Unverified50% confidenceFactExact time
RWKV由Peng Bo于2023年提出，通过线性注意力加时间衰减的WKV算子实现类似效果，已扩展到14B参数规模
1
Sources
50%
Confidence
Long-term
Relevance
8/25/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedvLLM was open-sourced by UC Berkeley in 2023 and increased LLM inference throughput by 2–4x through its PagedAttention technology.66% similarUnverifiedYaRN由Peng等人于2023年提出，通过对RoPE旋转频率进行非均匀缩放来扩展上下文窗口64% similarUnverifiedGPTQ introduced layer-wise calibration datasets in 2022, and AWQ in 2023 reduced precision loss by protecting weight channels corresponding to salient activations.63% similarUnverifiedThe k-quant quantization scheme was introduced in mid-2023 by ikawrakow, a core developer of the llama.cpp project61% similarUnverifiedMulti-Head Latent Attention（MLA）由 DeepSeek-V2 于2024年提出60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/798162API
curl https://kongchang.com/api/v1/knowledge/claims/798162MCP
get_claim(id=798162)