next two:
โข Part 2: training recipe and the data curation behind it.
โข Part 3: releasing GRACE, a benchmark judged by a video-aware judge.
this all started as a side quest while shipping Marlin with @HappyyPablo
๐ค Model: https://t.co/wEK417BWzn
๐ฎ Hosted Demo: https://t.co/w8zf2ayM8y
Shipping a 3-post series on how we built Marlin-2B ๐.
Part 1: The Map Was Wrong ๐บ๏ธ, just dropped.
we evaluated Marlin on CaReBench + DREAM-1K, the numbers felt off, so we built a video-aware judge and ran it on the benchmarks themselves.
Blog link: https://t.co/HnkoGOXZiF
Marlin ๐ is now supported by @ModelScope2022
Super happy to see so much love from the Chinese developer community for marlin. They liked it so much they pinned it for more visibility ๐ฅฐ
cool new release: a tiny open video VLM that understands what happens in videos and when ๐
Marlin-2B (Apache 2.0!) can caption clips into timestamped events, or find a natural-language moment inside the video (can see a ton of cool use cases with it)
Made a Hugging Face demo for it โฌ๏ธ
Update for everyone tracking this: Marlin-2B has been relicensed to Apache 2.0.
The BSL 1.1 LICENSE is replaced with standard Apache 2.0 text. NOTICE preserves attribution to Qwen3.5-2B-Base. HF metadata now reports license:apache-2.0.
https://t.co/2Smf5Mg5ul
The license was a mistake โ fair critique, now fixed. Truly open source ๐
open sourcing Marlin-2B ๐
a tiny VLM to extract structured information from videos
Marlin is finetuned for two questions devs want to ask in their videos: what is happening, and when?
Best open model in its weight class, competitive with Gemini-2.5-flash at only 2B params ๐งต
open sourcing Marlin-2B ๐
a tiny VLM to extract structured information from videos
Marlin is finetuned for two questions devs want to ask in their videos: what is happening, and when?
Best open model in its weight class, competitive with Gemini-2.5-flash at only 2B params ๐งต
AI agents can negotiate $4,200 off a car purchase.
But they can't buy a $4 domain.
The bottleneck isn't intelligence. It's trust with money.
MPP(Machine Payment Protocol) has 88 services. The internet has millions.
Who's solving this?
We launched quietly yesterday with a short post - no big announcements, no ads. Just testing the waters.
Honestly, we expected zero downloads. Instead, 74 people took the time to try Nemo. It might not sound huge, but for us, itโs a powerful signal.