{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreifuksogyp7diwmxajtqgb5d4ixzfkq7u5gip47eecshkluhqmhlxy",
"uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mfd2pvwiela2"
},
"path": "/t/gpt-oss-20b-vision-adding-multimodal-to-gpt-oss-with-a-novel-multi-scale-approach-trained-on-a-single-dgx-spark/173659#post_2",
"publishedAt": "2026-02-20T14:10:35.000Z",
"site": "https://discuss.huggingface.co",
"textContent": "I’d say the proof-of-concept looks solid, and with more GPU time, this could get really powerful. Props for being so honest about what works and what doesn’t - makes it feel realistic.",
"title": "GPT-OSS-20B-Vision: Adding multimodal to GPT-OSS with a novel multi-scale approach, trained on a single DGX Spark"
}