Buy Crypto
Pay with
USD
USD
Buy & Sell
Hot
Supports Visa, Mastercard, SEPA & more
P2P
0 Fees
Flexible trading, zero fees
Gate Card
Use your crypto for payments worldwide
Markets
Trade
Basic
Advanced
Futures
Futures
Access hundreds of perpetual contracts
CFD
Gold
One platform for global traditional assets
Options
Hot
Trade European-style vanilla options
Unified Account
Maximize your capital efficiency
Demo Trading
Introduction to Futures Trading
Learn the basics of futures trading
Futures Events
Join events to earn rewards
Demo Trading
Use virtual funds to practice risk-free trading
U.S. Stocks
CFD
U.S. stock CFD derivatives
Futures
High leverage, 24/7 trading
Tokenized Stocks
Backed by real stock assets
GUSD
Mint GUSD for Treasury RWA yields
Earn
Launch
CandyDrop
tag
Collect candies to earn airdrops
Launchpool
Quick staking, earn potential new tokens
HODLer Airdrop
Hold GT and get massive airdrops for free
Pre-IPOs
Unlock full access to global stock IPOs
Alpha Points
Trade on-chain assets and earn airdrops
Futures Points
Earn futures points and claim airdrop rewards
Investment
Square
Square
Post, share, and explore crypto trends
Live
moments live
Live crypto market analysis
Chat
Chat with crypto traders
News
What is happening in crypto
flower_head
More
Promotions
AI
Others
TradFi
Rewards
Square
Latest
Hot
News
Profile

Δ-Mem: Efficient Online Memory for Large Language Models

MeNews
robot
Abstract generation in progress
ME News Report, May 16 (UTC+8), researchers introduced Δ-Mem, an efficient online memory system specifically designed for large language models. This system significantly reduces memory usage by storing and updating only the incremental changes in model activations, rather than the full activation states. Experiments show that Δ-Mem can reduce memory consumption by up to 70%, while maintaining nearly the same quality of model outputs. This approach helps deploy and run large-scale language models in resource-constrained environments, enhancing their feasibility in online inference and continual learning scenarios. (Source: AiHot)
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 10
  • 2
  • Share
Comment
Add a comment
Add a comment
DeltaSmile
· 1h ago
Lossless output quality is the most critical, and many compression schemes sacrifice too much precision.
View OriginalReply0
PickingUpAirdropsInTheFog
· 6h ago
The ability to continuously learn and improve is underestimated; it is very important for personalized models.
View OriginalReply0
InvisibleMarketMaker
· 6h ago
A 70% reduction in memory is indeed impressive, but in online scenarios, could the computational overhead of incremental updates become a new bottleneck?
View OriginalReply0
ColdStartUnderTheAurora
· 6h ago
Someone is finally seriously addressing the memory wall issue of LLMs; looking forward to the follow-up work.
View OriginalReply0
HotAirBalloonCrossingMountains
· 6h ago
It feels like sparse attention can be stacked and used together, with dual compression.
View OriginalReply0
PettyLp
· 6h ago
The online memory system is very accurately positioned, filling the gap in the reasoning phase.
View OriginalReply0
CheckTheBlockchainBefore
· 6h ago
How compatible are parameter-efficient fine-tuning methods like LoRA and curiosity?
View OriginalReply0
ProtocolPaladin
· 6h ago
If this direction is open-sourced, the community can come up with many creative ideas.
View OriginalReply0
PerpPessimist
· 6h ago
What evaluation dataset is used for the experiment? GLUE or more complex reasoning tasks.
View OriginalReply0
TokenomicsMechanic
· 6h ago
Is 70% the maximum or the average? Do different model sizes show significant performance differences?
View OriginalReply0
View More

  • Pinned

Sitemap