FRAMEWIREIndonesiaUpdated Aug 14Live wire
0:00 / 0:00

Not off to a good start.

This is using UnslothAI's application and recommended 19GB, Qwen 3.8 27B model. 12.1 tok/s and many times needing to tell it to continue after numerous response reached token limit and after having to turn off all thinking. I haven't used Unsloth much

Rich · Atom Tan StudioAug 14
0:00 / 0:00

Qwen 3.8 27B Tetris completed🙌🙌

Thank you for the RTA match. 👉The cause of the repeated mistakes in tool call was probably thinking. If you turn it off, you can finish the race in one shot! I plan on using it carefully after this.

junAug 149
0:00 / 0:00

Qwen 3.8 27b Pagoda test compared to 3.8 max and 3.6.

BF16 results! Looks great!

Loktar 🇺🇸Aug 1494
0:00 / 0:00

In a span of 3 days: Grok 4.6, Muse Glimmer, Nemotron 3.5 Lightning and DeepSeek V4 Pro 0813 dropped.

Few advances worth noting: First post-training on these checkpoints are showing incredible results in a short amount of time. It could be overfitting to benchmarks but this kind

Caleb EomAug 142