Primary sourcearXiv cs.AI
View sourceEvent ·Multi-source
FUTRANSFORMER:Hopper GPU 上的 FP8 prefill 与 INT4 decode 推理实现
FUTRANSFORMER 是一个面向 NVIDIA Hopper GPU 的 CUDA 实现,针对 Llama 风格 Transformer 块做推理优化。prefill 阶段采用 FP8 输出与延迟缩放,在单张 H100 上最高比编译后的 PyTorch FP8 快 1.48 倍,达到 1.29 PFLOP/s。decode 阶段用 INT4 投影权重和 INT4 KV-cache,在 8B 与 70B 形状、batch 1-32 范围内比 PyTorch BF16/INT4 块快 1.3-2.4 倍。Llama-3.1-8B 单步 decode 耗时 2.8 ms,而编译版 transformers 实现需要 8.1 ms。量化质量方面,WikiText-2 上 INT4 权重路径将困惑度从 6.24 升至 6.99(GPTQ),旋转 INT4 cache 额外增加 0.20。
Current read
一篇把 Llama 推理在 H100 上抠到极致的 CUDA 论文,decode 比 PyTorch 快最多 2.4 倍,做推理优化的值得看代码思路。
11 sources42° AI heatLast updated 2026/10/16 17:22:01
Evidence
11 sourcesRelated source 1PyTorch
View sourceRelated source 2vLLM
View sourceRelated source 3Meta AI
View sourceRelated source 4karminski-牙医 (AI工具)
View sourceRelated source 5@atomic_chat_hq
View sourceRelated source 6AIGCLINK
View sourceRelated source 7NVIDIA AI
View sourceRelated source 8Artificial Analysis
View sourceRelated source 9marktechpost
View sourceRelated source 10ollama
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.