Unverified50% confidenceBenchmarkExact time
On the hard tier, recall@5 improved from 0.000 to 0.955 (under default blending weights), and reached 1.000 when using the lower-bound variant.
1
Sources
50%
Confidence
Long-term
Relevance
9/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThroughput peaks at approximately 115 tokens per second around 5 draft tokens and then falls back, not varying linearly with the acceptance rate.73% similarUnverifiedOn dependency tasks, memory raised the success rate from a leakage-verified baseline of 0.00 to 0.55–1.00.72% similarUnverified1个基点等于0.01个百分点，500个基点即利率上升5个百分点70% similarUnverifiedThe ratio of input to output tokens surged from 4.52:1 to 13:1.68% similarUnverifiedThe output of the low-rank residual adapter is added back to the original activation values in residual form, and its size is typically only 0.1%–1% of the original layer weights.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/940003API
curl https://kongchang.com/api/v1/knowledge/claims/940003MCP
get_claim(id=940003)