Viewing enriched Twitter post
Just found out that flashattention use look up table+linear approximation to accelerate softmax. We have tried similar strategies on work down at MSR two years before on A100. And get over 20% end-to-end speed up of self attention. If you are interested in how to parallel look up table and linear approximation to replace non-linear function. Please take a look at our paper. Paper: https://t.co/ShiHZS6RKz #FlashAttention4 #AI #Mlsys
{
"media": [
{
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2041554712386764983/media_0.jpg",
"media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2041554712386764983/media_0.jpg",
"type": "photo",
"filename": "media_0.jpg"
},
{
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2041554712386764983/media_1.jpg",
"media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2041554712386764983/media_1.jpg",
"type": "photo",
"filename": "media_1.jpg"
}
],
"processed_at": "2026-04-08T00:23:53.271633",
"pipeline_version": "2.0"
}{
"type": "tweet",
"id": "2041554712386764983",
"url": "https://x.com/GT_HaoKang/status/2041554712386764983",
"twitterUrl": "https://twitter.com/GT_HaoKang/status/2041554712386764983",
"text": "Just found out that flashattention use look up table+linear approximation to accelerate softmax.\n\nWe have tried similar strategies on work down at MSR two years before on A100. And get over 20% end-to-end speed up of self attention.\n\nIf you are interested in how to parallel look up table and linear approximation to replace non-linear function. Please take a look at our paper.\n\nPaper: https://t.co/ShiHZS6RKz\n\n#FlashAttention4 #AI #Mlsys",
"source": "Twitter for iPhone",
"retweetCount": 6,
"replyCount": 0,
"likeCount": 49,
"quoteCount": 0,
"viewCount": 4655,
"createdAt": "Tue Apr 07 16:32:28 +0000 2026",
"lang": "en",
"bookmarkCount": 36,
"isReply": false,
"inReplyToId": null,
"conversationId": "2041554712386764983",
"displayTextRange": [
0,
279
],
"inReplyToUserId": null,
"inReplyToUsername": null,
"author": {
"type": "user",
"userName": "GT_HaoKang",
"url": "https://x.com/GT_HaoKang",
"twitterUrl": "https://twitter.com/GT_HaoKang",
"id": "1717265172593889281",
"name": "Hao Kang",
"isVerified": false,
"isBlueVerified": true,
"verifiedType": null,
"profilePicture": "https://pbs.twimg.com/profile_images/1923449943777476608/-CQFDAKi_normal.jpg",
"coverPicture": "",
"description": "",
"location": "Atlanta",
"followers": 229,
"following": 209,
"status": "",
"canDm": false,
"canMediaTag": true,
"createdAt": "Wed Oct 25 19:41:43 +0000 2023",
"entities": {
"description": {
"urls": []
},
"url": {}
},
"fastFollowersCount": 0,
"favouritesCount": 730,
"hasCustomTimelines": true,
"isTranslator": false,
"mediaCount": 20,
"statusesCount": 97,
"withheldInCountries": [],
"affiliatesHighlightedLabel": {},
"possiblySensitive": false,
"pinnedTweetIds": [
"2023804981954973904"
],
"profile_bio": {
"description": "@GeorgiaTech cs phd student | efficient machine learning | ex-intern @MSFTResearch @UCLA @ZJU_China",
"entities": {
"description": {
"hashtags": [],
"symbols": [],
"urls": [],
"user_mentions": [
{
"id_str": "0",
"indices": [
0,
12
],
"name": "",
"screen_name": "GeorgiaTech"
},
{
"id_str": "0",
"indices": [
70,
83
],
"name": "",
"screen_name": "MSFTResearch"
},
{
"id_str": "0",
"indices": [
84,
89
],
"name": "",
"screen_name": "UCLA"
},
{
"id_str": "0",
"indices": [
90,
100
],
"name": "",
"screen_name": "ZJU_China"
}
]
},
"url": {
"urls": [
{
"display_url": "haokang-timmy.github.io",
"expanded_url": "https://haokang-timmy.github.io/",
"indices": [
0,
23
],
"url": "https://t.co/4Ff8hXf07u"
}
]
}
}
},
"isAutomated": false,
"automatedBy": null
},
"extendedEntities": {
"media": [
{
"allow_download_status": {
"allow_download": true
},
"display_url": "pic.twitter.com/fcuBSgPJ45",
"expanded_url": "https://twitter.com/GT_HaoKang/status/2041554712386764983/photo/1",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"orig": {
"faces": []
}
},
"id_str": "2041552473333395456",
"indices": [
280,
303
],
"media_key": "3_2041552473333395456",
"media_results": {
"id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARxVDShrW3AACgACHFUPMb1WgLcAAA==",
"result": {
"__typename": "ApiMedia",
"id": "QXBpTWVkaWE6DAABCgABHFUNKGtbcAAKAAIcVQ8xvVaAtwAA",
"media_key": "3_2041552473333395456"
}
},
"media_url_https": "https://pbs.twimg.com/media/HFUNKGtbcAA2TwN.png",
"original_info": {
"focus_rects": [
{
"h": 706,
"w": 1261,
"x": 133,
"y": 0
},
{
"h": 706,
"w": 706,
"x": 448,
"y": 0
},
{
"h": 706,
"w": 619,
"x": 492,
"y": 0
},
{
"h": 706,
"w": 353,
"x": 625,
"y": 0
},
{
"h": 706,
"w": 1394,
"x": 0,
"y": 0
}
],
"height": 706,
"width": 1394
},
"sizes": {
"large": {
"h": 706,
"w": 1394
}
},
"type": "photo",
"url": "https://t.co/fcuBSgPJ45"
},
{
"allow_download_status": {
"allow_download": true
},
"display_url": "pic.twitter.com/fcuBSgPJ45",
"expanded_url": "https://twitter.com/GT_HaoKang/status/2041554712386764983/photo/1",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"orig": {
"faces": []
}
},
"id_str": "2041554003280994307",
"indices": [
280,
303
],
"media_key": "3_2041554003280994307",
"media_results": {
"id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARxVDoyjVyADCgACHFUPMb1WgLcAAA==",
"result": {
"__typename": "ApiMedia",
"id": "QXBpTWVkaWE6DAABCgABHFUOjKNXIAMKAAIcVQ8xvVaAtwAA",
"media_key": "3_2041554003280994307"
}
},
"media_url_https": "https://pbs.twimg.com/media/HFUOjKNXIAMy-wq.png",
"original_info": {
"focus_rects": [
{
"h": 374,
"w": 668,
"x": 0,
"y": 29
},
{
"h": 436,
"w": 436,
"x": 0,
"y": 0
},
{
"h": 436,
"w": 382,
"x": 0,
"y": 0
},
{
"h": 436,
"w": 218,
"x": 7,
"y": 0
},
{
"h": 436,
"w": 668,
"x": 0,
"y": 0
}
],
"height": 436,
"width": 668
},
"sizes": {
"large": {
"h": 436,
"w": 668
}
},
"type": "photo",
"url": "https://t.co/fcuBSgPJ45"
}
]
},
"card": null,
"place": {},
"entities": {
"hashtags": [
{
"indices": [
412,
428
],
"text": "FlashAttention4"
},
{
"indices": [
429,
432
],
"text": "AI"
},
{
"indices": [
433,
439
],
"text": "Mlsys"
}
],
"symbols": [],
"urls": [
{
"display_url": "arxiv.org/pdf/2412.08585",
"expanded_url": "https://arxiv.org/pdf/2412.08585",
"indices": [
387,
410
],
"url": "https://t.co/ShiHZS6RKz"
}
],
"user_mentions": []
},
"quoted_tweet": {
"type": "tweet",
"id": "2029572926035702061",
"url": "https://x.com/togethercompute/status/2029572926035702061",
"twitterUrl": "https://twitter.com/togethercompute/status/2029572926035702061",
"text": "Together Research has produced FlashAttention, ATLAS, ThunderKittens and more.\n\nThis week at AI Native Conf: seven more releases, all coming to production soon.\n\nThread →\n\n#ainativeconf #ainativecloud https://t.co/XXIXMRRiLe",
"source": "Twitter for iPhone",
"retweetCount": 14,
"replyCount": 1,
"likeCount": 106,
"quoteCount": 10,
"viewCount": 29585,
"createdAt": "Thu Mar 05 15:01:07 +0000 2026",
"lang": "en",
"bookmarkCount": 47,
"isReply": false,
"inReplyToId": null,
"conversationId": "2029572926035702061",
"displayTextRange": [
0,
200
],
"inReplyToUserId": null,
"inReplyToUsername": null,
"author": {
"type": "user",
"userName": "togethercompute",
"url": "https://x.com/togethercompute",
"twitterUrl": "https://twitter.com/togethercompute",
"id": "1592266692528197632",
"name": "Together AI",
"isVerified": false,
"isBlueVerified": false,
"verifiedType": "Business",
"profilePicture": "https://pbs.twimg.com/profile_images/2028879663221882886/sMzEKnKc_normal.jpg",
"coverPicture": "https://pbs.twimg.com/profile_banners/1592266692528197632/1772467486",
"description": "",
"location": "San Francisco, CA",
"followers": 54195,
"following": 396,
"status": "",
"canDm": false,
"canMediaTag": true,
"createdAt": "Mon Nov 14 21:23:18 +0000 2022",
"entities": {
"description": {
"urls": []
},
"url": {}
},
"fastFollowersCount": 0,
"favouritesCount": 2074,
"hasCustomTimelines": true,
"isTranslator": false,
"mediaCount": 774,
"statusesCount": 2576,
"withheldInCountries": [],
"affiliatesHighlightedLabel": {},
"possiblySensitive": false,
"pinnedTweetIds": [
"2030030777870684358"
],
"profile_bio": {
"description": "Accelerate inference, model shaping, and pre-training on a research-optimized platform.",
"entities": {
"description": {
"hashtags": [],
"symbols": [],
"urls": [],
"user_mentions": []
},
"url": {
"urls": [
{
"display_url": "together.ai",
"expanded_url": "http://together.ai",
"indices": [
0,
23
],
"url": "https://t.co/ymnk7vEStt"
}
]
}
}
},
"isAutomated": false,
"automatedBy": null
},
"extendedEntities": {
"media": [
{
"allow_download_status": {
"allow_download": true
},
"display_url": "pic.twitter.com/XXIXMRRiLe",
"expanded_url": "https://twitter.com/togethercompute/status/2029572926035702061/photo/1",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"orig": {
"faces": []
}
},
"id_str": "2029569657271947264",
"indices": [
201,
224
],
"media_key": "3_2029569657271947264",
"media_results": {
"id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARwqetmkmxAACgACHCp90rYbsS0AAA==",
"result": {
"__typename": "ApiMedia",
"id": "QXBpTWVkaWE6DAABCgABHCp62aSbEAAKAAIcKn3SthuxLQAA",
"media_key": "3_2029569657271947264"
}
},
"media_url_https": "https://pbs.twimg.com/media/HCp62aSbEAAkRZl.jpg",
"original_info": {
"focus_rects": [
{
"h": 630,
"w": 1125,
"x": 75,
"y": 0
},
{
"h": 630,
"w": 630,
"x": 435,
"y": 0
},
{
"h": 630,
"w": 553,
"x": 474,
"y": 0
},
{
"h": 630,
"w": 315,
"x": 593,
"y": 0
},
{
"h": 630,
"w": 1200,
"x": 0,
"y": 0
}
],
"height": 630,
"width": 1200
},
"sizes": {
"large": {
"h": 630,
"w": 1200
}
},
"type": "photo",
"url": "https://t.co/XXIXMRRiLe"
}
]
},
"card": null,
"place": {},
"entities": {
"hashtags": [
{
"indices": [
172,
185
],
"text": "ainativeconf"
},
{
"indices": [
186,
200
],
"text": "ainativecloud"
}
],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"quoted_tweet": null,
"retweeted_tweet": null,
"isLimitedReply": false,
"communityInfo": null,
"article": null
},
"retweeted_tweet": null,
"isLimitedReply": false,
"communityInfo": null,
"article": null
}