Steve0927/qwen3-embedding-0.6b-lrat-fullv1-broad-extrap-a125

Qwen3 Embedding 0.6B LRAT Full-v1 to Broad Extrapolation a1.25

This checkpoint extends the aggregate training direction from the team's full-v1 retriever to broad-g16-v2:

full-v1 + 1.25 * (broad-g16-v2 - full-v1)

Both endpoints start from Qwen/Qwen3-Embedding-0.6B and were trained only on the organizer-provided LRAT trajectories and corpus. This is a global parameter direction; no A-board query, qrel, per-query failure, or external corpus was used to construct it. No weights from another pretrained checkpoint are used.

The checkpoint was selected by aggregate fixed-seed evaluation with the released BrowseComp-Plus Agent and Judge workflow using Qwen3.5-4B. Local scores are used only for checkpoint selection and are not claimed as official results.