Problem
Only ~16 of the available environments are published to the pip index at hub.primeintellect.ai/ob1/simple/. The rest require prime env install, which doesn't integrate with standard Python dependency management (pyproject.toml, uv, pip).
This makes it impossible to declare environment dependencies in pyproject.toml. For example, wordle is not on the pip index — you can't uv add wordle or pip install wordle from the Prime index.
Why this matters
When integrating verifiers into a training framework, we want all deps in pyproject.toml so that uv sync gives you a working environment. Having to run prime env install <name> as a separate step outside of the standard packaging flow is fragile and hard to automate (CI, containers, Slurm jobs, etc.).
What would help
- Publish all environments to the pip index (not just 16)
- Or document which envs are on the index vs require
prime env install, and the reasoning behind the split
Related: #254
Problem
Only ~16 of the available environments are published to the pip index at
hub.primeintellect.ai/ob1/simple/. The rest requireprime env install, which doesn't integrate with standard Python dependency management (pyproject.toml, uv, pip).This makes it impossible to declare environment dependencies in pyproject.toml. For example,
wordleis not on the pip index — you can'tuv add wordleorpip install wordlefrom the Prime index.Why this matters
When integrating verifiers into a training framework, we want all deps in pyproject.toml so that
uv syncgives you a working environment. Having to runprime env install <name>as a separate step outside of the standard packaging flow is fragile and hard to automate (CI, containers, Slurm jobs, etc.).What would help
prime env install, and the reasoning behind the splitRelated: #254