🐦 Twitter Post Details

Viewing enriched Twitter post

@asapzzhou

(1/n) Tiny-A2D: An Open Recipe to Turn Any AR LM into a Diffusion LM Code (dLLM): https://t.co/Nv7d1t8Qin Checkpoints: https://t.co/rpibkb2Xfq With dLLM, you can turn ANY autoregressive LM into a diffusion LM (parallel generation + infilling) with minimal compute. Using this recipe, we built a 🤗collection of the smallest diffusion LMs that work well in practice. Key takeaways: 1. Finetuned on Qwen3-0.6B, we obtain the strongest small (~0.5/0.6B) diffusion LMs to date. 2. The base AR LM matters: Investing compute in improving the base AR model is potentially more efficient than scaling compute during adaptation. 3. Block diffusion (BD3LM) generally outperforms vanilla masked diffusion (MDLM), especially on math-reasoning and coding tasks.

Media 2
Media 3

📊 Media Metadata

{
  "media": [
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1998098118827770210/media_0.mp4?",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1998098118827770210/media_0.mp4?",
      "type": "video",
      "filename": "media_0.mp4"
    },
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1998098118827770210/media_1.jpg?",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1998098118827770210/media_1.jpg?",
      "type": "photo",
      "filename": "media_1.jpg"
    },
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1998098118827770210/media_2.jpg?",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1998098118827770210/media_2.jpg?",
      "type": "photo",
      "filename": "media_2.jpg"
    }
  ],
  "processed_at": "2025-12-09T14:52:32.855947",
  "pipeline_version": "2.0"
}

🔧 Raw API Response

{
  "type": "tweet",
  "id": "1998098118827770210",
  "url": "https://x.com/asapzzhou/status/1998098118827770210",
  "twitterUrl": "https://twitter.com/asapzzhou/status/1998098118827770210",
  "text": "(1/n) Tiny-A2D: An Open Recipe to Turn Any AR LM into a Diffusion LM\n\nCode (dLLM): https://t.co/Nv7d1t8Qin\nCheckpoints: https://t.co/rpibkb2Xfq\n\nWith dLLM, you can turn ANY autoregressive LM into a diffusion LM (parallel generation + infilling) with minimal compute. Using this recipe, we built a 🤗collection of the smallest diffusion LMs that work well in practice.\n\nKey takeaways:\n\n1. Finetuned on Qwen3-0.6B, we obtain the strongest small (~0.5/0.6B) diffusion LMs to date.\n\n2. The base AR LM matters: Investing compute in improving the base AR model is potentially more efficient than scaling compute during adaptation.\n\n3. Block diffusion (BD3LM) generally outperforms vanilla masked diffusion (MDLM), especially on math-reasoning and coding tasks.",
  "source": "Twitter for iPhone",
  "retweetCount": 49,
  "replyCount": 4,
  "likeCount": 204,
  "quoteCount": 5,
  "viewCount": 14763,
  "createdAt": "Mon Dec 08 18:31:29 +0000 2025",
  "lang": "en",
  "bookmarkCount": 133,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "1998098118827770210",
  "displayTextRange": [
    0,
    278
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "asapzzhou",
    "url": "https://x.com/asapzzhou",
    "twitterUrl": "https://twitter.com/asapzzhou",
    "id": "1961300478870130691",
    "name": "Zhanhui Zhou @ NeurIPS",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1988573455462301696/71yi_tYI_normal.jpg",
    "coverPicture": "",
    "description": "CS PhD @berkeley_ai",
    "location": "Berkeley, California",
    "followers": 476,
    "following": 35,
    "status": "",
    "canDm": false,
    "canMediaTag": true,
    "createdAt": "Fri Aug 29 05:30:53 +0000 2025",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {
        "urls": [
          {
            "display_url": "github.com/ZHZisZZ",
            "expanded_url": "https://github.com/ZHZisZZ",
            "url": "https://t.co/jkkZ2zXY7A",
            "indices": [
              0,
              23
            ]
          }
        ]
      }
    },
    "fastFollowersCount": 0,
    "favouritesCount": 125,
    "hasCustomTimelines": false,
    "isTranslator": false,
    "mediaCount": 13,
    "statusesCount": 32,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "1998098118827770210"
    ],
    "profile_bio": {},
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "display_url": "pic.x.com/PoIuRH1D63",
        "expanded_url": "https://x.com/asapzzhou/status/1998098118827770210/video/1",
        "id_str": "1998098067246260225",
        "indices": [
          279,
          302
        ],
        "media_key": "13_1998098067246260225",
        "media_url_https": "https://pbs.twimg.com/amplify_video_thumb/1998098067246260225/img/1SYO19oO0Nvr9Sou.jpg",
        "type": "video",
        "url": "https://t.co/PoIuRH1D63",
        "additional_media_info": {
          "monetizable": false
        },
        "ext_media_availability": {
          "status": "Available"
        },
        "sizes": {
          "large": {
            "h": 1080,
            "w": 1920,
            "resize": "fit"
          },
          "medium": {
            "h": 675,
            "w": 1200,
            "resize": "fit"
          },
          "small": {
            "h": 383,
            "w": 680,
            "resize": "fit"
          },
          "thumb": {
            "h": 150,
            "w": 150,
            "resize": "crop"
          }
        },
        "original_info": {
          "height": 1080,
          "width": 1920,
          "focus_rects": []
        },
        "allow_download_status": {
          "allow_download": true
        },
        "video_info": {
          "aspect_ratio": [
            16,
            9
          ],
          "duration_millis": 6594,
          "variants": [
            {
              "content_type": "application/x-mpegURL",
              "url": "https://video.twimg.com/amplify_video/1998098067246260225/pl/3FVYzemwkUwCdDQH.m3u8"
            },
            {
              "bitrate": 256000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/1998098067246260225/vid/avc1/480x270/w_2Rkwk3S2OakbCV.mp4"
            },
            {
              "bitrate": 832000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/1998098067246260225/vid/avc1/640x360/4mJ5OlHaq9BQEdZE.mp4"
            },
            {
              "bitrate": 2176000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/1998098067246260225/vid/avc1/1280x720/X9KRi12gTat7KovI.mp4"
            },
            {
              "bitrate": 10368000,
              "content_type": "video/mp4",
              "url": "https://video.twimg.com/amplify_video/1998098067246260225/vid/avc1/1920x1080/sHqEnsZGVo5Nf6EU.mp4"
            }
          ]
        },
        "media_results": {
          "result": {
            "media_key": "13_1998098067246260225"
          }
        }
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "hashtags": [],
    "symbols": [],
    "timestamps": [],
    "urls": [
      {
        "display_url": "github.com/ZHZisZZ/dllm",
        "expanded_url": "https://github.com/ZHZisZZ/dllm",
        "url": "https://t.co/Nv7d1t8Qin",
        "indices": [
          83,
          106
        ]
      },
      {
        "display_url": "huggingface.co/collections/dl…",
        "expanded_url": "https://huggingface.co/collections/dllm-collection/tiny-a2d",
        "url": "https://t.co/rpibkb2Xfq",
        "indices": [
          120,
          143
        ]
      }
    ],
    "user_mentions": []
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "article": null
}