Intern-Decision-4B released as multimodal structured decision model
Intern-Decision-4B is a multimodal structured decision model fine‑tuned from Qwen3.5‑4B. It takes a shared state, a schema of named questions and optional images, and returns an answer distribution for each question in a single forward pass. The model maps options to single‑token symbols, reads logits before each placeholder, applies a softmax over allowed symbols and uses the checkpoint’s…
Key points
- Intern-Decision-4B is fine‑tuned from Qwen3.5‑4B and supports text and image inputs
- Accepts up to 16 questions, 62 options each, and eight images per request, with an 8,192 token limit
- Default temperature 1.992418 was fitted on 1,728 calibration cases and 1,693 validation cases
The released code provides a DecisionEngine class for Python 3.12+, loading the model on CUDA and handling image resizing. Requests may contain up to 16 questions, 62 options per question, and eight images, with a token limit of 8,192. Latency numbers are measured on a single RTX 4090 but not disclosed. The model’s license follows the original Qwen terms, and the weights are available via Hugging Face under the internlm/Intern-Decision‑4B repository.
Model page: Intern-Decision-4B →
internlm/Intern-Decision 4B and 0.8B
huggingface.co · 26 September 2026
Loading the full article…
This text was published by huggingface.co. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Reddit discussionreddit.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- ChatGPT-6 Astra cracks 1941 Enigma message in two days, Crypto Cellar reports · 1 src
- Anthropic releases Claude Opus 5.5 with 40% lower costs · 12 src
- Meta opens early access for new Muse AI features · 1 src
- GPT-6 Astra scores 99.9% on ARC-AGI-3, OpenAI says · 6 src
- OpenAI reportedly preparing to preview GPT-6 Cyber model within days · 5 src
Comments
via GitHub Discussions