thaitea/laya-vision-smolvlm-256m-score
visual-question-answering model by thaitea
Older snapshot: check the original source for current information.
From the publisher
Model card & documentation
Source previewRead the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.
Typed, calibrated decisions about an image plus optional text, in one forward pass with no text generation. Give it a picture, some context and a set of questions; it answers each as a multiple choice (choice), a yes/no probability (noul) or a graded level on a rubric you write (score), with probabilities you can act on at face value. This is the recommended Laya Vision checkpoint, pinned to the run cauldron-score-2ep-bidir-full (the moving copy is thaitea/laya-vision): a SmolVLM-256M backbone whose answer options attend to each other, trained on 19 closed-form subsets of The Cauldron and on four rubric-scored image datasets, so that the score head is trained for the first time. Laya Vision is an independent fork of Laya that replaces its ModernBERT text encoder with a small vision-language model and keeps Laya's predict(state, questions) API, RLCD training objective and temperature calibration. A score question's criteria is the rubric, level…
Read the full model card ↗ · Preview checked 2026-11-03T22:11:30.999Z
Inside the original model card — Document outline
- Laya Vision
- Usage
- Results
- score questions (new)
- Official VQA splits
- Cauldron holdouts
- Training
- Limitations
- License
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Documentation belongs to its respective authors. Reported project/model license: cc-by-nc-sa-4.0. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
Model overview
thaitea/laya-vision-smolvlm-256m-score is a visual-question-answering model repository published by thaitea on Hugging Face. The source reports the laya library.
This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.
Model facts
- Task
- visual-question-answering
- Library
- laya
- Recent downloads (30 days)
- 0
- Cumulative likes
- 0
- Hugging Face trending score
- 0
- Reported safetensors parameters
- 236,562,310
- Architecture
- Not reported
- License
- cc-by-nc-sa-4.0
- Access
- Not gated by Hugging Face
- Created
- 2026-09-22T06:24:49.000Z
- Last modified
- 2026-09-22T06:25:04.000Z
Compatibility and lineage
Reported languages: en
Reported base models: HuggingFaceTB/SmolVLM-256M-Instruct
Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.
Use and evaluate
Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.
No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.
Source and freshness
Source: Hugging Face Hub. Metadata observed 2026-09-22T06:37:17.952Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.