一、Llama 4实测(真是水平如何)
-
strawberry中有几个r,回答:2个R

-
9.9与9.11哪个大

-
“Llama 4”反着输出结果:4 amallL

-
表格抽取,extract Swin-T/TNT-S、Transformer ‘s Throughput -
Llama 4给出了SwinV2-T的结果


-
Llama 4编程,Llama 4 Maverick——Python六边形测试失败

-
来自karminski的Llama 4编程能力真是总结:402B的Llama-4-maverick相当于QwQ-32B水平,选DeepSeek-V3-0324更香。
https://x.com/karminski3

-
Llama4 编程评测排名

-
Llama4 长上下文深度理解效果is bad


网友整理了Llama 4三个不同尺寸模型本地部署所需要的资源,仅3台512GB的M3 Ultra就本地拥有一个2T参数的Llama 4 Behemoth多模态模型,冲!

三、Llama 4体验link(不想本地,还想免费玩)
英伟达:
https://build.nvidia.com/meta/llama-4-maverick-17b-128e-instruct
OpenRouter:
https://openrouter.ai/chat?models=meta-llama/llama-4-maverick:free
竞技场:
https://lmarena.ai/ -> llama-4-maverick-03-26-experimental
抱抱脸:
https://huggingface.co/spaces/openfree/Llama-4-Maverick-17B-Research
https://x.com/sagnikPatra19/status/1908974601566138581
llama4 code: https://x.com/karminski3/status/1909029187392086221
code: https://aider.chat/docs/leaderboards/
llama4 长上下文: https://www.reddit.com/r/singularity/comments/1jsxpjc/fictionlivebench_for_long_context_deep/
(文:PaperAgent)