Kimi K3, previewed: inside the first open 3T-class model
• 6
AI & ML interests
None defined yet.
Recent Activity
View all activity
Articles
JonnaMat
updated 19
models 1 day ago
embedl/gemma-3-270m-it-FlashHead
0.3B • Updated • 2 • 3
embedl/gemma-3-1b-it-FlashHead
1.0B • Updated • 3
embedl/gemma-3-1b-it-FlashHead-W4A16
1.0B • Updated • 3
embedl/Qwen3.5-9B-W4A16-FlashHead
Image-Text-to-Text • 9B • Updated • 41 • 1
embedl/Qwen3.5-9B-FlashHead
Image-Text-to-Text • 10B • Updated • 9 • 2
embedl/Qwen3.5-4B-FlashHead
Image-Text-to-Text • 5B • Updated • 4 • 2
embedl/Qwen3.5-2B-FlashHead
Image-Text-to-Text • 2B • Updated • 6 • 2
embedl/Qwen3.5-27B-FlashHead
Image-Text-to-Text • 28B • Updated • 2
embedl/Qwen3.5-0.8B-FlashHead
Image-Text-to-Text • 0.9B • Updated • 22 • 1
embedl/Qwen3-1.7B-FlashHead
2B • Updated • 82 • 3
embedl/Qwen3-1.7B-FlashHead-W4A16
2B • Updated • 63 • 3
embedl/Qwen3-0.6B-FlashHead
0.6B • Updated • 130 • 4
embedl/Llama-3.2-3B-Instruct-FlashHead
3B • Updated • 4
embedl/Llama-3.2-3B-Instruct-FlashHead-W4A16
4B • Updated • 4
embedl/Llama-3.2-1B-Instruct-FlashHead
1B • Updated • 24 • 4
embedl/Llama-3.2-1B-Instruct-FlashHead-W4A16
1B • Updated • 6
embedl/Cosmos-Reason2-8B-W4A16-FlashHead
Image-Text-to-Text • 3B • Updated • 124 • 1
embedl/Cosmos-Reason2-32B-W4A16-FlashHead
Image-Text-to-Text • 6B • Updated • 33 • 1
embedl/Cosmos-Reason2-2B-W4A16-Edge2-FlashHead
Image-Text-to-Text • 2B • Updated • 961 • 9
Post
3142
🚗 The reasoning backbone quadruples from 8B to 32B , while the action expert remains at 2.3B!
👀 We took a closer look at the architectural evolution from nvidia/Alpamayo-1.5-10B to nvidia/Alpamayo2-Super .
Read the analysis here:
https://huggingface.co/blog/JonnaMat/alpamayo2-super
Our analysis explores some implications of this design choice, especially from a distillation perspective where keeping the expert compact could be key for efficient deployment. 🧠
👀 We took a closer look at the architectural evolution from nvidia/Alpamayo-1.5-10B to nvidia/Alpamayo2-Super .
Read the analysis here:
https://huggingface.co/blog/JonnaMat/alpamayo2-super
Our analysis explores some implications of this design choice, especially from a distillation perspective where keeping the expert compact could be key for efficient deployment. 🧠
- 3 replies