Unverified50% confidenceFactExact time
34B 80层模型比27B推理时间增加约18%
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
Qwen3.6非官方衍生模型：34B深度扩展与14B极致压缩解析
bilibiliAI观模者6/21/2026
Related Claims
Unverified在35B参数模型中GEMM内核往往占据总计算时间的60-80%69% similarUnverified目前Cursor调用的模型上下文长度仅为20万token65% similarUnverifiedH100内存子系统的动态功耗在大模型推理中占比可达整卡功耗的20%～35%64% similarUnverified如果模型跑BF16训练但DCGM_FI_PROF_PIPE_TENSOR_ACTIVE指标长期低于30%，基本可以断定算力被严重浪费63% similarUnverified对于参数量大于30B的模型，4-bit量化造成的性能损失通常在基准测试中下降1-3%，但对小模型影响更显著62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/40889API
curl https://kongchang.com/api/v1/knowledge/claims/40889MCP
get_claim(id=40889)