---
library_name: transformers
---

# MLLMcheckpoint

Public checkpoints for evaluating 70 vision encoders with three language models: SmolLM2, Qwen3, and Qwen2.5.

Each model is stored at `<llm>/<encoder>/` with two variants:

- `base`: pretrained multimodal checkpoint
- `instruct`: finetuned instruction-following checkpoint

The repository contains 210 MLLMs and 2968 checkpoint files (1046023518806 bytes). See `manifest.tsv` for the complete inventory.