探索 | Thread Easy - Twitterスレッドを展開 | リーダー・要約・作成

we are just s used to expecting these runs to be RNG that if something doesn't work, going to fp16 isn't an idea we try. though good ones understand fp32 supremacy and use it already when casting layernorm, attention reductions, logits => softmax even when weights are bf16.

RL razor paper link for the information shared in ss - https://t.co/uq9f816ng5 highlighted by @ChinmayKak sometime back to me

tokenbender

Sat Nov 01 03:45:49

we are just s used to expecting these runs to be RNG that if something doesn't work, going to fp16 isn't an idea we try. though good ones understand fp32 supremacy and use it already when casting layernorm, attention reductions, logits => softmax even when weights are bf16.

RL razor paper link for the information shared in ss - https://t.co/uq9f816ng5 highlighted by @ChinmayKak sometime back to me

tokenbender

Sat Nov 01 03:45:49

I thought I was super money-motivated — until I ran a small business and saw how much it eats up your time, focus, and peace. That’s when I started second-guessing it.

Growth Coach｜Helping creators build their personal brand on X 增长教练｜帮助创作者在X上打造个人品牌公众号：PandaTalk8 X 增长群群主

Mr Panda

Sat Nov 01 03:42:11

Fascinating analysis. Trump had to take one step back on Chyna to focus on crushing the immediate threat to the overall campaign (Canada). The day of the rake is coming closer…

We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

Sat Nov 01 03:41:10

RT @marc_louvion: I made $66,040 in October 2025. 🧑💻 CodeFast — $20.7K ⚡️ ShipFast — $16.8K 📈 DataFast — $16.3K ⭐️ TrustMRR — $8.6K 🐥 Twi…

🧑‍💻 https://t.co/Y30jsaHwz9 $20K/m ⚡️ https://t.co/vatLDmi9UG $17K/m 📈 https://t.co/3EDxln5mdi $16K/m ⭐️ https://t.co/MZc8tG9xWi $8K/m 🧬 https://t.co/SfrVXVtmdA $.5K/m 🍜 https://t.co/r07EpGSYJ2 $0K/m 🧾 https://t.co/7olaOzV8Xd $0/m +18 https://t.co/4zCWHGJp1S

Marc Lou

Sat Nov 01 03:40:12

I still remember back in 2018 when I was doing my PhD in AI systems research, our entire lab didn’t even have 10 4090 GPUs. Now I can access thousands of H100s, H200s, and B200s, wild.

Co-founder & CTO @hyperbolic_labs cooking fun AI systems. Prev: OctoAI (acquired by @nvidia) building Apache TVM, PhD @ University of Washington.

Yuchen Jin

Sat Nov 01 03:36:56

探索

Newest first — browse tweet threads

探索

Newest first — browse tweet threads

we are just s used to expecting these runs to be RNG that if something doesn't work, going to fp16 isn't an idea we try. though good ones understand fp32 supremacy and use it already when casting layernorm, attention reductions, logits => softmax even when weights are bf16.

we are just s used to expecting these runs to be RNG that if something doesn't work, going to fp16 isn't an idea we try. though good ones understand fp32 supremacy and use it already when casting layernorm, attention reductions, logits => softmax even when weights are bf16.

I thought I was super money-motivated — until I ran a small business and saw how much it eats up your time, focus, and peace. That’s when I started second-guessing it.

Fascinating analysis. Trump had to take one step back on Chyna to focus on crushing the immediate threat to the overall campaign (Canada). The day of the rake is coming closer…

RT @marc_louvion: I made $66,040 in October 2025. 🧑💻 CodeFast — $20.7K ⚡️ ShipFast — $16.8K 📈 DataFast — $16.3K ⭐️ TrustMRR — $8.6K 🐥 Twi…

I still remember back in 2018 when I was doing my PhD in AI systems research, our entire lab didn’t even have 10 4090 GPUs. Now I can access thousands of H100s, H200s, and B200s, wild.