Xiaomi and TileRT_AI release MiMo-V2.5-Pro-UltraSpeed, claiming over 1,000 tokens per second on a 1-trillion-parameter MoE
The release uses FP4 quantization and DFlash speculative decoding.
133001714349.3K
Sentiment
Sentiment building, check back later.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS3.6KBOOKMARKS6LIKES21
Gavin Baker@GavinSBaker
@zephyr_z9 @fi56622380 Not the first useful one but certainly nice to see!
Zephyr@zephyr_z9
This is super big I think this is the first useful speculative decoding method deployed on a big quasi frontier model Massive unlock @fi56622380
5hViews 3.6KLikes 21Bookmarks 6
RETWEETS14
Zephyr@zephyr_z9
This is super big I think this is the first useful speculative decoding method deployed on a big quasi frontier model Massive unlock @fi56622380
6hViews 44.6KLikes 274Bookmarks 136
You Jiacheng@YouJiacheng
super cool
1hViews 1KLikes 5Bookmarks 1