Unverified50% confidenceBenchmarkExact time
MTP推测解码在Apple Silicon编码基准上把每秒Token数从50.2提到95，接近翻倍
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/26/2026
First Seen
Valid until: 12/25/2026
Sources
Related Entities
Related Claims
Unverified在 RTX Pro 6000（96GB 显存）上，官方 FP8 版配合内置 MTP 投机解码可跑到约 120 tokens/秒，质量与 16-bit 几乎无差异67% similarUnverifiedMTP-LX with 4bit quantization achieves 40+ tokens per second generation speed on Mac, approximately double the speed of the MLX 6bit solution67% similarUnverifiedMicroGPT 在苹果M5芯片上跑出了每秒千万级token（10M tps）的吞吐量61% similarVerified在RTX 5090上，Muse Glimmer生成速度从每秒75个Token提升到233个Token，提速约3.1倍61% similarUnverifiedUP主在4G显存电脑上实测MiniCPM-2B生成速度达每秒57个token，首字生成仅0.22秒61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/952783API
curl https://kongchang.com/api/v1/knowledge/claims/952783MCP
get_claim(id=952783)