Unverified50% confidenceBenchmarkExact time
In the experiment, the 27B model scored 61.0% on LiveMath when skills were written by the 4B model, whereas it scored 56.3% when it wrote the skills itself
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn mixed-precision training, each FP16 parameter occupies 2 bytes of GPU memory, and each FP32 parameter occupies 4 bytes of GPU memory.63% similarUnverified2.4万亿参数模型在BF16精度下考虑KV Cache和激活值后实际约需1.4-1.5TB显存62% similarUnverifiedMoE模型中'35B'指总参数量，每个token只激活少量Expert（通常2-4个），实际计算量可能仅相当于6-7B的Dense模型62% similarUnverified早期Codex在HumanEval基准测试中以37%的准确率独立解决编程题，达到当时最高水平62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/828181API
curl https://kongchang.com/api/v1/knowledge/claims/828181MCP
get_claim(id=828181)