Viewing enriched Twitter post
Almost done benchmarking, write-up coming tomorrow - but wanted to share some important findings right away: Tested @Alibaba_Qwen QwQ from 3 to 8 bit EXL2 in MMLU-Pro, and by raising max_tokens from default 2K to 8K, smaller quants got MUCH better scores. They need room to think! https://t.co/H9VkSiiYPP
{
"score": 0.86,
"scored_at": "2025-08-09T13:46:07.555483",
"import_source": "network_archive_import",
"links_checked": true,
"checked_at": "2025-08-10T10:32:59.403027",
"media": [
{
"type": "photo",
"url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1863331342479438175/media_0.png?",
"filename": "media_0.png"
},
{
"id": "",
"type": "photo",
"url": null,
"media_url": "https://pbs.twimg.com/media/Gdvg_JVXsAA5SIr.png",
"media_url_https": null,
"display_url": null,
"expanded_url": null
}
],
"reprocessed_at": "2025-08-12T15:26:59.667008",
"reprocessed_reason": "missing_media_array",
"original_structure": "had_both"
}{
"user": {
"created_at": "2023-09-11T22:37:51.000Z",
"default_profile_image": false,
"description": "🐦⬛🐺 AI Evangelist, Engineer, Consultant, Researcher; ASI achieved internally: 👩🏻🦰🐦⬛🐺 @AI_AmyRavenwolf (Artificial Sassy Intelligence) 🤖💁♀️🧠",
"fast_followers_count": 0,
"favourites_count": 15371,
"followers_count": 2572,
"friends_count": 510,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 73,
"location": "🇩🇪",
"media_count": 333,
"name": "Wolfram Ravenwolf 🐺🐦⬛",
"normal_followers_count": 2572,
"possibly_sensitive": false,
"profile_banner_url": "https://pbs.twimg.com/profile_banners/1701364361711808512/1732670123",
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1706727481753923584/jBUq13SE_normal.jpg",
"screen_name": "WolframRvnwlf",
"statuses_count": 4816,
"translator_type": "none",
"url": "https://t.co/AQHQZorYDj",
"verified": true,
"withheld_in_countries": [],
"id_str": "1701364361711808512"
},
"id": "1863331342479438175",
"conversation_id": "1863331342479438175",
"full_text": "Almost done benchmarking, write-up coming tomorrow - but wanted to share some important findings right away: Tested @Alibaba_Qwen QwQ from 3 to 8 bit EXL2 in MMLU-Pro, and by raising max_tokens from default 2K to 8K, smaller quants got MUCH better scores. They need room to think! https://t.co/H9VkSiiYPP",
"reply_count": 13,
"retweet_count": 40,
"favorite_count": 254,
"hashtags": [],
"symbols": [],
"user_mentions": [
{
"id_str": "1753339277386342400",
"name": "Qwen",
"screen_name": "Alibaba_Qwen",
"profile": "https://twitter.com/Alibaba_Qwen"
}
],
"urls": [],
"media": [
{
"media_url": "https://pbs.twimg.com/media/Gdvg_JVXsAA5SIr.png",
"type": "photo"
}
],
"url": "https://twitter.com/WolframRvnwlf/status/1863331342479438175",
"created_at": "2024-12-01T21:16:25.000Z",
"#sort_index": "1863331342479438175",
"view_count": 38768,
"quote_count": 3,
"is_quote_tweet": true,
"is_retweet": false,
"is_pinned": false,
"is_truncated": false,
"quoted_tweet": {
"user": {
"created_at": "2023-09-11T22:37:51.000Z",
"default_profile_image": false,
"description": "🐦⬛🐺 AI Evangelist, Engineer, Consultant, Researcher; ASI achieved internally: 👩🏻🦰🐦⬛🐺 @AI_AmyRavenwolf (Artificial Sassy Intelligence) 🤖💁♀️🧠",
"fast_followers_count": 0,
"favourites_count": 15371,
"followers_count": 2572,
"friends_count": 510,
"has_custom_timelines": false,
"is_translator": false,
"listed_count": 73,
"location": "🇩🇪",
"media_count": 333,
"name": "Wolfram Ravenwolf 🐺🐦⬛",
"normal_followers_count": 2572,
"possibly_sensitive": false,
"profile_banner_url": "https://pbs.twimg.com/profile_banners/1701364361711808512/1732670123",
"profile_image_url_https": "https://pbs.twimg.com/profile_images/1706727481753923584/jBUq13SE_normal.jpg",
"screen_name": "WolframRvnwlf",
"statuses_count": 4816,
"translator_type": "none",
"url": "https://t.co/AQHQZorYDj",
"verified": true,
"withheld_in_countries": [],
"id_str": "1701364361711808512"
},
"id": "1862643071625646168",
"conversation_id": "1862643071625646168",
"full_text": "WTF! What sorcery is this, @Alibaba_Qwen? I kept benchmarking - and not only does the 4.25-bit version get the same score as the 8-bit (what?), using Qwen2.5-Coder-0.5B as a draft model for speculative decoding sped it up from 27 to 42 tk/s AND it scored even higher (whaaat?)! 🤯 https://t.co/tNKvpoAqKq",
"reply_count": 23,
"retweet_count": 79,
"favorite_count": 596,
"hashtags": [],
"symbols": [],
"user_mentions": [
{
"id_str": "1753339277386342400",
"name": "Qwen",
"screen_name": "Alibaba_Qwen",
"profile": "https://twitter.com/Alibaba_Qwen"
}
],
"urls": [],
"media": [
{
"media_url": "https://pbs.twimg.com/media/GdluzJeWYAA7djb.png",
"type": "photo"
}
],
"url": "https://twitter.com/WolframRvnwlf/status/1862643071625646168",
"created_at": "2024-11-29T23:41:29.000Z",
"#sort_index": "1863331342479438000",
"view_count": 101334,
"quote_count": 7,
"is_quote_tweet": true,
"is_retweet": false,
"is_pinned": false,
"is_truncated": false
},
"startUrl": "https://x.com/wolframrvnwlf/status/1863331342479438175"
}