List your company

Sequential Decision Making on WebShop (post-training evaluation)

14.8Rollout LengthStarting policy
6.50088.655410.8112.9646Jun 30, 2026

Evaluation results

MethodLinks
Starting policyBackbone=Qwen2.5-7B-Instruct2026.0614.8
GRPOBackbone=Qwen2.5-7B-Instruct, Training Protocol=GRPO2026.068
TRIAGEBackbone=Qwen2.5-7B-Instruct, Training Protocol=TRIAGE2026.066.82