🐦 Twitter Post Details

Viewing enriched Twitter post

@iScienceLuvr

MDPO: Overcoming the Training-Inference Divide of Masked Diffusion Language Models "We propose a novel Masked Diffusion Policy Optimization (MDPO) to exploit the Markov property diffusion possesses and explicitly train the model under the same progressive refining schedule used at inference. MDPO matches the performance of the previous state-of-the-art (SOTA) method with 60x fewer gradient updates, while achieving average improvements of 9.6% on MATH500 and 54.2% on Countdown over SOTA when trained within the same number of weight updates. Additionally, we improve the remasking strategy of MDLMs as a plug-in inference replacement to overcome the limitation that the model cannot refine tokens flexibly. This simple yet effective training-free strategy, what we refer to as RCR, consistently improves performance and yields additional gains when combined with MDPO"

Media 1

📊 Media Metadata

{
  "media": [
    {
      "type": "photo",
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1958105587235147991/media_0.jpg?",
      "filename": "media_0.jpg"
    }
  ],
  "processed_at": "2025-08-20T12:48:14.880044",
  "pipeline_version": "2.0"
}

🔧 Raw API Response

{
  "type": "tweet",
  "id": "1958105587235147991",
  "url": "https://x.com/iScienceLuvr/status/1958105587235147991",
  "twitterUrl": "https://twitter.com/iScienceLuvr/status/1958105587235147991",
  "text": "MDPO: Overcoming the Training-Inference Divide of Masked Diffusion Language Models\n\n\"We propose a novel Masked Diffusion Policy Optimization (MDPO) to  exploit the Markov property diffusion possesses and explicitly train the  model under the same progressive refining schedule used at inference.  MDPO matches the performance of the previous state-of-the-art (SOTA)  method with 60x fewer gradient updates, while achieving average  improvements of 9.6% on MATH500 and 54.2% on Countdown over SOTA when  trained within the same number of weight updates. Additionally, we  improve the remasking strategy of MDLMs as a plug-in inference  replacement to overcome the limitation that the model cannot refine  tokens flexibly. This simple yet effective training-free strategy, what  we refer to as RCR, consistently improves performance and yields  additional gains when combined with MDPO\"",
  "source": "Twitter for iPhone",
  "retweetCount": 3,
  "replyCount": 1,
  "likeCount": 11,
  "quoteCount": 0,
  "viewCount": 1548,
  "createdAt": "Wed Aug 20 09:55:26 +0000 2025",
  "lang": "en",
  "bookmarkCount": 4,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "1958105587235147991",
  "displayTextRange": [
    0,
    277
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "iScienceLuvr",
    "url": "https://x.com/iScienceLuvr",
    "twitterUrl": "https://twitter.com/iScienceLuvr",
    "id": "441465751",
    "name": "Tanishq Mathew Abraham, Ph.D.",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1913710019729821696/Qge4zx6u_normal.jpg",
    "coverPicture": "https://pbs.twimg.com/profile_banners/441465751/1738204246",
    "description": "",
    "location": "",
    "followers": 80199,
    "following": 1243,
    "status": "",
    "canDm": true,
    "canMediaTag": true,
    "createdAt": "Tue Dec 20 03:45:50 +0000 2011",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 104894,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 2445,
    "statusesCount": 17835,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [
      "1952221233648718307"
    ],
    "profile_bio": {
      "description": "CEO @SophontAI |\nPhD at 19 (2023) |\nFounder, ex CEO @MedARC_AI |\nex Research Director Stability AI | \nBiomed. engineer @ 14 |\nTEDx talk➡https://t.co/xPxwKTq6Qb",
      "entities": {
        "description": {
          "urls": [
            {
              "display_url": "bit.ly/3tpAuan",
              "expanded_url": "https://bit.ly/3tpAuan",
              "indices": [
                136,
                159
              ],
              "url": "https://t.co/xPxwKTq6Qb"
            }
          ],
          "user_mentions": [
            {
              "id_str": "0",
              "indices": [
                4,
                14
              ],
              "name": "",
              "screen_name": "SophontAI"
            },
            {
              "id_str": "0",
              "indices": [
                52,
                62
              ],
              "name": "",
              "screen_name": "MedARC_AI"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "sophontai.com",
              "expanded_url": "https://sophontai.com",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/uQ936JTZf1"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "allow_download_status": {
          "allow_download": true
        },
        "display_url": "pic.twitter.com/CJ4b2hDd49",
        "expanded_url": "https://twitter.com/iScienceLuvr/status/1958105587235147991/photo/1",
        "ext_media_availability": {
          "status": "Available"
        },
        "features": {
          "large": {},
          "orig": {}
        },
        "id_str": "1958105579504947200",
        "indices": [
          278,
          301
        ],
        "media_key": "3_1958105579504947200",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARsslqZrGlAACgACGyyWqDfb0NcAAA==",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAABCgABGyyWpmsaUAAKAAIbLJaoN9vQ1wAA",
            "media_key": "3_1958105579504947200"
          }
        },
        "media_url_https": "https://pbs.twimg.com/media/GyyWpmsaUAAWAGS.jpg",
        "original_info": {
          "focus_rects": [
            {
              "h": 786,
              "w": 1404,
              "x": 0,
              "y": 0
            },
            {
              "h": 1404,
              "w": 1404,
              "x": 0,
              "y": 0
            },
            {
              "h": 1601,
              "w": 1404,
              "x": 0,
              "y": 0
            },
            {
              "h": 1820,
              "w": 910,
              "x": 0,
              "y": 0
            },
            {
              "h": 1820,
              "w": 1404,
              "x": 0,
              "y": 0
            }
          ],
          "height": 1820,
          "width": 1404
        },
        "sizes": {
          "large": {
            "h": 1820,
            "w": 1404
          }
        },
        "type": "photo",
        "url": "https://t.co/CJ4b2hDd49"
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {},
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "article": null
}