Intern-Decision: Efficient Multimodal Decision Making with Autoregressive Masked Language Modeling
Intern-Decision's 4B model hits 90.02% decision accuracy with 44 ms latency on an RTX 4090.
Intern-Decision is a family of 0.8B, 2B, and 4B multimodal models that emit structured choices and probabilities from text and optional images using autoregressive masked language modeling over a JSON skeleton. The 4B model reaches 90.02% unweighted average accuracy across seven suites, versus 88.74% for Jev v1.13.0, with 44.16 ms mean latency on one RTX 4090 for a fixed 289-token request. Temperature scaling cuts expected multiclass Brier from 0.628 to 0.550 on a 96-case pilot. Routing 42 of 111 Jevbench-Hard items to Atria-Dawn-Preview raises accuracy from 82/111 to 97/111.
- Intern-Decision offers 0.8B, 2B, and 4B multimodal decision models.
- The 4B model scores 90.02% unweighted average accuracy versus 88.74% for Jev v1.13.0.
- On an RTX 4090, 4B mean latency is 44.16 ms for a 289-token, three-field request.
- Handoff of 42 hard items to Atria-Dawn-Preview lifts accuracy from 73.87% to 87.39%.
Full article218 words · extracted from huggingface.co · click to collapse
Interactive agents repeatedly choose among a finite set of actions, labels, or ordinal outcomes. Generating a free-form response for each choice adds latency and complicates the interface between a model and its environment. We present Intern-Decision, a family of 0.8B, 2B, and 4B multimodal decision models that return structured choices and candidate probabilities from text and optional images. Our autoregressive masked language modeling objective places one decision placeholder per field in a complete assistant JSON skeleton. Ground-truth symbols appear only in supervision, while all fields are evaluated in a single causal forward pass using the backbone's native language-model head. The 4B model achieves a 90.02% unweighted average accuracy across seven evaluation suites, compared with 88.74% for Jev v1.13.0. For a fixed 289-token request with three fields on one RTX 4090, mean latency is 44.16 ms for the 4B model and approximately 33--34 ms for the smaller models. On a separate 96-case known-distribution pilot, temperature scaling reduces expected multiclass Brier from 0.628 to 0.550 and expected calibration error from 0.213 to 0.089. A confidence-based handoff study on 111 public Jevbench-Hard items improves accuracy from 82/111 (73.87%) to 97/111 (87.39%) by routing 42 items to Atria-Dawn-Preview. We also describe qualitative game, visual interface, and browser demonstrations. These results support compact decision models as a practical component of larger agent systems.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2610.22131