Unverified50% confidenceFactExact time
Speed figures in llmfit come from three sources: locally benchmarked results, results measured by others under similar conditions, and purely formula-based estimates.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/16/2026
First Seen
Valid until: 12/15/2026
Sources
Related Entities
Related Claims
Unverified通义千问3.6 27B在语义三次全对率和格式正确率上均领先Muse Glimmer66% similarUnverifiedIn strict format scoring tests, TwIL-LM's 1.7B model reportedly outperforms models 5 to 15 times its size.64% similarUnverifiedThe performance gap between a carefully quantized 4-bit model and the original FP16 model is typically within 1–3 percentage points.63% similarUnverified常用的量化校准算法包括MinMax、Entropy（基于KL散度）和Percentile三种63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/925040API
curl https://kongchang.com/api/v1/knowledge/claims/925040MCP
get_claim(id=925040)