More qubits v.s. Better qubits #443

#ideathon-quantumcomputing

#ideathon-quantumcomputing

Ideathon

 #ideathon-quantum

#ideathon-quantum

Ideathon

Jay Gambetta announced the "first European quantum data center is coming to Ehningen, Germany in 2024" yesterday. IBM Quantum also unveiled the Osprey chip with up to 433 qubits at the end of 2022. Yes, we see more qubits, and for quantum computation, we also prefer better qubits. Today, in arXiv, a paper (https://arxiv.org/abs/2306.03939v1) focuses on the "IBM Quantum System One" in Ehningen and finds up to seven qubits are capable to violate the local hidden variables model.

0 comments

CC0 (Public Domain)

Can quantum computer substantially boost the LLM training? #300

#ideathon-quantumcomputing

#ideathon-quantumcomputing

Ideathon

#ideathon-quantum

#ideathon-quantum

Ideathon

 #ideathon-LLM

#ideathon-LLM

Ideathon

LLMs, e.g. chatGPT really caused a sensation recently. But, the training cost also excludes most companies and players. Can we train LLMs using quantum computers? How about the complexity? In recent arxiv paper (2303.03428, https://arxiv.org/abs/2303.03428v2), the authors propose an efficient solution for generic (or stochastic) gradient descent algorithms, scaling as O(T^2*polylog(n)), where T is the number of iteration and n is the number of parameter, as long as the models are both sufficiently dissipative and sparse, with small learning rates. The authors also benchmark the solution using ResNet with different parameters and find it's possible to obtain quantum enhancement after model pruning. Even though, we do not know about the parameters in GPT-3.5 or GPT-4, but if it's dissipative and sparse just like ResNet, it's possible to REALLY decrease the hugh cost of training. It's really a promising application for quantum computation.

1 comments

CC0 (Public Domain)