Oriel — AI & technology worth noticing, every day.

LiquidAI/d1-2B

AI model · https://huggingface.co/LiquidAI/d1-2B

First seen by Oriel on .

AI summary

The Hugging Face card presents d1-2B, a 1.92B-parameter multimodal decision model that produces calibrated typed answers in one forward pass without token generation.

  • Built on LFM3-VL-2B, it accepts text or JSON, pictures, and video, with multiple named questions answered together in one call.
  • Answer types include classifications, probabilities, extracted text, numbers, dates, scores, quantity intervals, and object bounding boxes.
  • It targets routing, moderation, intent classification, document extraction, judging, reranking, and visual inspection rather than conversational chat.
  • The model leads the listed open d1 models on the decision board, averaging 70.2 across shared public benchmarks and exceeding d1-3B on most decision tests.
  • Warm one-question latency is 5.7 ms on an NVIDIA H100 and 3.8 ms on an AMD MI325X; reliability decreases beyond roughly 64K tokens of state.

#multimodal models #decision models #computer vision #document extraction

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Models: 248, latest observation ; +148 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Models: 9,009, latest observation ; +5,370 over observed period since

Aliases

SourceExternal IDURL
Hugging Face ModelsLiquidAI/d1-2Bhttps://huggingface.co/LiquidAI/d1-2B Read discussion