Unverified50% confidenceBenchmarkExact time
The unified Radix Cache added an Agentic-Aware tail-optimized LRU eviction strategy, which reduced p99 ITL by 43.9% at concurrency 32 in an agentic trajectory test on DeepSeek-V4-Pro FP4 with 4-card B300.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/3/2026
First Seen
Valid until: 1/1/2027
Sources
Related Entities
Related Claims
UnverifiedRedis 4.0引入LFU策略，使用Morris概率计数器以8位空间记录访问频次，并通过lfu-decay-time配置衰减因子60% similarUnverifiedThe models participating in this comparative evaluation include Fable 5, DXU E4GA, DXU E4 Slash N731 official release, Qwen 3 MiniMax 0720, and DeepSeek V4 Pro.58% similarUnverifiedThe base model (no LoRA) was configured with 30 steps, CFG 4, er_sde sampler, and simple scheduler57% similarUnverifiedFP8量化将模型权重从FP16压缩为8位浮点表示，在NVIDIA Hopper架构（H100/H200）上可获得原生加速，且相比INT8对精度损失更小56% similarUnverifiedThis paper adopts an INT8+bf16 mixed-precision strategy: the original model weights use INT8 quantization (via the LLM.int8() method from the bitsandbytes library), while the LoRA components use bf16 with fp32 master weights.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/969383API
curl https://kongchang.com/api/v1/knowledge/claims/969383MCP
get_claim(id=969383)