Side effect of blocking Chinese firms from buying the best NVIDIA cards: top models are now explicitly being trained to work well on older/cheaper GPUs. The new SoTA model from @Kimi_Moonshot uses plain old BF16 ops (after dequant from INT4); no need for expensive FP4 support.
🚀 "Quantization is not a compromise — it's the next paradigm." After K2-Thinking's release, many developers have been curious about its native INT4 quantization format. 刘少伟, infra engineer at @Kimi_Moonshot and Zhihu contributor, shares an insider's view on why this choice…
Finally wrapped up the hardware half of Nand2Tetris, took me a week. It had been on my to-do list for way longer than I’d like to admit.
Deepseek engineers so cracked they bypassed cuda
United States Trends
- 1. Michigan 132K posts
- 2. Ohio State 52.2K posts
- 3. Underwood 7,926 posts
- 4. Ryan Day 6,895 posts
- 5. #GoBucks 10.4K posts
- 6. Stoops 5,194 posts
- 7. Sherrone Moore 2,624 posts
- 8. Julian Sayin 5,057 posts
- 9. #TheGame 4,561 posts
- 10. Clemson 7,674 posts
- 11. #GoBlue 9,554 posts
- 12. Jeremiah Smith 8,527 posts
- 13. Bo Jackson 2,511 posts
- 14. Fortnite 210K posts
- 15. Brutus 16.7K posts
- 16. Vicario 4,033 posts
- 17. Beamer 1,593 posts
- 18. Kentucky 21K posts
- 19. Ann Arbor 4,946 posts
- 20. Malachi Toney 3,348 posts
Something went wrong.
Something went wrong.