🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

it appears there is no straightforward way to access the attentions of Qwen2.5 VL vision encoder since it uses a fused attention mechanism

tuple in mandarin is 元组

variable in mandarin is 变量

dimension in mandarin is 维度

function in mandarin is 函数

parameter in mandarin is 参数

instance in mandarin is 实例

vector in mandarin is 向量

token in mandarin is 词元

feed forward network in mandarin is 前馈网络

with the built-in function vars, I can view all the values of all attributes too

我用内置函数vars可以查看所有属性的值

autoregressive in mandarin is 自回归

what happens when one over-relies on vibe coding is that one stops practicing problem solving, and figuring out how to actually solve the problem the cognition is offloaded to the llm it's stupid

i just experienced this myself

when crafting prompts, is it better to have more concise prompts or more verbose prompts?

rotary position embedding in mandarin is 旋转位置编码

wow, i crafted a prompt in which the llm doesn't give me the answer, but instead hints/guides me to it a blurse, though like 90% blessing and 10% curse (no more neatly presented dinner plate)

talk about cognitive offloading; been a bit since i've exercised my cognition

for some reason, gemini split singular chinese characters into their components, and indivually gave their pinyin