task <id> | Get details for a specific task |
Manage optimization sessions.
| Subcommand | Description |
|---|---|
| forge session list | List all optimization sessions |
| forge session show <id> | Show details for a session |
| forge session resume <id> | Resume a paused session |
| forge session export <id> | Export session data to a file |
| forge session delete <id> | Delete a session |
| forge session clean | Clean old sessions |
Manage Forge CLI configuration.
| Subcommand | Description |
|---|---|
| forge config show | Display current configuration |
| forge config get <key> | Get a specific configuration value |
| forge config set <key> <value> | Set a configuration value |
| forge config reset | Reset configuration to defaults |
| forge config edit | Open interactive configuration wizard |
| forge config path | Show configuration file path |
Check backend connectivity and GPU status.
Update Forge CLI to the latest version.
During optimization, you can use these keyboard shortcuts:
| Key | Action |
|---|---|
| Ctrl+C | Cancel optimization |
| p | Pause/resume |
| s | Show status |
| ? | Show help |
| Variable | Description |
|---|---|
| FORGE_API_URL | Override API URL |
| FORGE_ANALYTICS | Set to false to disable analytics |
| NO_UPDATE_CHECK | Set to skip update check |
| NO_COLOR | Disable colored output |