Solo GRPO and Tandem RLVR policies from three single-seed experiments. Each Tandem policy is the trained senior; its frozen junior is not included. The reproduction checkpoints are selected by validation pass@4; the others are final step-200 checkpoints. Validation uses each arm's own rollout protocol, so GRPO and Tandem scores are not directly comparable. Base revisions, selection evidence and SHA-256 hashes are in checkpointmanifest.json. The Qwen3.5 checkpoints keep the Qwen35ForConditionalGeneration architecture (text-only training) and are stored in FP32; dtype="auto" loads BF16.
Organization
Tandem RLVR collaboration
cge7-tandem-rlvr
Models in Library1
Datasets in Library0
Models on Hugging Face1
Followers3
Compare the published evidence
This profile connects the 1 model and dataset record shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Reinforcement learning.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.