JITMEM stores complete raw trajectories, retrieves the top-k trajectories with BM25 over task descriptions, synthesizes an ephemeral task-conditioned payload with a curator, executes with a frozen LLM, and admits judged-successful trajectories to a streaming memory bank; only the curator is trained, using GRPO and same-task benchmark reward. · CiteArk
Not assessedNo independent reproduction scheduledMethodclaim-method-001
JITMEM stores complete raw trajectories, retrieves the top-k trajectories with BM25 over task descriptions, synthesizes an ephemeral task-conditioned payload with a curator, executes with a frozen LLM, and admits judged-successful trajectories to a streaming memory bank; only the curator is trained, using GRPO and same-task benchmark reward.
Source: paper:PDF pp. 4-5, Section 3 and Figure 1
Reported and observed measurements
No structured measurement is attached to this Claim.
Assessments (0)
No immutable Assessment has been published for this Claim yet.
