0

Openthoughts Tblite RL Env (Community)

Fresh

OpenThoughts TBLite Harbor dataset environment using v1 tasksets and harnesses.

Type
RL Env
License
apache-2.0
Published
May 2026

Cite

Notes

Only stored in your browser.

OpenThoughts TBLite

Harbor environment for the latest openthoughts/openthoughts-tblite Harbor dataset, wired through the v1 HarborTaskset and packaged v1 harnesses.

Each task uses the prebuilt Prime sandbox image:

team-clyvldofb0000gg1kx39rgzjq/openthoughts-tblite-<task>:latest

Run

uv pip install -e ./environments/openthoughts_tblite

Select harnesses in eval TOML through the v1 package id:

[eval.harness]
id = "verifiers.v1.packages.harnesses.opencode"
max_turns = 4

Arguments

ArgumentDefaultDescription
taskset.datasetopenthoughts/openthoughts-tbliteHarbor dataset ID.
taskset.task_namesNoneOptional task-name allowlist.
taskset.agent_timeout_seconds900.0Fallback agent timeout when a task does not set [agent].timeout_sec.
taskset.verifier_timeout_seconds900.0Fallback verifier timeout when a task does not set [verifier].timeout_sec.
taskset.timeout_multiplier1.0Multiplies each task's sandbox lease, agent command timeout, and verifier timeout.