Skip to main content

​
Overview

Legacy vf-* commands are still available for backward compatibility. These commands are equivalent to their prime counterparts but use a different command structure.
New projects should use the prime command namespace. Legacy commands may be deprecated in future versions.

​
Command Mapping

Legacy CommandModern EquivalentDescription
vf-setupprime lab setupInitialize workspace
vf-initprime env initCreate environment template
vf-installprime env installInstall environment package
vf-build(internal)Build OpenEnv Docker image
vf-evalprime eval runRun evaluations
vf-tuiprime eval tuiView results in TUI
vf-gepaprime gepa runRun GEPA prompt optimization
vf-rlprime rl runRun RL training (local)
vf-train(deprecated)Old training command
vf-vllm(internal)vLLM server utilities
prime-rl(standalone)prime-rl training harness

​
Usage Examples

​
vf-setup

Equivalent to prime lab setup: 
vf-setup
vf-setup --prime-rl
vf-setup --skip-install
See prime lab setup for full documentation.

​
vf-init

Equivalent to prime env init: 
vf-init my-env
vf-init my-env --path ~/envs
vf-init my-env --multi-file
vf-init my-env --openenv
See prime env init for full documentation.

​
vf-install

Equivalent to prime env install: 
vf-install my-env
vf-install primeintellect/gsm8k
vf-install my-env --path ~/envs
vf-install gsm8k --from-repo
vf-install gsm8k --from-repo --branch dev
See prime env install for full documentation.

​
vf-build

Internal command for building OpenEnv Docker images. Called automatically by prime env push for OpenEnv environments. 
vf-build my-env
vf-build my-env --path ~/envs
This command:
  1. Locates proj/ directory with OpenEnv project
  2. Detects contract type (gym vs. mcp) from server/app.py
  3. Builds Docker image using prime images push
  4. Waits for build completion
  5. Writes .build.json with image metadata
Options:
​
env
string
required
Environment ID (e.g., my-env)
​
--path
string
default:"./environments"
Base path for environments directory. The command looks for {path}/{env_id}/proj/.Aliases: -p
Output: 
Building image for 'my-env' with context='./environments/my_env/proj' dockerfile='server/Dockerfile' image='my-env:latest'
Image build completed with status=ready
Wrote ./environments/my_env/proj/.build.json with image='registry.primeintellect.ai/username/my-env:latest' port=8000 app='server.app:app' contract='gym'

​
vf-eval

Equivalent to prime eval run: 
vf-eval gsm8k -m gpt-4.1-mini -n 10
vf-eval gsm8k -m gpt-4.1-mini -n 100 -r 5 -t 1024 -T 0.7
vf-eval configs/eval/benchmark.toml
See prime eval run for full documentation.

​
vf-tui

Equivalent to prime eval tui: 
vf-tui
See prime eval tui for full documentation.

​
vf-gepa

Runs GEPA (Genetic Prompt Optimization Algorithm) for automated prompt improvement: 
vf-gepa my-env -m gpt-4.1-mini
vf-gepa my-env -m gpt-4.1-mini -c configs/gepa/custom.toml
GEPA iteratively evolves prompts to maximize reward on your environment. See the GEPA documentation for details.

​
vf-rl

Runs local RL training with the verifiers trainer: 
vf-rl configs/rl/gsm8k.toml
For production training, use Hosted Training or prime-rl instead. vf-rl is for quick local experiments only.

​
vf-train

Deprecated. Use vf-rl, prime-rl, or Hosted Training instead.

​
vf-vllm

Internal command for vLLM server management. Not intended for direct use.

​
prime-rl

Standalone training framework. Installed via prime lab setup --prime-rl. 
uv run prime-rl configs/prime-rl/wiki-search.toml
See prime-rl documentation for details.

​
Migration Guide

​
From vf-* to prime

Update your workflows to use modern commands: Before: 
vf-setup
vf-init my-env
vf-install my-env
vf-eval my-env -m gpt-4.1-mini -n 10
vf-tui
After: 
prime lab setup
prime env init my-env
prime env install my-env
prime eval run my-env -m gpt-4.1-mini -n 10
prime eval tui

​
Scripts and CI/CD

Update scripts to use the prime command: 
#!/bin/bash
# Old
# vf-eval gsm8k -m gpt-4.1-mini -n 100 -s

# New
prime eval run gsm8k -m gpt-4.1-mini -n 100 -s

​
Aliases (Optional)

If you prefer the old command names, create aliases: 
# Add to ~/.bashrc or ~/.zshrc
alias vf-setup='prime lab setup'
alias vf-init='prime env init'
alias vf-install='prime env install'
alias vf-eval='prime eval run'
alias vf-tui='prime eval tui'

​
Why the Change?

The prime command namespace provides:
  1. Better organization - Logical grouping (lab, env, eval)
  2. Consistency - Matches other Prime Intellect tools
  3. Discoverability - prime --help shows all subcommands
  4. Extensibility - Easier to add new command groups

​
Deprecation Timeline

  • v0.8.x - Both vf-* and prime commands supported (current)
  • v0.9.x - vf-* commands emit deprecation warnings
  • v1.0.x - vf-* commands may be removed
We recommend migrating to prime commands now to avoid disruption.

​
Support

If you encounter issues migrating to the new commands:

​
Summary

Legacy vf-* commands still work but are being phased out in favor of the prime command namespace. Update your workflows to use:
  • prime lab setup instead of vf-setup
  • prime env {init,install,push} instead of vf-{init,install,build}
  • prime eval {run,tui} instead of vf-{eval,tui}
The functionality is identical; only the command structure has changed.

Build docs developers (and LLMs) love

Get started for freeTalk to us