Viewing enriched Twitter post
The Sparse Frontier Explores the efficiency-accuracy trade-offs of sparse attention in Transformer LLMs. Findings: - Efficiency trade-off: For very long sequences, big models with high, sparse attention outperform smaller, dense ones. - Phase sensitivity: Decoding can… https://t.co/agTfnfTUx9
{
"media": [
{
"type": "image",
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1915762761705177357/media_0.jpg?",
"filename": "media_0.jpg"
}
],
"nlp": {
"sentiment": "neutral",
"topics": [
"sparse attention",
"Transformer LLMs",
"efficiency-accuracy trade-offs"
],
"entities": [],
"summary": "The tweet discusses the efficiency-accuracy trade-offs of sparse attention in Transformer LLMs, highlighting findings related to model performance on long sequences.",
"language": "en",
"processed_at": "2025-08-07T12:20:58.028522"
}
}{
"data": {
"threaded_conversation_with_injections_v2": {
"instructions": [
{
"type": "TimelineClearCache"
},
{
"type": "TimelineAddEntries",
"entries": [
{
"entryId": "tweet-1915762749059342714",
"sortIndex": "1953521741485572096",
"content": {
"entryType": "TimelineTimelineItem",
"__typename": "TimelineTimelineItem",
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1915762749059342714",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjozNDQ4Mjg0MzEz",
"rest_id": "3448284313",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": true,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": true,
"can_media_tag": true,
"created_at": "Fri Sep 04 12:59:26 +0000 2015",
"default_profile": false,
"default_profile_image": false,
"description": "Building with AI agents @dair_ai • Prev: Meta AI, Galactica LLM, Elastic, PaperswithCode, PhD • I share insights on how to build with AI Agents ⬇️",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "dair-ai.thinkific.com",
"expanded_url": "https://dair-ai.thinkific.com/",
"url": "https://t.co/JBU5beHQNs",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 31703,
"followers_count": 258925,
"friends_count": 643,
"has_custom_timelines": true,
"is_translator": false,
"listed_count": 4291,
"location": "",
"media_count": 3787,
"name": "elvis",
"normal_followers_count": 258925,
"pinned_tweet_ids_str": [
"1953502698801623159"
],
"possibly_sensitive": false,
"profile_banner_url": "https://pbs.twimg.com/profile_banners/3448284313/1565974901",
"profile_image_url_https": "https://pbs.twimg.com/profile_images/939313677647282181/vZjFWtAn_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "omarsar0",
"statuses_count": 15324,
"translator_type": "regular",
"url": "https://t.co/JBU5beHQNs",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"professional": {
"rest_id": "1460724937073770505",
"professional_type": "Creator",
"category": [
{
"id": 713,
"name": "Science & Technology",
"icon_name": ""
}
]
},
"tipjar_settings": {
"is_enabled": false,
"cash_app_handle": "£elvisomarsaravia"
}
}
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1915762749059342714"
],
"editable_until_msecs": "1745592007000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "13829",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://typefully.com/\" rel=\"nofollow\">Typefully</a>",
"note_tweet": {
"is_expandable": true,
"note_tweet_results": {
"result": {
"id": "Tm90ZVR3ZWV0OjE5MTU3NjI3NDg4OTk5MjYwMTY=",
"text": "The Sparse Frontier\n\nExplores the efficiency-accuracy trade-offs of sparse attention in Transformer LLMs.\n\nFindings:\n\n- Efficiency trade-off: For very long sequences, big models with high, sparse attention outperform smaller, dense ones.\n\n- Phase sensitivity: Decoding can tolerate greater sparsity, especially in larger models, than prefilling while still keeping accuracy.\n\n- No one-size-fits-all: Sparsification strategies vary by task and phase; even modest sparsity can hurt at least one benchmark.\n\n- Generalization: Newly derived scaling laws for sparse attention back up these trends beyond the tested settings.",
"entity_set": {
"hashtags": [],
"symbols": [],
"urls": [],
"user_mentions": []
}
}
}
},
"legacy": {
"bookmark_count": 68,
"bookmarked": false,
"created_at": "Fri Apr 25 13:40:07 +0000 2025",
"conversation_id_str": "1915762749059342714",
"display_text_range": [
0,
273
],
"entities": {
"hashtags": [],
"media": [
{
"display_url": "pic.x.com/agTfnfTUx9",
"expanded_url": "https://x.com/omarsar0/status/1915762749059342714/photo/1",
"id_str": "1915762744881790976",
"indices": [
274,
297
],
"media_key": "3_1915762744881790976",
"media_url_https": "https://pbs.twimg.com/media/GpYoEPqbQAAej4z.png",
"type": "photo",
"url": "https://t.co/agTfnfTUx9",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 891,
"w": 928,
"resize": "fit"
},
"medium": {
"h": 891,
"w": 928,
"resize": "fit"
},
"small": {
"h": 653,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 891,
"width": 928,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 928,
"h": 520
},
{
"x": 19,
"y": 0,
"w": 891,
"h": 891
},
{
"x": 73,
"y": 0,
"w": 782,
"h": 891
},
{
"x": 241,
"y": 0,
"w": 446,
"h": 891
},
{
"x": 0,
"y": 0,
"w": 928,
"h": 891
}
]
},
"media_results": {
"result": {
"media_key": "3_1915762744881790976"
}
}
}
],
"symbols": [],
"timestamps": [],
"urls": [],
"user_mentions": []
},
"extended_entities": {
"media": [
{
"display_url": "pic.x.com/agTfnfTUx9",
"expanded_url": "https://x.com/omarsar0/status/1915762749059342714/photo/1",
"id_str": "1915762744881790976",
"indices": [
274,
297
],
"media_key": "3_1915762744881790976",
"media_url_https": "https://pbs.twimg.com/media/GpYoEPqbQAAej4z.png",
"type": "photo",
"url": "https://t.co/agTfnfTUx9",
"ext_media_availability": {
"status": "Available"
},
"features": {
"large": {
"faces": []
},
"medium": {
"faces": []
},
"small": {
"faces": []
},
"orig": {
"faces": []
}
},
"sizes": {
"large": {
"h": 891,
"w": 928,
"resize": "fit"
},
"medium": {
"h": 891,
"w": 928,
"resize": "fit"
},
"small": {
"h": 653,
"w": 680,
"resize": "fit"
},
"thumb": {
"h": 150,
"w": 150,
"resize": "crop"
}
},
"original_info": {
"height": 891,
"width": 928,
"focus_rects": [
{
"x": 0,
"y": 0,
"w": 928,
"h": 520
},
{
"x": 19,
"y": 0,
"w": 891,
"h": 891
},
{
"x": 73,
"y": 0,
"w": 782,
"h": 891
},
{
"x": 241,
"y": 0,
"w": 446,
"h": 891
},
{
"x": 0,
"y": 0,
"w": 928,
"h": 891
}
]
},
"media_results": {
"result": {
"media_key": "3_1915762744881790976"
}
}
}
]
},
"favorite_count": 113,
"favorited": false,
"full_text": "The Sparse Frontier\n\nExplores the efficiency-accuracy trade-offs of sparse attention in Transformer LLMs.\n\nFindings:\n\n- Efficiency trade-off: For very long sequences, big models with high, sparse attention outperform smaller, dense ones.\n\n- Phase sensitivity: Decoding can… https://t.co/agTfnfTUx9",
"is_quote_status": false,
"lang": "en",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 1,
"reply_count": 9,
"retweet_count": 21,
"retweeted": false,
"user_id_str": "3448284313",
"id_str": "1915762749059342714"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "Tweet"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet"
}
}
},
{
"entryId": "tweet-1915762761705177357",
"sortIndex": "1953521741485572086",
"content": {
"entryType": "TimelineTimelineItem",
"__typename": "TimelineTimelineItem",
"itemContent": {
"itemType": "TimelineTweet",
"__typename": "TimelineTweet",
"tweet_results": {
"result": {
"__typename": "Tweet",
"rest_id": "1915762761705177357",
"has_birdwatch_notes": false,
"core": {
"user_results": {
"result": {
"__typename": "User",
"id": "VXNlcjozNDQ4Mjg0MzEz",
"rest_id": "3448284313",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": true,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": true,
"can_media_tag": true,
"created_at": "Fri Sep 04 12:59:26 +0000 2015",
"default_profile": false,
"default_profile_image": false,
"description": "Building with AI agents @dair_ai • Prev: Meta AI, Galactica LLM, Elastic, PaperswithCode, PhD • I share insights on how to build with AI Agents ⬇️",
"entities": {
"description": {
"urls": []
},
"url": {
"urls": [
{
"display_url": "dair-ai.thinkific.com",
"expanded_url": "https://dair-ai.thinkific.com/",
"url": "https://t.co/JBU5beHQNs",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 31703,
"followers_count": 258925,
"friends_count": 643,
"has_custom_timelines": true,
"is_translator": false,
"listed_count": 4291,
"location": "",
"media_count": 3787,
"name": "elvis",
"normal_followers_count": 258925,
"pinned_tweet_ids_str": [
"1953502698801623159"
],
"possibly_sensitive": false,
"profile_banner_url": "https://pbs.twimg.com/profile_banners/3448284313/1565974901",
"profile_image_url_https": "https://pbs.twimg.com/profile_images/939313677647282181/vZjFWtAn_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "omarsar0",
"statuses_count": 15324,
"translator_type": "regular",
"url": "https://t.co/JBU5beHQNs",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"professional": {
"rest_id": "1460724937073770505",
"professional_type": "Creator",
"category": [
{
"id": 713,
"name": "Science & Technology",
"icon_name": ""
}
]
},
"tipjar_settings": {
"is_enabled": false,
"cash_app_handle": "£elvisomarsaravia"
}
}
}
},
"card": {
"rest_id": "https://t.co/YMqQQI1uAY",
"legacy": {
"binding_values": [
{
"key": "thumbnail_image",
"value": {
"image_value": {
"alt": "arXiv logo",
"height": 144,
"width": 144,
"url": "https://pbs.twimg.com/card_img/1949758743958425600/kRc27OMH?format=jpg&name=144x144_2"
},
"type": "IMAGE"
}
},
{
"key": "description",
"value": {
"string_value": "Sparse attention offers a promising strategy to extend long-context capabilities in Transformer LLMs, yet its viability, its efficiency-accuracy trade-offs, and systematic scaling studies remain...",
"type": "STRING"
}
},
{
"key": "domain",
"value": {
"string_value": "arxiv.org",
"type": "STRING"
}
},
{
"key": "thumbnail_image_large",
"value": {
"image_value": {
"alt": "arXiv logo",
"height": 420,
"width": 420,
"url": "https://pbs.twimg.com/card_img/1949758743958425600/kRc27OMH?format=jpg&name=420x420_2"
},
"type": "IMAGE"
}
},
{
"key": "thumbnail_image_original",
"value": {
"image_value": {
"alt": "arXiv logo",
"height": 1000,
"width": 1000,
"url": "https://pbs.twimg.com/card_img/1949758743958425600/kRc27OMH?format=jpg&name=orig"
},
"type": "IMAGE"
}
},
{
"key": "site",
"value": {
"scribe_key": "publisher_id",
"type": "USER",
"user_value": {
"id_str": "808633423300624384",
"path": []
}
}
},
{
"key": "thumbnail_image_small",
"value": {
"image_value": {
"alt": "arXiv logo",
"height": 100,
"width": 100,
"url": "https://pbs.twimg.com/card_img/1949758743958425600/kRc27OMH?format=jpg&name=100x100_2"
},
"type": "IMAGE"
}
},
{
"key": "thumbnail_image_x_large",
"value": {
"image_value": {
"alt": "arXiv logo",
"height": 1000,
"width": 1000,
"url": "https://pbs.twimg.com/card_img/1949758743958425600/kRc27OMH?format=png&name=2048x2048_2_exp"
},
"type": "IMAGE"
}
},
{
"key": "thumbnail_image_alt_text",
"value": {
"string_value": "arXiv logo",
"type": "STRING"
}
},
{
"key": "vanity_url",
"value": {
"scribe_key": "vanity_url",
"string_value": "arxiv.org",
"type": "STRING"
}
},
{
"key": "thumbnail_image_color",
"value": {
"image_color_value": {
"palette": [
{
"rgb": {
"blue": 255,
"green": 255,
"red": 255
},
"percentage": 94.17
},
{
"rgb": {
"blue": 105,
"green": 116,
"red": 124
},
"percentage": 4.33
},
{
"rgb": {
"blue": 46,
"green": 21,
"red": 170
},
"percentage": 1.26
},
{
"rgb": {
"blue": 131,
"green": 116,
"red": 203
},
"percentage": 0.23
}
]
},
"type": "IMAGE_COLOR"
}
},
{
"key": "title",
"value": {
"string_value": "The Sparse Frontier: Sparse Attention Trade-offs in Transformer LLMs",
"type": "STRING"
}
},
{
"key": "card_url",
"value": {
"scribe_key": "card_url",
"string_value": "https://t.co/YMqQQI1uAY",
"type": "STRING"
}
}
],
"card_platform": {
"platform": {
"audience": {
"name": "production"
},
"device": {
"name": "Android",
"version": "12"
}
}
},
"name": "summary",
"url": "https://t.co/YMqQQI1uAY",
"user_refs_results": [
{
"result": {
"__typename": "User",
"id": "VXNlcjo4MDg2MzM0MjMzMDA2MjQzODQ=",
"rest_id": "808633423300624384",
"affiliates_highlighted_label": {},
"has_graduated_access": true,
"is_blue_verified": false,
"profile_image_shape": "Circle",
"legacy": {
"can_dm": true,
"can_media_tag": true,
"created_at": "Tue Dec 13 11:23:26 +0000 2016",
"default_profile": false,
"default_profile_image": false,
"description": "News from https://t.co/enurGFxpcS, a free distribution service and an open archive for scholarly articles.\n\nFor help with arXiv, see https://t.co/LcWuhM0BOl",
"entities": {
"description": {
"urls": [
{
"display_url": "arXiv.org",
"expanded_url": "http://arXiv.org",
"url": "https://t.co/enurGFxpcS",
"indices": [
10,
33
]
},
{
"display_url": "arxiv.org/help",
"expanded_url": "https://arxiv.org/help",
"url": "https://t.co/LcWuhM0BOl",
"indices": [
133,
156
]
}
]
},
"url": {
"urls": [
{
"display_url": "arxiv.org",
"expanded_url": "https://arxiv.org/",
"url": "https://t.co/DHMkdi4lF9",
"indices": [
0,
23
]
}
]
}
},
"fast_followers_count": 0,
"favourites_count": 948,
"followers_count": 41367,
"friends_count": 187,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 487,
"location": "Ithaca, NY",
"media_count": 109,
"name": "arXiv.org",
"normal_followers_count": 41367,
"pinned_tweet_ids_str": [],
"possibly_sensitive": false,
"profile_banner_url": "https://pbs.twimg.com/profile_banners/808633423300624384/1481635469",
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1365352170267299840/IzvjKckL_normal.jpg",
"profile_interstitial_type": "",
"screen_name": "arxiv",
"statuses_count": 1106,
"translator_type": "none",
"url": "https://t.co/DHMkdi4lF9",
"verified": false,
"want_retweets": false,
"withheld_in_countries": []
},
"tipjar_settings": {}
}
}
]
}
},
"unmention_data": {},
"edit_control": {
"edit_tweet_ids": [
"1915762761705177357"
],
"editable_until_msecs": "1745592010000",
"is_edit_eligible": false,
"edits_remaining": "5"
},
"is_translatable": false,
"views": {
"count": "3686",
"state": "EnabledWithCount"
},
"source": "<a href=\"https://typefully.com/\" rel=\"nofollow\">Typefully</a>",
"legacy": {
"bookmark_count": 6,
"bookmarked": false,
"created_at": "Fri Apr 25 13:40:10 +0000 2025",
"conversation_id_str": "1915762749059342714",
"display_text_range": [
0,
23
],
"entities": {
"hashtags": [],
"symbols": [],
"timestamps": [],
"urls": [
{
"display_url": "arxiv.org/abs/2504.17768",
"expanded_url": "https://arxiv.org/abs/2504.17768",
"url": "https://t.co/YMqQQI1uAY",
"indices": [
0,
23
]
}
],
"user_mentions": []
},
"favorite_count": 15,
"favorited": false,
"full_text": "https://t.co/YMqQQI1uAY",
"in_reply_to_screen_name": "omarsar0",
"in_reply_to_status_id_str": "1915762749059342714",
"in_reply_to_user_id_str": "3448284313",
"is_quote_status": false,
"lang": "zxx",
"possibly_sensitive": false,
"possibly_sensitive_editable": true,
"quote_count": 0,
"reply_count": 0,
"retweet_count": 0,
"retweeted": false,
"user_id_str": "3448284313",
"id_str": "1915762761705177357"
},
"quick_promote_eligibility": {
"eligibility": "IneligibleNotProfessional"
}
}
},
"tweetDisplayType": "SelfThread"
},
"clientEventInfo": {
"component": "tweet",
"element": "tweet"
}
}
},
{
"entryId": "cursor-bottom-1953521741485572085",
"sortIndex": "1953521741485572085",
"content": {
"entryType": "TimelineTimelineCursor",
"__typename": "TimelineTimelineCursor",
"value": "DAAFCgABGxxNrHY___MLAAIAAAAwRW1QQzZ3QUFBZlEvZ0dKTjB2R3AvQUFBQUFJYWxpZ1U1VnZCRFJxV0tCSHptNkY2CAADAAAAAgAA",
"cursorType": "Bottom"
}
}
]
},
{
"type": "TimelineTerminateTimeline",
"direction": "Top"
},
{
"type": "TimelineTerminateTimeline",
"direction": "Bottom"
}
],
"metadata": {
"reader_mode_config": {
"is_reader_mode_available": true
},
"scribeConfig": {
"page": "ranked_replies"
}
}
}
}
}