Viewing enriched Twitter post
What is speculative decoding? Speculative decoding is an inference optimization that uses a fast, small "draft" model to quickly propose several future tokens, which are then verified in parallel by a larger, slower "target" model. This process significantly speeds up the token generation of large language models (LLMs) by allowing the generation of multiple tokens per step, without sacrificing the quality of the output text Learn more at https://t.co/qYAJJ8Snin
{
"media": [
{
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2066578531010728271/media_0.jpg",
"media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2066578531010728271/media_0.jpg",
"type": "photo",
"filename": "media_0.jpg"
},
{
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2066578531010728271/media_1.jpg",
"media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2066578531010728271/media_1.jpg",
"type": "photo",
"filename": "media_1.jpg"
}
],
"processed_at": "2026-06-16T17:18:19.377107",
"pipeline_version": "2.0"
}{
"type": "tweet",
"id": "2066578531010728271",
"url": "https://x.com/NielsRogge/status/2066578531010728271",
"twitterUrl": "https://twitter.com/NielsRogge/status/2066578531010728271",
"text": "What is speculative decoding?\n\nSpeculative decoding is an inference optimization that uses a fast, small \"draft\" model to quickly propose several future tokens, which are then verified in parallel by a larger, slower \"target\" model. \n\nThis process significantly speeds up the token generation of large language models (LLMs) by allowing the generation of multiple tokens per step, without sacrificing the quality of the output text\n\nLearn more at https://t.co/qYAJJ8Snin",
"source": "Twitter for iPhone",
"retweetCount": 44,
"replyCount": 8,
"likeCount": 327,
"quoteCount": 0,
"viewCount": 27457,
"createdAt": "Mon Jun 15 17:48:11 +0000 2026",
"lang": "en",
"bookmarkCount": 276,
"isReply": false,
"inReplyToId": null,
"conversationId": "2066578531010728271",
"displayTextRange": [
0,
275
],
"inReplyToUserId": null,
"inReplyToUsername": null,
"author": {
"type": "user",
"userName": "NielsRogge",
"url": "https://x.com/NielsRogge",
"twitterUrl": "https://twitter.com/NielsRogge",
"id": "133818617",
"name": "Niels Rogge",
"isVerified": false,
"isBlueVerified": false,
"verifiedType": null,
"profilePicture": "https://pbs.twimg.com/profile_images/1828883496066060288/n0OUAXhz_normal.jpg",
"coverPicture": "https://pbs.twimg.com/profile_banners/133818617/1636026056",
"description": "",
"location": "Belgium",
"followers": 20787,
"following": 727,
"status": "",
"canDm": true,
"canMediaTag": false,
"createdAt": "Fri Apr 16 18:18:57 +0000 2010",
"entities": {
"description": {
"urls": []
},
"url": {}
},
"fastFollowersCount": 0,
"favouritesCount": 1902,
"hasCustomTimelines": true,
"isTranslator": false,
"mediaCount": 823,
"statusesCount": 3359,
"withheldInCountries": [],
"affiliatesHighlightedLabel": {},
"possiblySensitive": false,
"pinnedTweetIds": [
"1575416431793119232"
],
"profile_bio": {
"description": "ML Engineer @ml6team @huggingface. @KU_Leuven grad. General interest in machine & deep learning. Making AI more accessible for everyone!",
"entities": {
"description": {
"user_mentions": [
{
"id_str": "",
"indices": [
12,
20
],
"name": "",
"screen_name": "ml6team"
},
{
"id_str": "",
"indices": [
21,
33
],
"name": "",
"screen_name": "huggingface"
},
{
"id_str": "",
"indices": [
35,
45
],
"name": "",
"screen_name": "KU_Leuven"
}
]
},
"url": {
"urls": [
{
"display_url": "nielsrogge.github.io",
"expanded_url": "http://nielsrogge.github.io",
"indices": [
0,
23
],
"url": "https://t.co/5ZROAxhmdm"
}
]
}
}
},
"isAutomated": false,
"automatedBy": null
},
"extendedEntities": {
"media": [
{
"display_url": "pic.twitter.com/WNYGOy0QjX",
"expanded_url": "https://twitter.com/NielsRogge/status/2066578531010728271/photo/1",
"ext_master_playlist_only": [],
"ext_media_availability": {
"status": "Available"
},
"ext_playlists": [],
"features": {
"large": {
"faces": []
},
"orig": {
"faces": []
}
},
"id_str": "2066578222800678914",
"indices": [
276,
299
],
"media_key": "3_2066578222800678914",
"media_results": {
"id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARyt9fHGl7ACCgACHK32OYlX0U8AAA==",
"result": {
"__typename": "ApiMedia",
"id": "QXBpTWVkaWE6DAABCgABHK318caXsAIKAAIcrfY5iVfRTwAA",
"media_key": "3_2066578222800678914"
}
},
"media_url_https": "https://pbs.twimg.com/media/HK318caXsAIqDLk.jpg",
"original_info": {
"focus_rects": [
{
"h": 1726,
"w": 3082,
"x": 0,
"y": 0
},
{
"h": 1986,
"w": 1986,
"x": 0,
"y": 0
},
{
"h": 1986,
"w": 1742,
"x": 0,
"y": 0
},
{
"h": 1986,
"w": 993,
"x": 351,
"y": 0
},
{
"h": 1986,
"w": 3082,
"x": 0,
"y": 0
}
],
"height": 1986,
"width": 3082
},
"sizes": {
"large": {
"h": 1320,
"w": 2048
}
},
"type": "photo",
"url": "https://t.co/WNYGOy0QjX"
}
]
},
"card": null,
"place": {},
"entities": {
"hashtags": [],
"symbols": [],
"urls": [
{
"display_url": "paperswithcode.co/methods/specul…",
"expanded_url": "https://paperswithcode.co/methods/speculative-decoding",
"indices": [
447,
470
],
"url": "https://t.co/qYAJJ8Snin"
}
],
"user_mentions": []
},
"quoted_tweet": {
"type": "tweet",
"id": "2066560651942863297",
"url": "https://x.com/lmsysorg/status/2066560651942863297",
"twitterUrl": "https://twitter.com/lmsysorg/status/2066560651942863297",
"text": "🚀 New blog: The next generation of speculative decoding: DFlash and Spec V2 \n\nDFlash + Spec V2 hit >4.3X baseline throughput for LLM inference, now the default speculative decoding engine in SGLang! Together with @modal and https://t.co/ZXetBKIRym, our jointly-released DFlash drafter for Qwen 3.5 397B-A17B beats both baseline and native MTP in every setting we benchmarked:\n 1️⃣ >4.3X baseline & 1.5X native MTP throughput (concurrency 1, HumanEval, 8xB200)\n 2️⃣ Block diffusion drafter: a full token block in one forward pass\n 3️⃣ KV injection: target-model features fed into every draft layer’s KV cache for higher acceptance\n 4️⃣ Spec V2 overlap scheduler: +33% end-to-end\n\nRead the code, deploy a DFlash server, and start experimenting!",
"source": "Twitter for iPhone",
"retweetCount": 67,
"replyCount": 12,
"likeCount": 397,
"quoteCount": 12,
"viewCount": 77622,
"createdAt": "Mon Jun 15 16:37:08 +0000 2026",
"lang": "en",
"bookmarkCount": 248,
"isReply": false,
"inReplyToId": null,
"conversationId": "2066560651942863297",
"displayTextRange": [
0,
279
],
"inReplyToUserId": null,
"inReplyToUsername": null,
"author": {
"type": "user",
"userName": "lmsysorg",
"url": "https://x.com/lmsysorg",
"twitterUrl": "https://twitter.com/lmsysorg",
"id": "1822588444046249984",
"name": "LMSYS Org",
"isVerified": false,
"isBlueVerified": true,
"verifiedType": null,
"profilePicture": "https://pbs.twimg.com/profile_images/1823251762243121152/0cgIY6yc_normal.jpg",
"coverPicture": "",
"description": "",
"location": "US",
"followers": 15542,
"following": 199,
"status": "",
"canDm": false,
"canMediaTag": true,
"createdAt": "Sun Aug 11 10:58:54 +0000 2024",
"entities": {
"description": {
"urls": []
},
"url": {}
},
"fastFollowersCount": 0,
"favouritesCount": 1404,
"hasCustomTimelines": true,
"isTranslator": false,
"mediaCount": 292,
"statusesCount": 1134,
"withheldInCountries": [],
"affiliatesHighlightedLabel": {},
"possiblySensitive": false,
"pinnedTweetIds": [
"2066560651942863297"
],
"profile_bio": {
"description": "Large Model Systems Organization: Join our Slack: https://t.co/vzYOTP4w6C. We developed SGLang https://t.co/OjwQadINKU, Chatbot Arena (now @arena), and Vicuna!",
"entities": {
"description": {
"urls": [
{
"display_url": "slack.sglang.io",
"expanded_url": "https://slack.sglang.io",
"indices": [
50,
73
],
"url": "https://t.co/vzYOTP4w6C"
},
{
"display_url": "sglang.io",
"expanded_url": "https://sglang.io",
"indices": [
95,
118
],
"url": "https://t.co/OjwQadINKU"
}
],
"user_mentions": [
{
"id_str": "",
"indices": [
139,
145
],
"name": "",
"screen_name": "arena"
}
]
},
"url": {
"urls": [
{
"display_url": "lmsys.org",
"expanded_url": "https://lmsys.org/",
"indices": [
0,
23
],
"url": "https://t.co/icnD5fVXFw"
}
]
}
}
},
"isAutomated": false,
"automatedBy": null
},
"extendedEntities": {
"media": [
{
"allow_download_status": {
"allow_download": true
},
"display_url": "pic.twitter.com/pMLitReOBa",
"expanded_url": "https://twitter.com/lmsysorg/status/2066560651942863297/photo/1",
"ext_master_playlist_only": [],
"ext_media_availability": {
"status": "Available"
},
"ext_playlists": [],
"features": {
"large": {
"faces": [
{
"h": 147,
"w": 147,
"x": 481,
"y": 235
}
]
},
"orig": {
"faces": [
{
"h": 147,
"w": 147,
"x": 481,
"y": 235
}
]
}
},
"id_str": "2066555418881769472",
"indices": [
280,
303
],
"media_key": "3_2066555418881769472",
"media_results": {
"id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARyt4TRTG0AACgACHK3l9r3aocEAAA==",
"result": {
"__typename": "ApiMedia",
"id": "QXBpTWVkaWE6DAABCgABHK3hNFMbQAAKAAIcreX2vdqhwQAA",
"media_key": "3_2066555418881769472"
}
},
"media_url_https": "https://pbs.twimg.com/media/HK3hNFMbQAA5kLl.jpg",
"original_info": {
"focus_rects": [
{
"h": 629,
"w": 1123,
"x": 121,
"y": 0
},
{
"h": 629,
"w": 629,
"x": 368,
"y": 0
},
{
"h": 629,
"w": 552,
"x": 406,
"y": 0
},
{
"h": 629,
"w": 315,
"x": 525,
"y": 0
},
{
"h": 629,
"w": 1438,
"x": 0,
"y": 0
}
],
"height": 629,
"width": 1438
},
"sizes": {
"large": {
"h": 629,
"w": 1438
}
},
"type": "photo",
"url": "https://t.co/pMLitReOBa"
}
]
},
"card": null,
"place": {},
"entities": {
"hashtags": [],
"symbols": [],
"urls": [
{
"display_url": "z-lab.ai",
"expanded_url": "http://z-lab.ai",
"indices": [
224,
247
],
"url": "https://t.co/ZXetBKIRym"
}
],
"user_mentions": [
{
"id_str": "1551987185372512263",
"indices": [
213,
219
],
"name": "Modal",
"screen_name": "modal"
}
]
},
"quoted_tweet": null,
"retweeted_tweet": null,
"isLimitedReply": false,
"communityInfo": null,
"article": null
},
"retweeted_tweet": null,
"isLimitedReply": false,
"communityInfo": null,
"article": null
}