[ AI ] IBM granite, NVIDIA robotics, IBM Langflow, watsonx, LMCache, LiteRT, Self driving car, Mamba model, RAG
IBM granite: Time series from sensor, stock market
* Forcast* Anomaly detection - classification high medium risk, data synthesis
Embedding AI - 1M params - edge device - small AI
Send insight
---
NVIDIA
* Jackson Thor robotic
* Nemo
----
* IBM Langflow : open source
* Watsonx (n8n)
Enterprise : guardrails / log, tool using policy
-----
LMcache : open source
KC cache
Prompt -> key value (RAGs)
Same context ( e.g. same docs)
Reduce operating costs
CacheGen : make cache cheap to move and store
CacheBlend : reuse chunk, not just prefix
----
Google มี framework ที่ optimize AI บน mobile phone นั่นคือ LiteRT : google pytorch
Quantisation เท่าไหร่บน device นี้
Model explorer: quantize แล้ว error < 5%
XMNpack ใช้ CPU ลดลง
Google AI edge portal : benchmark AI on real device
----
Self driving car
* Nvidia Alpamayo 2 model
* Carla simulation
Random forest -> multiple scenarios -> json -> LLM
Car talk to mobile phone (cross the road)
-----
Mamba model (recursive) - alternative to transformer
รันได้ใน
Quantisation เท่าไหร่บน device นี้
Model explorer: quantize แล้ว error < 5%
XMNpack ใช้ CPU ลดลง
Google AI edge portal : benchmark AI on real device
----
Self driving car
* Nvidia Alpamayo 2 model
* Carla simulation
Random forest -> multiple scenarios -> json -> LLM
Car talk to mobile phone (cross the road)
-----
Mamba model (recursive) - alternative to transformer
รันได้ใน
ความคิดเห็น