horizon-rl/strands-sglang

SGLang model provider for Strands Agents for on-policy agentic RL training.

60
/ 100
Established

Provides token-level rollout capture (token IDs, logprobs, loss masks) directly from SGLang's `/generate` endpoint, eliminating retokenization drift for on-policy RL training. Integrates with the Strands Agents SDK framework and enforces strict, deterministic tool-call parsing without post-processing heuristics. Designed for seamless training pipelines with frameworks like slime, exposing complete token trajectories needed for policy gradient methods.

50 stars and 4,156 monthly downloads. Available on PyPI.

Maintenance 13 / 25
Adoption 16 / 25
Maturity 22 / 25
Community 9 / 25

How are scores calculated?

Stars

50

Forks

4

Language

Python

License

Apache-2.0

Last pushed

Mar 13, 2026

Monthly downloads

4,156

Commits (30d)

0

Dependencies

4

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/agents/horizon-rl/strands-sglang"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.