Unverified50% confidenceFactExact time
Replay data refers to general dialogue samples mixed in at a certain ratio during domain fine-tuning, the purpose of which is to prevent the model from overfitting on narrow-domain data and losing its general dialogue ability (catastrophic forgetting).
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRetention is one of the core metrics for evaluating language model state updates, referring to whether a model, after completing target field updates, can avoid modifying or forgetting fields that the conversation did not ask to change.69% similarUnverifiedExisting evaluation benchmarks tend to fix scenarios and phrasing, allowing models to pass tests by memorizing standard answers while failing in real interactions where phrasing varies—a phenomenon known as benchmark overfitting.69% similarUnverifiedMulti-turn dialogue is implemented by re-inputting all previous conversation history to the model with each request, rather than the model having genuine memory capability.68% similarVerifiedIn multi-turn conversations, before each new response is generated, the system must re-input all prior conversation history into the model, causing token consumption to accumulate and grow with each turn68% similarVerifiedLarge models do not have memory capability by default; a ChatMemory mechanism must be introduced to remember the content of past conversation turns.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1018231API
curl https://kongchang.com/api/v1/knowledge/claims/1018231MCP
get_claim(id=1018231)