Note 33 public model repositories. Vision encoders, ViTs, and understanding-side projectors. Component scope and incomplete architecture documentation are noted per model.

Note 18 public model repositories. Discrete and continuous visual tokenizers, VAEs, and reconstruction modules, with decoder availability and latent conventions documented.

Note 8 public model repositories. Diffusion transformers, conditional decoders, and generation models, with their conditioning, codec, and sampling requirements.

Note 12 public model repositories. Visual subsystems from unified understanding and generation models, including pixel interfaces, flow modules, and mixed component bundles.

Note 2 public model repositories. Scientific image and weather interfaces, with native dependencies and explicitly scoped component validation.

Note 3 public model repositories. CLIP-style representations and embedding experiments. Architecture mismatches and incomplete dual-encoder documentation are noted.

Note 6 public model repositories. Detection, referring-expression grounding, and segmentation models, with weight packages distinguished from deployment-code repositories.

Note 11 public model repositories. Original nanoVLM experiments across vision and language backbones, including all public model variants.

Note 23 public model repositories. Controlled nanoVLM Factory experiments across vision backbones, language models, and multimodal projectors.

Note 8 public model repositories. Public Wenwu-Duobao Qwen-VL and LLaVA variants, organized by model family, language, and checkpoint version.

Note 12 public model repositories. SparkLM pretraining, fine-tuning, TinyStories, and GPT-2 experiments. Repositories with pending weights are explicitly marked.

Note 12 public model repositories. Distilled, pruned, quantized, and fine-tuned language or vision-language models, including Qwen, Youtu-FT, and Marvis.

Note 2 public model repositories. Hunyuan experimental checkpoints for mathematical reasoning and code generation.

Note 6 public model repositories. Public repositories whose audited file listings contain no recognized weight formats. Code or documentation may still be available.

Note 1 public model repository. Audio generation components, including local DiTs, with native conditioning, history, sampling, and codec dependencies.