[KongchangAI]
ConceptReference Policy Optimization

RPO

An objective function that adds a supervised fine-tuning retention term to DPO. It mitigates the alignment tax by anchoring the probability mass of correct outputs under the reference policy.

Timeline (last 90 days)

Nov 5

After preference alignment on Qwen3.5-4B using length-normalized DPO plus an SFT retention term (the RPO objective), the performance change on the structured extraction task was -0.55 percentage points, passing the TOST equivalence test within the ±5 percentage point equivalence interval.

Unverified50%
Oct 12

The two key metrics of disaster recovery architecture are RTO (Recovery Time Objective) and RPO (Recovery Point Objective).

Unverified50%
Oct 9

RTO (Recovery Time Objective) refers to the maximum acceptable duration from failure occurrence to service restoration, while RPO (Recovery Point Objective) refers to the maximum tolerable time window of data loss.

Unverified50%
Oct 6

Oracle's RPO at the end of fiscal year 2024 exceeded $98 billion.

Unverified50%
Oct 6

RPO (Remaining Performance Obligation) is a forward-looking revenue metric commonly used in the cloud computing industry, referring to the contracted amount that customers have signed but not yet recognized.

Unverified50%
Sep 17

RPO (Recovery Point Objective) measures the acceptable data loss time window and determines backup frequency.

Unverified50%
Sep 15

For business scenarios with extremely high data integrity requirements, such as finance and e-commerce, the recovery granularity of PITR directly determines the maximum data loss (RPO) in the event of a disaster.

Unverified50%

All Facts (7)

Source Articles