tollMa

190并且身材很曼妙/喜欢有生命力的美女/

[tollMa @zzbx3]

最近看片的两个性癖
1.男人不受控制的被撸出来
2.女人被强制喷出来

[tollMa @zzbx3]

@tibo_maker My Codex Usage page shows only about 800M tokens used, yet my entire weekly 20× Pro allowance was wiped out. This is clearly a serious bug. Support keeps giving generic explanations instead of investigating the actual usage logs. Please fix this immediately！

[tollMa @zzbx3]

@thsottiaux My Codex Usage page shows only about 800M tokens used, yet my entire weekly 20× Pro allowance was wiped out. This is clearly a serious bug. Please fix this immediately and review the account: [email protected].

[tollMa @zzbx3]

@cherylnatsu 谁在本地搞ai啊...不都是ssh上去吗？不懂ai可以不发言

[tollMa @zzbx3]

@Donuts_dog0 还在吗老师

[tollMa @zzbx3]

@lix728191433001 胸很大，但是腰很细那种。有图吗？

[tollMa @zzbx3]

@Donuts_dog0 面包什么时候更点4i

[Boris Cherny @bcherny]

I'm Boris and I created Claude Code. I wanted to quickly share a few tips for using Claude Code, sourced directly from the Claude Code team. The way the team uses Claude is different than how I use it. Remember: there is no one right way to use Claude Code -- everyones' setup is different. You should experiment to see what works for you!

[karminski-牙医 @karminski3]

一分钟学会估算大模型显存需求

刚有同学在评论区问我VibeVoice-ASR-9B需要多少现存运行, 于是我花点时间给大家写个特别简单的大模型用多少显存的判断方法：

首先确定大模型的参数量，比如一个模型是30B的。那就看它的精度，一般如果是16bit精度格式, 比如FP16, BF16, 那么就正好是60GB, 因为8bit一个字节, 16bit正好一个参数需要两个字节, 30B的参数正好是 30B×2Byte = 60GB. 那么举一反三, 如果是 8bit 精度, 比如 FP8, 那就正好是30GB, 如果是4bit量化格式, 那就再减半, 是15GB.

现在我们复杂一些, 上面说的是Dense模型的情况, 模型无论多大都要加载到显卡里(咱们不考虑参数卸载到内存的情况).

那MoE模型怎么算呢？MoE模型有两部分参数：共享参数和专家参数。比如DeepSeek-V3是671B总参数，但实际推理时只激活37B参数。关键在于你用什么推理框架：

[tollMa @zzbx3]

@lspjswmc 感谢隐藏🙈

[tollMa @zzbx3]

@lspjswmc 蹲

[tollMa @zzbx3]

@Cc_cccc_ccccc 那就是你卡了

[tollMa @zzbx3]

@Cc_cccc_ccccc 我听你们都不卡

[tollMa @zzbx3]

@Cc_cccc_ccccc 不好意思上哈哈

[tollMa @zzbx3]

#新头像

[tollMa @zzbx3]

@QingYinJiaYi 好腿