Unverified50% confidenceFactExact time
After the first compaction, size estimation uses the actual messages sent; previously it overestimated by about 7x.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/26/2026
First Seen
Valid until: 1/24/2027
Sources
Related Entities
Related Claims
UnverifiedDistilBERT将参数量压缩至6600万（原版1.1亿），推理速度提升60%，精度损失不足3%63% similarUnverifiedAfter fine-tuning the 406M model specifically for the 2,000-word retrieved segment input format, the previously lost 6.30 points were fully recovered.62% similarUnverifiedCommon speedups from speculative decoding range between 1.5x and 3x, and DeepSeek's reported 30% reduction in verification volume and 4% to 7% throughput gain are relatively conservative engineering returns.62% similarUnverifiedSpeculative decoding technology makes the speed approximately 1.6 to 1.8 times faster, and in actual tests the large model accepted the small assistant's guesses approximately 79% of the time61% similarUnverifiedComparing real usage data from 7 days before and after, the author found token input dropped 62% and the most bloated session fell from 974K tokens to 267K tokens, but the author acknowledges that the work content differed between the two weeks, so the figures reflect only a trend rather than an exact year-over-year comparison, and the test was run only once.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098166API
curl https://kongchang.com/api/v1/knowledge/claims/1098166MCP
get_claim(id=1098166)