BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240626T180034Z
LOCATION:Level 2 Lobby
DTSTART;TZID=America/Los_Angeles:20240625T180000
DTEND;TZID=America/Los_Angeles:20240625T190000
UID:dac_DAC 2024_sess236_RESEARCH1048@linklings.com
SUMMARY:Hyft: A Reconfigurable Softmax Accelerator with Hybrid Numeric For
 mat for both Training and Inference
DESCRIPTION:Work-in-Progress Poster\n\nTianhua Xia (Independent) and Sai Q
 ian Zhang (New York University)\n\nThe attention mechanism is a pivotal el
 ement within the Transformer architecture, making a substantial contributi
 on to its exceptional performance. Within this attention mechanism, Softma
 x is an imperative component that enables the model to assess the degree o
 f correlation between various segments of the input. Yet, prior research h
 as shown that Softmax operations can significantly increase processing lat
 ency and energy consumption in the Transformer network due to their intern
 al nonlinear operations and data dependencies.\nIn this work, we proposed~
 \textit{Hyft}, a hardware efficient floating point Softmax accelerator for
  both training and inference. Hyft aims to reduce the implementation cost 
 of different nonlinear arithmetic operations by adaptively converting inte
 rmediate results into the most suitable numeric format for each specific o
 peration, leading to reconfigurable accelerator with hybrid numeric format
 .\n\nTopic: AI, Autonomous Systems, Cloud, Design, EDA, Embedded Systems, 
 IP, Security
END:VEVENT
END:VCALENDAR